QCL Lecture Long

Intersubband Optoelectronics
Jerome Faist
ETH Zurich
Z urich, September 17, 2009
Contents
List of gures xiv
List of tables xv
List of variables and symbols xv
1 Introduction: intersubband optoelectronics as an example of quantum en-
gineering 1
1.1 Quantum engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Basic requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Tools of the mid-infrared . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3.1 Transparent materials . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.2 Reectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.3 Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.4 Fibers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.5 Polarizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.6 Spectrometers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Notes and acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Technology 5
2.1 Molecular Beam epitaxy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Metal Organic Vapor Phase epitaxy . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Reactive Ion etching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Quantum Cascade laser processing . . . . . . . . . . . . . . . . . . . . . . . 6
3 Electronic states in semiconductor quantum wells 9
3.1 Band structure of semiconductors in the kp approximation: origin of the
eective mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.1.1 Basic approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.1.2 Beyond the perturbation expansion . . . . . . . . . . . . . . . . . . . 12
3.1.3 Example: a two-band Kane model . . . . . . . . . . . . . . . . . . . . 12
3.2 Envelope function approximation . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2.1 Multiband case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
i
ii CONTENTS
3.2.2 One Band model: the Ben-Daniel Duke . . . . . . . . . . . . . . . . . 15
3.2.3 Two band model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2.4 Formal derivation of the 2x2 model . . . . . . . . . . . . . . . . . . . 16
3.3 Hartree potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Building blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4.1 The single quantum well . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4.2 The coupled well system . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.4.3 The superlattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.5 In-plane dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.6 Full model: the valence band . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4 Intersubband absorption 29
4.1 Interaction Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Intersubband and interband transition . . . . . . . . . . . . . . . . . . . . . 30
4.3 Selection rules and absorption geometries . . . . . . . . . . . . . . . . . . . . 30
4.4 Absorption strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.4.1 Absorption for a 2D system . . . . . . . . . . . . . . . . . . . . . . . 32
4.4.2 Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.4.3 Gain and loss cross sections . . . . . . . . . . . . . . . . . . . . . . . 33
4.5 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.6 Sum rule in absorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.7 Absorption in a quantum well: a two-band model . . . . . . . . . . . . . . . 37
4.8 Absorption linewidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.8.1 Non-parabolicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.8.2 Disorder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.9 Stark-tuning of intersubband absorption . . . . . . . . . . . . . . . . . . . . 42
5 Intersubband scattering processes 45
5.1 Spontaneous emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2 Scattering rate by bulk optical phonons . . . . . . . . . . . . . . . . . . . . . 47
5.2.1 Temperature dependence . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3 Quasi-elastic intersubband process . . . . . . . . . . . . . . . . . . . . . . . . 50
5.3.1 Acoustic phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.3.2 Electron-electron scattering . . . . . . . . . . . . . . . . . . . . . . . 52
5.3.3 Impurity scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.4 Comparison with experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.4.1 Interband pump and probe . . . . . . . . . . . . . . . . . . . . . . . . 56
5.4.2 Intersubband pump probe . . . . . . . . . . . . . . . . . . . . . . . . 57
5.4.3 Intersubband saturation experiments . . . . . . . . . . . . . . . . . . 57
5.4.4 Intersubband electroluminescence . . . . . . . . . . . . . . . . . . . . 57
5.4.5 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
CONTENTS iii
6 Detectors 61
6.1 Fundamentals of infrared detection: noise, BLIB.. . . . . . . . . . . . . . . . 61
6.2 Quantum Well Infrared Photoconductor (QWIP) . . . . . . . . . . . . . . . 61
7 Mid-infrared waveguides 63
7.1 Dielectric waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
7.2 Dispersion of doped layers, Drude approximation . . . . . . . . . . . . . . . 63
7.3 Metal-based waveguides for the Terahertz . . . . . . . . . . . . . . . . . . . 63
7.4 Far-eld computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
8 Quantum Cascade lasers I: fundamentals 65
8.1 Historical perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
8.2 Active region: fundamental concepts . . . . . . . . . . . . . . . . . . . . . . 66
8.2.1 Injection/relaxation region . . . . . . . . . . . . . . . . . . . . . . . . 66
8.2.2 Cascading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
8.3 Intersubband versus interband lasers . . . . . . . . . . . . . . . . . . . . . . 69
8.4 Rate equation analysis, threshold condition, slope eciency . . . . . . . . . 70
8.5 Optimization of the active region: intersubband toolbox . . . . . . . . . . . 71
8.5.1 Tunneling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
8.5.2 Optical phonon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
8.5.3 Phase space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
8.5.4 Escape time, Bragg reection and upper level connement . . . . . . 78
8.5.5 Injection eciencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
8.6 Optimization of the active region: dierent designs . . . . . . . . . . . . . . 81
8.6.1 Designs: general trends . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.6.2 Three quantum well active region . . . . . . . . . . . . . . . . . . . . 81
8.6.3 Double phonon resonance . . . . . . . . . . . . . . . . . . . . . . . . 83
8.6.4 Boundtocontinuum active regions . . . . . . . . . . . . . . . . . . . 85
8.7 Cascading: scaling with the number of periods . . . . . . . . . . . . . . . . 87
8.7.1 Threshold current density . . . . . . . . . . . . . . . . . . . . . . . . 87
8.7.2 slope eciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.8 Temperature dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.8.1 intersubband physics . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.8.2 Backlling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.8.3 Self-heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.9 Doping of the active region . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
8.10 Fundamental limits of quantum cascade lasers . . . . . . . . . . . . . . . . . 95
8.11 Short wavelengths: discontinuity and strain compensation . . . . . . . . . . 98
8.11.1 Conduction band discontinuity and performance . . . . . . . . . . . . 98
8.11.2 Strain-compensated In
x
Ga
1x
As/Al
y
In
1y
As/InP material system . . 99
iv CONTENTS
9 Quantum Cascade laser II: Mode control 107
9.1 Fabry Perot cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
9.2 Distributed feedback cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
9.2.1 Multilayer approach, Bragg reection condition . . . . . . . . . . . . 109
9.2.2 coupled mode analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 112
9.2.3 Fabrication geometries . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Surface grating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Buried grating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
9.2.4 Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Dynamical behavior and linewidth . . . . . . . . . . . . . . . . . . . 120
Cavity pulling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
9.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
9.4 External cavities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
10 Quantum Cascade lasers III: applications 125
10.1 Energy deposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.1.1 Wallplug eciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.1.2 High power operation . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.2 Telecommunications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.2.1 high frequency modulation . . . . . . . . . . . . . . . . . . . . . . . . 126
10.2.2 Telecommunications experiments . . . . . . . . . . . . . . . . . . . . 126
10.3 Gas sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.3.1 Gas absorption lines: pressure and temperature depdendence . . . . . 126
10.3.2 General system consideration . . . . . . . . . . . . . . . . . . . . . . 126
10.3.3 multipass cells and fringe noise . . . . . . . . . . . . . . . . . . . . . 126
10.3.4 interpulse modulation . . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.3.5 intrapulse modulation . . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.3.6 Wavelength modulation . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.3.7 Photoacoustic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.4 Liquid sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
11 Quantum Cascade lasers IV: Transport models 127
11.1 Classical models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
11.1.1 Drude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
11.1.2 Esaki-Tsu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
11.2 Transport as a result of intersubband transitions . . . . . . . . . . . . . . . . 127
11.2.1 Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
11.2.2 Diagonal transition lasers . . . . . . . . . . . . . . . . . . . . . . . . 127
11.2.3 With charge conservation and the optical eld . . . . . . . . . . . . . 127
11.2.4 Computing the electron distribution in the subband . . . . . . . . . . 127
11.3 Resonant tunneling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
CONTENTS v
11.3.1 Problem of coherence . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
11.3.2 Density matrix approach: Kazarinov, Willenberg . . . . . . . . . . . 129
11.3.3 second-order current . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
11.3.4 Full model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
11.4 Non-equilibrium Greens function theory . . . . . . . . . . . . . . . . . . . . 129
12 Quantum Cascade Lasers V: Device properties and characterization 131
12.1 Electrical charateristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
12.2 Gain and loss measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
12.3 Electron temperature measurements: hot electrons and hot phonons. . . . . 131
12.4 Active region non-linearities . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
12.5 Conned phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
13 Intersubband processes II (collective excitations) 133
13.1 Depolarization shift. (Dielectric model and connection to the oscillator model)133
13.2 Intersubband plasmon, linewidth versus charge density . . . . . . . . . . . . 133
13.3 Raman spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
13.4 Optical microcavities and cavity polaritons (Alessandro, Carlo, ..) . . . . . . 133
14 Interlevel transition in Quantum dots 135
14.1 Fabrication: self-assembled Stransky Krastanov growth . . . . . . . . . . . . 135
14.2 Interband spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
14.3 Intraband spectroscopy techniques . . . . . . . . . . . . . . . . . . . . . . . 135
14.4 The intersubband polariton in quantum dots. (Ferreira, Luke Wilson, ...) . . 135
15 Intersubband non-linearities and non-linear devices 137
15.1 Non-linear intersubband polarization (Jacob Kurghin, Ghilad Almoghi, Carlo
Sirtori,) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
15.2 Second harmonic generation structures (step well, coupled wells) . . . . . . 137
15.3 Trade-o between power and non-linear power; phase matching condition . . 137
15.4 Non-linear generation of THz (Carlo) . . . . . . . . . . . . . . . . . . . . . . 137
15.5 Quantum Cascade lasers with non-linear elements: mid-IR (Claire) and THz
(Belkin) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
15.6 NIR-THz mixing in QCL (Carlo) . . . . . . . . . . . . . . . . . . . . . . . . 137
16 Appendix: Designs 139
16.1 Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
16.1.1 3QW 10.3um diagonal . . . . . . . . . . . . . . . . . . . . . . . . . . 139
16.1.2 3QW 5.3um vertical . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
16.1.3 Two-phonon resonance at 9um . . . . . . . . . . . . . . . . . . . . . . 139
16.1.4 Bound-to-continuum at 9um . . . . . . . . . . . . . . . . . . . . . . . 139
16.1.5 Bound to continuum at 16um . . . . . . . . . . . . . . . . . . . . . . 139
16.1.6 Broad bound-to-continuum . . . . . . . . . . . . . . . . . . . . . . . . 140
vi CONTENTS
List of Figures
2.1 TEM image of a QCL active region . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 STM image of the active region of graded and abrupt interface design . . . . 6
2.3 Processing steps for the fabrication of a Fabry-Perot cavity quantum cascade
laser. a) MBE or MOCVD growth of the active region. b) Dening the ridge
by photolithography ( spin resist, expose with UV light and develop). c) The
ridge is etched in a wet etching solution. d) Si
3
N
4
is deposited by PECVD. e)
Open the insulating oxide on top of the ridges. f) Top and bottom metalization. 7
2.4 Schematic drawing of the facet of a processed device. . . . . . . . . . . . . . 8
3.1 GaAs band structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Comparison for various model for the energy-dependent eective mass . . . . 17
3.3 Computed connement energy of electrons in the conduction band as a func-
tion of well width. Bound states are shown with full lines, resonances with
dotted lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4 Energy states of a quantum well computed with a two-band model, and com-
pared with a one-band model (dashed lines). The growing importance of
non-parabolicity as one moves away from the gap is clearly apparent. . . . . 21
3.5 A coupled quantum well system. . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.6 A coupled quantum well system. . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.7 Schematic band structure of a superlattice. . . . . . . . . . . . . . . . . . . 23
3.8 Computed dispersion of a 100A InGaAs well and 30A AlInAs barrier super-
lattice. A conduction band discontinuity of E
c
= 0.52eV as well as a Kane
energy of 18.3eV was assumed in an eective two-band model. For clarity, the
zero of the energy scale has been set to the bottom of the InGaAs quantum
wells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.9 Dispersion of the rst miniband (green), compared to a sinus function (red).
The result of the nite superlattice are also plotted as full squares . . . . . . 24
3.10 Square of the wavefunctions for a nite superlattice formed by eight 100A
InGaAs wells separated by 30A AlInAs barriers. . . . . . . . . . . . . . . . 25
3.11 Dispersion of the state of a quantum well in the valence band. . . . . . . . . 26
3.12 Schematic description of the origin of the valence band dispersion in the quan-
tum well showing schematically the eects of connement and interactions. 27
4.1 Experimental geometries allowing the measurements of intersubband transitions 31
vii
viii LIST OF FIGURES
4.2 Photocurrent in a quantum well photoconductor as a function of the polar-
ization. The residual responsivity for the TE polarization is less than 0.2%
than for the electric eld normal to the layers. (From H.C.Liu et al, APL 1998) 31
4.3 Intersubband absorption in an 65A thick quantum well. (From West and
Eglash, APL) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4 Intersubband absorption between two bound states . . . . . . . . . . . . . . 35
4.5 Intersubband absorption in a multiquantum well designed for triply resonant
non-linear susceptibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.6 Intersubband absorption from a bound state to a continuum . . . . . . . . . 36
4.7 Comparison of the intersubband absorption for various structures, with energy
levels schematically drawn close to the curves . . . . . . . . . . . . . . . . . 37
4.8 Lineshape functions in various limiting cases. . . . . . . . . . . . . . . . . . . 40
4.9 Transmission through a 15nm thick InAs/AlSb mulitquantum well system.
Because of the non-parabolicity of this system, a broad (about 20meV wide)
peak would be expected, in contrast with the narrow experimental line ob-
served. This narrowing was interpreted by Warburton et al. as a result of
depolarization shift. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.10 Linewidth of the absorption of quantum wells of various widths. The dotted
line shows the compute value of the broadening caused by the optical phonon
scattering. The decrease in linewidth as a function of quantum well width is a 42
4.11 Experimental measurement of the broadening of the intersubband transition
and mobility as a function of temperature, compared with the model proposed
by Unuma et al. [1]. Optical and acoustic phonon intra- and inter-subband
processes are also included in the model, as indicated. . . . . . . . . . . . . . 43
5.1 Schematic description of the relevant scattering mechanism in the two relevant
cases. The simplest case is the situation occuring in the Mid-Infrared (de-
picted on the left) where the LO phonon scattering is dominant. In contrast,
in the terahertz, at low temperature, a number of non-radiative processes
should be taken into account including acoustic phonons, electron scattering
and other elastic processes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.2 Radiative lifetime of a square quantum well as a function of the transitiion
energy. A constant value of the oscillator strenght of 23, corresponding to an
InGaAs/AlInAs quantum well, has been assumed . . . . . . . . . . . . . . . 46
5.3 Scattering by an optical phonon between two subbands. An electron with
initial energy Ei and initial wavevector k
i
is scattered to the lower subband
at k
f
, therefore losing (or gaining) an energy equal to
LO
in the process. . 47
5.4 Computed electron lifetime (left vertical axis) due to optical phonon scat-
tering as a function of transition energy in a square quantum well, whose
width is indicated in the rigth vertical axis. Zero initial kinetic energy in the
upper subband, and zero temperature was assumed. The lled squares are
experimental measurements, as indicated. . . . . . . . . . . . . . . . . . . . . 48
LIST OF FIGURES ix
5.5 Computed electron lifetime due to optical phonon scattering as a function
of the initial kinetic energy. A square, 220

Athick InGaAs/AlInAs quantum
well was considered. The equivalent electron temperature is shown in the top
horizontal axis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.6 Computed temperature dependence of the intersubband lifetime for a 100A
thick InGaAs/AlInAs quantum well . . . . . . . . . . . . . . . . . . . . . . . 50
5.7 Intersubband relaxations processes in a quantum well with an energy separa-
tion smaller than the optical phonon energy E
21
<
LO
. In this conguration,
various processes may play a role: optical phonon emission (op), electron-
electron scattering (EE), ionized-impurity scattering or interface roughness
scattering (II), acoustic-phonon emission (AC). . . . . . . . . . . . . . . . . 51
5.8 Various electron-electron scattering processes. The 2, 2 1, 1 is the domi-
nant electron-electron scattering term, as the Auger-like term 2, 2 2, 1 was
shown to vanish in symmetric quantum wells. The 2, 1 2, 1 will lead to
thermalization between subband while the 2, 2 2, 2 or 1, 1 1, 1 will tend
to thermalize the subband themselves. . . . . . . . . . . . . . . . . . . . . . 52
5.9 The electron-electron scattering time for the 2, 2 1, 1 process presented
as a function of well width for an innite quantum well at zero temperature
for dierent screening models: no screening, constant screening length, static
single subband screening for and ideal 2D system with no z-dependence (i.e.
[A
2,21,1
[ = 1), static single subband screening including q
- dependence form
factor. The Fermi energy of the excited state E
f2
is equal to 10meV and the
initial electron state wave vector

k
i
is equal to zero. Adapted from Ref [2],
with the correction suggested by [?], with permission . . . . . . . . . . . . . 54
5.10 Techniques for the measurement of the intersubband lifetime. . . . . . . . . . 56
5.11 Lifetime as a function of pumping density. . . . . . . . . . . . . . . . . . . . 60
8.1 Schematic band structure of the device considered by R. Kazarinov and R.
Suris. A superlattice is under a suciently strong electric eld such that the
ground state of a well is above the rst excited state of the following well . . 66
8.2 a) Schematic conduction band diagram of a quantum cascade laser. Each stage
of the structure consists of an active region and a relaxation/injection region.
Electrons can emit up to one photon per stage. b) General philosophy of the
design. The active region is a three-level system. The lifetime of the 3 2
transition has to be longer than the lifetime of level 2 to obtain population
inversion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
8.3 Highly simplied picture of the eective potential in a quantum cascade laser.
At zero eld, the current is blocked by the injection regions that prevent the
current from owing. With the application of a strong electric eld, these
injection regions are attened by the electric eld and the current ows
along the electronic cascade, generating the photons. . . . . . . . . . . . . . 68
8.4 Population inversion between the two upper levels of a ladder of three states.
The relevant non-radiative process are also indicated. . . . . . . . . . . . . . 70
x LIST OF FIGURES
8.5 Schematic illustration of the behavior of the population inversion and the
photon ux with injected current. . . . . . . . . . . . . . . . . . . . . . . . . 72
8.6 Two states coupled by a tunnel barrier. The lifetime of the upper state is
controlled by the thickness of the coupling barrier . . . . . . . . . . . . . . . 73
8.7 Photon-assisted tunneling transition. Plotted are the oscillator strength f =
2m
0
2
z
2
E
ij
and optical phonon scattering rate as a function of barrier width for
system consisting of a 4nm and 1.5nm InGaAs well coupled by an AlInAs bar-
rier. The additional points are the two values for a 6nm thick single quantum
well exhibiting the same transition energy . . . . . . . . . . . . . . . . . . . 74
8.8 Product of the oscillator strength and the lifetime as a function of tunnel
barrier thickness. This number is a gure of merit of the gain cross section.
The value for L
b
= 0 is the one computed for a single quantum well 6nm thick,
as in Fig. ??. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
8.9 Engineering a population inversion using an optical phonon resonance. In
a ladder of three states, where the two lower ones are spaced by an optical
phonon energy, the n=3 lifetime will naturally be longer than the n=2 because
of the smaller exchanged wavevector for the optical phonon emission between
the n=2 and n=1 states. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
8.10 Engineering the lifetimes. In a superlattice, a population inversion builds up
at the edges of the minigap because of the phase space for scattering out from
upper state of the lower miniband is much larger than the phase space to
scatter in the same state. a) real-space and (b) reciprocal space picture of the
energy bands. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
8.11 Product of the oscillator strength and the upper state lifetime (left) and
branching ratio
3
/
32
(right) for a 60/5nm GaInAs/AlInAs superlattice with
an number of period N. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
8.12 Computed transmission of a injector designed to provide maximum reection
at the energy of the excited state, as shown. . . . . . . . . . . . . . . . . . . 79
8.13 Comparison of the electroluminescence eciency of a structure designed with
and without a Bragg reector facing the upper state. . . . . . . . . . . . . . 80
8.14 Schematic description of the relevant levels and injection eciencies in a cas-
cade laser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.15 Schematic diagram of the rst generation of quantum cascade lasers. . . . . 82
8.16 Schematic of the parameters optimized in the active regions. . . . . . . . . . 82
8.17 Schematic potential prole and design of the active region of a quantum cas-
cade laser based on three quantum wells. a) Two quantum well before the
application of the electric eld. b) The electric eld bring the two ground
states in resonance and yields a splitting equal to the optical phonon energy.
A third thinner well is added upstream. If the state of this well is resonant
with the excited state of the coupled well, the resulting transition is diagonal
(c), if the well is thinner and the resonance is above, the transition is vertical
(d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
LIST OF FIGURES xi
8.18 Examples of active regions based on a three quantum wells, using lattice-
matched InGaAs/AlInAs material. Structure a) is designed for a wavelength
of = 10.3m using the concept shown in Fig. 8.17 c), that leads to a
diagonal transition with a longer lifetime. Structure b) is designed for =
5.3m and exhibits a vertical transition, as explained in Fig. 8.17 d) [3]. . . 84
8.19 Schematic band diagram of a two-phonon resonance gain region designed for
operation at 9m. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.20 Low-temperature (T = 80K) luminescence spectra of the active region based
on a two-phonon resonance. Calculated transition energies are shown on the
top horizontal axis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.21 Oscillator strength (right axis) and product of oscillator strength and upper
state lifetime (left axis) as a function of applied electric eld for a superlattice
consisting of 5 period of a 59
AInGaAs quantum well and a 9
AAlInAs barrier 87
8.22 Three rst miniband edges as a function of position for a chirped superlattice
(a) and for a bound-to-continuum structure c). . . . . . . . . . . . . . . . . 88
8.23 a) Schematic conduction band diagram of one stage of a bound-to-continuum
active region under an applied electric eld of 3.5 10
4
V/cm. The moduli
squared of the relevant wavefunctions are shown. b) luminescence spectrum
of the active region at 300 and 80K, as indicated. The applied bias is 9V .
Lower curve: computed oscillator strength of the various transitions from the
upper state. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.24 Overlap as a function of the number of period . . . . . . . . . . . . . . . . 90
8.25 Slope eciency as a function of the number of periods in a series of other-
wise identical quantum cascade lasers. Inset: slope eciency per stage.(From
ref. [4]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.26 Temperature dependence of the light-versus current characteristics of a quan-
tum cascade laser based on a bound-to-continuum transition operating at
= 9m.(From ref. [5]). Inset: Threshold current density as a function of
temperature, tted by the expression 8.8.23 with a T
0
of 190K . . . . . . . . 92
8.27 a) Electrons thermally excited from the injector populate the lower state. The
critical energy is the dierence between the Fermi energy of the injector
and the lower state. b) Threshold current density versus temperature for
a series of quantum cascade lasers with an identical active region based on
a vertical transition operating at = 4.6m but with injectors of various
lengths showing dierent values of
inj
. . . . . . . . . . . . . . . . . . . . . 93
8.28 Comparison of the electroluminescence from a two otherwise identical samples
with dierent doping proles. In the sample with the setback, only the injector
region is doped. In the reference sample, the area marked in green in the
schematic band structure in the inset is also doped. . . . . . . . . . . . . . . 94
xii LIST OF FIGURES
8.29 The threshold current density J
TH
in open squares and the maximum current
density J
NDR
in solid squares as a function of sheet densities corresponding
to the doping study (n67, n72, n69, n71). In dashed lines, tted curves corre-
sponding to an extrapolated doping oset of n
offset
= 1.0 10
11
cm
2
added
to the active region doping. Open and solid circles represent the threshold
and the maximum current densities respectively of a laser grown after a longer
MBE running time (n120). . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
8.30 Wallplug eciency as a function of photon energy. Filled squares, experimen-
tal values in continuous wave, from references [6, 7, 8, 9, 10, 11, 12]. Solid
disk: pulsed values from references [13, 14, 10, 15, 16]. The solid line rep-
resents the prediction from the simplied formula with the key parameters
derived from [15]. The inuence of a shorter or longer in-plane dephasing
time
//
is shown by the two dotted curves. Long dashed lines represent the
prediction from the model scaled by a factor 0.4. . . . . . . . . . . . . . . . 97
8.31 Characteristic temperature T
0
for various quantum cascade lasers, as function
of the photon energy and for various material systems. A clear relationship
between discontinuity and temperature behavior is is apparent. . . . . . . . . 99
8.32 Characteristic temperature T
0
for various quantum cascade lasers, as function
of the photon energy and for various material systems. A clear relationship
between discontinuity and temperature behavior is is apparent.(Redrawn from
data from [17] ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
8.33 Band gap versus lattice constant for the main semiconductor materials, to
the exception of group III nitrides. Of particular interest for quantum cas-
cade lasers are the GaAs/Al
x
Ga
(1x)
As, the In
x
Ga
1x
As/Al
y
In
1y
As/InP, the
InAs/AlSb heterostructure material system . . . . . . . . . . . . . . . . . . 100
8.34 Conduction band discontinuity as a function of In content in the quantum
well material, for a strain-compensated material with the well thickness rep-
resenting 40% of the total stack thickness. Full line, result of the computa-
tion, dashed line: value adjusted to the measured discontinuity of the lattice
matched material. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
8.35 Electron eective mass of In
1x
Ga
x
As material as a function of the Ga content
x (in contrast to the text where x labels the Indium content). Full line:
unstrained material. Dotted, dashed lines: in-plane, respectively out of plane
mass for material grown on InP. From Suwagara et al., . . . . . . . . . . . . 104
8.36 Conduction band structure of a strain-compensated quantum cascade laser
designed for short wavelength operation. The predicted emission wavelength
was = 3.16m using the eective mass of the unstrained material. Measured
wavelength were between 3.4m and 3.6m, compatible with a much heavier
mass as predicted by theory. . . . . . . . . . . . . . . . . . . . . . . . . . . 105
8.37 Emission spectrum of a short wavelength quantum cascade laser at various
temperatures, as indicated. . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
LIST OF FIGURES xiii
8.38 Threshold current density versus temperature for the device operating at =
3.6m. The relatively low T
0
observed (85K) is an indication of leakage paths
above the barriers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
9.1 Subthreshold luminescence of a short wavelength quantum cascade laser as
a function of increasing current, from 100mA to 800mA by steps of 100mA.
The temperature is 20K and the threshold current is at 950mA. The spectra
are shown with two dierent horizontal scales a) and b). The shift of the
individual Fabry-Perot resonances is caused by the device heating. . . . . . 108
9.2 Bragg reection condition for gratings of various order N a) First order grating,
only backscattering is possible. b) For N=2, either backscattering or vertical
emission is possible. c) In the third order grating (N=3), emission is only
possible inside the substrated because of the large refractive index dierence
between the guided mode (n
eff
= 3.2) and the vacuum. . . . . . . . . . . . 110
9.3 Computed transmission for a nite quarter wave stack (640 pairs) with a
periodicity of 1.56m of average refractive index 3.2 and a refractive index
step of n = 1.5 10
2
. A stop band of width about 4cm
1
appears in the
transmission. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
9.4 Computed gain for a 1mm long device, with the same parameters as the one
of Fig. 9.3. Symmetric facets have been assumed . . . . . . . . . . . . . . . . 112
9.5 Distribution of the optical eld in the cavity of the laser as a function of the
product L. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9.6 Distribution of the optical eld in the cavity of the laser for an overcoupled
laser with a L = 4.7. This overcoupling enables very low mirror losses but
at the cost of a lower eciency. . . . . . . . . . . . . . . . . . . . . . . . . . 114
9.7 a) Scanning electron micrograph of a distributed feedback quantum cascade
laser with a grating on the surface. b) Computed eective index, losses and
mode prole of the waveguide mode for three values of the etch depth: 0, 0.3
and 0.5m. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
9.8 a) Schematic cut through a waveguide design that uses air as the top cladding
and a lateral injection through a doped InGaAs layer. b) Scanning electron
micrograph of a nished device. . . . . . . . . . . . . . . . . . . . . . . . . 115
9.9 a)Light and bias versus current characteristic of distributed feedback quantum
cascade laser with a air/semiconductor top cladding and a lateral injection.
b) Representative spectra of the device at various operation temperatures. . 116
9.10 a) Atomic force microscopy of a buried grating prior to the regrowth step.
b) Scanning electron micrograph of the active region of a device after the
regrowth was performed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
9.11 a) Computed mode for a 10m wavelength with (black line) and without (blue
line) 200nm thick grating. The computed change of refractive index is n =
1.5 10
2
while the losses are not aected. c) Real part of the mode eective
index as a function of etched depth. As expected from equation 9.2.32, the
relationship is linear. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
xiv LIST OF FIGURES
9.12 Single mode optical power versus injected current for a distributed feedback
quantum cascade laser combining a buried grating and a buried heterostruc-
ture waveguide. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
9.13 Temperature tuning of the a buried grating device operating in continuous
wave above room temperature. . . . . . . . . . . . . . . . . . . . . . . . . . . 119
9.14 Mode position of a distributed feedback quantum cascade laser as a func-
tion of both the temperature of the submount and the electrical power dissi-
pated.(From ??) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
9.15 Single mode spectra of a distributed feedback quantum cascade laser as a func-
tion of time during a current pulse. The chirp is thermally driven. (From [18]) 121
9.16 Measured linewidth as a function of pulse length. The point labeled Aero-
dyne is the result of a spectral t of the absorption line of NH
3
at 967cm
1
that did yield a linewidth of 0.012cm
1
. Adapted from [19] . . . . . . . . . 122
9.17 Measured half width at half maximum of the emission spectra of a quantum
cascade laser as a function of pulse length. The linewidth is obtained by tting
a vibrational molecular line, and the pulse length is a increasing function of
the voltage on the pulser, indicated in the horizontal axis. Adapted from [?] 123
List of Tables
3.1 Fundamental parameters for various III-V semiconductors . . . . . . . . . . 11
5.1 Summary of lifetime measurements for transition energies above the optical
phonon energy in GaAs quantum wells. Early measurements were inaccurate
due to the strong pumping needed. ISB: intersubband, IB: interband. Only
low temperature measurements (T=10K) are reported . . . . . . . . . . . . . 59
5.2 Summary of some lifetime measurements for transition energies below the
optical phonon energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
8.1 Some elastic constants, energies and deformation potentials relevant for In-
GaAs/AlInAs material system. (from ??). . . . . . . . . . . . . . . . . . . . 102
9.1 Measured temperature tuning coecients of distributed feedback quantum
cascade lasers, as reported in various references. The tuning coecient is
sensitive to temperature but in a very good approximation independent on
the wavelength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
xv
xvi LIST OF TABLES
variables used
Physical constants:
0
vaccuum permettivity
q electron charge
k Boltzmann constant
m
0
Electron mass
P Kanes matrix element
E
P
Kane energy
Variables:
Absorption coecient (conventionally in cm
1
)
2D
Absorbed fraction by a 2D system (uniteless)
E Energy
Energy
c Electric eld
Dielectric constant
Chemical potential (energy units)
[), (x) Electron wavefunction
Lifetime
n
refr
Refractive index
n
i
Sheet density in subband i
N
i
Volume density in subband i
Chapter 1
Introduction: intersubband
optoelectronics as an example of
quantum engineering
1.1 Quantum engineering
Intersubband transitions in quantum wells are transitions between states created by quan-
tum connement in ultra-thin layers of semiconductors. Their unique physical properties,
such as a atomic-like density of states, as well as the fact that they can be manufactured
in the technology mature III-V semiconductors makes them an attractive building block
for mid-infrared optoelectronics. In fact, modulators, detectors and lasers were all demon-
strated using this technology, and are becoming mainstream devices for a new generation of
optoelectronics for sensing and telecommunication applications.
The developement of intersubband optoelectronics is a good example of the power of quan-
tum engineering: the same way the understanding of Maxwells equations lead rst to
explanation of the electrical phenomena, and then to their engineering; quantum mechanics
is moving from the realm of explanation to the one of engineering.
1.2 Basic requirements
A basic knowledge of solid-state physics and of quantum electronics. Level: master/graduate.
1.3 Tools of the mid-infrared
The intersubband optoelectronics has developed the mid-infrared portion of the spectrum.
As compared to the visible or the near-infrared, this portion of the spectrum is based on
rather dierent tools and techniques. In this chapter, we briey review the dierent aspects
of the mid-infrared concerning materials, detector and sources.
1
2
CHAPTER 1. INTRODUCTION: INTERSUBBAND OPTOELECTRONICS AS AN
EXAMPLE OF QUANTUM ENGINEERING
1.3.1 Transparent materials
Traditional insulators: Glass (SiO
2
) and Saphire (Al
2
O
3
).
Salts (NaCl, KCl, .)
II-VI semiconductors: ZnSe, ZnS
III-V semiconductors: GaAs
IV semiconductors: C, Si, Ge
1.3.2 Reectors
Dielectric constant of metals. subsectionDetectors
Thermal detectors: Pyroelectric, thermoelectric, bolometers, Golay cells
Photoconductors: Mercury Cadmium Telluride photoconductors, InSb, PbS, ..
1.3.3 Sources
Blackbody source
CO
2
laser
Interband lasers
1.3.4 Fibers
In general, the optical bers in the mid-infrared have general performance levels much below
what is achieved in the near-IR by the glass bers.
Fluorides (seem the most advanced). Limited to the short wavelength portion of the
mid-IR
Chalcogenides
Silver Halides. Very wide transparency range, low loss, but high chemical reactivity.
Holllow bers. High loss, limited bandwidth.
1.3.5 Polarizers
Brewster plates
Grid polarizers
1.4. BIBLIOGRAPHY 3
1.3.6 Spectrometers
Grating spectrometer
Fourier Transform Infrared Spectrometer
1.4 Bibliography
At this stage there is not real textbook covering the topic. Some textbooks covers aspects
of the material presented in this lecture:
Power point presentations They will be made available in pdf format before the lecture.
E. Rosencher and B. Winter Optoelectronics . Publisher: Cambridge University
Press (1998). Good coverage of the fundamental properties. Include the basic descrip-
tion of the intersubband transitions.
G. Bastard Wave mechanics applied to semiconductor heterostructures (Les edi-
tions de physique). Very good description of the computation of electronic state in
semiconductor heterostructures (sections 3-5).
H. Schneider and H.C. Liu Quantum Well Infrared Photodetectors (Springer se-
ries in optical sciences) The reference book on quantum well infrared detectors by the
leaders of the eld
Lecture notes
1.5 Notes and acknowledgements
Under this form, this script is meant for internal use since it does not always give proper
credit to the gures, especially for the ones taken from the books cited above. The author
would like to acknowledge Giacomo Scalari for his help in preparing this lecture.
4
CHAPTER 1. INTRODUCTION: INTERSUBBAND OPTOELECTRONICS AS AN
EXAMPLE OF QUANTUM ENGINEERING
Chapter 2
Technology
2.1 Molecular Beam epitaxy
Active region
Injector
50 nm
M
B
E

G
r
o
w
t
h
M
O
C
V
D

G
r
o
w
t
h
Figure 2.1: a) TEM image of a of the cleaved cross section of a quantum cascade laser active region
(sample S1840) composed by a part of a active region (grown by MBE) and an upper cladding layer
(grown by MOCVD). b) Enlargement of 3 periods of the active region
5
6 CHAPTER 2. TECHNOLOGY
Figure 2.2: STM image of the active region of (a) the structure with graded interfaces and
(b) the structure with abrupt interfaces. (c) shows the averaged line proles of the active
region. With courtesy of P.Oermans and P. M. Koenraad.
2.2 Metal Organic Vapor Phase epitaxy
2.3 Reactive Ion etching
2.4 Quantum Cascade laser processing
A laser consists of a gain medium inserted in a optical resonator cavity. In a quantum
cascade laser, as in most semiconductor lasers, the optical cavity is formed by the epitaxial
layers themselves. In the growth direction, optical connement is achieved by total inter-
2.4. QUANTUM CASCADE LASER PROCESSING 7
nal reection between the high refractive index gain region and the lower refractive index
substrate and cladding layers. Laterally, connement is achieved by dening a narrow (10
30wide) stripe by removing the semiconductor material. Processing of the laser structure
InP Substrate
Gain region
Cladding/contact
mask
MBE/MOCVD growth
a) b)
c) d)
e) f)
Spin resist / expose
Wet etch
S
3
N
4
CVD deposition
S
3
N
4
Dry etch
Top/substrate metalisation
Figure 2.3: Processing steps for the fabrication of a Fabry-Perot cavity quantum cascade
laser. a) MBE or MOCVD growth of the active region. b) Dening the ridge by pho-
tolithography ( spin resist, expose with UV light and develop). c) The ridge is etched in a
wet etching solution. d) Si
3
N
4
is deposited by PECVD. e) Open the insulating oxide on top
of the ridges. f) Top and bottom metalization.
proceeds through the steps outlined in Fig. 2.4. Thirty-ve periods of the active region
and relaxation region pairs are grown on a low-doped (n =1310
17
cm
3
) InP substrate by
Molecular Beam Epitaxy (MBE). The top cladding layer may be grown by MBE or Metal
Organic Chemical Vapor Deposition (MOCVD). Ridges are dened by photolithography and
wet chemical etching. An insulating layer of Silicon Nitride is then deposited by Plasma En-
hanced Chemical Vapor Deposition (PECVD). This Insulating layer is then opened on the
top of the ridges by a dry etching process. Contacts are then evaporated on the top and
bottom of the device for contacting.
8 CHAPTER 2. TECHNOLOGY
Figure 2.4: Schematic drawing of the facet of a processed device.
Chapter 3
Electronic states in semiconductor
quantum wells
3.1 Band structure of semiconductors in the kp ap-
proximation: origin of the eective mass
Band structure of the GaAs
Figure 3.1: GaAs band structure
9
10 CHAPTER 3. ELECTRONIC STATES IN SEMICONDUCTOR QUANTUM WELLS
3.1.1 Basic approximations
Behind the eective mass approximation lies a very powerful approach to the computation of
the band structure. It relies on the knowledge of the band structure at k = 0 and expanding
the wavefunctions in this basis. The Schroedinger equation for a crystal writes:
_
p
2
2m
0
+V (r) +

2
4m
2
0
c
2
(

V ) p
_
(r) = E(r) (3.1.1)
For simplicity, we drop the spin-orbit coupling term. The latter arises as a relativistic term:
the motion of the electon in the eld of the ion, the latter sees an equivalent magnetic eld
that operates on the angular momentum variable. Let us rst compute the action of p on
, written in terms of Bloch wavefunctions so that
k
(r) = e
i
kr
u
n
k
(r), (3.1.2)
we obtain:
p (e
i
kr
u
n
k
(r)) = i
(e
i
kr
u
n
k
(r)) (3.1.3)
=
ke
i
kr
u
n
k
(r) +e
i
kr
p u
n
k
(r) (3.1.4)
= e
i
kr
( p +
k)u
n
k
(r). (3.1.5)
Using the above relation, the Schrodinger equation is obtained for u
n
k
(r):
_
p
2
2m
0
+

m
k p +

2
k
2
2m
0
+V (r)
_
u
n
k
(r) = E
nk
u
n
k
(r) (3.1.6)
The Hamiltonian H = H
0
+W(
k) may be splitted into a k-independent

H
0
=
p
2
2m
0
+V (r) (3.1.7)
and k-dependent part
W(
k) =

2
k
2
2m
0
+

m
k p. (3.1.8)
The solution of the the equation
H
0
u
n0
(r) = E
n0
u
n0
(r) (3.1.9)
are the energies of the band structure at the point k = 0. The fundamental idea of the
k p approximation is to use the u
n0
(r) as a basis for the expansion of the wavefunction and
energies at nite k value. In the simplest cases, taking an interband transition across the
gap and looking at the conduction band, taking the second-order perturbation expansion:
E
c
(k) = E
c
(0) +

2
k
2
2m
0
+

2
k
2
m
0
m=c
[u
c,0
[p[u
m,0
)[
2
E
c
E
m
(3.1.10)
3.1. BAND STRUCTURE OF SEMICONDUCTORS IN THE KP APPROXIMATION:
ORIGIN OF THE EFFECTIVE MASS 11
GaN GaAs InP GaSb InAs InSb
E
g
(eV) 3.4 1.519 1.424 0.811 0.418 0.235
m
c
/m
0
0.17 0.0665 0.079 0.0405 0.023 0.0139
E
p
(eV) 20.2 22.71 17 22.88 21.11 22.49
r
vc
(
A) 2.55 6.14 5.67 11.5 21.5 39.5
0
(eV) 0.017 0.341 0.11 0.75 0.38 0.81
Table 3.1: Fundamental parameters for various III-V semiconductors
In the lowest order approximation, all other bands except for the valence band may be
neglected, in which case the dispersion can be written as (taking the zero energy at the top
of the valence band):
E
c
(k) = E
G
+

2
k
2
2m
0
+

2
k
2
m
0
p
2
cv
E
G
(3.1.11)
Dening the Kane energy E
P
= 2m
0
P
2
such that
E
P
=
2
m
0
[u
c,0
[p[u
v,0
)[
2
(3.1.12)
the dispersion of the conduction band can be written as:
E
c
(k) = E
c
+

2
k
2
2m
0
_
1 +
E
P
E
G
_
(3.1.13)
We then obtain the eective mass as:
(m
)
1
= (m
0
)
1
_
1 +
E
P
E
G
_
(3.1.14)
The Kane energy is much larger than the gap E
P
>> E
G
and is rather constant across the
III-V semiconductors. As a result, the eective mass in inversly proportional to the band
gap.
One should be careful that some authors use the denition
E
P
= 2m
0
P
2
(3.1.15)
(Bastard, for example), while others (Rosencher) use
E
P
= P
2
. (3.1.16)
At least they all use E
P
having the dimension of energy!. In the above expressions, the
spin-orbit term can be included by replacing the operator p by
= p +

4m
0
c
2
(

V ). (3.1.17)
3.1.2 Beyond the perturbation expansion
A very powerful procedure is to expand formally the solutions of Equ. 3.1.6 in the solutions
at k = 0, writing formally:
u
nk
(r) =
m
c
(n)
m
(k)u
m,0
(r) (3.1.18)
and restricting the sum to a limited relevant subset of bands. Improved accuracy can be
achieved by introducing more bands. In this basis, and projecting the equation onto the
state u
M,0
, the Hamilton equation is written as:
_
E
M0
+

2
k
2
2m
0
E
n
(k)
_
c
(n)
M
(k) +
m=M
H
kp
Mm
c
(n)
m
(k) = 0 (3.1.19)
where the kp Hamiltonian is
H
kp
Mm
=

m
0
k u
M,0
[p[u
m,0
) (3.1.20)
To the extend the matrix elements are known, equation 3.1.19 can be solved.
3.1.3 Example: a two-band Kane model
Let us write explicitely a two-band Kane model. The latter can be a fairly realistic model if
one is only interested at the eect of the valence band onto the conduction one, replacing the
three spin-degenerate valence bands (heavy hole, light hole, split-o) by an eective valence
band. The u
nk
can then be expressed as:
u
nk
= a
c
u
c0
+a
v
u
v0
(3.1.21)
Replacing this expansion into Eq. 3.1.19, we obtain the following matrix equation:
_
E
c
+

2
k
2
2m
0
m
0
k p
cv
m
0
k p
cv
E
v
+

2
k
2
2m
0
_
_
a
c
a
v
_
= E
_
a
c
a
v
_
. (3.1.22)
Assuming E
c
= 0, E
v
= E
G
, the solution of the matrix equation satises:
_
2
k
2
2m
0
E)(E
G
+

2
k
2
2m
0
E)

2
m
2
0
[
k p
cv
[
2
= 0. (3.1.23)
This second order equation in E can be of course solved directly; it is however more instructive
to write it under the form a pseudo eective mass equation (remember E
P
=
2
m
0
p
2
cv
:
E(k) =

2
k
2
2m
0
E
P
+E
G
+ 2E
E +E
G
. (3.1.24)
For k 0 the above expression reduces itself to the result of the perturbation expansion:
(m
)
1
= (m
0
)
1
_
1 +
E
P
E
G
_
. (3.1.25)
3.2. ENVELOPE FUNCTION APPROXIMATION 13
Equation 3.1.25 can be expressed in a somewhat simplied form:
m
(E) = m
(0)
_
1 +
E
E
G
_
(3.1.26)
that is commonly used in the literature.
3.2 Envelope function approximation
The problem we will try and solve now is the one of a heterosructure, in which two materials
A and B are sandwiched together. Of course such a material could be seen as a new material
by itself, and its band solved by ab initio techniques, but such a computation is very heavy,
time consuming and moreover does not give much physical insight into the result. The
envelope function approximation solves this problem in a very ecient and elegant manner.
It is widely used to predict the optical, electrical properties of semiconductor nanostructures.
3.2.1 Multiband case
At the core of the envelope function approximation is a generalization of the k p approx-
imation: it is postulated that the wavefunction can be written as a sum of slowly varying
envelop functions f
A,B
l
(r) that will modulate the Bloch function of the material, namely:
(r) =
l
f
A,B
l
u
A,B
l,k
0
(r). (3.2.27)
Behind the equation 3.2.27 is the idea that at each point, the wavefunction is described by
a k p decomposition and that this decomposition depends on the position. Furthermore, it
is assumed that
1. the envelop function f
A,B
l
(r) is slowly varying compared to the Bloch wavefunction, if
f
A,B
l
(r) is written in a Fourier decomposition, the wavevectors are close to the center
of the Brillouin zone
2. the Bloch functions are identical in both materials, i.e. u
A
n,k
0
(r) = u
B
n,k
0
(r). This also
implies that the interband matrix element S[p
x
[X) is equal in both materials.
It allows us to write te wavefunction as
(r) =
l
f
A,B
l
u
l,k
0
(r). (3.2.28)
Let us assume rst a quantum well, in which a layer of material A is sandwiched into a
barrier material B. Because of the in-plane translational invariance, the wavfuction may be
written as plane waves:
f
l
(r
, z) =
1
S
e
ik
l
(z) (3.2.29)
where z is choosen as the growth direction and k
= (k
x
, k
y
) is the in-plane wavevector.
Note that this convention there is confusion in the litterature as the sign may either mean
perpendicular to the plane of the layers or to the growth axis. The Hamiltonian is then
H =
p
2
2m
0
+V
A
(r)Y
A
+V
B
(r)Y
B
(3.2.30)
where the functions Y
A
(z) and Y
B
(z) turn on the potential in the respective layers. We
will develop our system close to k=0. To solve the system, we must:
1. Let H act upon (r)
2. multiply on the left by u
m0
(r)e
ik
m
(z)
3. integrate over space
We have to use the following relations. As the envelop function is slowly varying, we may
write _
cell
d
3
rf
l
f
m
u
l
u
m
= f
l
f
m
_
cell
d
3
ru
l
u
m
= f
l
f
m
lm
(3.2.31)
and take advantage of the fact that the band edge are eigen function of the Hamiltonian at
(k=0):
_
p
2
2m
0
+V
A,B
(r)
_
u
m,0
(r) =
A,B
m,0
u
m,0
(r). (3.2.32)
The derivation is rather tedious, but one should note the similarity with the normal k p
technique by considering the action of the operator p on the wavefunction:
p(e
k
l
(z)u
l
(r)) = (k
z
+p)e
ik
l
(z)u
l
(r) (3.2.33)
and we then may consider the substitution
p (k
i

z
+p) (3.2.34)
where it is understood that p then acts only on the Bloch part of the wavefunction. Using
the above substitution into the Hamiltonian, one nally get the following set of diferential
equation written in a matrix form:
D(z, i
z
) = (3.2.35)
where the elements of the matrix D are given by the equation
D
lm
=
_
A
l
Y
A
+
B
l
Y
B
+

2
k
2
2m
0

2
2m
0
2
z
2
_
l,m
+
k
m
0
l[p
[m)
i
m
0
l[p
z
[m)

z
(3.2.36)
and the matrix element l[p[m) =
_
cell
u
l
pu
m
d
3
r.
3.2.2 One Band model: the Ben-Daniel Duke
As an example, let us consider rst a pure one band model in the eective mass approxima-
tion. Setting
A
l
Y
A
+
B
l
Y
B
= V (z) (3.2.37)
we obtain the Schrodinger equation:
_

2
2m
2
z
2
+V (z)
_
(z) = (z) (3.2.38)
In this equation, the band is assumed to be parabolic with a curvature given by an eective
mass m
. For an isolated band such as the heavy hole band, it is a rather good approximation.
It can also be used successfully for thick quantum well structures in the conduction band,
when the connement energies are much smaller than the material energy gap. One rst
formal diculty arises because the mass of the barrier material is in general dierent than
the one of the quantum well. For this reason the eective mass should be considered as a
position-dependent quantity m
= m
(z). In that case, one shows that the proper boundary

conditions that will match the solutions are the interfaces are not by matching the envelope
function and its derivative, but by matching the envelope function
(z) (3.2.39)
and the quantity
(z)
m
z
. (3.2.40)
The latter condition can be shown to be equivalent to the conservation of the probability
current.
The one band model can be formally derived in the k p framework by adding the eect of
the remote bands in the matrix 3.2.36, the result of which is given (see Bastard) for a 8 band
model with the remote bands as
8
m=1
_
_
A
m
+V
m
(z) +

2
k
2
2m
0

2
2m
0
2
z
2
l,m
+
k
m
0
l[p
[m)
i
m
0
l[p
z
[m)

z
2
2
z
1
M
zz
lm
z

i
2
2
=x,y
_
k
1
M
z
lm
z
+

z
1
M
z
lm
k
+

2
2
,=x,y
k
1
M
lm
k
m
=
l
(3.2.41)
where eective mass parameters M
lm
are dened as
m
0
M
lm
=
2
m
0
l[p
[)
1

0
V
(z)
m[p
[) (3.2.42)
One sees that for a one band model, neglecting the in-plane dispersion, the term

2
2
z
1
M
zz
lm
z
is the only one remaining and will then change the eective mass.
3.2.3 Two band model
For the conduction band, a very nice model is the one in which one keeps one valence band,
creating a two-band model. For simplicity, let us look at the states at k
= 0 and neglect
the free electron term. We obtain the system of equation given by:
V
c
(z)f
c
i
m
0
p
cv
z
f
v
= f
c
(3.2.43)
i
m
0
p
cv
z
f
c
+V
v
(z)f
v
= f
v
(3.2.44)
Extracting f
v
from the second equation yields:
f
v
=
1
V
v
(z)
i
m
0
p
cv
z
f
c
(3.2.45)
replacing into the rst equation, after substitution, the following reslult is obtained.
2
[p
cv
[
2
m
0
z
1
V
v
(z)
z
f
c
+V
c
(z)f
c
= f
c
. (3.2.46)
Recalling the denition of the Kane energy E
P
=
2
m
0
p
2
cv
and dening an energy-dependent
eective mass:
1
m(, z)
=
1
m
0
E
P
V
v
(z)
(3.2.47)
we obtain nally a Schrodinger-like equation
2
2
z
1
m(, z)
z
f
c
+V
c
(z)f
c
= f
c
. (3.2.48)
This model is very useful to model the electronic states in the conduction band with the
inclusion of the non-parabolicity. It is very widely used in the study of intersubband transi-
tions.
3.2.4 Formal derivation of the 2x2 model
The 2x2 model can be formally derived from the 8x8 Kane Hamiltonian close to the point.
In this approximation, the heavy hole is decoupled from the other valence bands and one is
left with two equivalent 3x3 Hamiltonian (one for each spin direction):
H =
_
_
_
_
_
E
c
(z)
_
2
3
pcv
m
0
p
z

_
1
3
pcv
m
0
p
z
_
2
3
pcv
m
0
p
z
E
lh
(z) 0
_
1
3
pcv
m
0
p
z
0 E
so
(z)
_
_
_
_
_
(3.2.49)
acting on the three-dimentional vector of the envelop function = (
c
,
lh
,
so
), where c, lh
and so label the conduction, light-hole and split-o band edges, respectively. Notice that in
the Hamiltonian 3.2.49 we have also neglected the diagonal free electron terms p
2
z
/m
0
that
can be shown to contribute of terms of the order of E
g
/E
p
1. Moreover, as we are only
interested in the energy levels located above the edge of the conduction band, the problem
can be solved by using the second and third rows of
H = E (3.2.50)
to express the equation in the rst row as
p
z
1
2m(E, z)
p
z
c
+E
c
(z)
c
= E
c
(3.2.51)
but with now the energy- and position dependent eective mass given by
1
m(E, z)
=
1
m
0
_
2
3
E
p
E E
lh
+
1
3
E
p
E E
so
_
(3.2.52)
The solution of the dierential equation 3.2.51 give the conduction component
c
and the
energy of the stationary states. We must recall, however, that the totally stationary wave-
functions is given by the three components
c
,
lh
, and
so
, weighted by their corresponding
Bloch functions, so that the only knowledge of
c
is insucient for the complete physical
description of the stationary state. The 2x2 Hamiltonian can be recovered by a unitary
transformation, dening an eective valence band by E
v
= (2E
lh
+ E
so
)/3 and therefore
the gap is now an eective gap E
g
= E
g
+
0
/3, which has also a contribution coming
split-o edge
7
. In doing so, we neglect contributions on the wavefunctions and energies of
the order of (/[E
c
E
v
[)
2
= (/E
g
)
2
, with =
2
3

0
. For typical III-V semiconductors
used for QC lasers this factor is quite small: in GaAs (/E
g
)
2
GaAs
0.01 and in InGaAs
(lattice matched on InP) (/E
g
)
2
GaInAs
0.04 and therefore the corrections never exceeds
few percent. For devices based on InAs, however, this approximation is likely to fail as
E
g
in this material.
A feeling for these approximations can be obtained by comparing the various expressions for
the energy-dependent eective mass. This is done in Fig. ??
Figure 3.2: Comparison for various model for the energy-dependent eective mass
3.3 Hartree potential
For the carrier densities used (typically about 10
16
cm
3
), electron-electron interaction is not
a dominant term but must be taken into account if the electronic states are a good accuracy
is to be aimed at. The eects are especially strong in devices where the electrons are remote
from the ionized donors. It is usually treated in a mean-eld approximation by adding a
Hartree potential V
H
(z) to the Hamiltonian:
H =

2
2m
2
+V
QC
(z) +V
H
(z) , (3.3.53)
This term is computed from the local charge density
(z) = q
0
N
D
(z) q
0
i
n
i
[
i
(z)[
2
, (3.3.54)
where N
D
is the doping prole of ionized dopants,
i
(z) the probability density and n
i
is the
number of electrons per unit area in the i-subband. The potential V
H
(z) is computed from
(z) using Poissons equation:
2
V
H
(z)
z
2
=
(z)
0
. (3.3.55)
The electronic densities n
i
on the subbands are not known a priori, and in the general case
depend on the transport in the device. However, a good starting approximation is to assume
that the electron distribution is thermal, charaterized by Fermi distribution with a common
chemical potential in each period: and that charge neutrality is achieved in each period,
i.e.:
i
n
i
=
i
_
T
i
(E)f(E)dE (electrons) (3.3.56)
=
_
N
D
(z)dz = n
s
(impurities) (3.3.57)
where
T
i
(E) =
m
i
(E)
2
(E E
i
)
is the density of states of the i-subband, (E E
i
) is the Heavyside function, f(E) =
1
1+exp(
E
kT
)
is the Fermi distribution function, and n
s
is the total areal electron concentration,
which is equal to the impurities concentration. The Hartree potential depends on the solution
of Schrodingers equation; therefore Schrodingers and Poissons equations must be solved
iteratively until convergence is achieved.
3.4 Building blocks
When designing intersubband structures, it is important to acquire an intuitive grasp of
simple building block. This is done most easily in a simple one-band model. Solving for such
structure usually follows the following methodology:
3.4. BUILDING BLOCKS 19
1. The wavefunctions are found for the regions in which the potential is constant (or
assumed constant). They will be plane waves or decaying exponentials
2. Those solutions are then matched at each interfaces, using the boundary conditions
compatible with the Hamiltonian. For one band model, it means matching the value
of the wavefunction and of the derivative, divided by the mass, at the interfaces.
3. Boundary conditions are then imposed on the edge of the sample: decaying exponential
for bound states, or periodicity for a periodic structure
We will then review some of these results rapidely.
3.4.1 The single quantum well
The simplest case is of course the quantum well with innite barrier height. The wavefunc-
tions are simply sinus and cosinus functions (if the well is taken symmetric around zero) or
sinus only if the quantum well is lying in the interval [0,L]. In the latter case, the condition
on the value of k
k
n
= n
L
(3.4.58)
with n integer enables immediatly to write the energy as:
E
n
=

2
k
2
n
2m
=
n
2
2
2m
. (3.4.59)
Of course, in real structures one should taken into account the nite barrier height. Following
Bastard, we assume a potential with the form:
V
b
(z) =
_
V
b
if [z[ L/2
0 if [z[ > L/2
the bound energy solutions ( < 0) are solution of the transcendental equation:
k
w
tan(k
w
L
2
) =
b
(3.4.60)
for even states and
k
w
cotan(k
w
L
2
) =
b
(3.4.61)
for odd states. The wavector k
w
and
b
are dened by :
k
w
=
_
2m
2
( +V
b
) (3.4.62)
and
b
=
_
2m
2
. (3.4.63)
For positive energies, one does not obtain real bound states anymore, but resonances for
which the transmission function of the quantum well (corresponding to the transmission
probability of an incident electron at energy ) has a maximum has a maximum. The
transmission function can be computed and is given by
T() = [t()[
2
=
_
cos(k
w
L)
i
2
_
+
1
_
sin(k
w
L)
_
1
2
(3.4.64)
where is given by the weighted ratio of the wavevectors:
=
m
b
k
w
m
w
k
b
(3.4.65)
The following gures show the computation of such energy levels in various cases:
Figure 3.3: Computed connement energy of electrons in the conduction band as a function
of well width. Bound states are shown with full lines, resonances with dotted lines
A very attractive feature of the two-band model is that it can be implemented numerically
the same way as the one-band one by just introducing the energy-depedent eective mass.
This is of course also valid in the formula given above. The importance of introducing the
non-parabolicity via the interaction of the conduction and valence band is shown clearly in
the folloginw gure, where the energy states of a single quantum well are compared in the
case where the non-parabolicity is neglected (dotted line) and where it is taken into account
(solid lines). The discrepancy for the position of the highest level is 90meV, much larger
than any experimental uncertainty.
Figure 3.4: Energy states of a quantum well computed with a two-band model, and compared
with a one-band model (dashed lines). The growing importance of non-parabolicity as one
moves away from the gap is clearly apparent.
3.4.2 The coupled well system
The system to be considered here is two quantum well coupled through a tunnel barrier.
Of course this system can be modeled directly by solving the Hamiltonian equation of the
whole structure. We however wish to study this system in a thight-binding model in which
we use a base formed by the solution of the individual wells. Again, following Bastard, the
Hamiltonian to solve is:
H =
p
2
z
2m
+V
b
(z z
1
) +V
b
(z z
2
) (3.4.66)
Figure 3.5: A coupled quantum well system.
We solve the problem using the basis wavefunction of the isolated wells, that satisfy:
p
2
z
2m
+V
b
(z z
1
)(z z
1
) = E
1
(z z
1
). (3.4.67)
Neglecting higher excited states, the complete wavefunction can be expanded in terms of the
basis function using:
(z) = (z z
1
) +(z z
2
). (3.4.68)
Introducing the expansion in the Hamiltonian yields the following matrix equation:
_
E
1
+s (E
1
)r +t
(E
1
)r +t E
1
+s
__

_
= 0 (3.4.69)
Solving for E
1
yields the equation:
= E
1
t
1 r
+
s
1 r
(3.4.70)
where r is the overlap,
r =
1
(z z
1
)[
1
(z z
2
))
s is the shift
s =
1
(z z
1
)[V
b
(z z
1
)[
1
(z z
2
))
and t the transfer integrals
t =
1
(z z
1
)[V
b
(z z
1
)[
1
(z z
2
)).
This result can be expressed graphically, as shown in the following gure. The most impor-
tant term is then the transfer integral because it is responsible for the splitting between the
two states. The larger the transfert integral, the wider the spacing between the two states.
Figure 3.6: A coupled quantum well system.
3.4.3 The superlattice
In a superlattice, the potential V
b
(z) is dened as a sum of potential of the individual wells:
V
b
(z) =
n=
V
b
(z nd)
with each potential well being
V
b
(z nd) =
_
V
b
if [z nd[ L/2
0 if [z nd[ > L/2
Figure 3.7: Schematic band structure of a superlattice.
The wavefunctions are then given, for the quantum wells, by:
(z) = exp(ik
w
(z nd)) + exp(ik
w
(z nd)); [z nd[ L/2
For the barriers, for positive energies ( > 0), the wavefunction is then given by:
(z) = exp(ik
b
(z nd
d
2
)) + exp(ik
b
(z nd
d
2
)); [z nd
d
2
[ h/2.
In addition to the boundary conditions at both interfaces of the quantum well, we should also
impose the periodicity of the structure. As it is well-known, it will force the wavefunction
to follow Bloch theorem, that is
q
(z +nd) = exp(iqnd)
q
(z) (3.4.71)
for a periodicity d.
The resulting transcendental equation is given by:
cos(qd) = cos(k
w
L) cos(k
b
h)
1
2
_

1
_
sin(k
w
L) sin(k
b
h) (3.4.72)
where has the same meaning as for the single quantum well case. For negative energies,
i
and the trenscendental equation becomes:

cos(qd) = cos(k
w
L) cosh(
b
h)
1
2
_
_
sin(k
w
L) sinh(
b
h) (3.4.73)
The above equation denes a dispersion relation between the energy (appearing in k
w
and
k
b
) and q. An example of such a dispersion, showing clearly the minibands and minigaps, is
shown in Fig. 3.8. In particular, the increase in the miniband width with energy is clearly
apparent.
Figure 3.8: Computed dispersion of a 100A InGaAs well and 30A AlInAs barrier superlattice.
A conduction band discontinuity of E
c
= 0.52eV as well as a Kane energy of 18.3eV was
assumed in an eective two-band model. For clarity, the zero of the energy scale has been
set to the bottom of the InGaAs quantum wells
The superlattice dispersion of the rst miniband is compared to a sinusoidal t, showing the
that the latter is a very good approximation for narrow minibands.
0.005 0.01 0.015 0.02 0.025
Wavevector A
1
0.048
0.049
0.05
0.051
0.052
0.053
0.054
E
n
e
r
g
y
e
V
100A30A Superlattice dispersion

Figure 3.9: Dispersion of the rst miniband (green), compared to a sinus function (red).
The result of the nite superlattice are also plotted as full squares
The equivalent real space picture is shown in Fig. 3.10 where a the energies and wavefunctions
of a nite eight quantum well long superlattice
3.5. IN-PLANE DISPERSION 25
Figure 3.10: Square of the wavefunctions for a nite superlattice formed by eight 100A
InGaAs wells separated by 30A AlInAs barriers.
3.5 In-plane dispersion
For electronic states, the in-plane dispersion can be computed with various levels of accuracy.
The simplest approach is to assume that each subband has an eective mass given by its
mass at k=0. This approximation holds relatively well in the conduction band. However, it
completely fails in the valence band because the in-plane wavevector couple states belong-
ing to dierent bands and a complicated spectrum arises. For this reason, the analysis of
intersubband devices operating in the valence band is much more complicated.
Figure 3.11: Dispersion of the state of a quantum well in the valence band.
3.6 Full model: the valence band
To be accurate, the full kp envelope function model must be solved, that is the set of 8
dierential equations dened by the equation 3.2.35. Solving the full model is essential in
the valence band because no simplied model can easily be used if some degree of accuracy
is sought. It is observed that:
It is usualy convienient to use the growth direction as the quantization direction for
the angular momentum.
The connement potential lifts the degeneracy between the heavy and light hole bands,
because their dierent mass induce a dierent connement energy
The in-plane dispersion is highly non-parabolic because of the coupling between the
bands induced by the in-plane momentum. In particular, this coupling prevents any
crossing between the light-hole derived and the heavy hole-derived band (see dahsed
lines)
In some cases, the mass is inverted: the bottom of the LH1 band has a electron-
like character over some portion of reciprocal space because of the repulsion and its
proximity to the HH2 state.
This eect is shown schematically in Fig. 3.12. As a result, the computed band structure is
usually fairly complex, and yield results such as the ones shown in Fig. 3.11
3.6. FULL MODEL: THE VALENCE BAND 27
Figure 3.12: Schematic description of the origin of the valence band dispersion in the quan-
tum well showing schematically the eects of connement and interactions.
Chapter 4
Intersubband absorption
4.1 Interaction Hamiltonian
Typically two dierent Hamiltonian can be used to compute the absorption. The rst one
is the dipole given by
H
int
= q
R

c sin(t) (4.1.1)
The other one is derived from the potential vector

A by
H =
(
P q
A)
2
2m
(4.1.2)
Use Coulomb gauge (A = 0). For low intensity (neglect the term in A
2
), we obtain as an
interaction Hamiltonian:
H
int
=
q
m
0
A

P. (4.1.3)
Because of the large dierence between the light wavelength and the atomic dimension, the
spatial dependence of A(r) is neglected inside the matrix elements. This is called the dipole
approximation. The form commonly used is then:
i
[H
int
[
j
) =
q

A(r)
m

i
(r)[
P[
j
(r)) (4.1.4)
For a plane wave,

A is paralell to

c, for a wave polarized along z and propagating in the y
direction:
A(r, t) = A
0
e
z
e
i(kyt)
+A
0
e
z
e
i(kyt)
. (4.1.5)
It is convient to use A
0
being pure imaginary such both electric c and magnetic B elds are
real: that yield c = 2iA
0
and B = 2ikA
0
.
The matrix elements of the position and momentum operators are related to each other.
As the commutator between p and z is [z, p] = i, applying it to the kinetic term of the
Hamiltonian one can derive the relation:
n
[p[
m
) = im
0
nm
n
[z[
m
) (4.1.6)
This relation between the momentum and position matrix elements enables the use of either
one interaction Hamiltonian equivalently.
29
30 CHAPTER 4. INTERSUBBAND ABSORPTION
4.2 Intersubband and interband transition
The intersubband and interband transitions have a very dierent character. The distinction
can be easily understood when looking at the action of the momentum operator on two
wavefunction taken each on a dierent band:
i,j
(r) = f
i.j
(r)u
,
(r) (4.2.7)
where f
i,j
is the envelope and u
,
the Bloch part of the wavefunction. The matrix element
can be written then as
i
[p[
j
) = f
i
u
[p[f
j
u
) (4.2.8)
= f
i
[p[f
j
)u
[u
) +f
i
[f
j
)u
[p[u
) (4.2.9)
= f
i
[p[f
j
)
+f
i
[f
j
)u
[p[u
) (4.2.10)
where we have taken advantage of the slow variation of the envelope function compared to
the one of the Bloch part.The rst part of the last expression represents the intersubband
transition and will be non-zero when two envelope states are taken from the same band, the
second corresponds on the contrary to an interband transition matrix element. In this one-
band model, one should be careful to use the momentum matrix element with the eective
mass m
instead of the bare electron mass m

0
. One easy way to see this is to look at
the commutator [z, H] used to derive the relation between dipole and momentum matrix
elements; this one will involve the Hamiltonian with the eective mass and not the bare
electron mass.
4.3 Selection rules and absorption geometries
It can be shown that only the component of the electric eld in the growth direction will
induce a non-zero dipole matrix element. In fact, taking for the wavefunction the canonical
form:
f
i
=
1
S
e
ik
i
(z) (4.3.11)
then the matrix element corresponding to the x-component of the momentum p
x
yield:
(c
x
p
x
)
ij
=
i
[
j
)k
x
c
x
,k
= k
x
c
x
,k
i,j
(4.3.12)
which, because of the term
i,j
is zero for transitions between dierent states. As a result, only
the z-component of the electric eld couples to the intersubband transition. This has profound
experimental implications as it for example rules out absorption for light at normal incidence.
As a result, a number of geometries have been developed to measure the absorption, as shown
in Fig. 4.2.
4.3. SELECTION RULES AND ABSORPTION GEOMETRIES 31
Figure 4.1: Experimental geometries allowing the measurements of intersubband transitions
The validity of the selection rule was checked in GaAs/AlGaAs samples by H.C. Liu and
coworker in a photocurrent experiment. They found that for such samples, the absorption
by the TE polarization was less than 0.2% than the TM one.
Figure 4.2: Photocurrent in a quantum well photoconductor as a function of the polarization.
The residual responsivity for the TE polarization is less than 0.2% than for the electric eld
normal to the layers. (From H.C.Liu et al, APL 1998)
The absorption in TE is non-zero when considering the wavefunctions in the multiband case,
and is discussed shortly further in the text.
4.4 Absorption strength
4.4.1 Absorption for a 2D system
When looking at the transmission through of an essentialy two-dimentional system, one
should realize that the correct number to represent the absorption is unitless and represents
the fraction of the light absorption by the system. To compute the latter, one should start
from Fermis golden rule: the scattering rate, using Fermis golden rule, writes:
R =
1
=

2
[i[H
int
[f)[
2
(E
f
E
i
). (4.4.13)
The absorption of the electromagnetic wave by the quantum system is responsible for a decay
of the latter. Writing
2D
as the fraction of the wave absorbed by the system, we have that
the energy loss per area S is now
P/S = I
2D
(4.4.14)
and the intensity is related to the electric eld by
I =
1
2
0
n
refr
cc cos (4.4.15)
where n
refr
is the refractive index and c the light velocity. We assuming the beam arrives
at an angle to the normal; only the component of c in the z-direction will couple to the
intersubband system. Balancing the energy loss of the EM eld with the energy gained by
the quantum system yields:
2D
() =
2R
0
n
refr
cc
2
sin
2
cos
(4.4.16)
therefore:
2D
() =

0
n
refr
c
i,f
[i[ qZ[f)[
2
(E
f
E
i
)(k
f
k
i
)
sin
2
cos
(4.4.17)
where is now a density of state per unit area, the delta function forces the conservation of
the momentum in the light absorption. In the one band model, the electron subbands are
paralell and the absorption writes:
2D
() =
q
2
0
n
refr
c
z
2
ij
n
i
(E
f
E
i
)
sin
2
cos
(4.4.18)
where n
f
and n
i
are the populations (per unit area) of the nal and initial states, respectively
and z
ij
is the shorthand notation for the dipole matrix element. Scattering of the electrons
with lattice vibrations, as well as collisions with interface roughness does broadens the line in
practice and the delta function (E
f
E
i
) can be replaced by a normalized Lorentzian
line
/
(E
f
E
i
)
2
+
2
of half width and we have nally for the absorption:
2D
() =
q
2
z
2
ij
n
i
0
n
refr
c
(E
f
E
i
)
2
+
2
sin
2
cos
(4.4.19)
For the geometries shown in Fig. 4.2 one should multiply the above result by the number of
quantum wells in the stack and by the number of passes in the waveguide.
4.4. ABSORPTION STRENGTH 33
4.4.2 Waveguide
An other interesting case is the one of the waveguide where the absorption occurs in a guided
mode. In that case, the angle of incidence is = /2 and the expression 4.4.19 diverges.
In that case, it is more convenient to think of the absorption as being from a homogenous
material of thickness L, that is dene an absorption coecient =
2D
cos /L for a
homogenous material in which the light would be traveling. It is sucient to reinterpret the
above equation
() =
q
2
z
2
ij
N
i
0
n
refr
c
(E
f
E
i
)
2
+
2
(4.4.20)
with N
i
= n
i
/L is the volume density of the electron. The natural question is of course
what is the correct value of L ?. Is L the thickness of the well, the eective one of
the wavefunction, the one of the quantum well and the barrier?. In most cases, happily,
the particular choice of L as no practical inuence on the relevant physical quantity to be
measured. In fact, what is measured directly is not the material absorption but the one of
a guided mode. The modal loss of the waveguide
w
is related to the material one by an
overlap factor,
w
=
_
c
2
(x)(x)dx
_
c
2
(x)dx
= (4.4.21)
where the overlap factor is dened as a function of the electric eld prole in the waveguide
c(x) by
=
_
L/2
L/2
c
2
(x)dx
_
c
2
(x)dx
L
c
2
(4.4.22)
if

c
2
is the normalized intensity in the guide at the location of the quantum well. The linear
dependence of in L just cancels the inverse dependence of in the same quantity.
4.4.3 Gain and loss cross sections
When both lower and upper states are populated, the ground state density in equation 4.4.20
should be substituted by the dierence between ground and upper state densities. Rewriting
Equ. 4.4.20 using the wavelength
2
instead of

c
and writing the volume density N
i
= n
i
/L
per
as a function of the sheet density n
i
where L
p
is the period length of the structure we obtain:
() =
2q
2
z
2
if
(n
i
n
f
)
0
n
refr
L
p
(E
f
E
i
)
2
+
2
. (4.4.23)
The gain is dened simply as a negative absorption
g() = (). (4.4.24)
An important quantity is the peak gain cross section, or the maximum gain at resonance
( = E
f
E
i
) per unit of inverted carrier:
g
c
= g/(n
f
n
i
) =
2q
2
z
2
if
0
n
refr
L
p
(4.4.25)
4.5 Experimental results
The rst measurement of the intersubband absorption in a heterojunction quantum well (as
compared to the one in an inversion layer) is reported in Fig. 4.3. In this rst measurement,
the quantum well was a
Figure 4.3: Intersubband absorption in an 65A thick quantum well. (From West and Eglash,
APL)
4.5. EXPERIMENTAL RESULTS 35
Figure 4.4: Intersubband absorption between two bound states
In contrast to interband transition, where the interband matrix element r
cv
is the dominant
term, the atomic-like nature of the joint density of state as well as the tailorability of the
potential enables the fabrication of complex energy ladder structures.
Figure 4.5: Intersubband absorption in a multiquantum well designed for triply resonant
non-linear susceptibility
When the excited state is in a continuum, the absorption is broadened.
Figure 4.6: Intersubband absorption from a bound state to a continuum
4.6 Sum rule in absorption
Combining the completeness of the (eigen)states
n
[
n
)
n
[ = 1 (4.6.26)
and the relationship between position and momentum matrix element [z, p] = i, the sum
rule for the oscillator strength can be derived:
n
f
nl
= 1/m
(4.6.27)
where the oscillator strength is dened as:
f
ij
=
2m
0
2
[z
ij
[
2
(E
j
E
i
). (4.6.28)
Writing the sum rule this way emphasizes the eect of the eective mass in the enhancement
of the intersubband absorption that basically scales like 1/m
. In the k p approach of Kane,

the eective mass arises from the interaction between the conduction and valence bands, and
is therefore proportional to the bandgap. As a consequence, narrow gap semiconductor will
exhibit much larger intersubband transition strength than wide bandgap materials. This
important fact is obscured when a normalized oscillator strength is used f
= fm
in
place of the usual denition 4.6.28.
4.7. ABSORPTION IN A QUANTUM WELL: A TWO-BAND MODEL 37
Equation 4.6.27 also holds with the initial state being and excited state, in which case the
downward transitions have a negative sign. Since the sum must remain constant, the
upwards transitions have an oscillator strength that grow with the initial state index i. In
the simplest case of a harmonic oscillator, the oscillator strength f
i,i+1
between adjacent
states is simply proportional to the index i of the state:
f
i,i+1
= if
1,2
= i(m
/m
0
)
1
. (4.6.29)
This very important fact shows that transitions between excited states, as they occur natu-
rally in quantum cascade lasers, naturally yield larger intersubband absorption.
Figure 4.7: Comparison of the intersubband absorption for various structures, with energy
levels schematically drawn close to the curves
4.7 Absorption in a quantum well: a two-band model
The problem arising with the previous model is that it does not allow the introduction of non-
parabolicity. It is valid then only for connement energies much smaller than the bandgap
(E
n
, E
m
<< E
G
). One could be tempted to use the same approach when computing the
absorption in the multiband case, taking as the matrix element the envelope function of the
electron. This approach was used in some of the works, but lead to diculties since, strickly
speaking, the envelope functions of the conduction bands are no longer orthogonal to each
other (
m
[
n
) , =
nm
).A much better approach is to compute the matrix element directly
in the multiband model. If one wants to treat conduction band state, an eective two-band
model is perfectly valid. Let us then assume a two-component wavefunction:
1
= f
(1)
c
u
c
+f
1
v
u
v
(4.7.30)
2
= f
(2)
c
u
c
+f
2
v
u
v
(4.7.31)
Let us then evaluate the matrix element p
z
:
1
[p
z
[
2
) = f
(1)
c
u
c
+f
(1)
v
u
v
[p
z
[f
(2)
c
u
c
+f
(2)
v
u
v
)
= f
(1)
c
u
c
[p
z
[f
(2)
c
u
c
) +f
(1)
c
u
c
[p
z
[f
(2)
v
u
v
) +f
(1)
v
u
v
[p
z
[f
(2)
c
u
c
) +f
(1)
v
u
v
[p
z
[f
(2)
v
u
v
).
(4.7.32)
Let us consider the rst term, it yields:
f
(1)
c
u
c
[p
z
[f
(2)
c
u
c
) = f
(1)
c
[p
z
[f
(2)
c
)u
c
[u
c
) +f
(1)
c
[f
(2)
c
)u
c
[p
z
[u
c
)
= f
(1)
c
[p
z
[f
(2)
c
)
(4.7.33)
Similarely, the second term will yield
f
(1)
c
u
c
[p
z
[f
(2)
v
u
v
) = f
(1)
c
[f
(2)
v
)p
cv
(4.7.34)
Using similar derivations for the third and fourth terms, the result can be summarized in a
matrix form:
p =
_
p
z
p
cv
p
cv
p
z
_
(4.7.35)
acting on the components (f
c
, f
v
) of the wavefunction. Dropping the diagonal terms as
p
z
<< p
cv
, we nally obtain:
1
[p
z
[
2
) = f
(1)
c
[f
(2)
v
)p
cv
+f
(1)
v
[f
(2)
c
)p
cv
(4.7.36)
Using the relationship between f
c
and f
v
given by the equation 3.2.45 and using the denition
of the energy-dependent eective mass 3.2.47, we nally obtain:
1
[p
z
[
2
) = f
(1)
c
[
_
i

z
_
m
0
m(E, z)
+
m
0
m(E, z)
_
i

z
_
[f
(2)
c
) (4.7.37)
In this picture, both intersubband and interband transitions are treated on the same footing.
Using a generalized version of this model, it is possible to show (R.Yang, PRB 1994) that
transitions with in-plane electric eld couple states with opposite spins, and for a symmetric
quantum well for a transition between state of opposite parity, the ratio between the two
absorption is given approximately by
1
3
E
ij
(E
i
+E
g
+ )(E
j
+E
g
)
(4.7.38)
where E
ij
is the intersubband transition energy, is the spin-orbit splitting and E
g
the
fundamental energy gap of the material.
4.8. ABSORPTION LINEWIDTH 39
4.8 Absorption linewidth
As in atomic systems, nite upper state lifetime and inhomogeneities, transform the lineshape
of the optical transition from the delta function that arises in simplied approach into a
function with a nite energy width. In the preceding section, we assumed that the delta
function was replaced by a (normalize) lineshape function g() represented by a simple
Lorenzian linwidth with a broadening parameter .
g() =

( E
ji
)
2
+
2
(4.8.39)
However, in contrast to the atomic system where the homogenous broadening component
(collisions and radiative transition) can by well distinguished from the inhomogenous compo-
nents (velocity distribution due to thermal motion), in the quantum well case this distinction
is much less clear, as we will have to consider interface roughness scattering as the source
for a homogenous scattering. As such, the lineshape function has a very important role in
the modeling of quantum cascade lasers; for any given design, material choice, injection level
the densities and oscillator strength will be xed; the value of the gain will then be inversely
proportional to the linewidth as the lineshape function is normalized to unity.
In a naive approach, one would probably like to write a realistic lineshape function as the
convolution of a Lorenzian lineshape (given by the electron lifetime in the subband) over
a distribution of well width and the distribution of transition energies given by the non-
parabolicity. However, in such an approach, the lineshape should tend to a square, Lorenzian,
or Gaussian lineshapes in the following limiting cases, as shown in Fig. 4.8
Non-parabolicity: At low temperatures, for large electron densities and low bandgap
materials, where non-parabolicity should be dominant, a square-shaped lineshape should
be observed.
Disorder: For narrow wells, large bandgap materials, where disorder should be play a
dominant role, a Gaussian lineshape should be observed.
Lifetime: For systems in which a short lifetime would be expected, a Lorenzian line-
shape should be expected.
4.8.1 Non-parabolicity
Assuming the non-parabolicity is treated in the two band model, according to equation ??,
the broadening E is given by
E = E
F
E
21
E
G
(4.8.40)
where E
F
is the Fermi energy of the ground subband, E
21
is the intersubband transition
energy, E
G
is the bandgap. As an example, for a 15nm thick InAs/AlSb quantum well
(E
G
= 0.42eV ), for transition energy of 125meV, a sheet density of n
s
= 1 10
12
cm
2
Figure 4.8: Lineshape functions in various limiting cases.
corresponding to a Fermi energy of about 60meV, the predicted broadening is about 20meV.
The intersubband absorption was measured by Warburton et al in such a sample; and instead
of the square-top lineshape that would be expected from the simple argument, a narrow line
was observed. The authors eplained this eect by the depolarization shift. A simple picture
of this eect is achieved by considering the boundary condition for a TM wave incident
on the quantum well system. This boundary condition forces the displacement eld to be
continuous at the interface, i.e. the quantity c. Because of the dispersion induced by
the resonance, the (real part) of the dielectric constant is enhanced on the red side of the
transition, while it is depressed at the blue side. As a result, the eective electric eld
penetrating in the quantum well system is decreased on the red side of the transition while
it is enhanced on the blue side of it. The net result is a shift of the absorption peak to the
blue, called depolarisation shift. This denomination comes from the fact that the electrons
eectively screen the incident electric eld, preventing it from penetrating in the quantum
well system.
4.8.2 Disorder
In quantum well samples, the greatest source of disorder is the monolayer uctuations of the
quantum well interfaces. These will occur over a given correlation length
corr
. The rst
experimental indication that the linewidth was limited by interface roughness scattering
came from the experimental work of Campman and coworker who measured a series of
GaAs/AlGaAs quantum well samples of various width. They even measured also samples
where the well was not from a binary but the ternary systems AlGaAs and InGaAs. They
found Lorenzian lineshapes whose width did decreased with increasing width, as shown in
Fig. ??. A very successful model has been developed by Unuma et al. [1], based on a
statistical description of the interface that exhibits Gaussian uctuations of step height
4.8. ABSORPTION LINEWIDTH 41
Figure 4.9: Transmission through a 15nm thick InAs/AlSb mulitquantum well system. Be-
cause of the non-parabolicity of this system, a broad (about 20meV wide) peak would be
expected, in contrast with the narrow experimental line observed. This narrowing was in-
terpreted by Warburton et al. as a result of depolarization shift.
with a correlation length , such that:
(r)(r
)) =
2
exp
_
[r r
[
2
2
_
. (4.8.41)
The matrix element of scattering between an initial state in the subband m, at wavevector
k
to subband n at wavevector k is
mk
[H
l
[nk) =
_
d
2
rF
mn
(r)e
iqr
(4.8.42)
where the form factors F
mn
appear and express the inuence of the interfaces on the energy
levels:
F
mn
=
_
E
m
/L)(E
n
/L) (4.8.43)
so that the total intra-subband scattering rate (neglecting screening) may be written as:
IFR
intra
(E) =
m
2
(F
00
F
11
)
2
_

0
de
q
2
2
/4
(4.8.44)
where the wavevector q is expressed as:
q
2
= 2k
2
(1 cos ). (4.8.45)
Using this model, the authors were able to explain the dierence between the lifetime deduced
from the mobility and the one obtained from the linewidth of the intersubband absorption.
Apart from an additional angular dependence, the key dierence is that for the mobility,
the prefactor in the scattering rate is F
2
00
, whereas it is (F
00
F
11
)
2
for the intersubband
Figure 4.10: Linewidth of the absorption of quantum wells of various widths. The dotted
line shows the compute value of the broadening caused by the optical phonon scattering.
The decrease in linewidth as a function of quantum well width is a
absorption and [F
11
[ > [F
00
. This model was successfully extended to the case of many
interfaces by Tsujino [20] by introducing a correlation coecient c
i,j
between successive
interfaces.
IFR
intra
(E) =
m
i,j
c
ij
F
i
F
j
(4.8.46)
and where the form factors F
i
and F
j
are expressed as a function of the amplitude of the
wavefunction at the interface located a the position z
i
by:
F
i
= U(z
i
)
_
[f
n
(z
i
)[
2
[f
m
(z
i
)[
2
_
. (4.8.47)
Using this model, the author was able to correctly predict the linewidth of various structures
based on vertical and diagonal transitions and grown by various MBE reactors.
4.9 Stark-tuning of intersubband absorption
The addition of an electric eld to the structure will shift the levels. As compared to the
case of atoms and molecules, the eect is much stronger, for two reasons. First of all, the
wavefunction extends for a much larger distance even for the ground state, as the eective
mass of electrons is usually signicantly lighter than the bare mass. Using a second-order
perturbation approximation and taking into account only the rst two levels of a symmetric
quantum well, the Stark shift is written as:
E
12
(c) =
2q
2
0
z
2
12
c
2
E
12
(4.9.48)
As shown in the above equation, the Stark shift can easily as large as a few meV for a
structure with a transition at 100meV. A very interesting feature is however that quantum
4.9. STARK-TUNING OF INTERSUBBAND ABSORPTION 43
Figure 4.11: Experimental measurement of the broadening of the intersubband transition
and mobility as a function of temperature, compared with the model proposed by Unuma
et al. [1]. Optical and acoustic phonon intra- and inter-subband processes are also included
in the model, as indicated.
well structures need not be symmetric, in which case the rst-order term of the perturbation
theory dominates and the Stark shift is now:
E
12
(c) = q(z
22
z
11
)c (4.9.49)
where the term z
22
z
11
is the dierence between the center-of-mass of the probability
distribution between the upper and lower state. Using this eect, very large changes of the
transition energy can be achieved. Such tunability has been used in lasers and modulators.
Chapter 5
Intersubband scattering processes
As shown in Fig. 5, an electron in an excited subband can recombine to the lower one
through a variety of processes. It can change subband through spontaneous emission, in-
elastic scattering by a phonon, elastic scattering through an impurity or interface defect or
nally thought an electron-electrons scattering. In stark contrast to the interband case, the
radiative emission of photon is by large not the dominant intersubband mechanism even in
a perfectly clean material. As shown schematically in Fig. 5, the situation is very dierent
Figure 5.1: Schematic description of the relevant scattering mechanism in the two relevant
cases. The simplest case is the situation occuring in the Mid-Infrared (depicted on the
left) where the LO phonon scattering is dominant. In contrast, in the terahertz, at low
temperature, a number of non-radiative processes should be taken into account including
acoustic phonons, electron scattering and other elastic processes.
if one considers two subband that are spaced by an energy larger or smaller than the opti-
cal phonon energy. In the rst case, for an electron sitting in the excited subband, optical
phonon emission is always possible and lead to lifetimes in the order of 1ps. This process is
usually the dominant scattering mechanism as other processes that could compete with the
latter can be engineered away. For two subband separated than less than an optical phonon
energy, the situation is more complex and to some extend still plagued by a lack of good
45
46 CHAPTER 5. INTERSUBBAND SCATTERING PROCESSES
experimental data to contreversy.
5.1 Spontaneous emission
Between an initial state i and a nal state f with a non-zero optical matrix element z
if
,
spontaneous photon emission is possible and will occur with a rate given by
w
sp
=
q
2
0
n
refr
z
2
if
E
3
fi
3c
3
4
(5.1.1)
The energy dependence of the above equation can be better grasped when expressing it as
a function of the oscillator strength between the two states:
w
sp
=
q
2
0
6m
0
c
3
2
n
refr
f
if
E
2
fi
(5.1.2)
as the oscillator strength will be of the order of 1/m
for transition between the rst two

states in a potential well f
12
=
2
8
2
3
3
/m
0.96/m
for the case of the innite quantum

well, and f
12
= 1/m
for the parabolic well. As a result, the spontaneous emission rate is

fundamentally proportional to the square of the photon energy. In Fig. 5.1, the spontaneous
0.0150.02 0.03 0.05 0.07 0.1 0.15 0.2
Transition Energy eV
5. 10
8
1. 10
7
5. 10
7
1. 10
6
5. 10
6
0.00001
L
i
f
e
t
i
m
e
Radiative lifetime
Figure 5.2: Radiative lifetime of a square quantum well as a function of the transitiion
energy. A constant value of the oscillator strenght of 23, corresponding to an InGaAs/AlInAs
quantum well, has been assumed
lifetime is plotted as function of transition energy for a transition between the rst two
states in a InGaAs/AlInAs square well, assuming an oscillator strength f
12
= 0.043
1
. The
lifetime varies between 60ns at = 10m to 6s at = 100m. These long lifetimes are
to be compared with the very short non-radiative lifetime observed in the same system, and
explain the very low radiative eciency observed in the spontaneous emission.
5.2. SCATTERING RATE BY BULK OPTICAL PHONONS 47
5.2 Scattering rate by bulk optical phonons
The simplest treatement of optical phonon scattering in polar III-V semiconductors follows
an approach rst outlined by Price [21]. It is based on two important assumptions, usually
well veried in III-V structures:
The phonon scattering is dominated by the Frolich Hamiltonian, i.e. an interaction
between the electron and the piezoelectric potential created by the local lattice defor-
mation.
The optical phonon have very little dispersion, and therefore can be assumed to be
monoenergetic.
In this approach, an electron with an initial wavevector k
i
and energy E
i
in the upper
subband scatters to the lower subband at k
f
, therefore losing (or gaining, in the case of
optical phonon absorption) an energy equal to
LO
in the process. The process is shown
schematically in Fig. 5.2. The momentum Q exchanged in the process is
Q =
_
k
2
i
+k
2
f
2k
i
k
f
cos . (5.2.3)
The value of Q is on average larger for phonon absorption than for phonon emission because
the nal state is higher in the band. The scattering rate is obtained then by applying Fermis
k
i
k
f
Q
E
k
x
k
y
Figure 5.3: Scattering by an optical phonon between two subbands. An electron with initial
energy Ei and initial wavevector k
i
is scattered to the lower subband at k
f
, therefore losing
(or gaining) an energy equal to
LO
in the process.
golden rule on the envelope wavefunction. Almost all the treatements neglect the valence
band part of the wavefunction and treat the material as being a bulk material.
The scattering rate is then written as:
w
LO
=
m
e
2
LO
2
2
f
_
2
0
d
I
ij
(Q)
Q
(5.2.4)
where the eective dielectric constant entering the Frolich Hamiltonian is
1
P
=
1
+
1
s
, (5.2.5)
where
1
and
1
s
are the mid-infrared and the static dielectric function, respectively.
The form factor entering in Equ. 5.2.4 is computed using the envelope wavefunctions:
I
ij
(Q) =
_
dz
_
dz
i
(z)
j
(z)e
Q|zz
i
(z
)
j
(z
) (5.2.6)
The intersubband lifetime due to optical phonon emission, computed using Equ. 5.2.4, with
zero initial wavevector, is plotted as a function transition energy for a square single quantum
well in Fig. 5.4. As shown in Fig. 5.4, a strong reduction of the lifetime is predicted when
Figure 5.4: Computed electron lifetime (left vertical axis) due to optical phonon scattering
as a function of transition energy in a square quantum well, whose width is indicated in the
rigth vertical axis. Zero initial kinetic energy in the upper subband, and zero temperature
was assumed. The lled squares are experimental measurements, as indicated.
the two states are spaced resonantly with the optical phonon. Technically, this increase
in scattering rate appears when Q approaches zero at resonance. Note however that the
model used here, based on Fermis golden rule, is not able to compute this value exactly
at resonance. The global trend shown in Fig. 5.4 is rather robust, but will only hold for
small ratio of the in-plane kinetic energy to the intersubband energy. The computed lifetime
5.2. SCATTERING RATE BY BULK OPTICAL PHONONS 49
Figure 5.5: Computed electron lifetime due to optical phonon scattering as a function of the
initial kinetic energy. A square, 220

Athick InGaAs/AlInAs quantum well was considered.
The equivalent electron temperature is shown in the top horizontal axis
as a function of in-plane kinetic energy is shown in Fig. 5.5 for a square InGaAs/AlInAs,
220

Athick quantum well. The strong increase of lifetime with initial energy can easily be
understood by a study of Eqn. 5.2.4 and Fig. 5.2. As the electron kinetic energy is increased,
the average value of Q is increased as the integral in Eqn 5.2.4 runs on the circle of the
available nal k
f
values. This increase in lifetime should be kept in mind when estimating
population inversion in quantum cascade structures, as the electrons will exhibit a very large
kinetic energy after the rst optical phonon emission.
5.2.1 Temperature dependence
The above computation was done for spontaneous emission of optical phonon. When consid-
ering a system at nite temperature, absorption and stimulated emission of optical phonon
should be considered along with the spontanous emission factor. The phonon population
n
LO
is given by the Bose-Einstein factor:
n
LO
=
1
exp(
LO
/kT) 1
(5.2.7)
The total scattering rate is then obtained by separately adding the emission and absorption
processes:
w
tot
= (1 +n
LO
) w
(em)
LO
+n
LO
w
(abs)
LO
(5.2.8)
where w
(em)
LO
and w
(abs)
LO
are the optical phonon emission and absorption rate computed using
Eq. 5.2.4. As the phonon energy is larger than kT even at room temperature, the resulting
dependence of the scattering time is relatively weak, as the lifetime decreases by about a
factor of two when the temperature increases from zero to 400K.
50 100 150 200 250 300 350 400
Temperature K
0.2
0.4
0.6
0.8
1
L
i
f
e
t
i
m
e
p
s
Lifetime of a 10nm QW
Figure 5.6: Computed temperature dependence of the intersubband lifetime for a 100A thick
InGaAs/AlInAs quantum well
5.3 Quasi-elastic intersubband process
As shown in the preceding paragraphs, optical phonon scattering is the dominant scattering
mechansism between subbands separated by more than an optical phonon energy. Moreover,
it is a scattering process that is relatively easy to compute. In contrast, when the energy
spacing between subbands is reduced below the optical phonon energy, scattering by the
latter may be forbidden depending on the temperature and the electron densities and the
lifetime is now controlled by a competition between optical phonons and a number of other
processes, as shown in Fig. ??. Typically, while optical phonon scattering by electrons
thermally excited will always dominate at high temperatures (T > 60K), electron-electron
scattering or interface scattering will be the strongest scattering mechanisms, depending on
the structure and on the electron density. Only for extremely low densities, and for very
clean systems, will the acoustic phonon scattering be of any signicance.
5.3.1 Acoustic phonons
The computation of the acoustic phonon scattering rate is usually made using the same
approximations (one sound velocity c
s
, bulk phonons) as in the optical phonon case [21, 22].
The computation is however more cumbersome as the one has to integrate over the possible
values of the phonon wavevector, assuming conservation of energy and in-plane momentum
5.3. QUASI-ELASTIC INTERSUBBAND PROCESS 51
50
40
30
20
10
0
E
n
e
r
g
y

[
m
e
V
]
0
Wave vector
Figure 5.7: Intersubband relaxations processes in a quantum well with an energy separation
smaller than the optical phonon energy E
21
<
LO
. In this conguration, various processes
may play a role: optical phonon emission (op), electron-electron scattering (EE), ionized-
impurity scattering or interface roughness scattering (II), acoustic-phonon emission (AC).
for the scattering process between the intial k
i
and nal k
f
wavector of the electron. The
momentum of the phonon perpendicular to the interfaces is q
||
. The angle between the initial
and nal wavevector is . The total scattering rate for the phonon emission w
if
= 1/
if
writes
then [22]
w
if
=
c
0
4
2
dq
||
[f(q
||
[
2
_
2
0
d
_

0
dk
f
k
f
(1 +n
ac
)(E
i
E
f
+ (
2
/2m
)(k
2
i
k
2
f
))
(5.3.9)
where the energy of the phonon is obtained from the phonon wavevector q
||
= c
s
_
q
2
||
+[k
i
k
f
[
2
, (5.3.10)
c
0
= D
2
/2c
2
s
, (5.3.11)
and the form factor is computed
f(q
||
) =
_

i
(z)e
iq
||
z
f
(z)dz. (5.3.12)
D is the deformation potential for electrons, the density and c
s
the longitudinal velocity
of sound (8.6eV ,5.3g/cm
3
and 3700m/s for GaAs respectively).
As in the case of the optical phonon, n
ac
is the Bose-Einstein factor for the emission of
acoustic phonons, given by
n
ac
=
1
exp(/kT) 1
. (5.3.13)
The expression for the scattering rate for the absorption of acoustic phonon is very similar
to Equ. 5.3.9 replacing the term (1 + n
ac
) n
ac
and by changing the sign of the optical
phonon energy . The total scattering rate is then the sum of the scattering
rates for emission and absorption. The computation is more cumbersome because the Bose-
Einstein factor enters directly in the integral that must then be evaluated at each temperature
individually.
The lifetime for the emission of an acoustic phonon increases with quantum well thickness,
from a value of 80ps for a 8nm thick quantum well (E
21
= 112meV ) to a value of 240ps for
a 20nm thick quantum well (E
21
= 29meV ) in the GaAs-Al
0.3
Ga
0.7
As material system [22].
The lifetime increases further as the well width is increased, to a value of 400ps for a 28nm
(E
21
= 16meV ).
These long lifetimes have been very dicult to conrm experimentally. The main reason is
that it is dicult to design an experiment that will measure the lifetime of a single electron
in an excited subband and not the overall lifetime of a population. As is shown further in
the text, electron-electron scattering will shorten the lifetime as the electron density is above
a few 10
7
cm
2
5.3.2 Electron-electron scattering
Electron-electron (EE) is a process in which two electrons exchange energy and momentum.
This process may also be accompanied by a change of the subband index, and is usually
labeled by the index number of the intial and nal electron states, i.e. process klmn is an
interaction with carrier 1 scattering from sub-band k to m and carrier 2 from state l to
n. The various processes are shown schematically in Fig. 5.3.2. An important feature of
Figure 5.8: Various electron-electron scattering processes. The 2, 2 1, 1 is the dominant
electron-electron scattering term, as the Auger-like term 2, 2 2, 1 was shown to vanish
in symmetric quantum wells. The 2, 1 2, 1 will lead to thermalization between subband
while the 2, 2 2, 2 or 1, 1 1, 1 will tend to thermalize the subband themselves.
electron-electron scattering is the fact that total energy and momentum are conserved, and
as a result this process can only thermalize a carrier population but not cool it.
5.3. QUASI-ELASTIC INTERSUBBAND PROCESS 53
While process 1212 and 2121 have no inuence on the overall population of the subbands,
processes 2211 or 2221 are responsible for decreasing the upper state population. As a
summary, the inuence of the variou process is
2222 or 1111 (scattering within a single subband) will help establish a thermal equi-
librium within a single subband. In particular, for quantum cascade lasers with a large
transition energy (200meV) where electrons should emit a large number of optical
phonon before reaching the bottom of the band, the strength of this process will de-
termine wether the electron distribution will be a set of phonon replica or a thermal
one.
process 2121 or 1212 will tend to equilibrate the temperatures of the two subbands.
Again, for large transition energies, this process will tend to heat the upper state
population, since the lower one is heated by the large energy loss of the electrons.
process 2211 or 2221 will transfer electrons from the upper to the lower subbands. As
shown by the computations presented further in the text, this process are important
for low tansitions energies and medium densities.
A theoretical treatement of electron-electron scattering between subbands was performed [23]
in the limit of a innite square well potential. Unfortunatly, they focused their work on the
process 2221 that was shown later to be forbidden by parity in a symmetric structure [2].
Smet, Fonstad and Hu gave a thourough treatement of electron-electron scattering and
evaluated its importance as a function of temperature and intersubband transition energy [2].
They demonstrated that the process 2211 is the dominant scattering mechanism. Central to
the computation of the electron-electron scattering rate is the form factor for the transition:
A
i,jf,g
(q
) =
_

dz
_

dz
i
(z)
j
(z)e
q
|zz
f
(z
)
j
(z
) (5.3.14)
where q
= [
k
i
k
f
[ is the wavevector exchanged during the process. The transition rate
for electron-electron scattering can then be evaluated using Fermis Golden Rule and yields
the following expression:
W
i,jf,g
(
k
i
) =
e
4
2
2
r
2
0
_
d
2
k
j
_
d
2
k
f
_
d
2
k
g
[A
i,jf,g
(q
)[
2
2
sc
(q
, T)q
2
f
j
(
k
j
)(1 f
f
(
k
f
))(1 f
g
(
k
g
))
(E
f
(
k
f
) +E
g
(
k
g
) E
i
(
k
i
) E
g
(
k
g
))(
k
f
+
k
g
k
i
k
j
),
(5.3.15)
where
2
sc
is the screened dielectric function, f
i
(
k
i
) the Fermi distribution. Even after reduc-
tion of Equ. 5.3.15 using the delta functions, the total rate will require a three dimentional
integral for each initial wavevector

k
i
, causing a non-negligible computational burden. Com-
putation of the scattering rate as a function of transition energy and density is shown in
Fig. 5.3.2 In the same work, the temperature dependence of the electron-electron scattering
rate was also evaluated. The latter is virtually independent in temperature for reasonable
Figure 5.9: The electron-electron scattering time for the 2, 2 1, 1 process presented as a
function of well width for an innite quantum well at zero temperature for dierent screening
models: no screening, constant screening length, static single subband screening for and
ideal 2D system with no z-dependence (i.e. [A
2,21,1
[ = 1), static single subband screening
including q
- dependence form factor. The Fermi energy of the excited state E

f2
is equal to
10meV and the initial electron state wave vector

k
i
is equal to zero. Adapted from Ref [2],
with the correction suggested by [?], with permission
physical parameters. The reason for this behavior is because the integral 5.3.15 is domi-
nated by contributions in which there is no exchange of energy between the electrons and
for which therefore
q
elastic
=
_
2m
E
21
2
. (5.3.16)
Quantum cascade lasers operate at very low densities, even in the Terahertz region, satisfying
the condition that E
f2
<< E
21
. As a consequence, the Fermi factor appearing in Eq. 5.3.15
will not alter the result signicantly. An other consequence of this fact is that the strength of
the electron- electron scattering can be compared between structures by a mere computation
of the form factor 5.3.14 at q
elastic
.
A computation of the e-e scattering rate according to Eq. 5.3.15 is shown in Fig. 5.3.2 as a
function of transition energy, for assuming various screening approximations. The vertical
axis for the scattering rate was rescaled by a factor of four compared to the published
data, following a suggestion by a subsequent work of the same group ??. In this gure,
the initial Fermi energy is assumed to be E
f2
= 10meV , corresponding to a population
n
s
= 2.6 10
11
cm
3
. Under this assumption, the computed scattering rate for a 28nm
thick quantum well, corresponding to a transition energy of 15meV, is 8ps. The upper state
lifetime is half of this value, as each scattering event will promote two electrons to the lower
subband. However, as will be shown later in this text, the actual density of the upper state
is closer to n
s
= 2 10
9
cm
3
, a value about 100 times smaller.
Hyldgaard and Wilkins [24] gave a simplied expression for the electron-electron scattering
rate that, in addition of giving an order of magnitude estimate for the lifetime, has the
advantage of showing the dependence of the scattering rate on the physical parameters. In
5.4. COMPARISON WITH EXPERIMENTS 55
their expression, the scattering rate
ee
= /
ee
between two subband separated by an energy
E
21
is given by:
ee
=

ee
=
Ry
2
E
21
[U
0
(q
E
21
)[
2
I
P
(0) (5.3.17)
where U
0
(q
E
21
) is the unscreened interaction matrix element and I
P
(0) 0.785 a dimen-
sionless integrated phase space factor.
2
is the chemical potential of the upper subband
and is equal to
2
=

2
m
n
2
at low temperatures, with n
2
the upper state sheet density.
The key result is the linear dependence of the scattering rate in the electron density, as
well as its inverse dependence on the intersubband transition energy. Numerically, for a
coupled well system, and assuming a Fermi energy of E
f2
= 5meV , their computed lifetime
is = 0.94ps for a transition energy of about 10.6meV. Assuming the linearity between
scattering rate and density, this would extrapolate to a scattering rate of about 45ps for
a density of n
s
= 2 10
9
cm
3
. This value is not unreasonable for a square well at low
temperatures.
From this computations, it is quite clear that electron-electron scattering will play a signi-
cant role only for very clean structures at low temperatures.
5.3.3 Impurity scattering
Elastic scattering between subband may occur whenever the translational invariance of the
subband within the plane of the quantum well is lost. This may occur either through collision
with an ionized impurity or through scattering at an interface step.
Ionised impurity scattering is the strongest scattering mechanism limiting the low-temperature
mobility even in clean samples. Its inuence on the intersubband lifetime was evaluated by
Ferreira and Bastard [22]. It was found that a signicant scattering rate (compared to the
other scattering mechanism) is only found for relatively large doping levels.
5.4 Comparison with experiments
The measurement of the intersubband scattering time
ISB
a is a very challenging task. It
took a large number of experiments, spread over more than a decade, to reach a general
consensus on its value. Althought the rough estimate of the scattering rate is not in ques-
tion (
ISB
< 1ps for transitions larger than the optical phonon, about
ISB
10ps at low
temperature below the optical phonon energy) there is still a lack of systematic measurement
of its dependence with temperature, well width, electron density, and materials. The di-
culty arises from the combination of short lifetime, strong dependence on the experimental
conditions (temperature, density) and on the sample quality, especially for the low transition
energies.
The intersubband lifetime was measured by a number of techniques that one may separate
according to the dierent excitation and detection mechanisms used. As shown schematically
in Fig. 5.4.5, these are
Interband pump and probe
Intersubband pump and probe
Intersubband saturation experiments
Intersubband electroluminescence
Interband pump-probe Intersubband pump-probe
Intersubband saturation Electroluminescence
Figure 5.10: Techniques for the measurement of the intersubband lifetime.
Understanding the limitations of each of these techniques is very important when comparing
the experimental results.
5.4.1 Interband pump and probe
In this approach, an initial electron density is created in the excited state by rst pump pulse.
The population of the upper and lower levels is then probed by by a weaker probe pulse that
follows with a variable delay. In general, this technique presents two key diculties:
Interband pumping on the second excited state will always create a concommitant
population on the rst subband at the same energy. This population will the interact
with the upper state population.
Holes are also created and may aect the intersubband lifetime.
The most popular optical source, the Ti-saphire mode-locked laser, has a repetition
rate (about 80MHz) such that the population does not have enought time to recom-
bine totally across the gap between successive pulses. A parasitic population of the
lower state is then induced. A possibility to prevent this problem is to use p-doped
samples that exhibit much shorter electron radiative lifetimes (however, p-doping will
add additional scattering channels), or to sample the pulses with a lower repetition
rate.
Compared with the early results, the generalized use of high sensitivity detectors has allowed
to greatly decrease the pump and probe pulse energies.
5.4.2 Intersubband pump probe
In this technique, a population is created in the upper state by a pump photon tuned at the
energy of the transition. The decay of the upper state population is then measured by either
a measure of the recovery of the fundamental absorption, or by the decay of the absorption
from the excited state. This technique is free from many of the problem cited above and did
yield the most reliable results. It is, however, a technique that is not easy to implement as
it requires the generation and detection of short pulses in the mid- or far-infrared. This was
accomplished by either downconversion from the visible pulses or by direct generation by a
free electron laser.
5.4.3 Intersubband saturation experiments
A simpler technique consists in measuring the intensity dependent absorption. In a simple
two-level approximation, the latter will write as
(I) =
(0)
1 +
I
Isat
. (5.4.18)
Fitting the data will yield the saturation intensity I
sat
, that in turns will yield the lifetime
T
1
:
I
sat
=

2
c
0
n
refr
2q
2
0
z
2
12
T
1
T
2
(5.4.19)
where a Lorenzian line with dephasing time T
2
was assumed. The latter is usually tted from
the measurement of the lineshape for small pumping intensities. As shown by the preceding
equation, a drawback of this technique, besides the strong pump powers that it requires,
is the requirement to correctly evaluate the power density inside the sample as well as a
knowledge of the dipole matrix element z
12
.
5.4.4 Intersubband electroluminescence
Pump and probe as well as saturation measurements require relatively strong pump inten-
sities. An other drawback is that the measurement will be performed in a model system,
usually a single or coupled quantum well. In contrast, quantum cascade lasers will operate
in the regime of very low upper state densities (in general with a sheet density of about
n
s
10
9
cm
2
) and with an active region consisting of a complex arrangements of coupled
wells. For this reasons, a measurement of the intersubband luminescence intensity, although
by itself a very indirect measurement of the upper state lifetime, did prove itself quite valu-
able in some occasions. In particular, it is especially valuable to measure the variation of the
lifetime with an external parameter such as the electron density or temperature. As shown
further in the text, measurements of intersubband electroluminescence in vertical transitions
structures allowed an experimental verication of the relative strength of the scattering mech-
anisms. In particular, the square-root dependence of the electroluminescence intensity with
injected current is a strong indication that, in these structures, electron-electron scattering
is indeed the dominant scattering mechanism at low temperatures. In addition, the temper-
ature dependence of the luminescence intensity is consistent with optical phonon scattering
by thermally excited electrons being the dominant non-radiative scattering process at higher
temperatures. Same results were not observed as clearly in more complex active regions
based on chirped superlattices or bound-to-continuum transitions. In particular, the lumi-
nescence intensity as a function of current was found to be linear, supporting the hypothesis
that interface roughness scattering may dominate electron-electron scattering even at low
temperatures.
5.4.5 Experimental results
Because of an threshold for the onset of optical phonon emission, it was expected that a very
strong contrast would be observed when measuring intersubband lifetime below and above
the optical phonon energy. Indeed, the rst measurements were the time-resolved Raman
experiments of Oberli et al. [25] yielded, for a transition energy slightly below the optical
phonon energy (28meV) a lifetime of about 300ps; while the intersubband pump and probe
measurements of Seilmeier did yield a lifetime of 10ps for a transition energy of 150meV,
clearly above the optical phonon emission threshold. However, it was found out that both
measurements did not yield the true intersubband lifetime but rather the global relaxation
time of the hot electron plasma distributed thermally across the two transitions. When large
populations are excited, because of Pauli blocking, the electrons could not scatter back to
the ground subband because of the lack of available states.
In fact, time-resolved Raman experiments by Thatam et al ?? that yield an intersubband
lifetime of < 1ps for a 14.6nm thick GaAs active quantum well. Later measurements by
various technique, performed at low excitation levels agreed with the values computed using
optical phonon emission, as shown in Fig. 5.4. These values are summarized in Table 5.4.5.
In contrast, the lifetime for transition energies below the optical phonon energy, if limited by
the emission of acoustic phonons, would lead to very long lifetimes of the order of more than
100ps, raising the hope of intersubband lasers with very low threshold current densities.
Indeed, subsequent work [?, ?, ?] found scattering times spread from 20-750ps. Thanks to
the improvement of the detection techniques, the detection of the luminescence signal was
possible even for extremely low excitation densities. In a series of experiments, Hartig et
al. showed that the intersubband lifetime rst decreases with decreasing excitation density,
as Pauli blocking eects are reduced. The shortest value measured of 5ps, limited by the
temporal resolution of their experiment, is achieved for densities around 10
11
cm
2
. As shown
in Fig. ??, the lifetime then increases again when the density is further lowered and saturates
Mat E
12
(meV) (ps) Technique ref Year
GaAs SQW 150 10 ISB pump probe [26] 1987
GaAs SQW 150 0.9 ISB Saturation [?] 1988
GaAs SQW 53 < 1 IB Time Resolved Raman [27] 1989
GaAs SQW 125 0.65 IB luminescence and transport [?] 1990
GaAs SQW 112 0.65 ISB cw pump probe [28] 1993
InGaAs SQW 129 0.8 ISB cw pump probe [28] 1993
InGaAs SQW 200 1 ISB/IB pump probe [?] 1996
InGaAs SQW 260 1.3 ISB/IB pump probe [29] 1996
Table 5.1: Summary of lifetime measurements for transition energies above the optical
phonon energy in GaAs quantum wells. Early measurements were inaccurate due to the
strong pumping needed. ISB: intersubband, IB: interband. Only low temperature measure-
ments (T=10K) are reported
E
12
(meV) (ps) Technique Ref Year
27 325 Time-resolved PL [25] 1987
18 40 ISB pump probe [?] 1994
15 300 c.w. ISB pump probe [28] 1994
15 1200 Intersubband pump probe [?] 1995
15 30 Interband pump probe [30] 1998
Table 5.2: Summary of some lifetime measurements for transition energies below the optical
phonon energy
at about 30ps. These measurements were however performed in a coupled well system for
technical reasons.
An other approach has been intersubband measurements with quasi-c.w. excitation with a
CO
2
[28] or free electron laser [?]; detection was either by intersubband absorption [28] or
capacitive measurements [?]. Both groups used modulation-doped coupled quantum wells.
In the low temperature, low excitation limit, the measured values of lifetime range from
= 300ps [28] to 1ns [?].
M. Hartig et al., PRB 60(3) 1999
(Al)GaAs: /210/40/90/100/210/40/90/
ISBS due to carrier-carrier interaction
Figure 5.11: Lifetime as a function of pumping density.
Chapter 6
Detectors
6.1 Fundamentals of infrared detection: noise, BLIB..
6.2 Quantum Well Infrared Photoconductor (QWIP)
61
62 CHAPTER 6. DETECTORS
Chapter 7
Mid-infrared waveguides
7.1 Dielectric waveguides
7.2 Dispersion of doped layers, Drude approximation
7.3 Metal-based waveguides for the Terahertz
7.4 Far-eld computation
63
64 CHAPTER 7. MID-INFRARED WAVEGUIDES
Chapter 8
Quantum Cascade lasers I:
fundamentals
It is clear from the preceding chapters that a laser based on intersubband transitions will
benet from many interesting features of the this system. First of all, as for the detector, it
is based on an optical transition that both exhibits a strong oscillator strength (f 20 30
for common GaAs or InGaAs materials) that can be tuned over a wide range of frequencies
in the mid and far-infrared. However, it is also clear that such a laser has to overcome a
number of inherent challenges, the most serious one being the very short lifetime of the upper
state transition. Let us emphasize that, in contrast to the case of a light emitting diode,
such a short lifetime will not an inherent limitation to the eciency of such a device, it will
severely limit the available gain at a given pumping power density.
8.1 Historical perspective
The idea to use intersubband transitions to achieve optical gain traces back to the original
work of Kasarinov and Suris in the early seventies. In this paper, the examined the optical
response of a superlattice under the application a strong electric eld, such that the ground
state of one well would be situated just below the second excited state of the quantum well
downstream. The mechanism for population inversion and gain is fairly easy to understand:
the transition between the ground state of one well 1 to the excited state 2 of the well
downstream must proceed through a tunnel barrier. The strength of this interaction can
therefore be reduced by an arbitrary amount by just increasing the thickness of the tunnel
barrier. In contrast, the scattering between level belonging to the same quantum well, such
as the one occuring between the states n=2 and n=1 is expected to be much more ecient.
In such a system, population inversion is insured regardless of temperatue for any homogeous
electron distribution since the ground state of each well is also the upper state of the laser
transition.
One of the key problem that so far prevented this proposal to be demonstrated experimentally
is that the situation they described could hardly be obtained experimentally, because it would
mean operating the device in a unstable eld region of negative dierential resistance. In
65
66 CHAPTER 8. QUANTUM CASCADE LASERS I: FUNDAMENTALS
h
h
n
n+2
n+1
d
eFd
E
1
E
2
E
3
Figure 8.1: Schematic band structure of the device considered by R. Kazarinov and R. Suris.
A superlattice is under a suciently strong electric eld such that the ground state of a well
is above the rst excited state of the following well
fact, as shown very experimentally by K.K. Choi [31], a doped superlattice will break into
eld domains with, in each domain, energy states aligned in resonance. This conguration is
more stable because it enables a larger current to ow through the structure. For example,
in the case studied by Kazarinov and Suris, the real band aligement for a realistic structure
would be really that a part of the structure will exhibit a ground state of each well aligned
with the rst excited state of the following well, and the next part of the structure where
the ground state of each well is aligned with the second excited state.
8.2 Active region: fundamental concepts
8.2.1 Injection/relaxation region
The design of a successful intersubband laser had to simultaneously solve a number of prob-
lems:
The structure should be stable electrically
Population inversion should be engineered
Large enough gain should be achieved in the structure
To this end, an architecture was chosen in which the active structure does not anymore
consists of a simple periodic arrangement of quantum wells but consists now in a periodic
arrangement of a unit cell, each one consisting of a complex heterostructure potential con-
sisting of a number of well and barrier. As shown in Fig. ??, roughly speaking, each cell
can be divided in an gain region and an injection/relaxation region. The gain region is the
structure that will create and maintain the population inversion. As we will show, this active
region may operate following various designs. In general, the active region contains a ladder
of at least three states (or continuum of states), such that electrons are injected in the n=3
state and population inversion is maintained between the states n=3 and n=2. Assuming
8.2. ACTIVE REGION: FUNDAMENTAL CONCEPTS 67
3
2
1
32
32
>
2
Active region Relaxation/Injection
a)
b)
one period
Figure 8.2: a) Schematic conduction band diagram of a quantum cascade laser.
Each stage of the structure consists of an active region and a relaxation/injection
region. Electrons can emit up to one photon per stage. b) General philosophy
of the design. The active region is a three-level system. The lifetime of the
3 2 transition has to be longer than the lifetime of level 2 to obtain population
inversion.
that electrons are exclusively injected in the n=3 state, the population inversion requirement
is simply that
32
>
2
, (8.2.1)
i.e. the total lower state lifetime
2
is shorter than the time to inject electrons from the n=3
to the n=2 levels.
Following the gain region is the injection/relaxation region. The main purpose of this region
is to raise the energy of the electron compared to the band edge to enable injection by
resonant tunneling in the next period period. This function is conviniently achieved by
a sequence of alternating quantum wells and barriers with changing duty cycle, create an
eective graded gap that compensate the applied electric eld. a portion of the injection
region is also doped, providing the electron charge needed for the transport. This region also
provides an electron reservoir that will feed the carriers to the next period and provides
also a region where electrons can thermalize with the lattice. Finally, it is doped to act as
an electron reservoir, insuring that the integrated negative charge carried by the electrons is
compensated by the positive donors, even in the situation of strong injection, to prevent the
formation of space-charge domains. This is in contrast to the gain region that is left undoped
because the presence of dopants signicantly broadens the lasing transition by introducing
a tail of impurity states[32].
In conclusion, the presence of this injection/relaxation region will
prevent the formation of electrical domain,
reduce the applied electric eld by lengthening the period,
block electron escape from the upper state of the laser transition, by a suitable engi-
neering.
8.2.2 Cascading
A very attractive feature of lasers based on intersubband transitions is the fact that cascading
the active region periods is easy and almost come naturally. In a cascade laser, the gain
region consists of a large (typically N
p
= 10 100) number of repeat periods. Electrons are
then recycled from period to period as they cascade down the structure. Cascading has
two main advantages. Firstly, by increasing the size of the region in which gain occurs, it
will decrease the population density requirement for each individual active region period.
This allows a reduction of the threshold current density of the device. In addition, a single
electron is able to potentially emit a number N
p
of photons, increasing the slope eciency.
As shown in g. 8.3, both eects are obtained at the cost of a larger applied bias. In essence,
V = 0
qV ~ N (h + 3kT)
Figure 8.3: Highly simplied picture of the eective potential in a quantum cascade laser.
At zero eld, the current is blocked by the injection regions that prevent the current from
owing. With the application of a strong electric eld, these injection regions are attened
by the electric eld and the current ows along the electronic cascade, generating the photons.
in a cascade laser, one is exchanging a lower operating current for a larger operating voltage.
This enables a large reduction of the devices ohmic losses, especially in the mid- or far-
infrared were the photon energy is very small. In fact, as the slope eciency grows with
8.3. INTERSUBBAND VERSUS INTERBAND LASERS 69
N
p
, the device operating current for a given power is proportional to N
1
p
. The ohmic losses
(= RI
2
) for this device therefore decrease with the square of the number of period.
This advantage is of course valid for any semiconductor laser. The problem in applying such
an idea to interband near-infrared structures is the necessity of inserting a Zener tunnel
junction in the structure to enable re-injection of carriers across the gap. Such a tunnel
junction must be heavily doped, and is therefore usually responsible for large optical losses.
An elegant way around this problem is to work in a vertical cavity semiconductor laser, and
place the Zener junction at a node of the optical eld.
8.3 Intersubband versus interband lasers
Intersubband and interband lasers are vastly dierent devices. In an interband laser,
The population inversion is obtained naturally through a physical property of the
material (its energy gap).
A certain amount of current must be injected to reach material transparency. An
essential element of the optimization is therefore to minimze this transparency current.
The maximal current carried by the device is limited by either heating or catastrophic
damage on the facet.
In contrast, in an intersubband laser:
The population inversion must be established by a suitable design of the active region.
As a general rule there is no transparency current.
The device reaches a maximum current given by the doping of the device and and an
alignment of the band structure.
As a result, the optimization of the active region in both cases are very dierent. For
interband lasers, the reduction of the transparency current has been achieved through a
reduction of the active volume by conning the electron-hole pairs in quantum wells or even
quantum boxes. As a result, the transparency current fell from more than 10kA/cm
2
for a
p-n junction diode to 1 3kA/cm
2
for a double heterostructure active region to less than
100A/cm
2
for strained InGaAs quantum well lasers.
In contrast, in a quantum cascade laser, the population inversion must be maximized by
enginnering the lifetimes and oscillator strengths. In fact, mimimum threshold current (or
even power, taking into account the applied bias) is obtained for relatively large active regions
with 25-50 periods, as there are no direct penalty for increasing the active region thickness.
8.4 Rate equation analysis, threshold condition, slope
eciency
A key understanding of the requirements for the design of a QCL can be understood by
simply considering a simple atomic level analysis of the active region. For historical reasons,
the gain region is assumed to consist of three states, with laser action occurring between levels
n=3 and n=2, whereas the injector is assumed to consist only of one level, with a constant
population n
g
, aligned with the upper level of the next period. A schematic diagram of the
process connecting these three dierent states is shown in Fig. 8.4. Electrons are injected in
3
2
1
esc
31
32
J/q
injector
injector
Figure 8.4: Population inversion between the two upper levels of a ladder of three states.
The relevant non-radiative process are also indicated.
the n = 3 state from the ground state of the injector of the previous period at a rate equals
to J/q
0
. Electrons may scatter from this state to the lower n = 2 and n = 1 states with rates
1
32
and
1
31
respectively. The n = 3 lifetime, given by the relation
1
3
=
1
32
+
1
31
+
esc
may include an escape process into the continuum with a lifetime
esc
. We also assume that
electrons may be thermally activated to state n = 2 to an equilibrium population n
therm
2
. The
latter term may be either approximated by an activated behavior n
therm
2
= n
g
exp(/kT),
where n
g
is the sheet doping density of the injector and the energy dierence between the
Fermi level of the injector and the n = 2 state of the laser transition. Therefore, the rate of
change of the population of the three levels may be written as:
dn
3
dt
=
J
q
0
n
3
3
Sg
c
(n
3
n
2
) (8.4.2)
dn
2
dt
=
n
3
32
+Sg
c
(n
3
n
2
)
n
2
n
therm
2
2
(8.4.3)
dS
dt
=
c
n
refr
__
g
c
(n
3
n
2
)
tot
_
S +
n
3
sp
_
(8.4.4)
In equations 8.4.4,
tot
is the total modal loss (sum of the waveguide and mirror losses), the
fraction of the spontaneous emission coupled in the lasing mode, and
spon
the spontaneous
emission lifetime.
In considering the set of equations 8.4.4, it should be noted that the electron n and photon
S densities do not have the same units. In this work, we choose to dene the n
i
as sheet
densities per period and S the photon ux per period and unit active region width. However,
it should be remembered that these equations only represent one period of the active region,
8.5. OPTIMIZATION OF THE ACTIVE REGION: INTERSUBBAND TOOLBOX 71
whereas a quantum cascade laser will have many period sharing the same optical waveguide.
Usually, one considers the gain cross section g
c
appearing in Eq. 8.4.4 as the one of all the
periods in the active region put together, i.e. is proportional to the total overlap factor :
g
c
=
4q
2
0
n
refr
z
2
32
2
32
L
p
(8.4.5)
where z
32
is the dipole matrix element,
32
is the broadening of the transition, the wave-
length and L
p
the length of the period. If this convention is chosen, it should be noted that
the photon ux S is the one per period, not the total one for the whole active region.
The behavior below threshold is obtained by setting the derivatives and S to zero. In this
approach, the amplied spontaneous emission is neglected because
spon
>>
3
and only
becomes signicant extremely close to the threshold current. From the rst equation, we
get a relation between the upper state population and the electrical pumping: n
3
= J
3
/e.
Making use of the second, one obtains for the population inversion n = n
3
n
2
,
n =
J
3
q
0
_
1

2
32
_
n
therm
2
=
J
e
q
0
n
therm
2
(8.4.6)
where the eective lifetime
e
=
3
(1
2
/
32
) relates the population inversion to the
electrical pumping. The eective lifetime converges to the upper state lifetime for states
displaying a large ratio of lifetimes
2
/
32
<< 1. Note also that the population inversion can
only exist if
e
> 0, i.e. if
2
<
32
.
The threshold current density is reached when the modal gain g
c
n compensates for the
losses
tot
,
J
th
= q
0
tot
/g
c
+n
therm
2
e
(8.4.7)
or, expressing the gain cross section explicitly:
J
th
=
1
e
_
0
nL
p
(2
32
)
4q
0
z
2
32
(
tot
) +q
0
n
therm
2
_
. (8.4.8)
Above threshold, the gain is clamped and S increases linearly. By again setting the time
derivatives to zero and derivating by J, one can compute the slope eciency:
dP
dI
= N
p
h
m,1
dS
dJ
=
N
p
h
q
0
m,1
tot
e
+
2
(8.4.9)
where
m,1
is the front mirror loss. The behavior of S and n as a function of injected
current is shown schematically in Fig. 8.5.
8.5 Optimization of the active region: intersubband
toolbox
An analysis of Eq. 8.4.8 and 8.4.9 leads to the following conclusion: a low -threshold, high-
eciency laser will be obtained in a structure that achieves
Figure 8.5: Schematic illustration of the behavior of the population inversion and
the photon ux with injected current.
a large ratio of upper-state to lower-state lifetimes
32
/
2
,
a low waveguide loss
w
,
Narrow transition linewidth
32
,
long upper state lifetime
3
.
The multiple QC designs that exist simply reect the complicated trade-o that have to
be made when trying to fulll these conicting requirements. The required engineering of
the lifetimes and dipoles is achieved using an intersubband toolbox, whose ingredients are
combined to construct the various active region architectures.
8.5.1 Tunneling
A very interesting situation occurs when, as shown in Fig. 8.6, the initial and nal states
belong to two dierent quantum well coupled by a tunnel barrier. These states are localized
in their respective wells when the coupling energy
1
1
<< E
1
E
1
between the initial
state 1
and nal state 1 is smaller than the energy dierence between the uncoupled states
[
0
1
) and [
0
1
). In this situation, any scattering between the two state will contain the
coupling energy
11
. The latter is proportional to the tunneling probability exp(kL
b
)
where k =
_
2m
(V E)/ is the electrons wavevector L

b
is the barriers thickness. As
a result, the scattering rate an be lowered at will by simply increasing the tunnel barrier
thickness, regardless of the scattering mechanism.
This can easily be seen by applying rst order perturbation theory and computing the
scattering operator S between the perturbed initial [
1
) and nal states [
1
), taking for the
perturbed state
[
1
) = [
0
1
) +

1
1
E
1
E
1
[
0
1
) (8.5.10)
Figure 8.6: Two states coupled by a tunnel barrier. The lifetime of the upper state is
controlled by the thickness of the coupling barrier
and a similar expression for the perturbed ground state [
1
). The result yields:
1
[S[
1
) =

1
1
E
1
E
1
_
0
1
[S[
0
1
)
0
1
[S[
0
1
)
_
. (8.5.11)
The above expression is indeed directly proportional to
1
1
. In addition, as mentionned
already by Kazarinov and Suris in their seminal paper [33], it immediatly appears that a
scattering potential having the same matrix elements with both states 1 and 1 will experience
a destructive quantum interference, as the right hand side of Equ. 8.5.11 vanishes. Remote
impurities, that is, located far away from both wells, would induce such a potential. However,
most other scattering mechanisms, like optical phonons or interface defects, have typically
much shorter range and will act at least with a dierent phase on states [
0
1
) and [
0
1
).
The approximation used in equation 8.5.11 is valid as long as coupling to other quantum
states can be neglected. In most active regions however, the above expression must be
modied to take into account the fact that the upper state mainly couple to the excited state
of the downstream well. A very similar model may then be used based on a set of three
states with an upper state [
1
) coupled to the states of the following well.
[
1
) = [
0
1
) +

1
1
E
1
E
1
[
0
1
) +

1
2
E
1
E
2
[
0
2
) (8.5.12)
We will then assume that state 1 mostly couples to the excited state of the following well,
i.e. that
[

1
1
E
1
E
1
[ << [

1
2
E
1
E
2
[. (8.5.13)
The expression for the scattering matrix element of S is then even simpler:
1
[S[
1
) =

1
2
E
1
E
2
0
2
[S[
0
1
). (8.5.14)
This approximation justies the tunnel-assisted terminology used for the description of
the processes occuring in such coupled wells for the matrix elements between states 1 and
1 may be seen as the ones of a single quantum well multiplied by a tunneling coupling.
The upper state is a superposition of states 1 and 2. The scattering matrix element is
simply proportional to the amplitude componant of state [
2
) in the mixture of states. An
important limiting case is the one where states 1 and 2 are resonant, i.e. E
1
E
2
= 0. In
that case, the perturbation approach of equation 8.5.14 is not valid anymore; however, the
upper state wavefunction is composed at equal parts of state 1 and 2 with amplitude 1/
2.
As a consequence, the matrix elements will all be multiplied by the same factor.
Let us take as an example a system consisting of a 4nm and 1.5nm InGaAs well coupled by
an AlInAs barrier. The ground state of the two wells are separated by 230meV. As shown
in Fig. 8.7, both oscillator strength and intersubband lifetime depend strongly with the
tunnel barrier thickness. As shown in Fig. 8.7, the dependence expected by equation 8.5.14
0.1
1
10
100
0 20 40 60
0.01
0.1
1
10
()
O
s
c
i
l
l
a
t
o
r

s
t
r
e
n
g
t
h
L
b
L
b

~ exp(-2L
b
)
S
c
a
t
t
e
r
i
n
g

r
a
t
e

(
p
s
-
1
)
Figure 8.7: Photon-assisted tunneling transition. Plotted are the oscillator strength
f =
2m
0
2
z
2
E
ij
and optical phonon scattering rate as a function of barrier width for system
consisting of a 4nm and 1.5nm InGaAs well coupled by an AlInAs barrier. The additional
points are the two values for a 6nm thick single quantum well exhibiting the same transition
energy
with squared matrix elements varying exponentially in barrier thickness as exp(2L
b
) with
=
_
2m
(V E)/
2
is indeed observed in the limit of thick tunnel barriers.
As shown in Fig. 8.7, oscillator strength and intersubband rate have approximatly the same
dependence in barrier thickness. Since the gain cross section is proportional to the ratio
of these two numbers, the latter is not expected to greatly increase with barrier thickness.
However, a closer examination shows that, starting with values of a single quantum well,
the optical phonon scattering rate decreases reaches faster the asymptotic value. This eect
is shown in Fig. ?? where a computation of the product of the oscillator strength and the
lifetime is plotted as a function of the separation between the quantum wells for a xed
transition energy. This product increases from 18 to 35 as the diagonality of the transition
increases. This increase was one of the justication for the investigation of quantum cascade
lasers based on a photon-assisted tunneling transition [34].
0
10
20
30
40
50
0 20 40 60
L
b
(A)
f

(
p
s
)
L
b

Figure 8.8: Product of the oscillator strength and the lifetime as a function of tunnel barrier
thickness. This number is a gure of merit of the gain cross section. The value for L
b
= 0 is
the one computed for a single quantum well 6nm thick, as in Fig. ??.
8.5.2 Optical phonon
As was shown in the preceding chapter, for lasers operating in the mid-infrared, the lifetimes
are ultimatly limited by optical phonon emission, as the other parasitic processes such as
tunnel escape in the continuum can be designed away. A reduction of a lifetime may be
obtained by using a resonant optical phonon emission process. As shown in Chapt. ??, a
strong reduction of the intersubband lifetime is predicted when the transition approaches
the optical phonon value. Consider the structure shown in Fig. 5.4(a). In this three-level
system, the rst two states are separated by an optical phonon energy (34meV in InGaAs and
36meV in GaAs). Resonant optical phonon emission between these two states will reduce
the lifetime of the state n=2 to about 200fs. Intersubband transitions between the n=3 and
the n=2 states are less likely because they are non-resonant processes as soon as the emitted
photon energy is larger than the optical phonon. In other words, transitions between the
n=3 and the n=2 and n=1 states will proceed with a large wavevectors, whereas transitions
between the n=2 and n=1 proceed at nearly zero exchanged wavevectors.
8.5.3 Phase space
Finally, a third approach for modifying the intersubband lifetime is based on a phase space
argument. Consider a superlattice formed by the periodic alternance of wells and barriers.
As in its three-dimentional counterpart, this one-dimensional potential will create a set
Figure 8.9: Engineering a population inversion using an optical phonon resonance. In a
ladder of three states, where the two lower ones are spaced by an optical phonon energy,
the n=3 lifetime will naturally be longer than the n=2 because of the smaller exchanged
wavevector for the optical phonon emission between the n=2 and n=1 states.
of minibands and minigaps whose width can be adjusted by modifying the well a barrier
thicknesses, as shown in Figure 8.11. In an electrically pumped quantum cascade laser, the
carriers are injected in the lower state of the upper miniband, and population inversion and
optical gain is achieved at the edges of the minigap. Because of the nature of the electron-
Figure 8.10: Engineering the lifetimes. In a superlattice, a population inversion builds up
at the edges of the minigap because of the phase space for scattering out from upper state
of the lower miniband is much larger than the phase space to scatter in the same state. a)
real-space and (b) reciprocal space picture of the energy bands.
phonon interaction that depends on the exchanged wavevector, the intersubband lifetime in
a single quantum well will be almost identical (within 20%) to the one of a superlattice with
a minigap of identical size. For this reason, the scattering rate between levels at the edges
of the minigap will tend to decrease with the total number of wells in the superlattice, as
the total rate must be shared between more and more lower levels. In addition, the lower
state lifetime
2
is very short because electrons in the upper state of the lower miniband may
scatter to any point the the lower miniband very eciently.
Superlattice active regions do exhibit also an other advantage. In these structures, since the
optical transition occurs between excited states, its oscillator strength will be much larger
than the one taking place in a single quantum well with the same transition energy.
This behavior is illustrated in Fig. 8.11 where the product of the oscillator strength and the
upper state lifetime is plotted as a function of number of period for a superlattice consisting
of an alternance of 6nm thick InGaAs wells and 0.5nm thick AlInAs barriers. This gure of
merit increases from 22 to 50ps as the number of periods increases from 2 to 9. Similarely,
the ratio between the total upper state lifetime and the lifetime between the edges of the
minigap (called
3
and
32
by analogy to a three-level system) decreases from 0.5 to 0.11, close
to the value of 1/N expected for a model where the scattering rate is shared equally between
lower states. Fig. 8.11 show a continuous increase of the gure of merit f as a function of
0
20
40
60
2 4 6 8 10
0
0.2
0.4
0.6
0.8
1.0
f

(
p
s
)
3
2
N x
Number of periods
Figure 8.11: Product of the oscillator strength and the upper state lifetime (left) and branch-
ing ratio
3
/
32
(right) for a 60/5nm GaInAs/AlInAs superlattice with an number of period
N.
number of period. Actually, the concept of superlattice, strictly speaking, implies an innite
number of period. Two physical reason prevent, in practice, increasing the number of active
region period to a number larger than 8-10. First, as the number of period is increased, the
spacing of adjacent levels is decreasing, until the point where the states are closer than their
intrinsic broadening. Secontly, a related issue is the electrical injection that by denition
will be at an edge of the superlattice and that can only inject electron over a maximum of
a coherence length.
Intersubband lasers based on superlattice active regions have demonstrated high output
powers and high operating temperatures. Some of these phase space concepts may also be
used in structures based on discrete levels.
A very important feature of both the phase space argument and the tunnel eect do not
depend on the exact nature of the scattering mechanism. They will therefore also be ap-
plicable to other quantum cascade laser systems such as terahertz QC lasers and QC lasers
based non-polar materials such as Si/SiGe.
8.5.4 Escape time, Bragg reection and upper level connement
Stricktly speaking, under applied electric eld, all states of a quantum cascade structure are
in fact resonances with a nite lifetime caused by the nite probability of tunneling by auto-
ionisation. This auto-ionisation will translate into a nite escape time
esc
from the upper
state, reducing the total upper state lifetime. To circumvent this problem, the conned
states of the active region structure are computed by articially bounding the structure with
high barriers, transforming all the states into real bound states.
However, especially for states with large connement energy, the auto-ionisation into contin-
uum may be relevant to the upper state lifetime, occuring either directly or through coupling
with an intermediate state further down in the epilayer stack. Preventing this escape path
is especially important in the excited state of the active region, since this process will reduce
articially the upper state lifetime below the limit set by the optical phonon scattering.
This lifetime can be computed directly by invoquing Heisenbergs relationship and computing
the linewidth 2 of the resonance, leaving the structure unbounded.
= / (8.5.15)
The problem with this approach is that it will tend to underestimate the strength of the
process since it will assume an innite coherence length. In practice, this problem may be
circumvented by computing a structure truncated to a few periods.
To design around this problem, care must be take to design the injection/ relaxation region
with a minigap facing the upper state of the laser transition. Since the attenuation of the
wavefunction inside this region will be proportional to the width of the gap, it is usually a
good design rule to maximise the width of this minigap and to center it around the excited
state of the laser transition, as shown in Fig. ??. In fact, it is usually not a solution to just
increase the thickness of the escape barrier since this will reduce the escape rate of the lower
state as well. When this gap has to be build in the continuum, a design rule may be used
based on the Bragg reection condition[35]. In this approach, for each well/barrier pair, the
condition
k
w
l
w
+k
b
l
b
= (8.5.16)
is imposed, where k
w
=
2m
E/ and k
b
=
_
2m
(E
c
E)/ are the wavevector in the
quantum well and the quantum barrier, respectively. The individual well and barrier thick-
ness l
w
and l
b
may deviate from the exact quarter condition since we choose to keep the
eective conduction band edge of the injection region at under the applied eld. Math-
ematically, it translates into the requirement that the eective conduction band potential
V (x) of the injector at the position x, approximated by
V (x) = E
c
l
b
l
b
+l
w
(8.5.17)
exhibits an eective electric eld
V
x
= F that approximatively cancels the one applied
at threshold. In Fig. 8.12, the electron transmission coecient is plotted as a function of
energy for a injector designed to provide Bragg reection at the energy of the excited state
E
3
= 500meV . At this energy and given the applied electric eld of 85kV/cm, the electron
is barely conned in the well and has an energy above the barrier in the whole injection
region. A transmission coecient of 3 10
3
and 10
4
is found assuming that the electron
coherence is maintained over 3, respectively 5 periods, while a relativeley large transmission
coecient of 10
1
is achieved at the energy of the lower states. The escape time therefore
0.0001
0.001
0.01
0.1
1
10
0 0.2 0.4 0.6 0.8
3 wells
5 wells
"Minigap" "Miniband"
E
1
E
2
E
3
85kV/cm
Energy (eV)
T
r
a
n
s
m
i
s
s
i
o
n
Figure 8.12: Computed transmission of a injector designed to provide maximum reection
at the energy of the excited state, as shown.
depends critically on the number of periods. Here, we use for the escape time the expression
esc
= / = /(E
3
t) (8.5.18)
where t is the transmission coecient. For t = 1 10
4
, we obtain
esc
= 12ps, much larger
than the intersubband phonon scattering time. However, if the coherence length is reduced
to 3 well barrier periods, the computed escape time decreases to 0.4ps.
In Fig. 8.13, the electroluminescence intensity a structure with and without Bragg reector
are compared at low temperature. As demonstrated by a improvement of about a factor of
3 to 4 of the electroluminescence eciency, it is clear that the Bragg reector is eective in
signicanly reducing the electron escape time. Finally, let us note that as the temperature is
raised, the cohence length of the electron will be reduced, and electron may also be thermally
activated above the minigap, all of this increasing the electron escape rate.
0
4
8
12
0 0.2 0.4 0.6 0.8 1.0
0 5 10 15 20
0
4
8
12
Injection current (A)
O
p
t
i
c
a
l

p
o
w
e
r

(
m
W

s
t
e
r
a
d
i
a
n
s
-
1
c
m
-
2
)
O
p
t
i
c
a
l

p
o
w
e
r

(
n
W
)
Current density (kA/cm
2
)
graded gap injector
Bragg reflector injector
minigap
Figure 8.13: Comparison of the electroluminescence eciency of a structure designed with
and without a Bragg reector facing the upper state.
8.5.5 Injection eciencies
One could be tempted to conclude from the description above that optimization of a gain
region may be reduced to the simple optimization of the compute ratio of upper to lower
state lifetimes, combined with an optimization of the intersubband oscillator strength. This
picture is however oversimplied because it implicitly assumes that all the electrons are
initially injected in the upper state, and, furthermore, that they are all extracted immediatly
from the lowest state. In fact, the situation is much more complicated by the fact that both
of these processes (injection and extraction) are essentially non-ideal.
Let us assume, as shown schematically in Fig. 8.14, that a fraction
3
of the current is
injected into the upper state, and
2
into the lower state. Such a non-resonant injection
originates from elastic or inelastic scattering between the injector state and the lower states
of the active region. In this case, the expression for the material gain g reads:
g = g
c
n = g
c
J
q
0
_
3
(1

2
32
)
2
2
_
. (8.5.19)
The inuence of the injection eciencies
3
and
2
are clearly not symmetric. The gain
is simply scaled by the value of
3
, whereas, depending on the ratio of
2
/
3
, the value of
2
can have much more dramatic consequences of cancelling the gain altoghether. Clearly,
although the design should in principle aim at reaching injection eciencies as close to the
ideal values (
3
= 1 and
2
= 0) as possible, the detrimental eects of non-ideal injection
eciencies may be mitigated by a system with a large lifetime ratio (
2
/
3
). In the case
of a quantum cascade lasers based on a superlattice active region, equation (1) can still be
applied with the n = 3 state being the lower state of the upper miniband and the n = 2
8.6. OPTIMIZATION OF THE ACTIVE REGION: DIFFERENT DESIGNS 81
Injector Active
esc
21
1
2
3
32
esc
Figure 8.14: Schematic description of the relevant levels and injection eciencies in a cascade
laser
the upper state of the lowest miniband. Equation (1) shows clearly the strong dependence
of the gain on both injection eciencies and the ratio of the lifetimes: one must design a
structure which maximises the dierence
3
2
and minimises the ratio
2
/
32
.
8.6 Optimization of the active region: dierent de-
signs
8.6.1 Designs: general trends
As shown schematically in Fig. 8.15, after the demonstration of the rst working laser, the
studies rst aimed a testing the validity of the dierent components of the intersubband
toolbox individually, i.e. whether a quantum cascade laser could be build only based on
tunneling, optical phonon resonance, or phase space. In later studies, these dierent strate-
gies were combined in a single design in order to optimize the various components of the
threshold current density appearing in equation 8.4.8, as shown in Fig. 8.16. The result of
this optimization has lead to the focus, for the mid-infrared on roughly two or three designs,
that will be presented here.
8.6.2 Three quantum well active region
We will start our discussion of our dierent qc laser designs with an evolution of the original
structure designed for operation at 4.3m wavelength [36]. This design has some unique
features: rst of all it is the one which allows the shortest wavelength to be reached for a
given band discontinuity; by combining the optical phonon resonance with the possibility of a
diagonal transition it enables a good ratio of upper state to lower state lifetime. A schematic
description showing how such an active region design can be decomposed is shown in Fig. 8.17
where the bare states, the states of the quantum well before coupling by the tunnel barrier,
are shown along the dressed states. The active region is based on a pair of quantum well
(a), designed in such a way that the ground state of the thinner well is brought in resonance
with the ground state of the thicker well. The tunnel barrier is then adjusted such that
Figure 8.15: Schematic diagram of the rst generation of quantum cascade lasers.
isb
dn
esc
2
1
gain ~ (n
2
- n
1
)
Figure 8.16: Schematic of the parameters optimized in the active regions.
the energy splitting matches the optical phonon resonance Fig. 8.17(b). Because the excited
states gain energy faster with quantum well thickness, the excited states will not be resonant.
A third quantum well is then added upstream. Two possibilities exist. In the rst one, the
state of this additional quantum well is resonant with the n=3 state of the quantum well,
and will anticross with it Fig. 8.17(c). In this case, the transition is strongly diagonal and,
as a result, the lifetime will be approximatively twice the one of the vertical transition. If,
in the contrary, the state of this thin well is clearly above the n=3 state Fig. 8.17(d), the
transition will have the lifetime, oscillator strength of a vertical transition but the extension
of the wavefunction in the injection barrier will improve the injection eciency.
A schematic band diagram of both structures, designed for operation at = 10.3m and
= 5.2m respectively, are shown in Fig. 8.18. The choice of an diagonal transition at longer
wavelength allow to keep a good ratio of upper to lower state lifetime even for a relatively low
transition energy of 120meV. The diagonal nature of the laser transition is responsible for
the lifetime (1.6ps at low 80K) longer by a factor of two as compared to a vertical transition
at the same energy. In contrast, the lifetime of the upper state of the vertical transition
device is approximatively the same as the one of square quantum well. While the oscillator
strength of the diagonal structure (f = 16) is smaller than the one expected for a vertical
transition at the same energy, it is about the same as the one of the vertical transition at
E = h
LO

1
2
3
4
a)
b)
c)
1
2
3
4
1
2
3
4
d)
5
Figure 8.17: Schematic potential prole and design of the active region of a quantum cascade
laser based on three quantum wells. a) Two quantum well before the application of the
electric eld. b) The electric eld bring the two ground states in resonance and yields a
splitting equal to the optical phonon energy. A third thinner well is added upstream. If
the state of this well is resonant with the excited state of the coupled well, the resulting
transition is diagonal (c), if the well is thinner and the resonance is above, the transition is
vertical (d).
5.2m because of non-parabolicity eects. These devices operated above room temperature,
up to T = 330K with 200mW high peak power [3].
8.6.3 Double phonon resonance
One weakness of active regions based on three quantum well active is the extraction from
the ground level of the active region. In fact, studies on diagonal transitions showed that
the extraction from the active region, due to resonant tunneling, proceeded on a time scale
of about 2ps. This relatively long lifetime, compared to the fast scattering between level
n=2 and n=1 of the active region, enabled electrons to be scattered back to the n=2 state
by multiple scattering events between states. As a result the population inversion was more
dicult to achieve, degrading the high temperature and high power performances of these
devices.
One attempt to solve this problem has been the design based on so-called two-phonon ex-
traction. In the two-phonon resonance design, the extraction is obtained by a ladder of three
states separated by an optical phonon each. According to our computation, the addition
of this extra level decreases signicantly the population of the lower state. For a structure
described in the next paragraph emitting at 9m, the computed distribution show a
much reduced population of the lower laser state . A schematic conduction band diagram of
one period of the active region is shown in Fig. 8.19. The upper and lower lasing states are
Minigap
1
2
3
z
32
= 2.3 nm
3
= 1.6 ps
AlInAs barriers
InGaAs wells
E
c
= 0.52eV
Active region
Si: n = 2.5x10
17
cm
-3
Injector
(doped)
Minigap
Active region
Injector
(doped)
Minigap
a) b)
= 10.3 m = 5.3 m
z
32
= 1.6 nm
3
= 1.3 ps
Figure 8.18: Examples of active regions based on a three quantum wells, using lattice-
matched InGaAs/AlInAs material. Structure a) is designed for a wavelength of = 10.3m
using the concept shown in Fig. 8.17 c), that leads to a diagonal transition with a longer
lifetime. Structure b) is designed for = 5.3m and exhibits a vertical transition, as
explained in Fig. 8.17 d) [3].
the wavefunctions with numbers 4 and 3, respectively; and the ground state of the injector
is denominated with the letter g. As already mentioned in the preceding paragraph, the
active region was composed of 4 QWs which results in 3 coupled lower states (levels 1, 2, and
3). Each two of the latter (i.e. level 3 and 2, and level 2 and 1) were separated by one phonon
energy. This double phonon resonance yielded a short intersubband electron scattering time
(see values in caption of Fig. ??), and therefore an ecient extraction of the electrons into
the injector region. The upper lasing state exhibits a much longer intersubband electron
scattering time of
up
= (
1
43
+
1
42
+
1
41
)
1
= 0.54ps (at T = 300K) where the lifetimes
contain both emission and absorption processes. The relatively large dipole matrix element,
z
43
) = 3.0nm, conrms that the lasing transition is mainly a vertical one. Thanks to the
thin rst well, which reduces the overlap of the injector ground state g with the lower lasing
state wavefunctions 1, 2, and 3, the injection eciency was kept similarly high as in the
classical 3 QW design. The new design takes thus advantage of the good properties of both
the 3 QW design (high injection eciency) and the superlattice design (short lifetime of the
lower lasing state).
Low temperature (T = 80K) luminescence spectra were taken from a device in which the
light was extracted from a 45 degree wedge polished in the substrate. These spectra, taken
for various injection currents, are shown in Fig.8.21 along with the calculated transition
energies. All the transitions in the active region can easily be identied. Our computation
show that the dipole matrix elements between states 4 1 and 5 1 are always very small
(z
51
, z
41
< 0.1nm), regardless of applied electric eld. Peaks related to transitions from level
5 can only be observed in the strong injection regime , as observed experimentally. The
computed dipole matrix element z
42
decreases strongly with applied eld, as observed in
Fig. 8.21. In general, the luminescence peaks shown in Fig 8.21 conrms the location and
strength of the respective intersubband transitions.
Minigap
Minigap
1
2
3
4
5
Figure 8.19: Schematic band diagram of a two-phonon resonance gain region designed for
operation at 9m. .
8.6.4 Boundtocontinuum active regions
Besides the two phonon resonance approach, the boundtocontinuum approach provides an
other route for maintaining a high population inversion and low threshold current densities
even at high temperatures. The bound-to-continuum active region can be seen as an evolu-
tion of the active regions based on superlattices. As described above, a superlattice exhibits
two very attractive features: a very large oscillator strength and a favorable ratio of lifetimes
at the edge of the minigap .
However, active region of quantum cascade lasers operate under a strong applied electric
eld. The latter, applied on a periodic superlattice the electric eld applied to the structure
would at the same time break the miniband of the superlattice into a set of localized states,
losing all the favorable features of the extended states. An example of the eect of an applied
eld on the product of upper state lifetime and oscillator strength is shown in Fig. ??.
For this reason, the rst quantum cascade lasers based on a superlattice were doped homoge-
nously inside the active region to screen the eld. This technique presented the disadvantage
however of introducing additional broadening and loss caused by the ionized dopants. As a
result, operation at room temperature of these devices was strongly limited. A very ecient
technique to solve this problem was proposed by Tredicucci and consisted in chirping the
superlattice to articially compensate the applied electric eld. Numerically, it consisted in
changing the barrier and well widths of the superlattice such as to keep the energy of the
miniband edges constant even in the presence of the applied electric eld. This procedure
can be performed numerically by solving iteratively equation 3.4.72 for the well and barrier
thickness L
i
and h
i
, solving for the rst band edge occuring at qd = :
cos(k
w
L) cos(k
b
h)
1
2
_

1
_
sin(k
w
L) sin(k
b
h) = 1 (8.6.20)
where k
w
and k
b
are computed at an energy (z) =
0
+
j<i
q
0
(L
j
+ h
j
)c. The procedure
must be stopped when the excess energy at the upper band edge becomes close to the
discontinuity. The result of this procedure is shown for the chirped superlattice demonstrated
0.001
0.01
0.1
1
100 150 200 250
4-3 4-2 5-3 4-1 5-2
7V 40mA
9V 300mA
13V 1.5A
S1840
80K
Photon energy (meV)
I
n
t
e
n
s
i
t
y

(
a
r
b
.

u
n
i
t
s
)
Figure 8.20: Low-temperature (T = 80K) luminescence spectra of the active region based
on a two-phonon resonance. Calculated transition energies are shown on the top horizontal
axis
by A. Tredicucci ?? in Fig. 8.22 and compared to the same computed performed for a bound-
to-continuum structure.
In the latter design, shown schematically in Fig. 8.23, the active region spans the whole
period and consists of a chirped superlattice presenting a tilted lower miniband whose width
is maximum in the center and decreases on both sides close to the injection barriers. The
upper state is created in the rst minigap by a small well adjacent to the injection barrier.
Its wavefunction has a maximum close to the injection barrier and decreases smoothly in
the active region. This upper state is well-separated from the higher-lying states of the
superlattice, lying in its rst minigap. It therefore does not need to be conned by separating
the structure into an active region and an injection/relaxation region. Because of this large
energy separation (60meV ), the injection eciency
3
is not reduced by electron injection
into higher energy states of the superlattice[37, 38]. In addition, this conguration of the
wavefunction should enable an injection eciency comparable to the one of the 3QW design.
As a result of the localization of the upper state, the miniband selection rules are somewhat
relaxed, and the oscillator strength is now spread between the upper state and a number of
lower states. This feature is apparent in the electroluminescence spectrum show in Fig. 8.23
along with the computed matrix elements. This design architecture was found to oer many
interesting features:
High temperature performance: The devices shown in the above gure operated up to
150C with a very low dependence of the threshold current, translated in a T
0
parameter
as high as 190K
Long wavelength operation: Because of its good lifetime ratio, this design performs very
8.7. CASCADING: SCALING WITH THE NUMBER OF PERIODS 87
Figure 8.21: Oscillator strength (right axis) and product of oscillator strength and upper
state lifetime (left axis) as a function of applied electric eld for a superlattice consisting of
5 period of a 59
AInGaAs quantum well and a 9
AAlInAs barrier
well at long wavelengths. Above room temperature operation at wavelengths as long
as = 16m were achieved.
Broadband gain: Because a single upper state shares a number of lower state, this design
is very well adapted the achievement of gain over a broad wavelength range.
Non-linear generation: Bound-to-continuum active regions have shown to have a combi-
nation of mid-infrared gain and Terahertz susceptibility, enabling Terahertz generation
by intracavity non-linear mixing.
8.7 Cascading: scaling with the number of periods
As discussed above, the possibility to cascade the active regions is a key advantage of inter-
subband lasers. In the simple rate equation model, the number of periods appear in both
the threshold current density and slope eciency.
8.7.1 Threshold current density
The total overlap factor between the optical mode and the gain region, , is a function of
the number of periods of the structure,
=
Np
p
(8.7.21)
This expression can be simplied to =
p
N
p
, when covers only a central part of the
optical mode, i.e. when the optical intensity is constant over the active region periods and
therefore the
p
are almost identical. This approximation is valid as long as the active
region thickness is much smaller than the mode width. For typical InP-based waveguides,
100 150 200 250 300
Distance A
0
0.1
0.2
0.3
0.4
0.5
Bound to continuum
100 150 200 250 300
Distance A
0
0.1
0.2
0.3
0.4
0.5
Chirped superlattice
Figure 8.22: Three rst miniband edges as a function of position for a chirped superlattice
(a) and for a bound-to-continuum structure c).
it is veried for N
p
< 20 for lasers operating at 5m and N
p
< 30 for lasers operating
at 9m. In this regime, the threshold current density is simply inversely proportional
to the number of periods, as the reduction of the current density thus relies simply on a
geometrical increase of the gain region interacting with the optical mode. As the laser active
regions are electrically pumped in a series conguration, the amount of carriers injected per
period, therefore, do not vary with their number. Adding more period does not aect the
population inversion in each single stage, but it increases the applied voltage. This situation
is very dierent from that of conventional diode lasers, where by increasing the number of
quantum wells in the active region, one has also to increase the current to reach threshold
because a larger active region volume must reach transparency.
When more stages are then added,
p
decreases as the total overlap factor is naturally
constrained by the condition < 1. In practical application, a good practice is to keep the
number of period such that the total overlap factor is in the range 0.5 < < 0.8.
8.7.2 slope eciency
In a simple picture, the electrons are simply recycled from period to period, emitting a
photon at each step. As a result, and as predicted by equation 8.4.9, the slope eciency
of a quantum cascade laser is proportional to the number of stages. In a simplied model,
this remains true even when the optical eld is not constant over the various periods, i.e.
when the
p
are not constant. Intuitively, this can be understood by the fact that the slope
eciency inherently does not depend on .
However, a more detailed analysis shows that in the latter case, a hole burning occurs in the
8.8. TEMPERATURE DEPENDENCE 89
50 100 150 200 250
300K
80K
Photon Energy (meV)
a)
b)
Figure 8.23: a) Schematic conduction band diagram of one stage of a bound-to-continuum
active region under an applied electric eld of 3.5 10
4
V/cm. The moduli squared of the
relevant wavefunctions are shown. b) luminescence spectrum of the active region at 300 and
80K, as indicated. The applied bias is 9V . Lower curve: computed oscillator strength of
the various transitions from the upper state.
center of the waveguide and the slope eciency is renormalised by a factor
v
equal to
v
=
(
Np
i=1

i
)
2
N
p
Np
i=1
. (8.7.22)
This factor is
v
> 0.9 in common waveguides.
The dependence of the threshold current density, operating voltage and slope eciency has
been veried in experiments where quantum cascade lasers were grown and tested with a
number of stages varied between 3 and 75 [4]. As shown in Fig. 8.25, a very good agreement
between the simple theory shown here and the experiments was found.
8.8 Temperature dependence
A key feature of mid-infrared quantum cascade lasers, especially when compared to semi-
conductor lasers operating at similar wavelengths, is the weak dependence of the threshold
current density as a function of temperature. As in conventional semiconductor lasers, the
threshold current density J
th
(T) is a temperature-dependent parameter that can be conve-
niently tted by the expression
J
th
(T) = J
0
exp
_
T
T
0
_
(8.8.23)
where T
0
is a parameter that expresses the temperature dependence of the threshold current
density.
As shown in Fig. 8.26 where the light versus current plot of a a quantum cascade laser
based on a bound-to-continuum transition operating at = 9m is reported as a function of
temperature, the threshold current density varies according to equation 8.8.23 with a value
of T
0
as large as 190K. Both the large value of T
0
as well as the fact that the device operates
Figure 8.24: Overlap as a function of the number of period .
up to such a high temperature of 150C is unique to quantum cascade lasers. Interband
laser operating at similar wavelengths, based on lead salts, exhibited maximum operating
temperatures of 200K with T
0
values around 30K.
8.8.1 intersubband physics
Theoretically, the low temperature dependence of the threshold current density is expected
because of two inherent features of intersubband transitions First, unlike interband devices,
the joint density of states of intersubband transition is atomic-like. For this reason and as
discussed in detail in Chapter 5, if one neglects non-parabolicity, the temperature dependence
of the broadening of the gain curve originates only from the dependence of the in-plane
scattering mechanisms. As a result, the broadening parameter in the equation of the
threshold current 8.4.8 exhibits a weak temperature dependence. Experimentally, this can
be observed by the small change in the linewidth of the luminescence shown in Fig. 8.23 that
increases only from about 20meV at 80K to 24meV at 300K.
The second reason, also discussed in Chapter 5 is the fact that the main intersubband scat-
tering mechanism is the optical phonon scattering. For this reason, the upper state lifetime
is expected to decrease by a factor of approximatively two between cryogenic temperature
and 400K. If a design displaying a large ratio of upper to lower lifetime is choosen, the
threshold is expected to display approximately the same temperature dependence as the
e
appearing in equation 8.4.8.
8.8. TEMPERATURE DEPENDENCE 91
Figure 8.25: Slope eciency as a function of the number of periods in a series of otherwise
identical quantum cascade lasers. Inset: slope eciency per stage.(From ref. [4])
8.8.2 Backlling
An other important term in the expression of the threshold current density is the thermal
population of the lower state n
therm
2
. As shown schematically in Fig. 8.28 a), this population
originates from carriers thermally excited to the lower state from the injection region. The
population n
therm
2
must be compensated by the same upper state population to achieve
transparency, and therefore should be minimized in a given design. In a simplied model,
the thermal population can be approximated by a simple thermal activation term at an
electronic temperature T
el
n
therm
2
= n
s
exp(
inj
/kT
el
) (8.8.24)
where n
g
is the sheet density of the injector and
inj
the energy dierence between the lower
state and chemical potential of the injector. The eect of varying the value of
inj
for a given
design is shown in Fig. 8.28 b), where the threshold current density versus temperature for
a series of quantum cascade lasers with injectors of various lengths showing dierent values
of
inj
. The detrimental eect of a low value of
inj
on the high temperature operation is
especially apparent. Considered in detail, the computation of n
therm
2
is complex because the
electronic temperature T
el
is in general larger than the lattice one and is a strong function
of injected current. Therefore, assuming a constant value of n
therm
2
as a function of injected
current density J is a rather crude approximation.
8.8.3 Self-heating
To reach threshold of operation, quantum cascade devices must dissipate a power density
amounting to 20-50kW/cm
2
. For this reason, especially when operating in continuous wave,
one cannot assume an active region temperature T
act
equal to the one of the submount T
sub
.
In a simple model the active region temperature is assumed to be constant and related to
8
2
4
6
200 250 300 350 400 450
J (kA/cm
2
)
Current (A)
B
i
a
s

(
V
)
P
e
a
k

P
o
w
e
r

(
W
)
0
5
10
15
0 2 4 6 8
0 2 4 6 8
0
0.5
1.0
1.5
Temperature(K)
J
t
h

(
k
A
/
c
m
2
)
-30
0
30
60
90
120
150C
T
0
= 190K
Figure 8.26: Temperature dependence of the light-versus current characteristics of a quantum
cascade laser based on a bound-to-continuum transition operating at = 9m.(From ref. [5]).
Inset: Threshold current density as a function of temperature, tted by the expression 8.8.23
with a T
0
of 190K
the submount one by a single thermal resistance R
th
:
T
act
= T
sub
+R
th
UI (8.8.25)
where U and I are the operation voltage and current. Expressing the above relation at
threshold, and assuming that the threshold current density has the usual exponential depen-
dence in temperature with a characteristic temperature T
0
, we obtain
T
act
= T
sub
+
U
th
J
0
exp(T
act
/T
0
)
G
th
(8.8.26)
where the specic thermal conductance G
th
in Wcm
2
K
1
has been used. The above equa-
tion denes naturally a maximum for T
sub
, that can be obtained by setting dT
sub
/dT
act
= 0:
T
sub,max
= T
0
_
ln
_
T
0
G
th
J
0
U
th
_
1
_
(8.8.27)
The above equation denes the maximum operating temperature in continuous wave. A
large value of the latter requires the combination of a
a large value of T
0
,
a low intrinsic threshold power U
th
J
0
,
and a high specic thermal conductance G
th
.
The above criteria have driven the research in quantum cascade lasers. While the rst two
items drove the developement of new active region designs and low-loss waveguides, the last
point is concerned with thermal engineering of the active region.
8.9. DOPING OF THE ACTIVE REGION 93
Figure 8.27: a) Electrons thermally excited from the injector populate the lower state. The
critical energy is the dierence between the Fermi energy of the injector and the lower
state. b) Threshold current density versus temperature for a series of quantum cascade lasers
with an identical active region based on a vertical transition operating at = 4.6m but
with injectors of various lengths showing dierent values of
inj
.
8.9 Doping of the active region
Active region doping is a necessary evil in a quantum cascade laser, as to a large extend
the electrical stability of the system depends on having a structure displaying global electri-
cal neutrality in each period. In order to minimize the scattering induced by the presence of
ionised impurities, the latter are inserted as far as possible from the active region. The dra-
matic broadening eect of doping impurities inserted directly into the active region is shown
in Fig. ??. However, when the dopants are remote from the active region (in practice about
20nm away from it), no direct relationship between injector doping density and linewidth is
observed, in agreement with a linewidth dominated by interface roughness eect. We saw in
chapter 7 that the main loss mechanism of a waveguide is the free carrier absorption. In a
simple model where the optical losses are dominated by the active region one, we therefore
expect the waveguide losses to scale with the doping of the active region. Similarly, to the
extend that the lifetimes are not limited by electron-electron scattering and therefore are
independent of electron density, we expect the current-voltage characteristic to scale linearly
with doping.
Such a simple picture is well supported by the experiment. A device with an active region
based on a two-phonon resonance was grown and the doping of the injector was varied
between a sheet carrier density of n
s
= 1 10
11
cm
1
andn
s
= 2.5 10
11
cm
1
. As expected
from this simple model and shown in Fig. 8.29, both maximum operating current and the
threshold current increased linearly with doping. As the maximum current is linear in the
sheet carrier density, we can write it as charaterized by a single transport time at resonance
Figure 8.28: Comparison of the electroluminescence from a two otherwise identical samples
with dierent doping proles. In the sample with the setback, only the injector region is
doped. In the reference sample, the area marked in green in the schematic band structure
in the inset is also doped.
trans
:
J
max
=
n
s
q
0
trans
(8.9.28)
Therefore the slope of the J
max
versus n
s
yields the inverse transport lifetime. The transport
lifetime extrapolated from the data shown in Fig. 8.29 yields a transport time of
trans
= 4.3ps
. This time takes into account the sum of the time in the active region and the one of the
transport in the injector. Similarely, one can express the threshold current density 8.4.7 in
a simple form:
J
th
=
q
0
tot
g
c
eff
(8.9.29)
where we now included the ground state reabsorption into
tot
=
tot
+g
c
n
therm
c
. Expressing
the total losses
tot
as a linear expression in n
s
:
tot
=
m
+
w
+g
fc
n
s
(8.9.30)
where
m
are the mirror losses,
w
are the losses of the empty waveguide and g
fc
is the free
carrier cross section. As in the case of the maximum current, the slope of the J
th
versus n
s
characteristics yields
dJ
th
dn
s
=
q
0
g
fc
g
c
eff
. (8.9.31)
Assuming the eective upper state lifetime can be computed, the above expression enables
the computation of the ratio of the loss to gain cross section g
fc
/g
c
. As we will see in the next
8.10. FUNDAMENTAL LIMITS OF QUANTUM CASCADE LASERS 95
Figure 8.29: The threshold current density J
TH
in open squares and the maximum current
density J
NDR
in solid squares as a function of sheet densities corresponding to the doping
study (n67, n72, n69, n71). In dashed lines, tted curves corresponding to an extrapolated
doping oset of n
offset
= 1.0 10
11
cm
2
added to the active region doping. Open and solid
circles represent the threshold and the maximum current densities respectively of a laser
grown after a longer MBE running time (n120).
paragraph, this number as well as the ratio of
eff
/
trans
are the key numbers characterizing
the fundamental performance limitations of an active region. Applying this analysis on the
data of Fig. 8.29 and using
eff
= 0.55ps, on obtain a ratio of g
fc
/g
c
= 1.9 10
2
.
Of course this simple picture of the eect of doping breaks down if extremely large dop-
ing levels are used. In that case, the Hartree potential introduces large distortions in the
conduction band of the injector and signicantly alters the alignment between injector and
active region levels.
8.10 Fundamental limits of quantum cascade lasers
The simple atomic model developed in this chapter, combined with the considerations on
the doping dependence developed above, enables us to express the fundamental limitations
on the wallplug eciency of quantum cascade lasers. The wallplug eciency
wp
of a laser is
dened as the ratio of the extracted optical power to the injected electrical power. Neglecting
gain saturation and assuming a constant slope eciency, we have
wp
=
(J
max
J
th
)
J
max
dP/dI
U
. (8.10.32)
As in the preceding paragraph, we will assume that the maximum operating current can be
dened using a global transport time across one period
trans
dened by equation 8.9.28. The
operating voltage U at resonance can be easily written as
U = N
p
_
q
0
+
inj
_
. (8.10.33)
Using for the slope eciency the expression of equation 8.4.9 and for the threshold the
expression 8.4.7 we nally can combine these expressions together and get
wp
=
tr
tot
1
1 +
inj
/()
_
1

trans
eff
_

tot
n
s
N
p
g
c
+
n
therm
n
s
_
_
(8.10.34)
where
tr
is the dierential eciency of the laser transition, given in a three level model of
the active region by
tr
=

eff
eff
+
2
. (8.10.35)
For any given material and design, the parameters on which the design will be optimized
are the doping level n
s
, number of periods N
p
and the mirror loss
m
. We are obviously
interested at the maximum eciency achieved for optimized values of these parameters. As
in the previous paragraph, we assume that the waveguide losses are dominated by the doping
of the active region, i.e. by the term g
fc
n
s
in equation 8.9.30. We assume also that
inj
is large enought so that the thermal population of the lower laser state n
therm
is negligible.
Under those assumptions, the maximum value of the wallplug eciency, i.e. computed for
the value of the mirror loss
m
for which Eqn. 8.10.34 is maximum, takes an especially simple
form:
wp,max
=
tr
1
1 +
inj
/()
_
_
2
. (8.10.36)
In equation 8.10.36, we have introduced the reduced dimentionless upper state lifetime
=
up
/
tr
and dimentionless reduced cross section g
= g
c
/g
fc
. The fact that Eqn. 8.10.36
does not depend on either doping level or number of period justies a posteriori our
assumption that losses from the active region could be assumed to dominate.
In equation 8.10.36, the product

g
is a key parameter determining the wallplug e-

ciency: a value lower than one indicates a device that will never reach threshold as the free
carrier absorption will always be larger than the intersubband gain. In fact, the optimization
of an active region should of course try and maximize this parameter.
A large value of
is achieved in an active region where transport is very fast in the injec-

tor and most of the carriers are remaining on the upper state of the laser transition. An
inspection of the parameter g
demonstrates the very strong wavelength dependence of the

wallplug eciency. In fact, assuming a Lorenzian lineshape for the gain cross section char-
acterized by an in-plane lifetime broadening
//
and a normalized oscillator strength of the
laser transition f
= f
m
m
0
(where m
is the carriers eective mass); taking similarely the

free carrier absorption to orignate from intersubband absorption within the injector states
(assumed at a vanishing transition energy) characterized by a broadening lifetime
fc
, the
8.10. FUNDAMENTAL LIMITS OF QUANTUM CASCADE LASERS 97
Figure 8.30: Wallplug eciency as a function of photon energy. Filled squares, experimental
values in continuous wave, from references [6, 7, 8, 9, 10, 11, 12]. Solid disk: pulsed values
from references [13, 14, 10, 15, 16]. The solid line represents the prediction from the simplied
formula with the key parameters derived from [15]. The inuence of a shorter or longer in-
plane dephasing time
//
is shown by the two dotted curves. Long dashed lines represent
the prediction from the model scaled by a factor 0.4.
the dimentionless cross section writes:
g
=
1
2
//
fc
f
. (8.10.37)
The intrinsic
2
dependence of Eqn.8.10.37 stresses the increasing diculty of successfully
fabricating quantum cascade lasers operating at low frequencies. Expressing the normalized
cross-section g
using Eqn. 8.10.37 removes all explicit dependence of the wallplug eciency
from Eqn. 8.10.36 in the eective mass. Mathematically, it is simply because both the gain
cross-section and the free carrier absorption are proportional to the inverse eective mass. In
the mid-infrared, there is however a very strong indirect dependence of the wallplug eciency
in the eective mass through the dependence of both dephasing times
//
and
fc
in m
. Tthe
linewidth of an intersubband transition is limited by the interface roughness [39, 40]. Thus,
for any given interface structure (described by the step height and the correlation length
in Ref [39]) and band discontinuity E
c
, the lifetime
//
is proportional to (m
)
2
. The
eective mass dependence of the free carrier absorption is not as easy to express because
it depends on the dominant mechanism of scattering (interfaces, phonons, or impurities).
As shown in Fig. 8.30, the simplifed expression given by Eqn. 8.10.36 predicts well the
wavelength dependence of the wallplug eciency, although it tends to predicts value larger
than the ones achieved experimentally.
The experimental data points shown in Fig. 1 are somewhat skewed by the lack of reliable
data in pulsed mode, as the continous wave values are further limited by self-heating. Some
pulsed data are however especially valuable since they are results of optimization of the
wallplug eciency over mirror loss [16, 14]. In continous wave, the maximum eciency
must be achieved at the minimum possible current density in order to minimize self-heating
eects. In that case, both doping and mirror losses are minimized and the optimum value
depends strongly on the residual waveguide loss
scatt
.
8.11 Short wavelengths: discontinuity and strain com-
pensation
8.11.1 Conduction band discontinuity and performance
In quantum cascade lasers, the ability to tailor the photon energy is limited, on the high
energy side, by the size of the conduction band discontinuity between the two semiconductor
materials. In the case of the In
0.53
Ga
0.47
As/In
0.52
Al
0.48
As material which is lattice matched
to InP and has a conduction band discontinuity E
c
= 520meV, lasers with photon energies
up to h = 290meV (h = 0.56E
c
) were realized, although high temperature operation
(T > 300K) operation was obtained typically for photon energies h = 250meV (h =
0.48E
c
). The reason for this limitation is shown schematically in Fig. ??. Electrons from
the upper state of the laser transition can be thermally activated to continuum states. This
process is exponential in temperature and will severely limit the high temperature operation.
A rough estimate of this parameter can be obtained by assuming an escape time
esc
of the
form:
1
esc
=
1
cont
exp(E
act
/kT) (8.11.38)
where the escape rate in the continuum
1
cont
100fs is estimated from the width of a
typical continuum resonance. These quantities are very dicult to estimate because they
depend in a critical manner on the coherence length chosen. Under these assumptions, an
activation energy of E
act
= 85meV, usually chosen as the minimum value needed for a good
room temperature operation, yields an escape time of
esc
= 2.7ps,
These considerations are well supported by the experimental evidence. In Fig. 8.31, the char-
acteristic temperature T
0
is reported as a function of the emitted photon energy for quantum
cascade lasers fabricated using materials with various conduction band discontinuities. For
devices limited by optical phonon emission and exhibiting a good population inversion, the
characteristic temperature should lie between 400K (assuming a broadening independent
with temperature) and 170K (for typical measured values of the broadening of the electrolu-
minescence peak. From the data shown in Fig. 8.31, it is clear that the theoretical value of
T
0
can be achieved for large enough discontinuities, however a limited discontinuity rapidely
limits the high temperature operation.
An experiment [17] using devices fabricated from Al
x
Ga
(1x)
As/GaAs material system il-
lustrates very nicely this eect, and uses the fact that GaAs and Al
x
Ga
(1x)
As are lattice
matched for all Al contents x. In this work, two otherwise identical quantum cascade lasers
are compared, one using a barrier with an aluminium content of 33%, the other one with
45%. In Fig. 8.32 are reported the threshold current density versus temperature for both
8.11. SHORT WAVELENGTHS: DISCONTINUITY AND STRAIN COMPENSATION 99
Figure 8.31: Characteristic temperature T
0
for various quantum cascade lasers, as func-
tion of the photon energy and for various material systems. A clear relationship between
discontinuity and temperature behavior is is apparent.
devices. As expected, both devices operate with identical threshold currents at low tem-
perature. However, as the temperature is raised above about 100K, the device with the
lower conduction band discontinuity of about 300meV exhibits a much worst temperatue
performance characterized by a T
0
of 75K, while the device with the larger discontinuity of
390meV exhibited a T
0
of 166K, close to the theoretical value.
8.11.2 Strain-compensated In
x
Ga
1x
As/Al
y
In
1y
As/InP material
system
From the above considerations, it is clear that quantum cascade lasers operating at wave-
lengths shorter than 56m requires the use of a material system with a larger disconti-
nuity as the one (E
c
= 520meV) provided by the lattice-matched In
0.53
Ga
0.47
As/Al
0.48
In
0.52
As
material. For quantum cascade lasers based on GaAs, the limit is even much more severe
since, as shown in the previous paragraph, lack of conduction band discontinuity starts to
hamper the operation of quantum cascade lasers with wavelengths shorter than 9m.
For this reasons, a good deal of interest has been carried over other material systems that
would provide a larger discontinuity. The energy gap versus lattice constant of the main
III-IV-V semiconductor materials (to the exception of group III-nitrides) are reported in
Fig. 8.33. One very good candidate for short wavelength application is the InAs/AlSb ma-
terial system grown on InAs substrates. Both the very large conduction band discontinuity
(E
c
= 2.1eV and the very light eective conduction band mass m
/m
0
= 0.023 should
enable both very large optical gain and low short wavelength operation. In addition, the
lateral L and X valleys are also very high in energy (about 0.8eV) from the the bottom of
the conduction band of InAs. Indeed, devices grown using this material system hold the
present record for short wavelength operation ( = 2.9m) as well as very high performance
.
Figure 8.32: Characteristic temperature T
0
for various quantum cascade lasers, as func-
tion of the photon energy and for various material systems. A clear relationship between
discontinuity and temperature behavior is is apparent.(Redrawn from data from [17] )
.
Figure 8.33: Band gap versus lattice constant for the main semiconductor materials, to
the exception of group III nitrides. Of particular interest for quantum cascade lasers are
the GaAs/Al
x
Ga
(1x)
As, the In
x
Ga
1x
As/Al
y
In
1y
As/InP, the InAs/AlSb heterostructure
material system
at = 3.6m. Compared to the device based on the InP system, this material system suers
from less favorable refractive index and thermal conduction and the epitaxial growth is more
dicult to control. For this reason, an other very successful material system has been the
strain-compensated In
x
Ga
1x
As/Al
y
In
1y
As/InP.
The initial research in optoelectronics demonstrated that high crystalline quality, therefore
the absence of dislocations, was a prerequisite for the achievement of semiconductor lasers
with high performances and long lifetimes. Such a requirement also implied that the het-
erostructure materials had to be carefully lattice-matched so that bulk layers would not
develop dislocations due to strain between epitaxialy layers. It therefore came somewhat as
a surprise that thin layers of strained material could be grown with no detrimental eect on
material quality as long as the cumulated strain remained below the level needed to desta-
bilise the two-dimentional growth. Using this approach, strained InGaAs/GaAs quantum
wells were used to achieve interband lasers at wavelengths close to = 1m with excellent
performance. These results could be achieved because the active region material thickness
was kept very small, a few quantum wells of typical 10nm thickness. The active region of
quantum cascade laser is also based on thin quantum wells, and therefore is also compatible
with such an approach. If strain with one sign only is used, say for example compressive
strain for InGaAs/GaAs/AlGaAs material system, it is the cumulated strain required by the
large number of active region periods that will stronly limit the amount of strain that can be
introduced in each layer. If a material system is chosen were strain of both sign is possible,
one for the barrier and the other one for the quantum wells, imposing the condition of zero
accumulated strain on one active region period implies that an arbitrary number of active
region period can be grown.
Using compressively strained In
x
Ga
1x
As and tensile-strained Al
x
In
1x
As, such that the
total strain on an InP substrate is zero a material combination with a number of attractive
features.
The conduction band discontinuity can be increased by about 200-250meV as compared
to the lattice-matched case.
The - X discontinuity also increases with compressive strain in the InGaAs.
InP-based waveguides can still be used, with their low losses and high thermal con-
ductivity
The amount of strain can be chosen according to the target wavelength
Using this material system, very high performance quantum cascade lasers have been realized
from about 3.8m to 6m. The conduction band discontinuity can be tailored by
changing the In content in the well and barrier material in a concomitant way. The conduc-
tion band discontinuity can then be computed using van de Walle model solid theory [41].
In this approach, an absolute energy scale is rigorously dened from pseudopotential band
computation. Deformation potentials are then introduced to compute the new location of
the bands. Taking the total thickness of the well material h
w
and barrier material h
b
for one
a
0
(
A) G
001
(Mbar) E
v,av
(eV) E
c
(eV)
0
(eV) a
v
(eV) a
dir
c
(eV)
GaAs 5.65 2.522 -6.92 -5.29 0.34 1.16 -7.17
AlAs 5.65 2.656 -7.49 -4.27 0.28 2.47 -5.67
InAs 6.08 1.587 -6.67 -6.13 0.38 1.00 -5.08
InP 5.87 1.897 -7.04 -5.58 0.11 1.27 -5.04
Table 8.1: Some elastic constants, energies and deformation potentials relevant for In-
GaAs/AlInAs material system. (from ??).
period, imposing lattice matching to the substrate means that the lattice constant of the
heterostructure at mechanical equilibrium, given by
a
=
a
w
G
w
h
w
+a
b
G
b
h
b
G
w
h
w
+G
b
h
b
(8.11.39)
is equal to the lattice constant of InP (a
0
= 5.87
A). In Equ. ?? the indices w, b refer to

the well and barrier materials and will be refered by i = w, b in the following equations. As
each layer in now under biaxial strain, its lattice constant in the perpendicular direction also
changes and is now given by
a
i
= a
i
_
1 D
i
(a
/a
i
1)
. (8.11.40)
As a results, in-plane strain components are given by
i
=
a
i
a
i
1 (8.11.41)
and an out-of-plane component given by
i
=
a
a
i
1. (8.11.42)
In equation ?? the shear modulus G
i
is related to the elastic constants c
ij
by
G
i
= 2(c
i
11
+ 2c
i
12
)(1 D
i
/2) (8.11.43)
where, for a growth in the [001] direction, the values of D
i
are given by
D
001
i
= 2
c
12
c
11
. (8.11.44)
As the values of elastic constants are not tabulated for the alloys, the usual approach is
to interpolate those linearly from the values of the binaries. The key values needed for
the model presented here are listed in Table 8.1. The biaxial strain has an hydrostatic
component, responsible for the change of the crystal volume . The fractional change in
volume / is given by the trace of the strain tensor
= Tr(
) = (
xx
+
yy
+
zz
) = 2
i
+
i
. (8.11.45)
.
Figure 8.34: Conduction band discontinuity as a function of In content in the quantum well
material, for a strain-compensated material with the well thickness representing 40% of the
total stack thickness. Full line, result of the computation, dashed line: value adjusted to the
measured discontinuity of the lattice matched material.
Because of its symmetry, the minimum of the conduction band is only sensitive to the
hydrostatic component of the strain, and the change in the conduction band energy is given
by the deformation potential a
c
:
E
c
= a
c
. (8.11.46)
To compute the discontinuity, the conduction band positions are rst interpolated from the
binaries for the unstrained material, the values then corrected by the shift caused by the
strain using the above equation:
E
c
= E
b
c
+E
b
c
E
w
c
E
w
c
. (8.11.47)
The result of such computation for the In
x
Ga
1x
As/Al
x
In
1x
As material system is shown
in Fig. 8.34. As shown in Fig. 8.34 the discontinuity between the two materials can be
increased from 520meV to about 750-800meV depending on the amount of strain added to
each layer. The latter cannot be increased indenitely, as the critical thickness above which
the crystalline coherence cannot be maintained is strongly decreasing with increasing strain.
An other important consequence of strain is its strong eect on the eective mass of the
material. The hydrostatic component will reduce the bandgap, while the biaxial one will
split the heavy and light-holes degeneracy at the gamma point. As the eective gap of
both conduction-light hole and conduction split-o changes, so does the eective mass of the
conduction band as a result of the k p interaction. Fig. ?? shows the eective mass of the
Figure 8.35: Electron eective mass of In
1x
Ga
x
As material as a function of the Ga content
x (in contrast to the text where x labels the Indium content). Full line: unstrained material.
Dotted, dashed lines: in-plane, respectively out of plane mass for material grown on InP.
From Suwagara et al., .
In
x
Ga
1x
As material as a function of Indium content x, taken from the work of Suwagara
et al. . The mass relevant for the computation of the eignenstates, as well as for the
computation of the dipole matrix elements, is perpendicular mass m
. For increasing indium

content, the latter is actually increasing before decreasing again for Indium concentrations
larger than about 80%. Fig. ?? shows the schematic band diagram of a quantum cascade laser
designed with In
0.7
Ga
0.3
As/Al
0.6
In
0.4
As strained compensated on InP. The bandstructure was
computed using eective masses of unstrained material. As a result, the computed emission
wavelength = 3.16m is signicantly shorter than the one measured in the actual devices
= 3.4 3.6m. Using the masses from Suwagara et al, a value of 3.4m, much
closer to the experimental one is predicted. The laser emission spectrum of these devices is
shown in Fig. 8.37. The relatively low T
0
= 85K measured on these devices demonstrated
that the upper state exhibited a lifetime limited by thermoionic emission. Further work
has shown that the combination of very thin AlAs barriers with compressively strained
InGaAs quantum wells and AlInAs lattice-matched material was a very attractive material
choice. Lasers operating at cryogenic temperatures down to wavlengths as short as 3.1m
were achieved. For such short wavelength, the limitation in performance was attributed to
electron leakage into the lateral valley X of the InGaAs material.
1
2
3
Minigap
Miniband
Injection
barrier
Figure 8.36: Conduction band structure of a strain-compensated quantum cascade laser
designed for short wavelength operation. The predicted emission wavelength was = 3.16m
using the eective mass of the unstrained material. Measured wavelength were between
3.4m and 3.6m, compatible with a much heavier mass as predicted by theory.
Figure 8.37: Emission spectrum of a short wavelength quantum cascade laser at various
temperatures, as indicated.
Figure 8.38: Threshold current density versus temperature for the device operating at =
3.6m. The relatively low T
0
observed (85K) is an indication of leakage paths above the
barriers.
Chapter 9
Quantum Cascade laser II: Mode
control
9.1 Fabry Perot cavity
The active region of a quantum cascade laser is placed in the center of a mid-infrared
waveguide forming a long ridge. By cleaving the latter at a length of typically 1-3mm, a
Fabry-Perot cavity is formed. The cleaved facets exhibit an amplitude reectivity r =
R.
Imposing a stationary mode running in the cavity requires the eld to closes on itself after
a round-trip in the cavity:
r
2
exp(2iL) = 1 (9.1.1)
where
= n
eff
k
0
i
g
w
2
. (9.1.2)
is the propagation constant for a mode with a vacuum propagation constant k
0
= 2/ and
a net gain g
w
. Imposing the condition set by equation 9.1.1 for both the amplitude and
phase imposes the well-known conditions for the gain:
g =
1
L
ln(R) +
w
(9.1.3)
and the condition on the wavelength
=
2n
eff
L
N
(9.1.4)
where N is any positive integer. Taking into account the fact that the refractive index has in
general a wavelength dependence, the mode spacing between adjacent cavity modes writes
=

2
2L(n
eff

n
eff
)
(9.1.5)
Expressed in wavenumbers, the expression reads:
(
1
) =
1
2L(n
eff
+
1
n
eff
(
1
)
)
(9.1.6)
107
108 CHAPTER 9. QUANTUM CASCADE LASER II: MODE CONTROL
Figure 9.1: Subthreshold luminescence of a short wavelength quantum cascade laser as a
function of increasing current, from 100mA to 800mA by steps of 100mA. The temperature
is 20K and the threshold current is at 950mA. The spectra are shown with two dierent
horizontal scales a) and b). The shift of the individual Fabry-Perot resonances is caused by
the device heating.
where the term n
eff

n
eff
is also referred as the group index.

As show in Fig. 9.1 where a series of spectra of short wavelength quantum cascade laser is
shown as a function of injected current, the Fabry-Perot modes of a laser cavity are clearly
apparent much below threshold. Their contrast increases with injected current and their
location shifts with the change of n
eff
induced by the temperature change with injected
current. The latter eect can be analyzed using the expression for the transmission of a
Fabry-Perot cavity:
I(1/) =
_
1 R
R
_
b
1 +b
2
2b cos(4n
eff
L/)
(9.1.7)
where the parameter controlling the nesse of the Fabry-Perot cavity b is given by
b = Rexp((g
w
)L). (9.1.8)
For this reason, the measurement of the fringe contrast as a function of injected current
enables a measurement of the gain.
9.2 Distributed feedback cavity
As it is apparent in Fig. 9.1, for usual Fabry-Perot cavity lengths, the mode spacing (of the
order of 1cm
1
) is much narrower than the gain bandwidth of 100200cm
1
. For this reason,
even if because of mode competition, the laser operates on a single mode above threshold,
the exact location of the latter is almost impossible to predict with enough accuracy in order
to target a specic molecular vibration line.
9.2. DISTRIBUTED FEEDBACK CAVITY 109
9.2.1 Multilayer approach, Bragg reection condition
One solution used to solve this problem is to incorporate a distributed feedback inside the
active region of the device by creating a small, periodic modulation of the waveguide eective
index. In a rst approach, let us consider two regions of eective refractive index n
a
and n
b
of lengths a and b repeated with a periodicity = a +b. The coupling between forward and
backward propagation of the plane wave at frequency can be described by a 2x2 complex
matrix M describing the coecients of the plane waves at the nth layer with the one at the
n 1th layer:
_
a
n1
b
n1
_
=
_
A B
C D
__
a
n
b
n
_
(9.2.9)
where the coecients of the matrix M are given for a TE mode at normal incidence:
A = e
i
2na
a
_
cos(
2n
b
b) +
i
2
_
n
b
n
a
+
n
a
n
b
_
sin(
2n
b
b)
_
(9.2.10)
B = e
i
2na
a
_
i
2
_
n
b
n
a
n
a
n
b
_
sin(
2n
b
b)
_
(9.2.11)
C = e
i
2na
a
_
i
2
_
n
b
n
a
n
a
n
b
_
sin(
2n
b
b)
_
(9.2.12)
D = e
i
2na
a
_
cos(
2n
b
b)
i
2
_
n
b
n
a
+
n
a
n
b
_
sin(
2n
b
b)
_
(9.2.13)
where is the wavelength in vacuum. As the structure is periodic, the solutions must satisfy
Blochs theorem:
_
a
n
b
n
_
= e
iK
_
a
n1
b
n1
_
(9.2.14)
that nally yields the following dispersion relation
cos K = cos k
a
a cos k
b
b
1
2
_
n
b
n
a
+
n
a
n
b
_
sin k
a
a sin k
b
b (9.2.15)
where we have introduced the wavevectors k
a
=
2na
and k
b
=
2n
b
to emphasize the
similarity with the electronic dispersion relation obtained for the superlattice, in particular
when considering the two states in the continuum (equation 3.4.72). As in the electronic
case, gaps will open for frequencies corresponding to the Bragg condition
B
= 2n
eff
/N (9.2.16)
where N is an integer that characterize the order of the grating and n
eff
is the average
refractive index, assuming a weak modulation of the latter. The last equation can also be
written in terms of the propagation constant = n
eff
2
of the mode as:

2 = NG. (9.2.17)
where G =
2
is the reciprocal vector of the grating. The Bragg condition can be seen
as a conservation of the in-plane wavevector with the grating providing a integer number
Figure 9.2: Bragg reection condition for gratings of various order N a) First order grating,
only backscattering is possible. b) For N=2, either backscattering or vertical emission is
possible. c) In the third order grating (N=3), emission is only possible inside the substrated
because of the large refractive index dierence between the guided mode (n
eff
= 3.2) and
the vacuum.
of wavevectors NG and can be graphically described schematically by the vector diagrams
shown in Fig. 9.2. Of course, as the order of the grating increases, so does the number
of dierent scattering direction available for the light. Of particular interest are the rst
order grating, as only backscattering is allowed. The second order grating enables both the
frequency selection and the surface emission.
To use again the analogy with solid state systems, the Bragg reection creates a gap for
which the propagation is not possible. If the modulation is purely of the real part of the
refractive index, the gap is opened in the frequency domain; if the modulation is only on
the gain, the gap opens in the allowed propagation wavevector. Assuming again a small
refractive index change
[
n
a
n
b
n
a
+n
b
[ = [
n
n
[ << 1 (9.2.18)
and equal thicknesses (a = b) the width of the gap is shown to be
=
2
n
n
(9.2.19)
The computed transmission for a nite quarter wave stack with a periodicity of 1.56m of
average refractive index 3.2 and a refractive index step of n = 1.510
2
is shown in Fig. 9.3.
As expected, a gap appears centered on at the Bragg wavelength. The dierence between the
computed value of the stopband width 4.1cm
1
and the value obtained from equation 9.2.19
(3cm
1
) originates from the nite length of the stack. In fact, same computation performed
for the double number of periods yields a gap width of 3.3cm
1
.
The coupling constant of the grating is taken as the maximum value of K inside the gap,
i.e.
=
1
n
n
. (9.2.20)
The coupling constant describes the attenuation of the propagating wave in the gap center, it
is the solution for the imaginary part of K of equation 9.2.15 at the frequency corresponding
Figure 9.3: Computed transmission for a nite quarter wave stack (640 pairs) with a period-
icity of 1.56m of average refractive index 3.2 and a refractive index step of n = 1.510
2
.
A stop band of width about 4cm
1
appears in the transmission.
to the Bragg condition. The quantity 1/ is the attenuation length of the eld inside the
periodic grating. This quantity plays a major role in the evaluation of the performances of
distributed feedback lasers.
A laser with a periodic grating can only operate at the edge of this gap. In fact, the gap
opened in the versus k dispersion curve will force the derivative to be zero at the edge.
As a consequence, the group velocity
v
g
=

k
(9.2.21)
is zero at the band edge. For this reason, light at this band edge is stationary and therefore
a mode can be conned.
Using the multistack approach, one is also able to compute the threshold for the modes
of a distributed feedback laser. The transmission of the nite cavity is obtained by the
multiplication of the transmission matrices for each layer pair given by the elements 9.2.13
but with a complex refractive index given by n
a,b
ig

4
where g is the gain. The laser modes
are dened by the couples of values (, g) for which the transmission diverges. The result
of such a computation is shown in Fig. 9.4. As expected, the threshold corresponds to the
values predicted for a Fabry-Perot devices (g =
1
L
ln R = 13cm
1
) for the modes away from
the grating resonance, the modes on the side of the grating resonance have their threshold
gain reduced to about 6cm
1
.
Figure 9.4: Computed gain for a 1mm long device, with the same parameters as the one of
Fig. 9.3. Symmetric facets have been assumed
9.2.2 coupled mode analysis
Although the previous analysis enabled the computation of the threshold for realistic devices,
it is based on purely numerical simuations and therefore hides the physics of the device.
Kogelnick and Shank [42] have derived a coupled-wave analysis of distributed feedback lasers
that shows the physics of the system. Using their notation, the refractive index is assumed
to be modulated sinusoidally for both its real part
n(z) = n +n
1
cos(2
0
z) (9.2.22)
as well as for the loss
(z) = +
1
cos(2
0
z) (9.2.23)
where
0
=

= G/2 is the propagation constant of the grating. The eld inside the cavity
is assumed to be the sum of a right going wave and a left going wave:
E(z) = R(z) exp(i
0
z) +S(z) exp(i
0
z) (9.2.24)
where both functions R(z) and S(z) are assumed to vary slowly. Using the latter assumption,
a set of coupled dierential equation can be derived for R and S:
R
z
+ ( i)R = iS
S
z
+ ( i)R = iR
(9.2.25)
where the coupling constant is given by:
=
n
1
2n
+i
1
2
(9.2.26)
Figure 9.5: Distribution of the optical eld in the cavity of the laser as a function of the
product L.
and the normalized detuning is given as
=
n
c
. (9.2.27)
Note that the ratio of the coupling constant dened by equation 9.2.20 for a rectangular
refractive index prole and the one dened above is equal to
4
(taking n = 2n
1
) which is
exactly the Fourier component of a rectangular wave of unit amplitude. This model enables
the computation of the threshold gain in general, however without the eect of the facets.
The key results are the position of the modes for modulation of both the real part and the
imaginary part of the refractive index. It also enables to study the eect of the product L
on the internal optical eld along the cavity. This key result is shown in Fig. 9.5. The
solution of equation 9.2.25, using the appropriate boundary condition for a device spanning
symmetrically between L/2 to L/2 are given by the solution of the transcendental equation:
=
i
sinh L
(9.2.28)
where the parameter is given as a function of the gain and detuning as:
2
=
_
2
+ ( i)
2
(9.2.29)
Figure 9.6: Distribution of the optical eld in the cavity of the laser for an overcoupled laser
with a L = 4.7. This overcoupling enables very low mirror losses but at the cost of a lower
eciency.
Once equation 9.2.28 is solved, the functions R(z) and S(z) are given by :
R(z) = sinh (z +L/2) (9.2.30)
S(z) = sinh (z L/2) (9.2.31)
The solution for bandgap edge mode of a 2mm long laser operating at = 10.2m with
n
1
= 7.5 10
3
9.2.3 Fabrication geometries
Surface grating
The rst distributed feedback quantum cascade lasers were fabricated by etching the grating
directly onto the surface of the device onto a waveguide specially designed with a thin
AlInAs cladding. The interaction between the mode and the surface plasmon propagating
at the semiconductor-contact interface enabled a signicant modication of both refractive
index and loss and did therefore yield complex coupled DFBs. Single mode operation was
achieved at both 5.7 and 7.2 m. The modulation of the eective index can be achieved by
the eective index technique, in which the eective index of a propagating mode is computed
in both regions of the grating (i.e. in the etched and unetched parts). These number are
then used in a one-dimentional simulation of the structure as described above. For the
structure shown in Fig. ??, a refractive index step of n = 7 10
3
+ 3 10
4
is found
for an etch depth of 0.5m. The distributed feedback laser is predominantly index coupled
with a non-negligible loss component. Note also the fact that the coupling strength has a
superlinear dependence on the etch depth. However, for relatively deep gratings, eective
index techniques are not accurate and would tend to overestimate the losses. By using deep
Figure 9.7: a) Scanning electron micrograph of a distributed feedback quantum cascade laser
with a grating on the surface. b) Computed eective index, losses and mode prole of the
waveguide mode for three values of the etch depth: 0, 0.3 and 0.5m.
Figure 9.8: a) Schematic cut through a waveguide design that uses air as the top cladding
and a lateral injection through a doped InGaAs layer. b) Scanning electron micrograph of a
nished device.
etch with straight sidewalls, achieved with a dry etching technique, as well as a carefully
calibrated geometry, very good single mode operation was achieved for devices based on both
GaAs ?? and InP-based ??.
An other successful technique has been to use a device where air is used as the top cladding
layer and the current injection proceeds laterally through a doped (n = 210
18
cm
3
) InGaAs
conducting layer which is simultaneously the layer host to the grating. A schematic drawing
of the waveguide together with a scanning electron micrograph of a processed device are
shown in Fig. 9.9. Devices fabricated in this way exhibit a number of advantages:
The fabrication is relatively simple and does not require regrowth steps.
The grating is placed close to the peak of the guided mode, uses the large refractive
index step between the semiconductor material and air and is therefore ecient.
As the grating is etched in the doped material, a complex coupling is naturally achieved.
Figure 9.9: a)Light and bias versus current characteristic of distributed feedback quantum
cascade laser with a air/semiconductor top cladding and a lateral injection. b) Representative
spectra of the device at various operation temperatures.
The lateral losses induced by the contacts will help maintain fundamental transverse
mode even for wide devices.
Relatively large single mode peak powers (1W) could be achieved.
The top surface of the grating could be used as an access to the optical eld circulating
inside the device for intracavity detection.
However the relatively large width of the lasers, that enables the large peak powers, is also
responsible for the high drive currents of these devices. The light-versus current, as well as
some representative spectra of such a device are shown in Fig. ??. Single mode peak powers
as high as 0.8W and operation up to 120C were achieved [43].
Buried grating
As for the telecom devices, a very eective architecture is to etch the grating in the InGaAs
guiding layer above the active region, followed by an InP regrowth. An atomic force mi-
croscopy of a grating prior to the etching step, as well as a scanning electron microscope of
the active region of such a device is shown in Fig. 9.10. In such a device, the refractive index
step between InGaAs and InP ( n
etch
0.4) is close to the maximum of the waveguide.
Actually, using the variational approach, the change in eective index n
eff
as a function
of etching depth w can be immediately estimated:
n
eff
= n
etch
wI
norm
(z
grating
) (9.2.32)
where I
norm
(z
grating
) is the normalized intensity of the mode in the waveguide at the location
of the grating, i.e. satisfying
_

I
norm
(z)dz = 1. (9.2.33)
This approach is veried by an eective index computation on an example shown in g-
Figure 9.10: a) Atomic force microscopy of a buried grating prior to the regrowth step.
b) Scanning electron micrograph of the active region of a device after the regrowth was
performed.
Figure 9.11: a) Computed mode for a 10m wavelength with (black line) and without (blue
line) 200nm thick grating. The computed change of refractive index is n = 1.5 10
2
while the losses are not aected. c) Real part of the mode eective index as a function of
etched depth. As expected from equation 9.2.32, the relationship is linear.
Figure 9.12: Single mode optical power versus injected current for a distributed feedback
quantum cascade laser combining a buried grating and a buried heterostructure waveguide.
.
ure 9.10, where the mode prole and eective index are shown as a function of etch depth.
As expected from equation 9.2.32, a linear relationship is found between the eective index
and the etch depth, with the slope of this characteristic
dn
eff
dw
= 7.4 10
2
is reasonably
close to the one (6.8 10
2
) found by Equ. 9.2.32 assuming a average normalized intensity
at the interface (from Fig. 9.11) I
norm
= 0.17 .
Globally, the buried grating technology allows the fabrication of distributed feedback quan-
tum cascade lasers with the highest performance levels, at least in terms of low dissipation
and continuous wave operation. This is specially true with this technology is combined with
a buried heterostructure waveguide. Example of performance achieved with devices based
on this technology are shown in Fig. 9.12 and 9.13, demonstrating true uncooled operation
with tens of miliwatts of single mode operating power.
9.2.4 Tuning
Temperature
The operation frequency of a distributed feedback quantum cascade laser is xed by the
Bragg condition:
B
(T) = 2n
eff
(T)(T) (9.2.34)
As a result the tuning with temperature has two terms, and can be conveniently written as:
tun
=
1
B
d
B
dT
=
1
n
eff
dn
eff
dT
+
1
d
dT
(9.2.35)
Experimentally, this tuning coecient is
tun
= (6.5 9) 10
5
and is dominated by the
temperature tuning of the refractive index, as the second term is the thermal expansion of
InP and is an order of magnitude smaller (4.610
6
). As shown in table 9.1, this coecient
Figure 9.13: Temperature tuning of the a buried grating device operating in continuous wave
above room temperature.
Reference
tun
(m) Temperature (K)
[18] 6.8 10
5
7.8 220
[18] 6.5 10
5
5.3 220
[44] 6.5 10
5
10.16 200
[43] 7.8 10
5
5.3 330
[45] 7.1 10
5
4.9 260
[46] 8.9 10
5
7.8 370
unpublished
Table 9.1: Measured temperature tuning coecients of distributed feedback quantum cas-
cade lasers, as reported in various references. The tuning coecient is sensitive to tempera-
ture but in a very good approximation independent on the wavelength
is wavelength independent but increases with temperature. It is the reason why the tuning
coecient of continuous wave devices is larger than the one of pulsed ones at same holder
temperature, as the active region temperature is signicantly increased by self-heating in
devices that are operating in continuous wave.
Current
The same temperature coecient is also responsible for the tuning of the devices with injected
current. For slow variations of the thermal load in the device, the average temperature of
the waveguide T
wav
is related to the one of the holder by a thermal resistance:
T
wav
= T
sub
+R
th
UI. (9.2.36)
Note that the thermal resistance dened by the above equation is slightly lower than the
one dened by the similar equation for the active region 8.8.25, as the optical mode spreads
Figure 9.14: Mode position of a distributed feedback quantum cascade laser as a function of
both the temperature of the submount and the electrical power dissipated.(From ??)
over the claddings that are cooler thant the active region. As a result, the tuning coecient
can therefore be written as:
1
d
dP
elec
=
1
d
dT
dT
dP
elec
=
tun
R
th
. (9.2.37)
As expected, a strong tuning is achieved for a device with a large thermal resistance R
th
.
In short, a device that will provide a large current tuning exhibits a equally large dierence
between the active region and holder temperatures. In this respect, the opimization for max-
imum tuning is opposite to the one aiming at reaching the largest continuous wave operation
temperature. An example of tuning with both submount temperature and dissipated power
Dynamical behavior and linewidth
During short pulses current pulses, the waveguide temperature of a distributed feedback
quantum cascade laser will undergo a change in temperature and, as a result, induce a chirp
of the ouput frequency with time. The example of such a behavior is shown in Fig. ??
where a series of time-resolved spectra of a distributed feedback quantum cascade laser were
performed on a device driven by a 100ns long pulse at room temperature. The spectra were
aquired using a fast, room temperature mercury cadnium telluride detector and a gated
averager with a 3ns long gate. Because for such short time scales, the heat has no time
to diuse out of the active region, the thermal behavior of the device is dominated by the
specic heat of the active region and therefore the mode will shift linearly with time. This
is indeed the behavior observed in Fig. 9.15. The device remains single mode but tunes over
2cm
1
during the 100ns long pulse. This rapid chirp enables an intra-pulse spectroscopy
technique in which absorption features are detected using the very fast chirp of the laser
across the gas absorption line.
Figure 9.15: Single mode spectra of a distributed feedback quantum cascade laser as a
function of time during a current pulse. The chirp is thermally driven. (From [18])
As a result of this thermal chirp, the linewidth of a distributed feedback laser is expected
to grow linearly with pulse length. This is exactly the behavior observed experimentally, as
shown in Fig. ??. For this device driven at 1.25x the threshold, the linewidth is found to
grow linearly with time at a rate
d(1/)
dt
= 0.0075cm
1
ns
1
. Obviously, for very short pulses,
the linewidth is expected to grow again because of the time-bandwidth uncertainty relations.
Those can be expressed for the FWHM of the intensity of both the pulse and its Fourier
transform in frequency units as = 0.44 for a Gaussian pulse and = 0.88 for
a rectangular pulse. In a simple approach, the two contribution are simple summed and,
assuming rectangular pulses:
(
1
) = 0.88
1
c
+
d(1/)
dt
. (9.2.38)
The expression given by equation 9.2.38 ts relatively well the experimental data show in
Fig. 9.16, and predict a minimum linewdith of about 0.02cm
1
(assuming Gaussian pulses).
Distributed feedback quantum cascade lasers driven by short pulses and tuned by a sub-
threshold ramp have been used to perform high resolution spectroscopy, an approach pio-
neered by the group of Aerodyne research. A measurement of the laser linewidth can be
obtained from a tting the measured absorption of a known absorption line. Using this
technique, narrow linewidth of 0.012cm
1
have been measured on a similar device as the one
presented in gure 9.16. In fact, measurements on devices operating dierent wavelength
range have shown a linewidth as narrow as 0.0072cm
1
??. An example of such a measure-
ment is reported in Fig 9.17, where the measured linewidth is reported as function of the
voltage on the pulse, the latter acting as a surrogate for the pulse length. These results,
achieved by biasing the laser just above threshold, show clearly that under these condi-
tions, linewidth narrower that the ones predicted by equation 9.2.38 can be experimentally
Figure 9.16: Measured linewidth as a function of pulse length. The point labeled Aerodyne
is the result of a spectral t of the absorption line of NH
3
at 967cm
1
that did yield a
linewidth of 0.012cm
1
. Adapted from [19]
achieved. In addition, the pulse lengths for which the narrowest linewidth are predicted by
equation 9.2.38 are of 1.2ns, signicantly shorter than the ones available by the electronics
used in these experiments ( > 5ns). Clearly, the expression 9.2.38 is much too crude to
describe the experimental data in the regime of short pulses. A more corret theory should
include the build-up of the laser pulse from the spontanous emission or the thermal noise.
Assuming a typical mirror loss and a spontaneous of 5cm
1
, the buildup time of the photon
density in the cavity is about 20ps.
Cavity pulling
An important dierence between distributed feedback quantum cascade laser and their inter-
band counterparts is the almost total absence of carrier density-driven changes of refractive
index. In fact, the requirement of preserving global neutrality in a unipolar device forces the
number of electrons to remain constant. As a result, the only changes in refractive index that
can be expected are related to shifts of the carrier distribution or transition energies as a
function of applied voltage or temperature. In particular, a blue shift of the gain curve with
applied bias is expected to induce a concomitant blue shift of the mode. This phenomenon,
called cavity pulling in the literature of gas laser, has been observed in single mode quan-
tum cascade laser operating in the terahertz. It has never been observed in mid-infrared
devices for two reasons. First the large ratio of upper to lower state lifetimes necessary to
achieve high performance devices implies that the upper state density is pratically locked
to a constant value above threshold. As a result, the dierential resistance of the device de-
crease abruptly and the Stark tuning of the transition above threshold is minimum. Secondly
high performance operation is achieved by employing low loss waveguides. The concomitant
change in refractive index is therefore small. It is however not impossible to design a device
where such a tuning mechanism is designed in.
9.3. 123
Figure 9.17: Measured half width at half maximum of the emission spectra of a quantum
cascade laser as a function of pulse length. The linewidth is obtained by tting a vibrational
molecular line, and the pulse length is a increasing function of the voltage on the pulser,
indicated in the horizontal axis. Adapted from [?]
9.3
9.4 External cavities
Chapter 10
Quantum Cascade lasers III:
applications
For the applications, the key feature of quantum cascade lasers is the frequency range in
which they do emit. At this time, quantum cascade lasers emitting at wavelengths as short
as 2.9 m and as long as 250m (360m with magnetic eld) have been fabricated. By
covering most of the mid- and far-infrared spectrum, they enable applications where laser
emission, combined with small size and ease of operation is necessary. Compared to gas
lasers, they have the very nice
125
126 CHAPTER 10. QUANTUM CASCADE LASERS III: APPLICATIONS
10.1 Energy deposition
10.1.1 Wallplug eciency
10.1.2 High power operation
10.2 Telecommunications
10.2.1 high frequency modulation
10.2.2 Telecommunications experiments
10.3 Gas sensing
10.3.1 Gas absorption lines: pressure and temperature depden-
dence
10.3.2 General system consideration
10.3.3 multipass cells and fringe noise
10.3.4 interpulse modulation
10.3.5 intrapulse modulation
10.3.6 Wavelength modulation
10.3.7 Photoacoustic
10.4 Liquid sensing
Chapter 11
Quantum Cascade lasers IV:
Transport models
11.1 Classical models
11.1.1 Drude
11.1.2 Esaki-Tsu
11.2 Transport as a result of intersubband transitions
11.2.1 Formalism
11.2.2 Diagonal transition lasers
11.2.3 With charge conservation and the optical eld
11.2.4 Computing the electron distribution in the subband
The value of the tunneling time
esc
is dicult to estimate very accurately. Experimental
values obtained in QC structures yield
esc
= 2 3ps [47, 48], much larger than
21
and the
value (0.2ps) computed assuming the injection region behaves as a continuum.
A model of the room temperature electron kinetics in the active region, that included optical
phonon emission and absorption from all points in k-space (see Fig. ??) of the active region
subbands, demonstrated that this long escape time creates an eective bottleneck eect. In
this model, the electrons are interacting with a bath of optical phonons in thermal equilibrium
at a temperature T. Since we are neglecting electron-electron interactions, the electron may
gain or loose energy only in quanta of the optical phonon energy. Fig. ?? schematically
shows this model in a case where the transition energy corresponds to the emission of 4
optical phonons.
We label the available states by two indices (i, j) where i labels the state number and j the
position in the ladder of energies. j = 0 corresponds to the energy of the electron injection,
127
128 CHAPTER 11. QUANTUM CASCADE LASERS IV: TRANSPORT MODELS
j = 1 to the emission of one optical phonon, and so on. For an active region containing N
states (with N being the upper laser state), the rate equation for the lower states (i, j) (with
i < N) writes:
dn
i,j
dt
=
N1
k=1
(
n
k,j1
e
ik
+
n
k,j+1
a
ik
) +(j, 1)
n
N,0
e
Ni
+(j, 1)
n
N,0
a
Ni
n
i,j
_
N1
k=1
(
1
a
ik
+
1
e
ik
) +
1
esc
+(j, 1)
1
a
Ni
+(j, 1)
1
e
Ni
(11.2.1)
We assume that a cool electron distribution is injected in the upper laser level. Therefore
all the electrons are concentrated in j=0 in the upper level N. The rate equation for that
state writes:
dn
N,0
dt
=
J
q
0
+
N1
k=1
n
k,1
a
N,k
n
N,0
N1
k=1
(
1
e
N,k
+
1
a
N,k
) (11.2.2)
Solving equations 11.2.1 and 11.2.2 for a given injection current and temperature yields the
densities at each relevant point of the k-space. The result of such a computation, is shown
in Fig ??. This model nicely demonstrates the bottleneck eect of the long tunneling
escape time
esc
. At room temperature (T = 300K), the ratio of lower to upper state
population is n
2
/n
3
= 0.59 (see Fig. ??(a)), a value much larger than the one anticipated by
considering the lifetimes:
2
/
32
= 0.2. Even at cryogenic temperature (T = 77K), this ratio
n
2
/n
3
= 0.26 (see Fig. ??(b)) is larger than the one predicted from the lifetimes
2
/
32
= 0.1.
As expected, if the escape lifetime is set to a short value
esc
= 0.2ps, the ratio of population
is very small even at room temperature n
2
/n
3
= 0.136 (see Fig. ??(c)).
The eect of this bottleneck can be expressed as an eective lower state lifetime
eff
2
such
that n
2
/n
3
=
eff
2
/
32
. At T = 300K, we nd a value
eff
2
= 0.8ps.
The situation for the electron extraction is signicantly dierent in superlattice active region
designs. In those designs, the injection eciency in the lower state
2
is made very small by
a phase space argument: the n = 2 state is only one of a manifold of lower states forming the
lower miniband and thus injection into this very state is unlikely. Similarly,
eff
2
is expected
to be very small (
2

= 0.2ps) through very ecient intrasubband scattering. In addition,
no bottleneck problems are foreseen, since transport in the lower miniband provides a very
ecient electron extraction. However,
3
is not expected to be as large as in the 3QW case
since electron injection is not achieved through resonant tunneling [49].
T
11.3. RESONANT TUNNELING 129
11.3 Resonant tunneling
11.3.1 Problem of coherence
11.3.2 Density matrix approach: Kazarinov, Willenberg
11.3.3 second-order current
11.3.4 Full model
11.4 Non-equilibrium Greens function theory
130 CHAPTER 11. QUANTUM CASCADE LASERS IV: TRANSPORT MODELS
Chapter 12
Quantum Cascade Lasers V: Device
properties and characterization
12.1 Electrical charateristics
12.2 Gain and loss measurements
12.3 Electron temperature measurements: hot elec-
trons and hot phonons.
12.4 Active region non-linearities
12.5 Conned phonons
131
132
CHAPTER 12. QUANTUM CASCADE LASERS V: DEVICE PROPERTIES AND
CHARACTERIZATION
Chapter 13
Intersubband processes II (collective
excitations)
13.1 Depolarization shift. (Dielectric model and con-
nection to the oscillator model)
13.2 Intersubband plasmon, linewidth versus charge
density
13.3 Raman spectra
13.4 Optical microcavities and cavity polaritons (Alessan-
dro, Carlo, ..)
133
134 CHAPTER 13. INTERSUBBAND PROCESSES II (COLLECTIVE EXCITATIONS)
Chapter 14
Interlevel transition in Quantum dots
14.1 Fabrication: self-assembled Stransky Krastanov
growth
14.2 Interband spectroscopy
14.3 Intraband spectroscopy techniques
14.4 The intersubband polariton in quantum dots.
(Ferreira, Luke Wilson, ...)
135
136 CHAPTER 14. INTERLEVEL TRANSITION IN QUANTUM DOTS
Chapter 15
Intersubband non-linearities and
non-linear devices
15.1 Non-linear intersubband polarization (Jacob Kurghin,
Ghilad Almoghi, Carlo Sirtori,)
15.2 Second harmonic generation structures (step well,
coupled wells)
15.3 Trade-o between power and non-linear power;
phase matching condition
15.4 Non-linear generation of THz (Carlo)
15.5 Quantum Cascade lasers with non-linear elements:
mid-IR (Claire) and THz (Belkin)
15.6 NIR-THz mixing in QCL (Carlo)
137
138 CHAPTER 15. INTERSUBBAND NON-LINEARITIES AND NON-LINEAR DEVICES
Chapter 16
Appendix: Designs
16.1 Designs
16.1.1 3QW 10.3um diagonal
16.1.2 3QW 5.3um vertical
applied electric eld of 7.610
4
V/cm. The layer sequence of one period of the Al
0.48
In
0.52
As/Ga
0.47
In
0.53
structure, in nanometer, left to right and starting from the injection barrier is, 5.0/0.9/1.5/4.7/2.2/4.0/3.0
where the In
0.52
Al
0.48
As layers are in bold, In
0.53
Ga
0.47
As in Roman and underlined number
correspond to doped layer with Si to N
d
= 2 10
17
cm
3
16.1.3 Two-phonon resonance at 9um
The layer sequence of the structure, in nanometers, and starting from the injection barrier, is
as follows: 4.0/1.9/0.7/5.8/0.9/5.7/0.9/5.0/2.2/3.4/ 1.4/3.3/1.3/3.2/1.5/3.1/1.9/3.0/2.3/2.9/2.5/2.9/
nm. In
0.52
Al
0.48
As barrier layers are in bold, In
0.53
Ga
0.47
As well layers are in roman, and n-
doped layers (Si 4 10
17
cm
3
) are underlined
16.1.4 Bound-to-continuum at 9um
The layer sequence of one period of structure, in nanometers, left to right and starting from
the injection barrier is 4.0/2.0/0.7/6.0/0.9/5.9/1.0/5.2/1.3/3.8/1.4/3.5/ 1.5/3.3/1.6/3.1/1.9/3.1/2.3
where In
0.52
Al
0.48
0.53
Ga
0.47
As in Roman and underlined number cor-
respond to doped layer with Si to N
d
= 3 10
17
cm
3
.
16.1.5 Bound to continuum at 16um
The layer sequence of one period of structure, in nanometers, left to right and start-
ing from the injection barrier is 3.3/3.2/0.5/6.5/0.6/6.6/0.7/6.3/ 0.8/5.8/1.0/4.6/1.2/
4.4/1.4/4.4/1.7/4.2/2.0/4.1/2.2/4.0 where In
0.52
Al
0.48
0.53
Ga
0.47
As
in Roman and underlined number correspond doped regions Si, 3 10
17
cm
3
139
140 CHAPTER 16. APPENDIX: DESIGNS
16.1.6 Broad bound-to-continuum
The layer sequence of one period, in nanometers, starting from the injection barrier is
3.9/2.2/0.8/6/ 0.9/5.9/1/5.2/1.3/4.3/1.4/3.8/1.5/3.6/1.6/ 3.4/1.9/3.3/2.3/3.2/2.5/3.2/2.9/3.1
where
In
0.52
Al
0.48
0.53
Ga
0.47
As in roman and the doped layers (Si 2.310
17
cm
3
) are underlined.
Bibliography
[1] T. Unuma, M. Yoshita, T. Noda, H. Sakaki, and H. Akiyama, J. Appl. Phys. 93, 1586
(2003).
[2] J. Smet, C. Fonstad, and Q. Hu, J. Appl. Phys. 79, 9305 (1996).
[3] J. Faist, F. Capasso, C. Sirtori, D. Sivco, J. Baillargeon, A. Hutchinson, S. Chu, and
A. Cho, Appl. Phys. Lett. 68, 3680 (1996).
[4] C. Gmachl, F. Capasso, A. Tredicucci, D. Sivco, R. Kohler, A. Hutchinson, and A. Cho,
IEEE J. Select. Topics Quantum Electron. 5, 808 (1999).
[5] J. Faist, M. Beck, T. Aellen, and E. Gini, Appl. Phys. Lett. 78, 147 (2001).
[6] J. Yu, S. Darvish, A. Evans, J. Nguyen, S. Slivken, and M. Razeghi, Appl. Phys. Lett.
88, 041111 (2006).
[7] J. Yu, A. Evans, S. Slivken, S. Darvish, and M. Razeghi, IEEE Photon. Technol. Lett.
17, 1154 (2005).
[8] A. Evans, J. Yu, S. Slivken, and M. Razeghi, Appl. Phys. Lett. 85, 2166 (2004).
[9] A. Evans, J. Nguyen, S. Slivken, J. Yu, S. Darvish, and M. Razeghi, Appl. Phys. Lett.
88, 051105 (2006).
[10] L. Diehl, D. Bour, S. Corzine, J. Zhu, G. Hoer, M. Loncar, M. Troccoli, and F. Capasso,
Appl. Phys. Lett. 89, 081101 (2006).
[11] A. Evans, J. Yu, J. David, L. Doris, K. Mi, S. Slivken, and M. Razeghi, Appl. Phys.
Lett. 84, 314 (2004).
[12] L. Diehl, D. Bour, S. Corzine, J. Zhu, G. Hoer, M. Loncar, M. Troccoli, and F. Capasso,
Appl. Phys. Lett. 88, 201115 (2006).
[13] M. Rochat, D. Hofstetter, M. Beck, and J. Faist, Appl. Phys. Lett. 79, 4271 (2001).
[14] T. Gresch, M. Giovannini, N. Hoyler, and J. Faist, IEEE Photon. Technol. Lett. 18,
544 (2006).
141
142 BIBLIOGRAPHY
[15] T. Aellen, M. Beck, N. Hoyler, M. Giovannini, and J. Faist, J. Appl. Phys. 100, 043101
(2006).
[16] J. Nguyen, J. Yu, A. Evans, S. Slivken, and M. Razeghi, Appl. Phys. Lett. 89, 111113
(2006).
[17] H. Page, C. Becker, A. Robertson, G. Glastre, V. Ortiz, and C. Sirtori, Appl. Phys.
Lett. 75, 3529 (2001).
[18] J. Faist, C. Gmachl, F. Capasso, C. Sirtori, D. Sivco, J. Baillargeon, and A. Cho, Appl.
Phys. Lett. 70, 2670 (1997).
[19] D. Hofstetter, M. Beck, J. Faist, M. Nagele, and M. Sigrist, Opt. Lett. 26, 887 (2001).
[20] S. Tsujino, A. Borak, E. M uller, M. Scheinert, C. Falub, H. Sigg, D. Gr utzmacher,
M. Giovannini, and J. Faist, Appl. Phys. Lett. 86, 062113 (2005).
[21] P. Price, Ann. Phys. (San Diego) 133, 217 (1981).
[22] R. Ferreira and G. Bastard, Phys. Rev. B 40, 1074 (1989).
[23] S. Borenstain and J. Katz, Phys. Rev. B 39, 10852 (1989).
[24] P. Hyldgaard and J. Wilkins, Phys. Rev. B 53, 6889 (1996).
[25] D. Oberli, D. Wake, M. Klein, J. Klem, T. Henderson, and H. Morkoc, Phys. Rev. Lett.
59, 696 (1987).
[26] A. Seilmeier, H. Hubner, G. Abstreiter, G. Weimann, and W. Schlapp, Phys. Rev. Lett.
59, 1345 (1987).
[27] M. Tatham, J. Ryan, and C. Foxon, Phys. Rev. Lett. 63, 1637 (1989).
[28] J. Faist, F. Capasso, L. Pfeier, and K. West, Appl. Phys. Lett. 64, 872 (1994).
[29] S. Lutgen, R. Kaindl, M. Woerner, T. Elsaesser, A. A.Hase, and H. Kunzel, Phys. Rev.
B 54, R17343 (1996).
[30] M. Hartig, S. Haacke, P. Selbmann, B. Deveaud, R. Taylor, and L. Rota, Phys. Rev.
Lett. 80, 1940 (1998).
[31] K. Choi, B. Levine, R. Malik, J. Walker, and C. Bethea, Phys. Rev. B 35, 4172 (1987).
[32] J. Faist, F. Capasso, C. Sirtori, D. Sivco, A. Hutchinson, S. Chu, and A. Cho, Appl.
Phys. Lett. 65, 94 (1994).
[33] R. Kazarinov and R. Suris, Sov. Phys. Semicond. 6, 120 (1972).
[34] J. Faist, F. Capasso, C. Sirtori, D. Sivco, A. Hutchinson, and A. Cho, Nature 387, 777
(1997).
BIBLIOGRAPHY 143
[35] J. Faist, F. Capasso, C. Sirtori, D. Sivco, A. Hutchinson, and A. Cho, Appl. Phys. Lett.
66, 538 (1995).
[36] J. Faist, F. Capasso, D. Sivco, C. Sirtori, A. Hutchinson, and A. Cho, Science 264, 553
(1994).
[37] M. Troccoli, G. Scamarcio, V. Spagnolo, A. Tredicucci, C. Gmachl, F. Capasso, D. Sivco,
A. Cho, and M. Striccoli, Appl. Phys. Lett. 77, 1088 (2000).
[38] P. Kruck, H. Page, C. Sirtori, S. Barbieri, M. Stellmacher, and J. Nagle, Appl. Phys.
Lett. 76, 3340 (2000).
[39] T. Unuma, T. Takahashi, T. Noda, M. Yoshita, H. Sakaki, M. Baba, and H. Akiyama,
Appl. Phys. Lett. 78, 3448 (2001).
[40] K. Campman, H. Schmidt, A. Imamoglu, and A. Gossard, Appl. Phys. Lett. 69, 2554
(1996).
[41] C. Walle, Phys. Rev. B 39, 1871 (1989).
[42] H. Kogelnik and C. Shank, J. Appl. Phys. 43, 2327 (1972).
[43] D. Hofstetter, M. Beck, T. Aellen, and J. Faist, Appl. Phys. Lett. 78, 396 (2001).
[44] D. Hofstetter, J. Faist, M. Beck, A. Muller, and U. Oesterle, Appl. Phys. Lett. 75, 665
(1999).
[45] C. Gmachl, A. Straub, R. Colombelli, F. Capasso, D. Sivco, A. Sergent, and A. Cho,
IEEE J. Quantum Electron. 38, 569 (2002).
[46] A. Wittmann, M. Giovannini, J. Faist, L. Hvozdara, S. Blaser, D. Hofstetter, and
E. Gini, Appl. Phys. Lett. 89, 141116 (2006).
[47] J. Faist, F. Capasso, C. Sirtori, D. Sivco, A. Hutchinson, M. Hybertsen, and A. Cho,
Phys. Rev. Lett. 76, 411 (1996).
[48] S. Blaser, L. Diehl, M. Beck, J. Faist, U. Oesterle, J. Xu, S. Barbieri, and F. Beltram,
IEEE J. Quantum Electron. 37, 448 (2001).
[49] F. Capasso, A. Tredicucci, C. Gmachl, D. Sivco, A. Hutchinson, A. Cho, and G. Sca-
marcio, IEEE J. Select. Topics Quantum Electron. 5, 792 (1999).

QCL Lecture Long

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

QCL Lecture Long

Загружено:

Авторское право:

Доступные форматы

Intersubband Optoelectronics

AInGaAs quantum well and a 9

k) may be splitted into a k-independent

A) 2.55 6.14 5.67 11.5 21.5 39.5

(z). In that case, one shows that the proper boundary

and the trenscendental equation becomes:

100A30A Superlattice dispersion

instead of the bare electron mass m

. In the k p approach of Kane,

for transition between the rst two

for the case of the innite quantum

for the parabolic well. As a result, the spontaneous emission rate is

- dependence form factor. The Fermi energy of the excited state E

(V E)/ is the electrons wavevector L

AInGaAs quantum well and a 9

is a key parameter determining the wallplug e-

is achieved in an active region where transport is very fast in the injec-

demonstrates the very strong wavelength dependence of the

is the carriers eective mass); taking similarely the

A). In Equ. ?? the indices w, b refer to

. For increasing indium

is also referred as the group index.

of the mode as:

Вам также может понравиться