Вы находитесь на странице: 1из 16

Self-Consistent Monte Carlo Particle Transport with Model Quantum Tunneling Dynamics: Application to the Intrinsic Bistability of a Symmetric

Double-Barrier Structure
Published in Journal of Applied Physics 72, 5975 (1992)

R. E. Salvino and F. A. Buot Code 6864, Naval Research Laboratory Washington DC 20375

Abstract The intrinsic bistability in a symmetric resonant tunneling device (RTD) is simulated by the ensemble particle Monte Carlo technique, coupled with a simple model of the space- and time-dependent particle quantum dynamics inside the double-barrier region of the RTD. This model particle quantum dynamics is based upon the phase-time delay, which is obtained from a piecewise-linear-potential Airy function approach to the calculation of the transmission amplitude. An unambiguous hysteresis in the negative dierential resistance (NDR) region of the current-voltage (I V ) characteristic is observed for a symmetric AlGaAs/GaAs double-barrier structure. The dynamical accumulation of carriers in the well is seen to be the cause of this marked bistability/hysteresis. However, the plateau-like features of the I V curve are not resolved, although oscillations in the quantum well carrier density in the NDR are prominent. This article strongly suggests that a more accurate treatment of the space- and time-dependent particle quantum dynamics across the RTD is of paramount importance.

Introduction

Self-consistent ensemble particle Monte Carlo (MC) simulations incorporating space- and time-dependent quantum transport are expected to be a very useful tool for simulating the performance and reliability aspects of realistic and multidimensional quantum-based devices. As a rst step in 1

incorporating the space- and time-dependent particle dynamics of quantum transport in MC simulations, we have simulated the intrinsic bistability of a symmetric resonant tunneling device (RTD) by including a model dynamics for tunneling particles within the framework of traditional MC simulations. This diers from previous work [13] in two important respects: (1) the use of particle dynamics rather than probabilistic insertion of particles into, and extraction of particles from, the quantum region and (2) the use of a more accurate piecewise-linear-potential Airy function solution [46] to the time-independent Schrodinger equation rather than a piecewise-constantpotential plane wave approach. Our goal is to ll the need for a dynamical description of particle transport in MC for quantum tunneling structures. In developing the model of the particle quantum dynamics in the MC approach, we focus on a dynamical treatment of motion across the RTD to reproduce the plateau-like behavior in the current-voltage (I V ) curve found in more fundamental Wigner distribution function simulations [7, 8]. The phenomenon of bistability has been characterized as intrinsic and extrinsic in nature [914]. Intrinsic bistability is interpreted as due to charge accumulation in the resonant tunneling structure (the quantum well of the double-barrier system) and reveals itself as an hysteresis in the negative dierential resistance (NDR) region of the I V curve. This phenomenon is always observed for asymmetric double-barrier structures [14]; extrinsic bistability is interpreted as due to current oscillations in the NDR induced by the parasitic external circuit coupled to the RTD and is characterized by additional plateau-like features in the I-V curve. This behavior seems to always occur for symmetric double-barrier structures only [14]. Recent Wigner distribution function simulations [7,8] have shown, however, that current oscillations arise in the symmetric RTD in the NDR region of the I V curve that apparently have nothing to do with an external circuit. The role of the external circuit here is to simply maintain a constant voltage source, that is, the voltage-source circuit essentially shunts the high-frequency current. Dynamical bistability is identied at voltage biases for which the system initiates the attempt to go from the high current state to the low-current state (forward sweep in bias) or from the low-current state to the high-current state (reverse sweep in bias). It is likely that high-frequency oscillations occur in symmetric double-barrier structures only, since a strongly correlated transfer across the quantum well is required to maintain periodic oscillations. This necessitates that the width of each barrier be identical. The present MC simulation with the simple model particle quantum dynamics across the double-barrier structure captures the carrier accumulation aspect of bistability. It also captures, but does not fully resolve, the dynam2

ical oscillatory character of bistability that is also present in a symmetric RTD.

II

Computational Model

A schematic diagram of the resonant tunneling structure of a symmetric double-barrier RTD is shown in Figure 1. This structure has two 3 nm barriers of Alx Ga1x As, a 5 nm well of GaAs, and two 3 nm spacer layers of GaAs, all of which are completely undoped. The cathode and anode regions are GaAs with an impurity doping level of 1018 /cm3 . The entire system is 62 nm in length. The barrier heights are taken to be 0.3 eV and the temperature of the system is taken to be 77 K. The chemical potential is determined by inverting the density-chemical potential relationship,

n = nc F1/2 ( ) + 2.5T F3/2 ( ) , nc = (2gs / )(m kB T /2 2 )3/2 , =


c.

F ( ) is the Fermi-Dirac integral of order , is the nonparabolicity factor, T is the temperature in eV, gs = (2s + 1) is the spin degeneracy, m is the eective mass, kB is Boltzmanns constant, T is the temperature in degrees K, is the chemical potential, and c is the conduction-band minimum. The system is divided into mesh cells for the Poisson solver, each of which is 0.5 nm wide. Charge is assigned to the grid points using the nearest grid point (NGP) method. The doped regions initially contain 20 simulation particles randomly placed in each cell while the undoped regions are initially empty. The cold start begins at zero bias with 1800 simulation particles in the device. The system is allowed to evolve for a total time of 4 ps, 2000 time steps using a eld-adjusting time step of 2 fs. The nal conguration for a given bias is used as the initial conguration for the next bias. The maximum increment or decrement in bias is 0.02 V. The band structure is an eective-mass model with a nonparabolicity factor [16, 17]. A composite L/X valley constitutes the upper valley. It is assumed, based on the band structure of GaAs and AlGaAs, that only the lower valley sees the potential barrier of 0.3 eV, the upper L/X valley 3

Figure 1: Schematic diagram of the symmetric double-barrier resonant tunneling structure: 3 nm AlGaAs, 5 nm GaAs, and 3 nm AlGaAs surrounded by bulk GaAs layers.

seeing no barrier. The eective mass of the valley is 0.067 electron masses and the L/X eective mass is 0.35 electron masses. The nonparabolicity factor for the valley is 0.576/eV and is taken to be zero for the L/X valley. Traditional MC simulation [1620] describes the particle dynamics outside the quantum heterostructure. Freeight times are obtained from scattering rates in the usual way. The scattering mechanisms are [17, 21]: (1) acoustic phonon scattering treated in the elastic approximation, (2) polar optical phonon scattering, (3) equivalent valley phonon scattering, (4) nonequivalent valley phonon scattering, and (5) ionized impurity scattering treated in the Brooks-Herring approximation. The Poisson elds (electric potential and electric eld) are calculated self-consistently at every time step. The standard ohmic contact model at the electrodes is used [19]: if the cell adjacent to the contact is charge positive, then carriers are injected from a Fermi-Dirac source until the cell is charge neutral; if the cell is neutral or charge negative, then the cell is left alone. Deviation from the MC particle pushing algorithm occurs only when a particle crosses into the quantum heterostructure. The unique feature of our simulation is the replacement of the probabilistic insertion into, and extraction from, the quantum region [1,2] by a model 4

particle dynamics for the tunneling particles. The model dynamics is based upon the time delays, obtained from the phase of the complex transmission amplitude:

t = |t|ei T = |t| delay


2

(transmission amplitude),

(transmission coecient), = (tunneling time delay). E

The delay time, delay [2224], is added to the classical time of traversal, the distance traversed divided by the incident velocity, to obtain the total traversal time for the tunneling particle. This total traversal time, total , is then used to assign a velocity to the particle in the well by

vwell = lwell /total where lwell is the width of the well. The particle velocity in the barriers is assumed to be unaected by the tunneling time delay. Characteristic values for the transmission coecient, traversal times, and particle velocities in the quantum heterostructure as a function of the particle energy in the conduction band are shown in Figs. 2, 3, and 4, respectively. These curves give a pictorial representation of the model particle quantum dynamics used in the simulation. While the use of the phase time delay as a traversal time may still be controversial, it appears to capture some key features of the tunneling process [23, 24]. It is a reasonable starting point and it also has the added advantage of being relatively simple to implement. The use of a velocity obtained from the spatial gradient of the phase of the Airy function solutions (Bohm-like trajectories) is probably more accurate than the method presently employed here, although somewhat more computertime-intensive to implement, and is currently being explored. The philosophy of the calculation treats tunneling as a special scattering event. If the MC trajectory of a particle crosses into the double-barrier region, its freeight time is recomputed to be the time it takes the particle to reach the barrier edge. All particle dynamical attributes are updated with this recomputed time. The particle is assigned a transmission coecient and traversal times based upon the longitudinal component of the kinetic 5

Figure 2: A characteristic transmission coecient for a bias of 0.22 V. The rst resonant peak locates the lowest lying resonant state in the well. In our simulation, the tunneling particles are not energetic enough to sample the second broad peak in the transmission coecient.

energy. A test comparing a uniform random number to the transmission coecient is then performed to determine if the particle tunnels through the structure or not. If the particle is reected, the z component of the particle momentum is simply reversed and the reection is treated as an instantaneous event. If, on the other hand, the particle is transmitted, average (constant) velocities across the barriers and the well are assigned to the particle based upon the traversal times obtained from the time delays. It should be emphasized that the wavefunction is used only to calculate the transmission coecient and time delays, it is not used to assign particle positions as in previous work [13]: the model particle quantum dynamics fully determines the particle positions, and hence the particle density, in the quantum region. Scattering in the quantum region, the length of which is much less than the mean-free path, is assumed to be negligible. In our model dynamics, the average velocity in the well is smaller than the average velocity in the barrier on resonance, but is larger than the average velocity in the barrier o resonance for the range of bias simulated (Figure 4). This behavior is in qualitative agreement with the results of the Wigner-trajectory-average velocities in the barrier and well obtained from the Wigner distribution function approach [2527].

Figure 3: Characteristic traversal times for a bias of 0.22 V. The phase-time delay is incorporated into the well traversal time. The barrier traversal time is assumed to be unaected by the phase-time delay.

Figure 4: Characteristic average velocities in the barrier and in the well for a bias of 0.22 V. On resonance the particle moves more slowly in the well than in the barrier: o resonance the particle moves more quickly in the well than in the barrier.

The tunneling calculations are done self-consistently, updating the transmission amplitude at every time step along with the Poisson elds. The transmission amplitude calculation is based upon a piecewise-linear-potential Airy function approach to the time-independent Schrodinger equation [46]. The resonant tunneling structure (RTS) is divided into Airy mesh cells, which need not have any relation with the mesh cells for the Poisson solver. In each Airy mesh cell, the potential is tted to a linear form, V (z ) = V0 + F z and the exact Airy function solutions of the resulting Schrodinger equation are matched at the mesh boundaries. Nonparabolicity is not included in the transmission coecient test: E << 1 since the rst resonant peak lies below 0.1 eV for the structure investigated. The current is determined by the time derivative of the charge collected at the drain. Four methods are used: a two-point forward dierence, a twopoint central dierence, a three-point forward dierence, all averaged over the last 1000 time steps, and the slope of a linear least-squares t to the charge collected over the last 1000 time steps. There is little dierence in the calculations, except that the least-squares t values show somewhat less noise in the I V curve. The main features are similar for all current calculations, whether it includes the displacement current or only the particle current. The I V curve we present uses the linear least-squares t to the data, and also includes the displacement current.

III

Results

Our simulations display the two aspects of bistability in a symmetric RTD. Accumulation of charge in the well of the double-barrier structure and a dynamical oscillatory behavior in the NDR, although the oscillatory behavior is apparently not fully characterized at the source and drain. The charge accumulation in the well for the forward sweep and the lag in accumulation for the reverse sweep produce a marked hysteresis loop in the I V curve, measuring about 40 mV wide at its maximum extent (Figure 5). The unstable NDR region of the I V curve shows prominent carrier density oscillations in the double-barrier structure (Figure 6), and the accompanying oscillations in the location of the resonant-tunneling-transmission-peak energy with respect to the conduction-band minimum at the barrier edge of the emitter (Figure 7). This oscillatory behavior of the quantum well density, however, did not translate well into current oscillations at the electrodes: the suppression of current oscillations distorts the characteristic plateau region 8

Figure 5: The current-voltage characteristic for the symmetric double-barrier structure. The bistability and hysteresis are evident, measuring approximately 40 mV wide. Sustained density oscillations are observed in the 0.28 0.36 V range.

into a narrow cusp region (Figure 5). We believe this is due to numerical damping and expect the use of a higher order charge assignment and force determination scheme to improve the calculations. Oscillations in the carrier density in the quantum heterostructure are not observed outside the hysteresis loop, in agreement with the Wigner distribution function simulation results [7, 8]. Initially, the oscillations are observed as transients beginning at a bias around 0.26 V. As the bias approaches 0.32 V, the oscillations become long-lived and eventually remain throughout the duration of the run. These oscillations in the carrier density are maintained throughout the high-current region of the I V characteristic. At 0.36 V, the well discharges in less than 1 ps and the density in the well, maintaining a low value, loses its oscillatory character as does the left spacer (Fig. 6). It should be noted that the spacer and well oscillations are always out of phase, in agreement with the Wigner function calculations [7,8]. As mentioned, the static nonoscillatory behavior is characteristic of the I V regions outside of the hysteresis loop, for both low and high bias. The accumulation of charge in the quantum well is also reected in the dynamical behavior of the potential level in the well. For instance, at a bias of 0.356 V, the increasing carrier density in the well pulls the potential level up relative to the initial conguration (Figure 8). Increasing the bias

Figure 6: Spatially averaged carrier density in the left spacer region and in the well at forward bias sweep. The 0.356 V curves are for near peak current conditions, the 0.360 V curves are for low-current conditions. The time traces on the top refer to the left spacer layer, the time traces on the bottom refer to the quantum well.

Figure 7: First peak of the transmission coecient for the 0.22 V and the 0.356 V cases. Each initial curve is at t = 0, each nal curve is at t = 4 ps. The 0.22 V bias pushes the peak toward the conduction minimum of the emitter, the charge accumulation for the 0.356 V bias pulls the peak away from the conduction minimum towards Fermi level of emitter.

10

Figure 8: The conduction-band edge prole or potential prole for the 0.356 V case. Every third data point is shown. The initial prole is at t = 0, the nal prole is at t = 4 ps. The potential level in the well is pulled up by the charge accumulation in the well.

to 0.358 V shows little change. However, increasing to 0.360 V shows the bias discharges the well and the potential level is pushed down below the initial conguration (Figure 9). Comparison of the time-averaged potentials and densities shows the same eect. There is little dierence between the 0.356 and 0.358 V cases, but the 0.360 V time-averaged potential lies lower in energy than the other two since the time-averaged density in the well is smaller by about a factor of 4 (Figure 10). The steady-state well density lies lower by nearly two orders of magnitude. The behavior of the time-averaged carrier density stored in the well (Figure 11) is quite similar to the I V characteristic (Figure 5). This is strong evidence that the hysteresis in the I V curve is due to the accumulation of charge in the well in the present MC treatment. The eects discussed above are manifestations of the correlation of the Fermi level of the source and the resonant-energy level of the quantum well. This correlation is revealed by the location of the rst transmission-peak energy relative to the conduction-band edge of the emitter. The shifting of the peak of the transmission coecient gives an energy-parameter view of the tug of war between the applied bias and the charge accumulation in the well. The bias pushes the transmission peak toward the emitter conduction11

Figure 9: The conduction-band edge prole or potential prole for the 0.360 V case. Every third data point is shown. The initial prole is at t = 0, the nal prole is at t = 4 ps. The potential level in the well is pushed down as the bias discharges the well.

Figure 10: Time-averaged carrier density proles throughout the device at forward sweep. The 0.356 V case corresponds to near peak current conditions, the 0.360 V case corresponds to low current conditions. The vertical dashed lines indicate the positions of the barriers. The low count noise just to the right of the second barrier is an artifact of the depletion region created by the double-barrier structure. The time-averaging was carried out over the full length of the run, from t = 0 to t = 4 ps, including all transients.

12

Figure 11: The spatially averaged and time-averaged carrier density in the well as a function of the applied bias. Note the similarity in structure to the I V characterstic (Figure 5). The time-averaging is performed over the last 1000 time steps, from 2 to 4 ps, excluding all transients.

band minimum while the charge in the well provides a nonlinear feedback mechanism to pull back the transmission peak away from the conductionband minimum. For example, for the 0.22 V bias case (Figure 7), there is not enough charge in the well to screen the bias and the energy level is pushed toward the conduction-band edge of the emitter. On the other hand, for the 0.356 V bias case (Figure 7), the charge accumulation is sucient to successfully pull the transmission peak away from the conduction-band edge. For the catastrophic 0.360 V bias case, the resonant peak disappears below the bottom of the emitter conduction band, the well discharges in less than 1 ps, and the resonant peak never returns above the band minimum. The current stays low until the particles can sample the second broad peak in the transmission coecient. On the backward sweep, at a bias of 0.320 V, the resonant peak appears above the emitter conduction-band minimum in less than 0.5 ps and a rapid charging of the well results in a shift to a high-current state. This dynamical aspect of the transmission peak is a pictorial representation of the dynamical alignment of the resonant-energy level in the well with respect to the emitter conduction-band edge.

13

IV

Conclusions

We have incorporated a model particle quantum dynamics in MC simulations that appears to have captured some key features of the RTD problem. This simple model reproduces the bistability, hysteresis, and oscillatory behavior of the quantum well density of a symmetric double-barrier RTD found in more fundamental Wigner function simulations, and is the rst MC-based simulation to do so unambiguously. We have observed oscillations in the charge density in the quantum structure although, due to numerical damping eects, these oscillations did not appear to translate into well-dened current oscillations at the terminals. They result in a distorted plateau-like feature of the I V curve (Figure 5). However, the time-averaged values of the oscillating quantum well density closely track down the hysteresis behavior of the I V curve (Figure 11), in agreement with the Wigner distribution function approach [7, 8, 27]. We have also clearly depicted the dynamical alignment of the resonant-energy level of the quantum well with the emitter conduction-band edge by means of the dynamical behavior of the resonant-transmission-peak energy. It is the dynamical aspect of the alignment of the resonant level that produces the density oscillations in the quantum heterostructure. While the model dynamics may be too simple, the simulations clearly demonstrate the feasibility of the approach and strongly suggest the importance of the particle quantum dynamics across the RTD structure for understanding such devices. We intend to rene the numerical charge assignment scheme used in the Poisson solver and, more importantly, the particle quantum dynamics by calibrating with the results of a more accurate coupling of MC with the Wigner distribution function approach. This coupling is being pursued along two broad avenues: (1) matching the MC particle trajectory with a Wigner trajectory, and (2) matching lhe MC distribution function solution with the Wigner distribution function solution at appropriately chosen boundaries. As mentioned above, we also intend to use the gradient of the phase of the Airy function solutions to obtain the particle velocity at each point in the quantum region, as a function of particle energy, in a manner similar to the quantum potential approach of Bohm [28, 29]. This technique has been employed in accurately calculating tunneling times [30], and may be expected to yield an immediate improvement over the present approach.

14

Acknowledgments We would like to thank C. Moglestue for many helpful discussions regarding the MC method. His insight was a great asset in the enhancement of a 3D molecular dynamics simulation program to include the basic MC semiconductor device simulation routines. We would also like to thank K. L. Jensen for access to his Airy function tunneling program which was used as the basis for our tunneling subroutine. R. E. Salvino was a NRC/NRL Co-operative Research Associate. References [1] T. Baba and M. Mizuta, J. Appl. Phys. Lett. (Japan), 28, L1322 (1989). [2] T. Baba, M. Al-Mudares, and J.R. Barker, J. Appl. Phys. Lett. (Japan), 28, L1682 (1989). [3] K. Gallipulli, D. R. Miller, and D. P. Neikirk, IEDM Tech. Dig., 91, 511 (1991). [4] K. L. Jensen and A. K. Ganguly, Numerical Simulation of Field Emission and Tunneling: A Comparison of the Wigner Function and Transmission Coecient Approaches, preprint (1992). [5] K. F. Brennan and C. F. Summers, J. Appl. Phys., 61, 614 (1987). [6] C. M. Tan, J. Xu, and S. Zukotynski, J. Appl. Phys., 67, 3011 (1990). [7] F. A. Buot and K. L. Jensen, COMPEL, 10, 241 (1991). [8] K. L. Jensen and F. A. Buot, Phys. Rev. Lett., 66, 1078 (1991). [9] V. J. Goldman. D. C. Tsui, and J. C. Cunningham, Phys. Rev. Lett., 58, 1256 (1987). [10] T. C. L. G. Sollner, Phys. Rev. Lett., 59, 1622 (1989). [11] V. J. Goldman. D. C. Tsui, and J. C. Cunningham, Phys. Rev Lett., 59, 1623 (1989). [12] V. J. Goldman, D. C. Tsui, and J. C. Cunningham, Phys. Rev. B, 35, 9387 (1987).

15

[13] T. J. Foster. M. L. Leadbeater, L. Eaves, M. Hennini, O. H. Hughes, C. A. Payling, F. W, Sheard, P. F. Simmonds, G. A. Toombs, G. Hill, and M. A. Pate, Phys. Rev. B, 39, 6205 (1989). [14] L. Eaves, P. W. Sheard, arid G. A. Toombs. in Physics of Quantum Electron Devices, edited by F. Capasso (Springer, Berlin, 1990), pp. 107 - 146. See, in particular, the discussion on p. 138. [15] J. S. Blakemore, Semiconductor Statistics (Pergamon, Oxford, 1962), Appendix B, pp. 346 - 353. [16] R. W. Hockney and J. W. Eastwood, Computer Simulation Using Particles (Adam Hilger, Bristol, 1988), see Chap. 10. [17] W. Fawcett, A. D. Boardman, and S. Swain, J. Phys. Chem. Solids, 31, 1963 (1970). [18] C. Moglestue, Comp. Meth. Appl. Mech. Eng., 30, 173 (1982). [19] C. Moglestue, Rep. Prog. Phys., 53, 1333 (1990). [20] C. Moglestue, IEE Proc., 133, Pt. I, 35 (1986). [21] J. G. Ruch and W. Fawcett, J. Appl. Phys., 41, 3843 (1970). [22] D. S. Saxon, Elementary Quantum Mechanics (Holden-Day, San Francisco, 1968), pp. 155-158. [23] E. H. Hauge and J. A. Stovneng, Rev. Mod. Phys., 61, 917 (1989). [24] J. R. Barker, Physica B, 134, 22 (1985). [25] K. L. Jensen and F. A, Buot, IEEE Trans. Electron. Dev., 38, 2337 (1991). [26] K. L Jensen and F. A. Buot, Appl. Phys. Lett., 55, 669 (1989). [27] F A. Buot and K. L. Jensen, COMPEL, 10, 509 (1991). [28] F. A. Buot and K. L. Jensen, Self-Consistent Monte Carlo Particle Transport Including Space and Time-Dependent Quantum Tunneling, presented at IEEE First International Vacuum Microelectronics Conference, Williamsburg, Virginia (June 13-15, 1988). [29] C. Dewdney and B. J. Hiley, Found. Phys., 67, 27 (1982). [30] C. R. Leavens, Solid State Commun., 74, 923 (1990).

16

Вам также может понравиться