LETTERS

PUBLISHED ONLINE: 11 JANUARY 2016 | DOI: 10.1038/NNANO.2015.309

Tailoring high-temperature radiation and the

resurrection of the incandescent source
Ognjen Ilic1*, Peter Bermel2, Gang Chen3, John D. Joannopoulos1, Ivan Celanovic1 and Marin Soljai1
In solar cells, the mismatch between the Suns emission spectrum
and the cells absorption prole limits the efciency of such
devices1, while in incandescent light bulbs, most of the energy
is lost as heat2. One way to avoid the waste of a large fraction
of the radiation emitted from hot objects is to tailor the thermal
emission spectrum according to the desired application. This
strategy has been successfully applied to photonic-crystal emitters at moderate temperatures38, but is exceedingly difcult for
hot emitters (>1,000 K)914. Here, we show that a plain incandescent tungsten lament (3,000 K) surrounded by a cold-side nanophotonic interference system optimized to reect infrared light
and transmit visible light for a wide range of angles could
become a light source that reaches luminous efciencies
(40%) surpassing existing lighting technologies, and nearing a
limit for lighting applications. We experimentally demonstrate a
proof-of-principle incandescent emitter with efciency approaching that of commercial uorescent or light-emitting diode bulbs,
but with exceptional reproduction of colours and scalable power.
The ability to tailor the emission spectrum of high-temperature
sources may nd applications in thermophotovoltaic energy
conversion1518 and lighting.
Objects heated to very high temperatures radiate a broad spectrum of light across the visible and infrared wavelengths. At the
high end of the temperature range is the traditional incandescent
lament used for lighting. It is a suboptimal source of light
because most of the radiated power is emitted at wavelengths invisible to the human eye. At the same time, its extremely high temperature prohibits the use of nanoscale patterning to modify the
emission spectrum. Despite the fact that hot thermal emitters can
have much utilityas efcient light sources, among othersthe
ability to tailor their thermal radiation, particularly at 3,000 K,
remains a challenge.
Tailoring thermal radiation directly on the hot side is exceedingly
difcult, but we show here that the thermal emission spectrum of a
high-temperature emitter can be modied by incorporating optimized cold-side interference structures. Consider a thermal emitter
of emissivity (,T) sandwiched between two identical structures of
reectance R() and transmittance T(), separated by a small gap,
as shown in Fig. 1c. In general, the emissivity of a high-temperature
emitter depends on temperature and wavelength. By tracing the
reected radiation in the cavity surrounding the emitter, we show
(Supplementary Discussion) that the effective emissivity of this
emitter-tailoring-structure system can be expressed as



FT
(1 F)(1 + FR)
+
(1)
eff =
1 F 2 R(1 ) 1 F 2 R(1 )
where is the original emissivity of the thermal emitter and F is the
view factor characteristic to the geometry. Equation (1) highlights the

potential to tailor thermal emission by designing the surrounding

cold-side structure properties. In such a manner, the radiation spectrum of extremely high-temperature emitters can be modied
without the need for any structural patterning of the emitter
surface. The cavity effect due to the surrounding structure results
in a portion of the power emitted at unwanted wavelengths
being reabsorbed by the thermal emitter. Applying a similar
analysis allows us to express the fraction of reabsorbed power as
Pr/P0 = 2F 2R/(1 F 2R(1 )), where P0 is the original amount of
emitted power (Supplementary Discussion). This shows that the original spectrum of the thermal emitter is not just passively ltered, but
that the temperature of the emitter itself is tailored by the surrounding
photonic structure. The derivations above assume diffusive surfaces
and diffusive view factors; in the Supplementary Discussion we
extend our analysis to include specularly reecting surfaces and
show that, in our case, the two methods yield similar results.
We apply this concept to control the thermal emission spectrum
of an incandescent lament. Surrounding the lament with interference structures designed to transmit visible light and recycle infrared light for a wide range of emission angles dramatically improves
the efciency. The efciency of a lighting source (incandescent or
otherwise) can be dened as the ratio of the total emitted luminous
ux and the total supplied power. Because incandescent sources
emit essentially all supplied power into radiation, the luminous
efciency can be written as

P(T, )V()d
(2)
= 0
0 P(T, )d
where P(T, ) is the (hemispherical) spectral emissive power
of the emitter at temperature T, and V() is the photopic luminosity
function, which characterizes the spectral sensitivity of the human
eye19. The luminous efciency can theoretically assume any value
from 0 to 1; lighting sources, however, should also faithfully reproduce colours, requiring broadband emission in the visible spectrum.
The index that characterizes the quality of light is known as the
colour rendering index (CRI), which attains its peak value (100)
for high-temperature broadband emitters such as incandescent
light bulbs (Supplementary Fig. 3). An ideal lighting source would
then be a hypothetical high-temperature black body that emits
only at visible wavelengths. Such an emitter would have a luminous
efciency of approximately 4045% (see Methods).
In terms of luminous efciency, Fig. 1a shows that we could,
in principle, approach such high values by incorporating onedimensional photonic lms made of commonly deposited materials
and amenable to large-scale fabrication. The notion of increasing
the efciency of incandescent light bulbs by recycling infrared
radiation has been suggested previously2022. The novelty of our
approach is twofold. First, rejection of light at unwanted wavelengths (and subsequent reabsorption by the emitter) is made

1
Research Laboratory of Electronics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. 2 School of Electrical and Computer
Engineering, Birck Nanotechnology Center, Purdue University, West Lafayette, Indiana 47906, USA. 3 Department of Mechanical Engineering,
Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. * e-mail: ilico@mit.edu

NATURE NANOTECHNOLOGY | ADVANCE ONLINE PUBLICATION | www.nature.com/naturenanotechnology

2016 Macmillan Publishers Limited. All rights reserved.

LETTERS

NATURE NANOTECHNOLOGY

35

1.0
Power (bare emitter)
0.8

30
Normalized power

Luminous efficiency (%)

Optimized interference structure

Quarter-wave stack (chirped)
Three-material rugate stack
Four-material rugate stack

25
20
15
10

Front view

0.6

Th
em erm
it t al
er

a 40

DOI: 10.1038/NNANO.2015.309

Layers
0.4
Vis

0.2

Vis
IR IR

Efficiency (bare emitter)

0.0
0

50

100

150

200

250

300

No. of layers

50

100

150

200

No. of layers

250

300

Cross-section

Figure 1 | Potential of cold-side thermal emission tailoring. a, Projected luminous efciency of a tungsten thermal emitter (3,000 K) optimally enclosed by
interference structures of different design: quarter-wave stack (green), three- and four-material rugate stack (purple and orange), and stacks designed using
a combination of optimization techniques (red; see main text). For quarter-wave and rugate stacks, we introduce chirp to increase the bandwidth, and for
each number of layers (x axis) we explore the complete parameter space to nd the optimal design (see Methods). For all designs, the material refractive
indices are bounded by SiO2 (low) and TiO2 (high). In addition to these two indices, the designs in red also use intermediate values (for example, Al2O3 and
Ta2O5). b, The amount of power needed for the emitter to reach the temperature of 3,000 K when it is surrounded by the corresponding designs shown in a.
The power is normalized to the power needed to heat the plain emitter to the same temperature (dashed line). c, Sketch of the proposed scheme to tailor
the thermal emission of high-temperature objects; top: sketch of our experimental set-up (Supplementary Fig. 1).

Spectral emissive power (W mm2 m1)

2.5
Bare emitter
+ Optimized stack
+ Fabricated stack

2.0
(

Eye response)

1.5

1.0

0.5

0.0
0.5

1.0

1.5

2.0

2.5

Wavelength (m)

Figure 2 | Effective emissivity of an optimally enclosed structure.

The spectral emissive power (calculated) for an emitter (3,000 K) that is
plain tungsten (dashed) or tungsten sandwiched by interference structures
(solid; red and blue). The power is integrated over all angles (hemispherical
emission). The sensitivity of the human eye (luminosity function) is shown
in shaded purple. The structure optimized specically for luminous efciency
corresponds to the 300-layer design from Fig. 1a (red). For simplicity, the
fabricated design (blue) consists of 90 layers and only two materials
(SiO2 and Ta2O5).

possible with interference stacks that operate for a wide range of

both wavelengths and angles (unlike conventional photonic lms
that are usually designed for a single angle of interest). Second,
the special shape of the thermal emitter enables efcient reabsorption of unwanted radiation while allowing for resistive heating. Such
a parallel conguration is not sensitive to the alignment issues that
limit reabsorption and is compatible with thermal emitters such as
two-dimensional photonic crystals.
Here, we assume a planar emitter-tailoring-structure geometry,
where the emitter is a plain tungsten lament and the ltering structure is a one-dimensional layered stack of dielectric materials. The
2

tungsten emitter is at a uniform temperature T and its

optical properties are given by its temperature-dependent emissivity
(see Methods). The layered interference stack is composed of
common oxides of various refractive indices, from the low-index
silica (SiO2) to the high-index titania (TiO2). To explore the theoretical
potential of this scheme, we assume that no radiation leaks to the sides
(that is, near-unity view factor). Instead of separately optimizing the
interference structure against an arbitrary spectral target, we use the
knowledge of the effective emissivity of our combined structure
(equation (1)) to perform direct, system-level optimizations of the
total luminous efciency (equation (2)).
We perform this optimization for several different stack designs
including optimized quarter-wave (QW) stacks23, as well as more
advanced rugate-like stacks24and show that signicant gains in
efciency can be achieved for a reasonable number of layers in the
interference structure (Fig. 1). Another implication of the improved
efciency is that a smaller amount of power is needed to heat the
thermal emitter to the desired temperature. Figure 1b shows the
power consumed by the emitter enclosed between the interference
stacks normalized to the power consumed by the plain emitter at
the same temperature (3,000 K). For optimized designs, this ratio
can be lower than 20%.
Using a four-material rugate structure as a starting point, we
show that an extensive numerical optimization scheme (see
Methods) results in designs that reach very high luminous efciencies for a reasonable number of layers (Fig. 1a, red). For example, a
system in which the optimized interference stack consists of 300
layers of four materials (SiO2 , Al2O3 , Ta2O5 and TiO2) achieves
a very high luminous efciency (40%), surpassing (in luminous
efciency) alternative, commercially available, energy-efcient
lighting sources such as compact uorescent (510%) and lightemitting diode (LED; 413%) lamps2, as well as state-of-the-art
compact LED bulbs (1415%25, 29%26). The plotted line of optimized structures (red) shows that the increase in luminous efciency with any additional layer is no worse than linear.
However, the complexity of a global optimization problem consisting of hundreds of variables (layers and materials), makes it difcult to speculate about any global optima in the system. It is not
unlikely that similar or better performing stacks with fewer
layers could be designed in the future using some of the more

NATURE NANOTECHNOLOGY | ADVANCE ONLINE PUBLICATION | www.nature.com/naturenanotechnology

2016 Macmillan Publishers Limited. All rights reserved.

NATURE NANOTECHNOLOGY

LETTERS

DOI: 10.1038/NNANO.2015.309

Simulation prediction

Experimental result
1.0

2,250

2,000
0.8

0.6

1,500

1,250
0.4

Reectance

Wavelength (nm)

1,750

1,000
0.2

750

450

0.0
0

10

20

30

40

50

60

70

80

Incident angle (deg)

10

20

30

40

50

60

70

Incident angle (deg)

Figure 3 | Simulation and experimental measurements over a wide range of wavelengths and angles. Comparison between the simulated (left) and
measured (right) reectance spectra of the fabricated 90-layer interference structure (Fig. 2, blue line). The stack shows high transmission across the visible
spectrum and high reection in the infrared. The experimental data were restricted to 70 because of the sample size.
Front view (0)

Side view (45)

Spectral intensity enhancement

4
Emitter + stack:
Model
Measured

Emitter + stack:
Model
Measured

Emitter (bare)

Emitter (bare)

Eye response

Eye response
0

0
400

500

600

700

800

400

500

Wavelength (nm)

600

700

800

Wavelength (nm)

Figure 4 | Experimental demonstration of thermal emission tailoring. Measured (blue, solid) and modelled (grey, dashed-dot) spectral intensity of the
combined emitter-tailoring-structure system (Emitter + stack) normalized to the intensity of the plain emitter (black, dashed) that consumes the same
amount of power. The fabricated structure consists of 90 layers (see Fig. 3) and is made of two materials (SiO2 and Ta2O5) deposited on a silica substrate.
The spectrum is measured at 0 (front view) and 45 (side view). The sensitivity of the human eye (luminosity function) is shown in shaded purple. For the
same amount of input power, the luminous ux is 3.07 (2.90) times enhanced at 0 (45) compared with that of a plain emitter.

advanced techniques for optical thin-lm synthesis. Figure 2 compares the hemispherical (integrated over all angles) spectral emissive power of a plain emitter (bare tungsten) at 3,000 K with that of
the same emitter surrounded by the 300-layer optimized structures
from Fig. 1a. For reference, the shaded purple region corresponds
to the luminosity function of the human eye. Because we optimized for the total luminous efciency, the shape of the effective
emissivity curve is similar to the sensitivity prole of the human
eye (red line in Fig. 2): naturally, it is possible to optimize for a
different spectral prole, based on application and requirements.
As a result, the thermal emitter in this structure reaches its
target temperature (3,000 K) for one-tenth of the original input
power (Fig. 1b).
To demonstrate this approach, we designed a proof-of-concept
version. The interference structure, for simplicity, consists of only
two materials (SiO2 and Ta2O5) and 90 layers in total. The design
was obtained using the optimization techniques described above,

and the structure was arranged in a high-view-factor geometry. The

laments shape was specically designed to maximize reabsorption
(Fig. 1c), and comprises a thin sheet of polished tungsten, lasermachined into a closely packed, radiator-like structure that maximizes
the planar surface area while allowing for resistive heating. Figure 2
shows the spectral emissive power (simulation results) of the
emitter-tailoring-structure system (blue). We observe a close match
to the spectrum of the plain emitter in the visible and strong suppression of emission in the infrared. This is true for a wide range of wavelengths and angles: the average reectance of the fabricated structure
at 0 (45) is 92% (88%) for the 7002,000 nm wavelength range.
Figure 3 compares the simulated and measured reectance of the fabricated stack as a function of angle and wavelength. We observe strong
transmission across the visible spectrum and high reection in the
infrared, across a wide range of angles.
We characterize the performance of our structure by measuring the emitted spectral intensity at 0 and 45 angles and the

NATURE NANOTECHNOLOGY | ADVANCE ONLINE PUBLICATION | www.nature.com/naturenanotechnology

2016 Macmillan Publishers Limited. All rights reserved.

LETTERS

NATURE NANOTECHNOLOGY

amount of supplied electrical power. Figure 4 shows the measured

and modelled spectral intensities of the combined emittertailoring-structure system (labelled emitter + stack) normalized
to the intensity of the plain emitter that consumes the same
amount of power. As expected, we observe an enhancement of
emission in the visible spectrum, and suppression in the infrared.
To quantify this increase, we calculated the luminous intensity
enhancement , given by

0 Eemitter+stack (P0 , )V()d

0 Eemitter (P0 , )V()d

(3)

where E(P0 , ) is the spectral emitted ux measured by the detector

when the emitter is consuming electrical power P0 (135 W in the
experiment). From Fig. 4 we calculate = 3.07 (2.90) at 0 (45). The
estimated view factor from the setup is 0.95. We note that the calculation of involves a direct comparison of the two measured
spectra and, as such, is not sensitive to (nor does it require) the
value of temperature. Even though Fig. 4 shows mostly the spectrum
in the visible, the spectral enhancement (relative to the bare emitter)
in the visible wavelengths already tells us a great deal about the
infrared part of the spectrum. When the amount of electrical
power is the same, the stronger emission of light in the visible is
direct evidence of suppression of emission in the infrared, even
though emission in the infrared is not directly measured. The
very good match between our measurements at different angles
and our model (solid and dashed-dotted lines in Fig. 4) allows us
to evaluate the performance of our system. Matching the consumed
power and the view factor to our model, we estimate the luminous
efciency to be about 6.6% (see Methods). This value is already
approaching some commercially available compact uorescent
(CFL) bulbs and LED lights2. One advantage of our scheme is the
high quality of colour rendering for the fabricated lter; because
the source of the radiation is a blackbody-like emitter, the broadband tailoring ensures faithful reproduction of colours (see photograph in Supplementary Fig. 3), resulting in a CRI of 95 (to
calculate this number we followed Wyszecki and Stiles27). For comparison, compact uorescent bulbs have a CRI of 6588, and the
majority of LED xtures are in the 7090 range2. Another important
advantage of our scheme is scalability. Compact LED lights have so
far been limited in the total amount of light they can produce28, but
high-temperature thermal emitters allow a very high luminous ux
output, particularly relative to the size of the xture.
We have demonstrated a method to tailor high-temperature
thermal radiation that is scalable and can in principle be
implemented in more compact geometries. As fabrication
methods advance and deposition costs go down, stacks with a
higher number of layers (and materials), as well as two-dimensional
and three-dimensional photonic structures3,29,30 designed to
achieve more efcient light ltering may become appealing.
However, the ability to tailor thermal emission goes beyond just
efcient lighting. For example, developing a high-temperature
thermal emitter better matched to the absorption prole of a photovoltaic cell could have signicant implications for the performance
of solid-state thermophotovoltaic energy conversion schemes1518.
In addition, this concept can also be applied to thermal radiation
absorbers. More generally, these results show that such composite
systems provide control over the fundamental structural
properties of thermal emission and absorption: the designs that
we describe here are examples to showcase the immense potential
of nanophotonics principles for tailoring high-temperature
thermal radiation.

Methods
Methods and any associated references are available in the online
version of the paper.
4

DOI: 10.1038/NNANO.2015.309

Received 24 July 2015; accepted 25 November 2015;

published online 11 January 2016

References
1. Shockley, W. & Queisser, H. J. Detailed balance limit of efciency of pn
junction solar cells. J. Appl. Phys. 32, 510519 (1961).
2. US Department of Energy. Lighting Basics (accessed May 2015); http://energy.
gov/eere/energybasics/articles/lighting-basics
3. Greffet, J.-J. et al. Coherent emission of light by thermal sources. Nature 416,
6164 (2002).
4. De Zoysa, M. et al. Conversion of broadband to narrowband thermal emission
through energy recycling. Nature Photon. 6, 535539 (2012).
5. Inoue, T., De Zoysa, M., Asano, T. & Noda, S. Realization of dynamic thermal
emission control. Nature Mater. 13, 928931 (2014).
6. Raman, A. P., Anoma, M. A., Zhu, L., Rephaeli, E. & Fan, S. Passive radiative
cooling below ambient air temperature under direct sunlight. Nature 515,
540544 (2014).
7. Liu, X. et al. Taming the blackbody with infrared metamaterials as selective
thermal emitters. Phys. Rev. Lett. 107, 045901 (2011).
8. Han, S. E., Stein, A. & Norris, D. J. Tailoring self-assembled metallic photonic
crystals for modied thermal emission. Phys. Rev. Lett. 99, 053906 (2007).
9. Schlemmer, C., Aschaber, J., Boerner, V. & Luther, J. Thermal stability of microstructured selective tungsten emitters. AIP Conf. Proc. 653, 164173 (2003).
10. Sai, H., Kanamori, Y. & Yugami, H. High-temperature resistive surface grating
for spectral control of thermal radiation. Appl. Phys. Lett. 82, 16851687 (2003).
11. Nagpal, P. et al. Fabrication of carbon/refractory metal nanocomposites
as thermally stable metallic photonic crystals. J. Mater. Chem. 21,
1083610843 (2011).
12. Arpin, K. A. et al. Three-dimensional self-assembled photonic crystals with high
temperature stability for thermal emission modication. Nature Commun. 4,
2630 (2013).
13. Arpin, K. A., Losego, M. D. & Braun, P. V. Electrodeposited 3D tungsten
photonic crystals with enhanced thermal stability. Chem. Mater. 23,
47834788 (2011).
14. Rinnerbauer, V. et al. High-temperature stability and selective thermal emission
of polycrystalline tantalum photonic crystals. Opt. Express 21, 011482 (2013).
15. Lin, S. Y., Moreno, J. & Fleming, J. G. Three-dimensional photonic-crystal
emitter for thermal photovoltaic power generation. Appl. Phys. Lett. 83,
380 (2003).
16. Chubb, D. L. Fundamentals of Thermophotovoltaic Energy Conversion
(Elsevier, 2007).
17. Lenert, A. et al. A nanophotonic solar thermophotovoltaic device. Nature
Nanotech. 9, 126130 (2014).
18. Chan, W. R. et al. Toward high-energy-density, high-efciency, and moderatetemperature chip-scale thermophotovoltaic. Proc. Natl Acad. Sci. USA 110,
53095314 (2013).
19. Sharpe, L. T., Stockman, A., Jagla, W. & Jgle, H. A luminous efciency function,
V*(), for daylight adaptation. J. Vision 5, 948968 (2005).
20. Kostlin, H. & Almer, F. H. R. Incandescent lamp with infrared lter. US patent
4,017,758 (1977).
21. Goldstein, I. S., Fontana, R. P., Thorington, L. & Howson, R. P. The design,
construction and performance of an incandescent light source with a transparent
heat mirror. Lighting Res. Technol. 18, 9397 (1986).
22. Bergman, R. S. HalogenIR lamp development: a system approach. J. Illum. Eng.
Soc. 20, 1016 (1991).
23. Hecht, E. Optics 4th edn (Addison-Wesley, 2002).
24. Baumeister, P. Simulation of a rugate lter via a stepped-index dielectric
multilayer. Appl. Opt. 25, 26442645 (1986).
25. Next Generation Luminaires (NGL) Solid-State Lighting Design Competition
(2014), sponsored by the US Department of Energy; http://www.ngldc.org
26. Philips creates the worlds most energy-efcient warm white LED lamp Press
release (April 2013); http://www.newscenter.philips.com/main/standard/news/
press/2013/20130411-philips-creates-the-world-s-most-energy-efcient-warmwhite-led-lamp.wpd
27. Wyszecki, G. & Stiles, W. S. Color Science: Concepts and Methods, Quantitative
Data and Formulae 2nd edn (Wiley, 1982).
28. US Department of Energy, Ofce of Energy Efciency and Renewable Energy,
CALIPER Snapshot Reports (2014); http://energy.gov/eere/ssl/caliper-snapshotreports
29. Lin, S. Y. et al. A three-dimensional photonic crystal operating at infrared
wavelengths. Nature 394, 251253 (1998).
30. Fleming, J. G., Lin, S. Y., El-Kady, I., Biswas, R. & Ho, K. M. All-metallic
three-dimensional photonic crystals with a large infrared bandgap. Nature 417,
5255 (2002).

Acknowledgements
The authors thank P. Rebusco for critical reading and editing of the manuscript, and
J.J. Senkevich, W. Chan, S. Kooi, I. Wang, A. Cukierman, M. Harradon, A. Musabeyoglu,

NATURE NANOTECHNOLOGY | ADVANCE ONLINE PUBLICATION | www.nature.com/naturenanotechnology

2016 Macmillan Publishers Limited. All rights reserved.

NATURE NANOTECHNOLOGY

DOI: 10.1038/NNANO.2015.309

S. Johnson and Y. Xiang Yeng for valuable input. This work was partially supported by the
Army Research Ofce through the ISN (contract no.W911NF-13-D0001). The fabrication
part of the effort was supported by S3TEC, an Energy Frontier Research Center funded by
the US Department of Energy (DOE), Ofce of Science, Basic Energy Sciences (BES)
(award no. DE-SC0001299/DE-FG02-09ER46577).

Additional information

Author contributions

Competing nancial interests

All authors discussed the results and made critical contributions to the work.

LETTERS

Supplementary information is available in the online version of the paper. Reprints and
permissions information is available online at www.nature.com/reprints. Correspondence and
requests for materials should be addressed to O.I.

The authors declare no competing nancial interests.

NATURE NANOTECHNOLOGY | ADVANCE ONLINE PUBLICATION | www.nature.com/naturenanotechnology

2016 Macmillan Publishers Limited. All rights reserved.

LETTERS

NATURE NANOTECHNOLOGY

Methods

Thermal emitter properties and experimental setup details. The thermal emitter
was a sheet of refractory metal (tungsten sheet, 8085 m thick, obtained from
H.C. Starck), laser machined into a geometry that allowed for both resistive heating
as well as efcient reabsorption of reected light (Supplementary Fig. 1b). Before
laser cutting, the sheet of tungsten was covered with a thin layer (10 m) of
ultraviolet-curable, water-soluble, polymer (DaeCoat PCA-120713) to protect
against material ablation. Afterwards, the coating was removed in deionized water by
sonication. The optical properties of refractory metals strongly depend on
temperature as well as sample purity and preparation conditions. Supplementary
Fig. 2a shows the general dependence of tungsten emittance (normal) on
wavelength31. The crossover point, in the 11.5 m region, separates the spectral
regions where the emissivity increases/decreases with temperature.
In our calculations, we used tabulated data for normal spectral emittance31, in
200 K steps for a range of temperatures (1,6003,100 K). Emittance for angles other
than zero was calculated as follows: the high-temperature dielectric permittivity for
tungsten was obtained by tting the emittance data at each temperature to a sum of
DrudeLorentzian dielectric functions. Once the permittivity (as a function of
wavelength and temperature) was known, the angular dependence of absorptivity
was derived from Fresnel reection coefcients. Finally, from Kirchoffs law, the
(directional, spectral) emissivity equals the (directional, spectral) absorptivity. For
intermediate temperatures (temperatures between the 200 K steps) we obtained the
emissivity by linear interpolation.
Light emitted by the resistively heated tungsten lament was captured by two
detectors, at 0 and 45 angles with respect to the surface of the lament emitter
(Supplementary Fig. 1a). The detectors were calibrated to correct for any
nonlinearity in the measured signal and to capture the emitted light once the system
had reached optical equilibrium and the lament had reached its peak temperature/
resistance (1.5 s). The air in the setup was evacuated (P 1 106 torr)
and argon was added (P 1 104 torr). The one-dimensional photonic lm was
fabricated using ion-beam sputtering at Evaporated Coatings, with layers of Ta2O5
and SiO2 on a fused-silica substrate (2 2 cm). In general, coatings deposited using
ion-beam sputtering have particularly low absorption in the visible and the
near-infrared: the extinction coefcient for Ta2O5 was estimated to be <1 104
(550 nm32) and the tantala/silica coatings were used for antireection coatings in
precise gravitational wave detectors, with absorption loss on the order of ppm
(1,064 nm33). For wavelengths in the mid-infrared (>4 m), these materials can
become absorptive; however, we note from Fig. 2 that a 3,000 K tungsten emitter
radiates a very small fraction of light (3.8%) in the spectrum above 4 m.
Undesired absorption in the optically thick silica substrate may be further
reduced in glasses with particularly low infrared absorption such as Cornings
7979 or Heraeus Infrasil (in addition, the substrate thickness can be
reduced further).
Thermal emitter temperature estimation. Supplementary Fig. 2b shows the
dependence of tungsten resistivity on temperature34. The knowledge of resistivity,
together with the shape and size of the emitter, allowed us to estimate the
temperature from the measured resistance. First, we performed a nite-element heat
transfer analysis (COMSOL Multiphysics) to characterize the equilibrium
temperature distribution of the emitter, shown in Supplementary Fig. 2c. We observe
that only the main part of the emitter (the radiator-like structure) reaches very high
temperatures. Further analysis conrmed that most of the power was indeed
dissipated in this part of the emitter. In addition, we observed a highly uniform
temperature distribution between the central strips of the emitter. Values for the
high-temperature thermal conductivity and heat capacity of tungsten were obtained
from the literature35,36.
The measured resistance of the emitter in Fig. 4 of the main text was R = 2.90
when the interference structure surrounded the emitter and R = 2.62 when there
was no such structure (plain emitter). In both cases, the measured power delivered to
the emitter was the same (135 W) and included all emissions as well as losses in the
device. Taking into account the uncertainty and the variation in the emitter
thickness (8085 m), these values of resistance imply (based on the resistivity
dependence shown in Supplementary Fig. 2b) a temperature estimate for the central
part of the emitter of 2,9003,050 K for the case when the ltering structure is
present and an estimate of 2,7002,830 K when the emitter is bare. Another way to
estimate the temperature is to evaluate the energy of the emitted light, assuming the
literature value of temperature-dependent tungsten emittance31. It is important to
note that at high temperatures, resistivity and, particularly, emissivity data for
refractory metals is heavily dependent on the sample preparation process, the
presence of impurities and the surface nish31,34, which means that the margin of
error for our estimates may be signicant. With that in mind, our model estimates
(given the area of the emitter and the view factor of 0.95) that to consume 135 W of
power, the emitter would have a temperature of 2,950 K and the total luminous
efciency of the system would be 6.6%.
Because the high-temperature emitter is a plain, unpatterned material, it can
withstand extremely high temperatures without structural surface degradation. At
the temperatures required for efcient incandescent lighting, however, material
evaporation from the surface of the emitter may affect the optical properties of the
close-by interference structure. This can be particularly pronounced for pure metals

DOI: 10.1038/NNANO.2015.309

such as tungsten or tantalum, but efforts that include lament doping, the
introduction of halogen elements and high-molecular-weight inert gases could
reduce this effect. On the other hand, at the temperatures needed for
thermophotovoltaic energy conversion (such as selective radiators/absorbers1518) of
1,5002,000 Kstill very challenging for direct (hot-side) patterning of the thermal
emitter surfacesuch evaporation effects are drastically reduced.
Optimal efciency for lighting applications. Using equation (2) we can estimate
the luminous efciency of an ideal lighting source. A hypothetical black (or grey)
body at a temperature of 2,800 K(5,800 K) with its spectrum truncated to
wavelengths where the luminosity function is 1% of its peak value has a luminous
efciency of 40.1% (40.4%). This value depends on the choice of the cutoff, and as
such it should be mainly used as a guide.
Interference structure design and optimization. The optimizations of luminous
efciency shown in Fig. 1 were performed for various stack designs, including
quarter-wave as well as the three- and four-material rugate stacks. An optimized
quarter-wave stack consisting of two materials (SiO2 and TiO2) can be effective to a
certain extent at suppressing infrared radiation and allows for a higher luminous
efciency relative to the plain incandescent emitter (Fig. 1a, dashed). However, after
the rst several dozen periods, the quarter-wave stack no longer offers improved
efciency as additional layers are added. Another well-known issue with a quarterwave-stack design is the rejection of additional higher-order bands at 0/3, 0/5 and
so on, limiting their potential for very-broadband reection in the infrared spectrum
and transmission in the visible spectrum.
The purpose of a rugate-like interference structure24,40 is to suppress such
higher-order reection bands and thus allow for a wide rejection band in the infrared
(0.72.5 m) while maintaining low reection in the visible spectrum. This is
achieved through a sinusoidal refractive index prole. There are several ways to
implement such a prole: for example, the index of refraction n(x) or the dielectric
permittivity n 2(x) can be sinusoidally dependent on the structure optical thickness x.
However, it can be shown that the most complete suppression of the reectance
bands at higher harmonics is achieved when the natural logarithm of the refractive
index exhibits a sinusoidal prole40. Such a structure would have a refractive index
prole that satises ln(n(x)) = acos(4x/0) + b, where 0 is the design wavelength
and x is the optical thickness. Coefcients a and b ensure the correct limit to the
optical prole and are given by a = 1/2ln(nL/nH) and b = 1/2ln(nLnH) where nL and
nH denote the low and high refractive indices. In practice, the rugate lter prole is
obtained by using a discrete set of materials with refractive indices bounded by nL
and nH (corresponding to SiO2 and TiO2 in our case). For a four-material rugate
structure, the corresponding stacking of layers is [ABCDDCBA], where by denition
n(A) = nL , n(D) = nH and the optical thickness of each layer is 0/4. We increased the
bandwidth of the interference structure by introducing period chirping, dened as
Ti = Ti1 (1 + / 1)1/(Np 1) where Ti denotes the optical thickness of the ith period,
Np is the total number of layer periods and is the dimensionless chirp parameter.
Given the denitions above, for the quarter-wave stack and the three- and fourmaterial rugate structure designs from Fig. 1 of the main text, we swept the complete
parameter space to nd the optimal rugate design for a given number of layers. We
note that both three-material (Fig. 1a, pink) and four-material (Fig. 1a, orange)
rugate stacks represent a marked improvement over the quarter-wave stack. In
particular, when the total number of layers is small, the four-material design (with its
larger unit cell) performs worse than the three-material design; this is rectied for
stacks with more layers.
For custom optimization (Figs 1 and 2, red; Fig. 2, blue), we used rugate-like
designs as a starting point and applied other optimization methods37, including
needle optimization38. Specically, we used a combination of local, gradient-based,
optimization algorithms39, made possible by implementing a transfer matrix
approach that in addition to calculating luminous efciency also calculates the
necessary derivatives of the efciency (with respect to each layer thickness) at a
comparatively small computational cost. This is combined with a parallelized
(MPI) implementation of the calculation that efciently determines where a new
layer should be introduced or an existing layer removed to maximize the gure of
merit (luminous efciency) of the combined system (emitter + stack). For the
fabricated structure we also require an exceptionally high CRI. The combination
of these methods yields the designs with luminous efciency values plotted in
Figs 1 and 2.

References
31. Touloukian, Y. S. & DeWitt, D. P. Thermophysical Properties of Matter.
Thermal Radiative Properties: Metallic Elements and Alloys Vol. 7
(IFI/Plenum, 1970).
32. Demiryont, H., Sites, J. R. & Geib, K. Effects of oxygen content on the optical
properties of tantalum oxide lms deposited by ion-beam sputtering. Appl. Opt.
24, 490 (1985).
33. Harry, G., Bodiya, T. P. & DeSalvo, R. Optical Coatings and Thermal Noise in
Precision Measurement (Cambridge Univ. Press, 2012).
34. Desai, P. D., Chu, T. K., James, H. M. & Ho, C. Y. Electrical resistivity of selected
elements. J. Phys. Chem. Ref. Data 13, 10691096 (1984).
NATURE NANOTECHNOLOGY | www.nature.com/naturenanotechnology

2016 Macmillan Publishers Limited. All rights reserved.

NATURE NANOTECHNOLOGY

DOI: 10.1038/NNANO.2015.309

35. Powell, R. W., Ho, C. Y. & Liley, P. E. Thermal Conductivity of Selected

Materials (National Standard Reference Data Series 8, National Bureau of
Standards, 1966).
36. White, G. K. & Collocott, S. J. Heat capacity of reference materials: Cu and W.
J. Phys. Chem. Ref. Data 13, 12511257 (1984).
37. Larouche, S. & Martinu, L. OpenFilters: open-source software for the
design, optimization, and synthesis of optical lters. Appl. Opt. 47,
C219C230 (2008).

LETTERS

38. Tikhonravov, A. V., Trubetskov, M. K. & DeBell, G. W. Application of the needle

optimization technique to the design of optical coatings. Appl. Opt. 35,
54935508 (1996).
39. Johnson, S. G. The NLopt nonlinear-optimization package; http://ab-initio.mit.
edu/nlopt
40. Carniglia, C. K. in Laser-Induced Damage in Optical Materials: 1987 (eds
Bennet, H. E. et al.) 272277 (Special Publication 756, US National Institute of
Standards and Technology, 1988).

NATURE NANOTECHNOLOGY | www.nature.com/naturenanotechnology

2016 Macmillan Publishers Limited. All rights reserved.