Skip to content
LumenaryAcademy
Functional 75 min

Optical materials

How solids absorb, bend and emit light — the colours of semiconductors computed from their band gaps, LED and laser materials, the glass of optical fibre, quantum dots computed as particles in a sphere, and perovskite solar cells.

Code tested · 6AI-draftedAwaiting review by a materials scientist

Your guides: Chloe · The Materials Scientist, Margaret · The Tester, Gideon · The Failure Analyst

MotivationH

Light, through and out of solids

Cadmium sulfide is yellow, gallium phosphide orange, mercury sulfide (the pigment vermilion) red, and silicon a metallic grey-black, yet all four are clear crystals in the sense that matters to a physicist: they would be transparent if your eyes could see farther into the infrared. The internet runs through glass so clear that light crosses fifteen kilometres of it and half survives. A blue LED the size of a grain of salt, coated with a yellow powder, lights a room. Crystals a few nanometres across glow in colours set by their size alone, and won their discoverers the 2023 Nobel Prize in Chemistry. A family of crystals that can be printed from solution at low temperature is approaching the efficiency of silicon solar cells.

All of these follow from a few ideas: how the electrons and ions of a solid respond to an oscillating electric field (the dielectric response from Dielectric and ferroelectric materials, now at optical frequencies), and the band gaps of Electronic structure and bands. This lesson computes the colour of semiconductors from their gaps, follows light into fibre glass, shrinks a semiconductor until quantum confinement sets its colour, and asks how efficient a solar cell can possibly be. The devices themselves, modulators, detectors and photonic chips, are in Lumenary EE's photonics; molecular colour and spectroscopy are in Lumenary Chemistry's spectroscopy.

DefinitionsH

Refraction, reflection and absorption

Definition— complex refractive index

At optical frequencies a material is described by its complex refractive index n~=n+iκ\tilde n = n + i\kappa, related to the permittivity by n~2=εr\tilde n^2 = \varepsilon_r. The real part nn sets the phase speed c/nc/n, and so refraction; the extinction coefficient κ\kappa sets absorption.

Definition— absorption coefficient, Beer–Lambert law

Light of intensity I0I_0 entering an absorbing medium decays as I(x)=I0e−αxI(x) = I_0e^{-\alpha x}, with absorption coefficient α=4πκ/λ\alpha = 4\pi\kappa/\lambda. The absorption length 1/α1/\alpha is the depth at which the intensity has fallen to 1/e1/e.

Definition— normal-incidence reflectance

Light striking a flat surface between air and a medium at normal incidence is reflected with intensity fraction R=(n−1)2+κ2(n+1)2+κ2R = \dfrac{(n - 1)^2 + \kappa^2}{(n + 1)^2 + \kappa^2}.

Definition— direct and indirect band gaps

A gap is direct if the top of the valence band and the bottom of the conduction band occur at the same crystal momentum k\mathbf k, and indirect if they do not. A photon carries almost no crystal momentum, so in an indirect-gap material absorption or emission across the gap needs a lattice vibration (a phonon) to supply the difference, which makes it much weaker.

Some numbers. Silicon has n≈3.9n \approx 3.9 in the middle of the visible, so a bare silicon surface reflects (2.9/4.9)2≈35(2.9/4.9)^2 \approx 35 per cent of the light falling on it, which a solar cell cannot afford. A coating a quarter of a wavelength thick with n≈3.9≈2n \approx \sqrt{3.9} \approx 2, close to silicon nitride's, cancels the reflection at one wavelength by interference; texturing the surface into tiny pyramids gives light a second chance to enter. Absorption lengths differ even more: just above their gaps, direct-gap gallium arsenide and the halide perovskites absorb most light within a micrometre or less, while indirect-gap silicon needs tens to hundreds of micrometres in the near infrared. That is why silicon solar cells are made from wafers about 150 µm thick, and direct-gap absorbers from films a hundred times thinner.

ComputationC

The colour of a semiconductor, computed from its gap

A semiconductor absorbs every photon whose energy exceeds its gap and transmits (or, as a powder, scatters back) the photons below it. The absorption edge in wavelength is λg=hc/Eg=1239.84 nm/Eg[eV]\lambda_g = hc/E_g = 1239.84\ \text{nm}/E_g[\text{eV}]. So the colour we see is what is left of white light after everything shorter than λg\lambda_g has been removed. If λg\lambda_g is below 400 nm, nothing visible is absorbed and the crystal is colourless (or white as a powder); if it is beyond 700 nm, everything visible is absorbed and the material looks black, apart from whatever its surface reflects.

The program makes this quantitative. It takes a daylight-like spectrum (a 6500 K black body), removes everything above the gap, and converts what remains into the colour coordinates the eye uses, with an analytic fit to the CIE 1931 colour-matching functions published by Chris Wyman, Peter-Pike Sloan and Peter Shirley, and then to the sRGB colours of a computer screen.

python
import numpy as np

def trap(y, x):
    """Trapezoid-rule integral of y over x."""
    return float(np.sum((y[1:] + y[:-1]) * np.diff(x)) / 2)

# CIE 1931 colour-matching functions, from the multi-lobe Gaussian fit of Wyman, Sloan and Shirley (2013)
def g(lam, mu, inv_left, inv_right):
    t = (lam - mu) * np.where(lam < mu, inv_left, inv_right)
    return np.exp(-0.5 * t * t)

def cmf(lam):
    x = 0.362 * g(lam, 442.0, 0.0624, 0.0374) + 1.056 * g(lam, 599.8, 0.0264, 0.0323) - 0.065 * g(lam, 501.1, 0.0490, 0.0382)
    y = 0.821 * g(lam, 568.8, 0.0213, 0.0247) + 0.286 * g(lam, 530.9, 0.0613, 0.0322)
    z = 1.217 * g(lam, 437.0, 0.0845, 0.0278) + 0.681 * g(lam, 459.0, 0.0385, 0.0725)
    return x, y, z

lam = np.linspace(380, 780, 801)                     # nm
h, c, kB = 6.62607015e-34, 2.99792458e8, 1.380649e-23
T_light = 6500.0                                     # daylight-like illuminant: a 6500 K black body
L = 1 / (lam * 1e-9) ** 5 / (np.exp(h * c / (lam * 1e-9 * kB * T_light)) - 1)
xb, yb, zb = cmf(lam)
M = np.array([[3.2406, -1.5372, -0.4986], [-0.9689, 1.8758, 0.0415], [0.0557, -0.2040, 1.0570]])  # XYZ -> linear sRGB

def colour(T_spectrum):
    """Colour of light remaining after a filter with transmission T(lambda): hex sRGB and relative brightness."""
    XYZ = np.array([trap(L * T_spectrum * f, lam) for f in (xb, yb, zb)])
    rgb = M @ XYZ
    return rgb, XYZ[1]

white_rgb, white_Y = colour(np.ones_like(lam))
def describe(Eg_eV):
    edge = 1239.84 / Eg_eV                           # absorption edge, nm: shorter wavelengths are absorbed
    T = (lam > edge).astype(float)                   # an ideal sharp edge, thick sample
    rgb, Y = colour(T)
    rgb = np.clip(rgb / white_rgb, 0, 1)             # scale so that no absorption looks white
    srgb = np.where(rgb <= 0.0031308, 12.92 * rgb, 1.055 * rgb ** (1 / 2.4) - 0.055)
    hexcode = "#" + "".join(f"{int(round(255 * v)):02X}" for v in srgb)
    return edge, hexcode, Y / white_Y

print("material      gap/eV  edge/nm   sRGB      brightness")
for name, Eg in [("diamond", 5.47), ("ZnS", 3.6), ("ZnO", 3.3), ("CdS", 2.42), ("GaP", 2.26),
                 ("HgS", 2.0), ("CdSe", 1.74), ("GaAs", 1.42), ("Si", 1.12)]:
    edge, hexcode, Y = describe(Eg)
    print(f"{name:12} {Eg:6.2f}   {edge:6.0f}   {hexcode}    {Y:5.2f}")

print("\nthe cadmium pigment series, CdS -> CdS(1-x)Se(x): colour against gap")
for Eg in [2.6, 2.4, 2.3, 2.2, 2.1, 2.0, 1.9, 1.8]:
    edge, hexcode, Y = describe(Eg)
    print(f"  gap {Eg:4.2f} eV   edge {edge:4.0f} nm   {hexcode}   brightness {Y:4.2f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The computed colours match what you see. Diamond, zinc sulfide and zinc oxide come out white (#FFFFFF): their gaps lie in the ultraviolet, which is why zinc oxide and titanium dioxide are white pigments and sunscreens, absorbing ultraviolet and scattering all visible light. Cadmium sulfide, with an edge at 512 nm, comes out a strong yellow (#FFF300), gallium phosphide orange (#FFA700), mercury sulfide deep red (#AF0000, the colour of vermilion and of the mineral cinnabar), and cadmium selenide, gallium arsenide and silicon black. The second table shows something painters have used for a century: alloying cadmium sulfide with cadmium selenide lowers the gap smoothly, and the pigments run from lemon yellow through orange and red to maroon as the edge moves across the spectrum. That is band-gap engineering in a paint tube. (Cadmium pigments are now restricted in many uses because cadmium is toxic; the same physics underlies safer replacements, such as bismuth vanadate yellow.)

violetbluegreenyelloworangeredabsorbedtransmittedabsorption edge: 1240 / 2.42 eV = 512 nm380480580680780 nm3.26 eV1.59 eVphoton energy falls to the rightCdS: the light that remains looks yellow
Why cadmium sulfide is yellow. Its gap of 2.42 eV puts the absorption edge at 512 nm: violet, blue and some green light have enough energy to lift an electron across the gap and are absorbed (shaded); green-yellow, yellow, orange and red pass. The eye sees the mixture of what passes as yellow. A smaller gap moves the edge to the right, through orange and red, until everything visible is absorbed and the material looks black.

The program assumed a perfectly sharp edge and a thick sample. Real edges have a weak absorption tail below the gap (the Urbach tail, from disorder and lattice vibrations), indirect gaps such as GaP's absorb weakly near the edge, and in a pigment the particle size matters too, because scattering decides how far light travels before it returns.

Metals get their colours differently. Their free electrons reflect almost all light below the plasma frequency, which for most metals lies in the ultraviolet, so they look silvery. Gold and copper are coloured because their filled d bands lie only about 2 eV below the Fermi level: photons above that energy (blue and violet for gold, from green upwards for copper) lift d electrons into empty states and are partly absorbed, so the reflected light is yellow or reddish. In silver the d bands lie deeper, the interband absorption starts in the ultraviolet, and silver is white. Gold's d bands sit where they do partly because of relativistic effects on its heavy nucleus' electrons, so in a real sense gold's colour is relativistic.

ExplanationH

Materials for LEDs and lasers

Running the absorption backwards gives light. In a light-emitting diode, a forward-biased p–n junction injects electrons and holes into a thin active layer, and when they recombine across the gap each can emit a photon of energy close to EgE_g. Three materials questions decide whether it works well.

Is the gap direct? In an indirect-gap material such as silicon, an electron at the bottom of the conduction band has the wrong crystal momentum to drop straight into a hole, so it waits for a phonon, and long before that it usually recombines without light at a defect. Silicon LEDs are less efficient than GaAs ones by factors of thousands or more. All efficient LEDs and lasers use direct-gap compound semiconductors (Semiconductor materials).

What gap? The colour is set by composition. Indium gallium nitride, InGaN, covers violet, blue and green, its gap falling as indium replaces gallium; aluminium gallium indium phosphide, AlGaInP, grown on GaAs, covers red, orange and amber; AlGaN reaches into the ultraviolet for disinfection and curing; and InGaAsP on indium phosphide covers the 1.3 and 1.55 µm infrared of fibre communication. Between InGaN and AlGaInP lies the green gap, a range around 530–590 nm where both material systems become inefficient: InGaN because indium-rich layers grow with more defects and internal electric fields, AlGaInP because its gap turns indirect.

What to do about the defects? Blue InGaN LEDs grown on sapphire contain hundreds of millions of dislocations per square centimetre, and still convert well over half of the injected electrons into light; the leading explanation is that fluctuations in indium content trap carriers in small regions before they can reach a dislocation. White LEDs use a blue chip with a phosphor of cerium-doped yttrium aluminium garnet (YAG:Ce), which converts part of the blue to yellow, often with a second red phosphor for warmer light.

A laser needs more: population inversion, more electrons in the upper state than the lower, and an optical cavity to feed light back through the gain medium. Semiconductor lasers achieve inversion by heavy injection into quantum wells (the heterostructures of the previous lesson) and use cleaved crystal facets or built-in gratings as mirrors; they power fibre links, Blu-ray (405 nm GaN lasers), laser printers and the 3D sensors of phones. Solid-state lasers put active ions into a transparent host crystal or glass: neodymium in yttrium aluminium garnet (Nd:YAG, 1064 nm), titanium in sapphire (tunable from about 650 to 1100 nm, the workhorse of ultrafast science), erbium in silica fibre (amplifying the 1530–1565 nm band that carries the internet), and ytterbium-doped fibre lasers producing kilowatts for cutting and welding metal.

ComputationC

The glass of optical fibre

In 1966 Charles Kao and George Hockham, at Standard Telecommunication Laboratories in England, argued that the glass of the day lost light not because of anything fundamental but because of impurities, mainly iron and copper ions and water, and that purified glass could lose less than 20 dB per kilometre, enough for telecommunications. Four years later Robert Maurer, Donald Keck and Peter Schultz at Corning in the United States made a fibre with 17 dB/km; by 1979 a group at NTT in Japan had reached 0.2 dB/km at 1550 nm, close to the physical limit. Kao shared the 2009 Nobel Prize in Physics.

Fibre glass is made not by melting sand but by vapour deposition: silicon and germanium chlorides are oxidised in a flame or a heated tube to deposit layers of ultrapure silica soot, slightly enriched in germania where the core will be, which are sintered into a clear glass rod (a preform) and drawn into fibre 125 µm across. Starting from volatile chlorides leaves the transition-metal impurities behind, just as in purifying silicon. The germania raises the core's refractive index by about a third of a per cent, so light is guided by total internal reflection.

coatingcladding (silica)claddingcorelight innrefractive index profile
A step-index optical fibre (not to scale). A core a few micrometres in radius with a slightly higher refractive index than the surrounding cladding guides light by total internal reflection; a polymer coating protects the glass. Rays entering within the acceptance angle, set by the numerical aperture, are trapped. When the core is small enough, only a single mode propagates, which is what long-distance links use.

The program computes the optical properties that matter for a link: the refractive index of fused silica from the Sellmeier equation with Irving Malitson's 1965 coefficients, the group index (pulses travel at c/ngc/n_g), the material dispersion, the single-mode condition of a typical core, and the Rayleigh scattering loss.

python
import numpy as np

# Refractive index of fused silica: Sellmeier equation with Malitson's (1965) coefficients, wavelength in um
B = [0.6961663, 0.4079426, 0.8974794]
C = [0.0684043**2, 0.1162414**2, 9.896161**2]
def n_silica(lam_um):
    l2 = lam_um**2
    return np.sqrt(1 + sum(b * l2 / (l2 - c) for b, c in zip(B, C)))

c_light = 2.99792458e8
lam = np.linspace(0.6, 1.8, 120001)                  # um
n = n_silica(lam)
dn = np.gradient(n, lam)
d2n = np.gradient(dn, lam)
ng = n - lam * dn                                    # group index: pulses travel at c / ng
D = -(lam * 1e-6) / c_light * (d2n * 1e12) * 1e6    # material dispersion: s/m^2 -> ps/(nm km)
print(" wavelength/nm    n        group index   dispersion/(ps/(nm km))")
for target in [0.633, 0.85, 1.31, 1.55]:
    i = np.argmin(abs(lam - target))
    print(f"   {target * 1000:5.0f}       {n[i]:.5f}     {ng[i]:.5f}        {D[i]:7.2f}")
zero = lam[np.argmin(abs(D))]
print(f"material dispersion vanishes near {zero * 1000:.0f} nm")

# A step-index fibre: germania-doped core 0.36 % higher in index than the cladding
n_clad = n_silica(1.55)
n_core = n_clad * 1.0036
NA = np.sqrt(n_core**2 - n_clad**2)
a_max = 2.405 * 1.31 / (2 * np.pi * NA)              # single-mode for V = 2 pi a NA / lambda below 2.405
print(f"\nnumerical aperture {NA:.3f}: acceptance half-angle {np.degrees(np.arcsin(NA)):.1f} degrees in air")
print(f"single-mode at 1310 nm and above if the core radius is below {a_max:.2f} um")

# Loss: Rayleigh scattering from frozen-in density fluctuations falls as 1/lambda^4
A = 0.75                                             # dB km^-1 um^4, a typical value for silica fibre
print("\n wavelength/nm   Rayleigh loss/(dB/km)")
for l in [0.85, 1.31, 1.55]:
    print(f"   {l * 1000:5.0f}          {A / l**4:6.3f}")
for loss_per_km in [0.2, 3.0]:
    left = 10 ** (-loss_per_km * 80 / 10)
    print(f"after 80 km at {loss_per_km} dB/km: {100 * left:.3g} % of the light remains")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Each number explains a design choice of the telecommunications industry.

  • Dispersion. Different wavelengths travel at different speeds, which spreads a pulse. In silica the material dispersion crosses zero near 1273 nm; a standard single-mode fibre's own waveguide dispersion shifts the zero to about 1310 nm, which is why early long-distance links used 1310 nm. At 1550 nm the dispersion is about +22 ps per nanometre of spectral width per kilometre, and modern systems correct it digitally in the receiver.
  • Single mode. With a core index 0.36 per cent above the cladding, the numerical aperture is 0.12, and the fibre carries a single mode at 1310 nm and above if its core radius is below about 4.1 µm. Standard fibres have cores about 8–9 µm across: single-mode, so no pulse spreading between modes.
  • Loss. Glass is frozen liquid, and its frozen-in density fluctuations scatter light as 1/λ41/\lambda^4, like the air that makes the sky blue. Rayleigh scattering gives 1.4 dB/km at 850 nm but only 0.13 dB/km at 1550 nm. At longer wavelengths the tail of silica's infrared vibrational absorption rises, so the total loss has its minimum, about 0.15–0.2 dB/km, near 1550 nm; a water-related absorption peak near 1383 nm, once a serious problem, has been almost removed in modern "low-water-peak" fibre. After 80 km at 0.2 dB/km, 2.5 per cent of the light remains, which an erbium-doped fibre amplifier restores; at the 3 dB/km of window glass at its best, essentially nothing would.

Research continues on fibres with lower loss and higher capacity: multicore fibres, and hollow-core fibres that guide light in air inside a glass microstructure, which have recently been reported with losses comparable to or below those of the best solid fibres, and with about 30 per cent less delay, since light in air travels nearly one and a half times as fast as in glass.

ComputationC

Quantum dots: colour from size

Shrink a semiconductor crystal to a few nanometres and its gap grows. An electron and hole confined in a crystal smaller than their natural orbit size (the exciton Bohr radius, about 5 nm in CdSe) behave like particles in a box, whose energies rise as the box shrinks (Glasses and nanomaterials introduced nanoscale effects; quantum mechanics for chemists solved the box). Alexei Ekimov saw the effect in 1981 in nanocrystals of copper chloride grown inside glass, in Leningrad; Louis Brus at Bell Labs found it independently in colloidal suspensions in 1983 and explained it; and Moungi Bawendi's group at MIT made it practical in 1993 with a synthesis that produces nearly identical crystals by injecting precursors into a hot solvent. The three shared the 2023 Nobel Prize in Chemistry.

For a sphere of radius RR with impenetrable walls, the lowest s state has wavefunction ψ∝sin⁡(πr/R)/r\psi \propto \sin(\pi r/R)/r and energy ℏ2π2/(2m∗R2)\hbar^2\pi^2/(2m^*R^2), the particle-in-a-sphere result. Brus's model adds the confinement energies of the electron and hole to the bulk gap and subtracts their screened Coulomb attraction:

E(R)=Eg+ℏ2π22R2(1me∗+1mh∗)−1.786 e24πε0εrR.E(R) = E_g + \frac{\hbar^2\pi^2}{2R^2}\left(\frac{1}{m_e^*} + \frac{1}{m_h^*}\right) - \frac{1.786\,e^2}{4\pi\varepsilon_0\varepsilon_rR}.

The confinement term grows as 1/R21/R^2 and the Coulomb term only as 1/R1/R, so for small dots confinement wins. The program evaluates the model for cadmium selenide and then tests its most questionable assumption, the infinite walls, by solving the particle in a sphere with a finite barrier.

python
import numpy as np

HB2M = 0.0380998            # hbar^2 / (2 m_e) in eV nm^2
E2_4PIEPS0 = 1.439965       # e^2 / (4 pi eps0) in eV nm

# CdSe, commonly used values: bulk gap (eV), electron and hole effective masses (m_e), relative permittivity
Eg, me, mh, eps = 1.74, 0.13, 0.45, 10.6

def brus(R):
    """Lowest exciton energy of a sphere of radius R (nm): infinite walls (Brus, 1984)."""
    confinement = HB2M * np.pi**2 / R**2 * (1 / me + 1 / mh)    # particle in a sphere, ground state j0 zero = pi
    coulomb = 1.786 * E2_4PIEPS0 / (eps * R)
    return Eg + confinement - coulomb, confinement, coulomb

print("diameter/nm   confinement/eV   Coulomb/eV   gap/eV   emission/nm")
for d in [2, 3, 4, 5, 6, 8, 10]:
    E, conf, coul = brus(d / 2)
    print(f"    {d:3}          {conf:6.3f}         {coul:6.3f}     {E:6.3f}     {1239.84 / E:6.0f}")

# Walls are not infinite: a dot is surrounded by ligands or a shell, a barrier of finite height V0.
# For an s state in a spherical well, u = r psi satisfies u'' = -k^2 u inside and u'' = kappa^2 u outside,
# and matching u'/u at r = R gives k cot(kR) = -kappa. Solve for the electron's confinement energy.
def finite_sphere(R, m, V0):
    E = np.linspace(1e-6, V0 - 1e-6, 200001)
    k = np.sqrt(E * m / HB2M)
    kap = np.sqrt((V0 - E) * m / HB2M)
    f = k * np.cos(k * R) + kap * np.sin(k * R)        # zero where k cot(kR) = -kappa
    s = np.flatnonzero(np.sign(f[:-1]) != np.sign(f[1:]))
    return E[s[0]] if s.size else None

print("\nelectron confinement energy in CdSe (eV): infinite walls vs a 1.0 eV barrier")
for d in [2, 3, 4, 6]:
    R = d / 2
    inf = HB2M * np.pi**2 / (me * R**2)
    fin = finite_sphere(R, me, 1.0)
    print(f"  d = {d} nm:  infinite {inf:6.3f}   finite {fin:6.3f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

For dots 4 to 10 nm across, the model predicts emission running from blue-green (486 nm) through yellow and orange to red (674 nm) as the dots grow, approaching bulk CdSe's near-infrared edge. That is the familiar row of glowing vials, all the same compound, differing only in size. For smaller dots the model fails badly: it puts a 2 nm dot's gap at 5.2 eV, in the ultraviolet, while real 2 nm CdSe dots emit blue light. The second table shows one reason. With a realistic finite barrier, the electron's wavefunction leaks out of the dot and its confinement energy is much lower: 0.95 eV instead of 2.9 eV for a 2 nm dot. The bands also stop being parabolic at such high energies. In practice chemists size their dots with empirical calibration curves that link measured absorption peaks to sizes measured by electron microscopy.

1.62.02.53.03.4246810dot diameter (nm)lowest gap (eV)bulk CdSe 1.74 eV4 nm: about 486 nm, blue-green6 nm: about 598 nm, orangemodel unreliable below 3 nm
The lowest optical gap of CdSe quantum dots against diameter, from the Brus model computed above (solid curve), with the bulk gap of 1.74 eV (dashed). The gap rises steeply as the dots shrink below about 5 nm. Below about 3 nm the model overestimates the gap, because real barriers are finite and bands are not parabolic.

Quantum dots are now in products. Displays use them to convert blue backlight or blue OLED light into pure green and red, giving a wider range of colours; because cadmium is restricted in consumer electronics in many places, many such displays use indium phosphide dots instead. Dots are also used as fluorescent labels in biology, where their narrow, size-tuned emission lets many labels be read at once, and are being developed for infrared cameras, solar cells and single-photon sources for quantum communication.

ComputationC

Perovskite solar materials and the efficiency limit

How efficient can a solar cell be? In 1961 William Shockley and Hans-Joachim Queisser answered with an argument of beautiful generality, detailed balance. A cell with gap EgE_g absorbs only photons above EgE_g, and of each photon's energy it keeps only EgE_g; the excess is lost as heat. And because a good absorber is also a good emitter, the cell, warmed to room temperature and biased at a voltage VV, must itself radiate photons above EgE_g at a rate that grows as eqV/kBTe^{qV/k_BT}. The current is the difference between absorbed and emitted photons. The program computes the resulting maximum efficiency for a cell under unconcentrated sunlight, modelling the Sun as a 5778 K black body, and then for two cells stacked, with the wider gap on top.

python
import numpy as np

q, h, c, kB = 1.602176634e-19, 6.62607015e-34, 2.99792458e8, 1.380649e-23

E = np.linspace(0.3, 5.0, 9401)                      # photon energy grid, eV
dE = E[1] - E[0]

def photon_flux(T):
    """Black-body photon flux per unit area per eV leaving a flat surface (photons s^-1 m^-2 eV^-1)."""
    x = E * q / (kB * T)
    return 2 * np.pi / (h**3 * c**2) * (E * q) ** 2 / np.expm1(x) * q

def above(spectrum):
    """Integral of the spectrum over all photon energies above each grid energy."""
    return np.cumsum(spectrum[::-1])[::-1] * dE

# The Sun as a 5778 K black body, diluted by its small angular size (no atmosphere, no concentration)
f_sun = (6.957e8 / 1.496e11) ** 2
sun = f_sun * photon_flux(5778.0)
P_in = np.sum(sun * E * q) * dE                      # incident power, W/m^2
cell = photon_flux(300.0)                            # a cell at 300 K radiates like a black body above its gap
N_sun, N_cell = above(sun), above(cell)

def best(i_gap, N_abs=None):
    """Maximum power (W/m^2) of an ideal cell with gap E[i_gap]: J(V) = q (N_abs - N_cell exp(qV/kT))."""
    Jsc = q * (N_sun[i_gap] if N_abs is None else N_abs)
    J0 = q * N_cell[i_gap]
    V = np.linspace(0, E[i_gap], 2000)
    P = V * (Jsc - J0 * np.expm1(V / 0.025852))
    return P.max()

print(f"incident power: {P_in:.0f} W/m^2")
print(" gap/eV   efficiency (detailed balance)")
effs = []
for g in np.arange(0.6, 2.41, 0.1):
    i = int(round((g - 0.3) / dE))
    eff = best(i) / P_in
    effs.append((eff, g))
    print(f"  {g:4.2f}      {100 * eff:5.1f} %  " + "#" * int(round(100 * eff)))
eff, g = max(effs)
print(f"best single junction: {100 * eff:.1f} % at {g:.1f} eV")

# Two cells stacked, each with its own contacts: the top cell takes photons above its gap,
# the bottom cell those between the two gaps
best_pair = (0, 0, 0)
for g_top in np.arange(1.4, 2.21, 0.02):
    it = int(round((g_top - 0.3) / dE))
    top = best(it)
    for g_bot in np.arange(0.8, 1.41, 0.02):
        ib = int(round((g_bot - 0.3) / dE))
        bot = best(ib, N_abs=N_sun[ib] - N_sun[it])
        if top + bot > best_pair[0]:
            best_pair = (top + bot, g_top, g_bot)
P, gt, gb = best_pair
print(f"best two-gap stack: {100 * P / P_in:.1f} % with gaps {gt:.2f} and {gb:.2f} eV")
it, ib = int(round((1.68 - 0.3) / dE)), int(round((1.12 - 0.3) / dE))
pair = best(it) + best(ib, N_abs=N_sun[ib] - N_sun[it])
print(f"a 1.68 eV perovskite on 1.12 eV silicon, ideal: {100 * pair / P_in:.1f} %")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The single-junction limit peaks at about 31 per cent for gaps between 1.2 and 1.4 eV and falls off on both sides: small gaps waste most of each photon's energy as heat, large gaps let most photons pass. (With the real solar spectrum after passing through the atmosphere, the standard figure is about 33 per cent near 1.34 eV; the program's black-body Sun gives 1342 W/m², close to the measured 1361 W/m² above the atmosphere.) Silicon at 1.12 eV and gallium arsenide at 1.42 eV both sit near the top. Two stacked cells raise the limit to about 43 per cent, best with gaps near 1.8 and 0.9 eV, and a 1.68 eV top cell on silicon reaches 41 per cent ideally. That last pairing is the reason for the excitement about perovskites.

01020300.61.01.41.82.2band gap (eV)efficiency limit (%)SiGaAsperovskite
The detailed-balance (Shockley–Queisser) efficiency limit for a single-junction solar cell under unconcentrated sunlight modelled as a 5778 K black body, computed above, against band gap. Silicon (1.12 eV) and GaAs (1.42 eV) lie near the maximum of about 31 per cent; a wide-gap perovskite (about 1.7 eV) is less efficient alone but is well matched as the top cell of a tandem on silicon.

Halide perovskites have the same ABX₃ structure as the barium titanate of the previous lesson but with different ions: a large organic or caesium cation A (methylammonium, formamidinium or Cs⁺), lead (or tin) as B, and iodide, bromide or chloride as X. Methylammonium lead iodide has a direct gap of about 1.6 eV; replacing some iodide with bromide raises the gap continuously towards 2.3 eV, which makes a 1.7 eV top cell for silicon easy to design. They absorb strongly, so films half a micrometre thick suffice, and they can be deposited from solution or vapour at temperatures near 100 °C rather than the 1400 °C of silicon. Most remarkably, they are defect tolerant: the defects that form most easily in them create states near the band edges rather than deep in the gap, so carriers survive long enough to be collected even in films made quickly and cheaply.

Progress has been unusually fast. Tsutomu Miyasaka's group in Yokohama reported the first perovskite solar cells in 2009, at 3.8 per cent efficiency, using the perovskite as a dye in a liquid cell. In 2012 groups led by Nam-Gyu Park and Michael Grätzel, and by Henry Snaith, made solid-state cells near 10 per cent. As of 2025, certified single-junction perovskite cells had passed 26 per cent and perovskite-on-silicon tandems were close to 35 per cent in laboratory records, above the best silicon cells; the first commercial tandem modules shipped in 2024, and manufacturers in several countries run pilot lines. Records are tracked on the best research-cell efficiency chart maintained by the US national renewable-energy laboratory, which is the place to check current values.

ExplanationH

The professional view

Measurement. Optical materials are characterised by spectrophotometry (transmission and reflection), ellipsometry, photoluminescence (shining light and measuring the light re-emitted, which reveals gaps, defects and efficiency), and for LEDs and solar cells by quantum efficiency measured against calibrated standards. Solar-cell efficiencies count as records only when certified by a small number of accredited test centres under standard conditions (1000 W/m², a standard spectrum and 25 °C).

Computation. Band gaps and optical spectra are computed with the density functional methods of Density functional theory, with corrections (hybrid functionals or the GW method) because simple functionals underestimate gaps; databases such as the Materials Project are screened for new transparent conductors, phosphors and absorbers. Machine-learning searches for lead-free perovskites and stable wide-gap absorbers are among the examples in AI-assisted materials discovery.

Where it is going (as of 2026). Micro-LED displays, which need red, green and blue LEDs only micrometres across and run into the green gap and surface losses; ultraviolet AlGaN emitters for water disinfection; thin-film lithium niobate and silicon nitride photonic circuits (Lumenary EE's photonics); perovskite–silicon tandems moving into manufacturing; and structured materials that control light by their geometry rather than their chemistry, the Metamaterials of the frontier level.

Common misconceptionsH

Common misconceptions

Misconception

"A material's colour is the colour of light it absorbs." It is the colour of the light that remains. Cadmium sulfide absorbs violet and blue and looks yellow; the colour we see is made of the wavelengths it does not absorb.

Misconception

"An LED emits at exactly its band gap." The emission peak lies close to the gap, but it is broadened by the thermal spread of carriers (a few kBTk_BT), shifted by quantum-well confinement and strain, and in alloys such as InGaN by composition fluctuations. LED makers sort chips into colour bins because of this spread.

Misconception

"Smaller quantum dots are redder." The opposite: confinement raises the gap, so smaller dots emit bluer light, and larger dots approach the bulk material's colour.

Misconception

"With a good enough material, one solar cell could convert almost all sunlight." A single gap cannot use photons below it, and wastes the excess energy of those above it, as heat. Detailed balance limits a single junction under ordinary sunlight to about a third; going beyond requires several gaps (tandems), concentration, or as yet unproven schemes such as hot-carrier collection.

Exercises

Exercises

1. Optical materials

Warm-upHuman

Assuming an ideal sharp absorption edge at the band gap, which of these thick crystals absorb all visible light (400–700 nm) and so look black, apart from surface reflection? Select all that apply.

Choose all that apply

2. Optical materials

Warm-upHuman

Gallium phosphide has a band gap of 2.26 eV. At what wavelength is its absorption edge, in nanometres (nearest whole number)?

3. Optical materials

CoreHuman

In the near infrared, gallium arsenide has refractive index n=3.6n = 3.6 and negligible absorption. What percentage of light striking a bare, polished GaAs surface at normal incidence is reflected (nearest whole number)?

4. Optical materials

CoreHuman

Silicon's gap of 1.12 eV corresponds to a perfectly useful infrared wavelength (1107 nm). Why are silicon light-emitting diodes nevertheless extremely inefficient?

Choose one

5. Optical materials

CoreComputational

Pulses in a fibre travel at the group velocity c/ngc/n_g, where the group index is ng=n−λ dn/dλn_g = n - \lambda\,dn/d\lambda.

Using the Sellmeier equation for fused silica (Malitson's coefficients, provided in the starter), write group_index(lam_um) returning ngn_g at wavelength lam_um micrometres. Compute the derivative numerically with a central difference.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

6. Optical materials

StretchComputational

Using the Brus model from the lesson,

E(R)=Eg+ℏ2π22R2(1me∗+1mh∗)−1.786 e24πε0εrR,E(R) = E_g + \frac{\hbar^2\pi^2}{2R^2}\left(\frac{1}{m_e^*} + \frac{1}{m_h^*}\right) - \frac{1.786\,e^2}{4\pi\varepsilon_0\varepsilon_rR},

with CdSe values Eg=1.74E_g = 1.74 eV, me∗=0.13m_e^* = 0.13, mh∗=0.45m_h^* = 0.45, εr=10.6\varepsilon_r = 10.6, write brus_gap(d_nm) returning the gap in eV for a dot of diameter d_nm, and diameter_for(wavelength_nm) returning the diameter (nm) whose gap emits at that wavelength (λ=1239.84/E\lambda = 1239.84/E). Search diameters between 3 and 20 nm. Use ℏ2/2me=0.0380998\hbar^2/2m_e = 0.0380998 eV nm² and e2/4πε0=1.439965e^2/4\pi\varepsilon_0 = 1.439965 eV nm.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

7. Optical materials

Warm-upHuman

The detailed-balance limit on the efficiency of a single-junction solar cell, about 33 per cent under standard sunlight, is named after two physicists who published it in 1961. Give their surnames, joined by a hyphen or dash.

Further reading and resources