Semiconductor Diode Lasers

A semiconductor diode laser takes the two-band picture and the Fermi–Dirac occupation statistics of solid-state physics and drives them so far out of equilibrium that a slab of crystal amplifies light. In a gas or a ruby crystal the gain lives between sharp atomic levels; here it lives between two continua — the conduction and valence bands of a direct-gap semiconductor — and the pump is nothing more exotic than a current across a forward-biased p-n junction. Injected electrons recombine with holes across the gap and emit photons of energy near the bandgap $E_g$ . Because the bands are broad and continuous, so is the gain, and the emission wavelength is set by composition, temperature, and drive current rather than by an immutable atomic transition. Everything the laser chapter established is reused here, but wearing semiconductor clothing. Population inversion becomes a statement about two separate quasi-Fermi levels, $E_{Fc}$ for electrons and $E_{Fv}$ for holes; the gain condition becomes the Bernard–Duraffourg inequality $E_{Fc}-E_{Fv}>h\nu\ge E_g$ ; the threshold condition becomes a balance of modal gain against facet and internal loss; the resonator is simply the two cleaved crystal facets, made reflective by the index contrast with air. The one genuinely new idea is geometric: the double heterostructure, which uses bandgap steps to trap carriers and index steps to trap photons in the same sub-micron slab. Co-confining gain and light is what turned a cryogenic, pulsed curiosity into the room-temperature device that lights every fibre link, reads every disc, and cools every atom trap.

Key takeaways

A diode laser is a light-emitting p-n junction inside a resonator: forward bias injects electrons and holes into a shared active region, and their radiative recombination across the bandgap supplies gain near $h\nu\approx E_g$ .
Inversion is the Bernard–Duraffourg condition — the quasi-Fermi separation must beat the photon energy, $E_{Fc}-E_{Fv}>h\nu\ge E_g$ — so that stimulated emission outweighs interband absorption.
The double heterostructure co-confines carriers (bandgap steps) and photons (index steps) in a thin active layer, cutting threshold current density from $\sim10^{4}$ – $10^{5}$ to $\sim10^{3}\,\mathrm{A\,cm^{-2}}$ and enabling room-temperature CW operation.
Above threshold the gain clamps, so power rises linearly, $P_{out}=\eta_d(h\nu/e)(I-I_{th})$ , with wall-plug efficiencies now above 50–70%.
The cleaved facets ( $R\approx0.31$ from $n\approx3.5$ ) make a Fabry–Pérot cavity with closely spaced modes; a Bragg grating (DFB/DBR) selects a single mode for telecom, and material choice tunes the wavelength from GaN violet to InGaAsP mid-IR.
Key idea: gain in a diode laser is not between atomic levels but between the conduction and valence bands; the inversion criterion is the Bernard–Duraffourg condition $E_{Fc}-E_{Fv}>h\nu\ge E_g$ , and heterostructures win by co-confining carriers and photons to slash the threshold current.

Where this lives today

Every photon that carries the internet begins in a diode laser. In long-haul fibre links, InGaAsP distributed-feedback lasers at $1.55\,\mu\mathrm{m}$ sit at the transmit end of coherent transceivers, their single-mode Bragg-grating output modulated at tens of gigabaud and phase-encoded so one wavelength carries hundreds of gigabits. In the data center the geometry flips: vertical-cavity surface-emitting lasers (VCSELs) at $850\,\mathrm{nm}$ — a diode laser whose mirrors are epitaxial Bragg stacks above and below the active region rather than cleaved facets — are cheap, testable on-wafer, and array easily, so they dominate short-reach interconnects and flood-illuminate the infrared dots behind smartphone face recognition. At the precision end, the external-cavity diode laser (ECDL) rules atomic physics: a diode chip with an anti-reflection-coated facet, fed back by a diffraction grating, tunes across an atomic resonance with a linewidth well below a megahertz. These cool rubidium at $780\,\mathrm{nm}$ and caesium at $852\,\mathrm{nm}$ in magneto-optical traps, interrogate optical-clock transitions, and drive neutral-atom and trapped-ion quantum computers — the same Bernard–Duraffourg gain and rate-equation clamping analyzed above, in a device the size of a grain of rice.

The physical picture

Picture the conduction band as an almost-empty upper shelf and the valence band as an almost-full lower shelf, split by the bandgap $E_g$ . Forward-biasing the junction pumps electrons onto the upper shelf and drains the lower one, leaving holes. Once enough carriers are injected, at photon energies just above $E_g$ the upper states near the band edge are more likely occupied than the corresponding lower states, so a passing photon is more likely to stimulate a downward transition (emission) than to be absorbed. That population imbalance across two continua, not a pair of atomic levels, is the gain. The trick that makes it practical is a set of engineered walls. A double heterostructure sandwiches a thin low-gap active layer (GaAs) between higher-gap cladding (AlGaAs): the bandgap steps form a bathtub that keeps injected electrons and holes from diffusing away before they recombine, while the accompanying refractive-index step forms a waveguide core that keeps the optical mode overlapping the same slab. Carriers and light are trapped together in a volume a fraction of a micron thick, so a modest current builds the huge local carrier density — of order $10^{18}\,\mathrm{cm^{-3}}$ — that gain demands.

Figure

Band profile of a GaAs double heterostructure under forward bias — the active GaAs well has gap E_g ≈ 1.42 eV while the AlGaAs cladding gap is wider, so injected electrons ↓ and holes ○ pool together in the well and the quasi-Fermi split E_Fc − E_Fv exceeds hν ≈ E_g, the Bernard–Duraffourg gain condition

The light–current curve then tells the whole story. Below threshold the diode merely glows — broadband spontaneous emission, an ordinary LED. At threshold the round-trip gain equals the loss; above it the modal gain, and the carrier density that sets it, clamp, so every extra injected electron can no longer raise the population and is funnelled into stimulated emission instead — a power that is a straight line in current kinked sharply at $I_{th}$ . Because the gain slab is a fraction of a micron thick while the cavity is a few hundred microns long, the photon lifetime is a few picoseconds — short enough to modulate the current at many GHz, so the same device that makes the light also encodes the data.

Figure

Light–current characteristic of a diode laser — below the threshold current I_th ≈ 20 mA the junction only glows as an LED, then above threshold the gain clamps and optical power rises linearly with slope η_d ≈ 0.5 W per A, reaching P ≈ 50 mW at I = 120 mA

The core equations

The gain criterion is the semiconductor analogue of population inversion and the foundation for everything else. For a vertical transition between a valence state at $E_v$ and a conduction state at $E_c=E_v+h\nu$ , the emitting rate beats the absorbing rate only when the upper state is more likely filled than the lower; cast in quasi-Fermi statistics this gives a clean threshold on the quasi-Fermi-level separation.

E_{Fc}-E_{Fv} \;>\; h\nu \;\ge\; E_g \qquad (1)

where: $E_{Fc},\,E_{Fv}$ — quasi-Fermi levels of the conduction and valence bands under injection; $h\nu$ — photon energy; $E_g$ — bandgap. The reading: the injection-driven splitting of the two Fermi levels must exceed the photon energy, and since a transition needs real states at both ends, $h\nu$ can be no smaller than $E_g$ . Turning the inequality into an actual gain spectrum requires knowing how many transitions are available at each frequency and how strongly inverted they are. The material gain per unit length combines the joint density of states with the inversion factor $f_c-f_v$ , scaled by the radiative lifetime.

g(\nu) = \frac{c^2}{8\pi n^2 \nu^2 \tau_r}\,\rho(\nu)\,\big[f_c(\nu)-f_v(\nu)\big] \qquad (2)

where: $g(\nu)$ — material gain per unit length; $n$ — refractive index of the active material; $\tau_r$ — radiative recombination lifetime; $\rho(\nu)$ — reduced (joint) density of states; $f_c-f_v$ — inversion factor, positive only when (1) holds. The prefactor is Einstein's stimulated coefficient in disguise; the physics of inversion sits entirely in the bracket, which changes sign exactly at transparency. A laser needs the gain not merely positive but large enough to overcome loss on a round trip. Demanding that the field reproduce itself after one round trip of the Fabry–Pérot facet cavity fixes the threshold modal gain — the confinement-weighted gain the mode actually experiences.

g_{th} = \alpha_i + \frac{1}{2L}\ln\!\frac{1}{R_1 R_2} \qquad (3)

where: $g_{th}$ — threshold modal gain; $\alpha_i$ — internal (distributed) loss per unit length; $L$ — cavity length between facets; $R_1,R_2$ — facet power reflectivities. The second term is the mirror loss $\alpha_m$ : light leaking out of the ends is the useful loss, distinct from the parasitic $\alpha_i$ that is simply wasted. Those facet reflectivities cost nothing to make — they come free from cleaving the crystal, because a semiconductor with $n\approx3.5$ against air is already a decent mirror by Fresnel's normal-incidence formula.

R = \left(\frac{n-1}{n+1}\right)^2 \qquad (4)

where: $R$ — power reflectivity at the semiconductor–air facet; $n$ — refractive index (~3.5 for GaAs). For $n=3.5$ this gives $R=(2.5/4.5)^2\approx0.31$ : only 31%, yet ample, because the material gain runs to hundreds of $\mathrm{cm^{-1}}$ . Above threshold the device obeys a strikingly simple law. Because the gain clamps at $g_{th}$ , the carrier density and hence the spontaneous-recombination drain also clamp, so all injected current beyond $I_{th}$ becomes stimulated photons — a straight light–current line.

P_{out} = \eta_d\,\frac{h\nu}{e}\,(I - I_{th}) \qquad (5)

where: $P_{out}$ — emitted optical power; $\eta_d$ — differential (slope) quantum efficiency; $h\nu/e$ — photon energy per electron, in volts; $I,\,I_{th}$ — drive and threshold currents. The slope efficiency splits as $\eta_d=\eta_i\,\alpha_m/(\alpha_i+\alpha_m)$ : internal efficiency times the fraction of loss that escapes as output rather than being dissipated inside. The short cavity also sets the mode structure. Requiring an integer number of half-wavelengths between the facets spaces the longitudinal modes by a free spectral range that, for a diode's tiny length, is enormous.

\Delta\nu_L = \frac{c}{2 n_g L} \qquad (6)

where: $\Delta\nu_L$ — frequency spacing of adjacent longitudinal modes; $n_g$ — group index of the waveguide mode; $L$ — cavity length. For $L=300\,\mu\mathrm{m}$ and $n_g\approx4$ , $\Delta\nu_L\approx125\,\mathrm{GHz}$ (about $1.0\,\mathrm{nm}$ near $1.55\,\mu\mathrm{m}$ ) — wide, but the broad gain curve still admits several modes, so a plain Fabry–Pérot diode hops between them. To force a single mode, one abandons flat facet feedback for wavelength-selective feedback: a periodic index corrugation etched along (DFB) or beside (DBR) the waveguide reflects coherently only at the Bragg wavelength.

\lambda_B = 2\,n_{eff}\,\Lambda \qquad (7)

where: $\lambda_B$ — vacuum wavelength reflected by the grating; $n_{eff}$ — effective index of the guided mode; $\Lambda$ — first-order grating period. Only the mode satisfying (7) sees constructive back-reflection, so it wins the gain competition and lases single-mode — the spectral purity a dispersion-limited fibre link demands.

Derivation AThe Bernard–Duraffourg inversion condition

The goal is to derive the semiconductor gain criterion from the balance of stimulated emission against absorption and show it collapses to a single inequality on the quasi-Fermi-level separation. We treat a vertical transition between a valence state at $E_v$ and a conduction state at $E_c=E_v+h\nu$ : an incident photon can be absorbed (electron promoted up) or can stimulate emission (electron dropped down), and each rate is weighted by the occupation of the states involved. Net gain requires the emission rate to beat the absorption rate — so the whole question reduces to comparing two Fermi occupations, the direct analogue of atomic population inversion built from band statistics.

Step 1. Write the quasi-Fermi occupations, $f_c(E_c)=1/\big[1+e^{(E_c-E_{Fc})/k_BT}\big]$ and $f_v(E_v)=1/\big[1+e^{(E_v-E_{Fv})/k_BT}\big]$ , the probabilities that the upper and lower states hold electrons. Why: under injection each band thermalizes internally but not with the other, so each needs its own Fermi level — the split forward bias creates.

Step 2. Form the net stimulated rate. Emission needs the upper state full and lower empty, weight $f_c(1-f_v)$ ; absorption the reverse, $f_v(1-f_c)$ ; the difference is $f_c(1-f_v)-f_v(1-f_c)=f_c-f_v$ , so $g(\nu)\propto f_c-f_v$ . Why: the mixed $f_cf_v$ terms cancel, leaving a clean inversion factor that is positive exactly when the upper state is more occupied.

Step 3. Impose $f_c>f_v$ . Since the Fermi function decreases in its argument, this is equivalent to $E_c-E_{Fc}<E_v-E_{Fv}$ , i.e. $E_{Fc}-E_{Fv}>E_c-E_v=h\nu$ . A transition also needs real states, so $h\nu\ge E_g$ , giving $E_{Fc}-E_{Fv}>h\nu\ge E_g$ . Why: this converts occupation probabilities into an energy inequality the experimenter controls through injection, and pins the gain window between two energies — no gain below $E_g$ , none above the quasi-Fermi split.

Step 4. Read the physics. The quasi-Fermi separation is set by injection and is roughly the junction voltage times $e$ , so demanding it exceed $h\nu$ demands a forward bias of order the bandgap voltage. Why: it explains why a GaAs laser needs about $1.4\,\mathrm{V}$ and a GaN laser about $3\,\mathrm{V}$ — the drive voltage is set by the photon you want out. *Verified: reproduces the standard Bernard–Duraffourg result and Saleh and Teich Eq. (17.1-7ff); the factor $f_c-f_v$ changes sign exactly at transparency $E_{Fc}-E_{Fv}=h\nu$ .*

Go deeper · Derivation B · 2 minLight–current characteristic and slope efficiency from the rate equations

Derivation BLight–current characteristic and slope efficiency from the rate equations

The goal is to show that a working diode laser produces a straight light–current line above threshold and to identify the slope in terms of measurable loss ratios. We start from the coupled carrier and photon rate equations, look for the steady state, and exploit the single most important fact about a laser above threshold — that the gain clamps. The payoff, $\eta_d=\eta_i\,\alpha_m/(\alpha_i+\alpha_m)$ , separates the two ways a photon can be lost and tells a designer that every parasitic loss $\alpha_i$ is stolen directly from the output.

Step 1. Write the rate equations for carrier number $N$ and photon number $S$ : $\mathrm dN/\mathrm dt=\eta_i I/(eV)-N/\tau-v_g g S$ and $\mathrm dS/\mathrm dt=\Gamma v_g g S-S/\tau_p+\Gamma\beta N/\tau$ , with $V$ the active volume, $v_g$ group velocity, $\Gamma$ confinement factor, $\tau_p$ photon lifetime, $\beta$ spontaneous coupling. Why: the stimulated term $v_g g S$ appears with opposite signs in the two equations — it drains carriers to feed photons — which is the coupling that makes a laser a laser.

Step 2. Impose the photon steady state. With $S>0$ and $\beta$ negligible, $\mathrm dS/\mathrm dt=0$ forces $\Gamma v_g g=1/\tau_p$ , i.e. modal gain equals total loss $v_g(\alpha_i+\alpha_m)$ ; hence $g=g_{th}$ and, through $g(N)$ , the carrier density clamps. Why: the photon field acts as a servo — any rise of $N$ above threshold boosts stimulated emission, which pulls $N$ back down, so $g$ and $N$ cannot exceed threshold.

Step 3. Solve the carrier equation with $N$ clamped: the drain $N/\tau$ is fixed, so all injection beyond threshold flows into the stimulated term, $S=\Gamma\tau_p\,\eta_i(I-I_{th})/(eV)$ . The emitted power is the photon energy times the rate the useful loss empties the cavity, $P_{out}=h\nu\,V\,v_g\alpha_m\,S/\Gamma$ . Why: the excess current maps one-to-one onto stimulated photons — the origin of the linear curve — and only the mirror loss $\alpha_m$ delivers usable light.

Step 4. Substitute $S$ and collect terms: $P_{out}=\eta_i\,\dfrac{\alpha_m}{\alpha_i+\alpha_m}\,\dfrac{h\nu}{e}\,(I-I_{th})$ , so $\eta_d=\eta_i\,\alpha_m/(\alpha_i+\alpha_m)$ . Why: the loss ratio is the fraction of each recombination photon that escapes as output, which is why lowering $\alpha_i$ or raising $\alpha_m$ steepens the curve — up to the point where higher $\alpha_m$ raises the threshold. *Verified: standard rate-equation result matching Coldren and Corzine and Saleh and Teich; recovers $\eta_d\to\eta_i$ as $\alpha_i\to0$ and $\eta_d\to0$ as $\alpha_m\to0$ .*

Go deeper · Derivation C · 2 minThreshold current density and the heterostructure advantage

Derivation CThreshold current density and the heterostructure advantage

The goal is to expose in one line of algebra why the double heterostructure and the quantum well were revolutionary. A carrier balance at threshold relates the threshold current density to the active-layer thickness and gives $J_{th}\propto d$ : because the game is to reach a fixed threshold carrier density with the least current, thinning the active layer directly buys a lower threshold — until the mode outgrows the layer and the confinement factor fights back. This single proportionality organizes the whole history of the device, from cryogenic homojunctions to room-temperature quantum wells.

Step 1. Balance carriers per unit area at steady state: the injection rate reaching the active layer $\eta_i J/e$ must equal the recombination rate $d\,n/\tau$ for a layer of thickness $d$ , density $n$ , lifetime $\tau$ , so $\eta_i J/e=d\,n/\tau$ . At threshold $n=n_{th}$ , the density where $\Gamma g(n_{th})=g_{th}$ , giving $J_{th}=\dfrac{e\,d}{\eta_i}\,\dfrac{n_{th}}{\tau}$ . Why: in steady state every injected carrier must recombine, and $n_{th}$ is fixed by the cavity, so the only geometric knob left in $J_{th}$ is the thickness $d$ .

Step 2. Read $J_{th}\propto d$ . A homojunction confines nothing, so carriers spread over $d\sim1$ – $2\,\mu\mathrm{m}$ ; a double heterostructure squeezes them into $d\sim0.1$ – $0.2\,\mu\mathrm{m}$ ; a single quantum well reaches $d\sim10\,\mathrm{nm}$ . Each step down in $d$ is a step down in threshold. Why: the recombination that must be fed scales with the carrier volume, so a thinner slab needs proportionally less current to reach the same density.

Step 3. Respect the confinement caveat. As $d$ shrinks below the mode size the confinement factor $\Gamma$ falls, so the material gain needed to hit $g_{th}=\Gamma g$ rises; there is an optimum thickness, and multiple quantum wells recover $\Gamma$ by stacking several thin wells. Why: gain and confinement pull opposite ways as $d\to0$ , which is why real devices settle on a few nanometre-scale wells rather than one vanishingly thin one.

Step 4. Attach the history. Homojunction GaAs lasers needed $J_{th}\sim5\times10^{4}\,\mathrm{A\,cm^{-2}}$ and ran only pulsed at 77 K; double heterostructures brought $J_{th}$ to $\sim10^{3}\,\mathrm{A\,cm^{-2}}$ , enabling room-temperature CW operation in 1970. Why: that hundredfold drop is the difference between a laboratory curiosity and the device in every fibre transmitter — exactly what the 2000 Nobel Prize recognized. *Verified: consistent with the historical DH threshold reduction and Saleh and Teich §17.3; the $J_{th}\propto d$ scaling underlies the homojunction-to-quantum-well progression.*

Limiting cases

Case	Result	Meaning
Below threshold (LED regime)	$I<I_{th}$	Spontaneous emission dominates; power grows slowly and the spectrum is broad — the device is a light-emitting diode, not a laser.
Transparency	$E_{Fc}-E_{Fv}=h\nu$	Net gain is exactly zero — stimulated emission balances absorption; below this carrier density the material absorbs, above it, it amplifies.
Bulk versus quantum well	$L_z\lesssim\lambda_{dB}$	When the active thickness approaches the electron de Broglie wavelength the 3D $\sqrt{E}$ density of states becomes a 2D staircase, lowering transparency current and sharpening the gain.
High temperature	$I_{th}(T)=I_0\,e^{T/T_0}$	Threshold rises empirically with characteristic temperature $T_0$ (about 120 K for GaAs, 50–70 K for InGaAsP telecom lasers) as carrier spreading and Auger loss worsen.

Common trapThe p-n junction itself is the gain medium, and threshold is just a diode turn-on current

Both halves mislead. Gain does not live in the bulk junction the way resistance lives in a resistor; it lives in a sub-micron active layer where the two quasi-Fermi levels straddle the photon energy, and only while Bernard–Duraffourg holds. Nor is threshold a property of the diode I–V curve, which turns on near the bandgap voltage regardless of lasing. Threshold is a current-density phenomenon: $J_{th}$ is fixed by cavity and material, so narrowing the stripe lowers the threshold current at the same $J_{th}$ — which is why real lasers are narrow stripes, not broad slabs. And the kink in the light–current curve is gain clamping, not a diode knee: below it an LED, above it a laser, the straight line the fingerprint of stimulated emission taking over.

Worked examples

1 · What colour does the bandgap set? (confidence builder)

A GaAs active layer has bandgap $E_g=1.42\,\mathrm{eV}$ at room temperature. What is the emission wavelength? Then find the bandgap a quaternary alloy needs to emit at the $1.55\,\mu\mathrm{m}$ telecom loss minimum.

Step 1. Emission is near the band edge, $h\nu\approx E_g$ , so $\lambda\approx hc/E_g$ . With the convenient constant $hc=1240\,\mathrm{eV\,nm}$ , $\lambda=1240/1.42=873\,\mathrm{nm}$ — GaAs emits in the near infrared at about $0.87\,\mu\mathrm{m}$ , the wavelength of the first practical diode lasers and of CD pickups.

Step 2. Invert for telecom: to emit at $1550\,\mathrm{nm}$ the material needs $E_g=1240/1550=0.80\,\mathrm{eV}$ , far below GaAs. No binary sits there, which is why the telecom workhorse is quaternary InGaAsP grown lattice-matched to InP with its composition tuned to place $E_g$ at $0.80\,\mathrm{eV}$ . (A GaN violet diode at $405\,\mathrm{nm}$ needs $3.06\,\mathrm{eV}$ — a wide-gap nitride.) GaAs emits at 873 nm ≈ 0.87 µm; a 1.55 µm telecom laser needs $E_g\approx0.80$ eV, supplied by InGaAsP on InP. Unlike an atomic laser, you dial the wavelength by choosing the bandgap.

2 · Threshold gain of a cleaved Fabry–Pérot laser

A GaAs laser has cavity length $L=300\,\mu\mathrm{m}=0.030\,\mathrm{cm}$ , uncoated facets ( $R_1=R_2=0.31$ ), internal loss $\alpha_i=10\,\mathrm{cm^{-1}}$ , and confinement factor $\Gamma=0.30$ . Find the mirror loss, the threshold modal gain, and the material gain the active layer must supply.

Step 1. Mirror loss from equation (3): $\alpha_m=\dfrac{1}{2L}\ln\dfrac{1}{R_1R_2}$ . Here $R_1R_2=0.31\times0.31=0.0961$ , so $\ln(1/0.0961)=\ln(10.41)=2.34$ , and $\dfrac{1}{2L}=\dfrac{1}{0.060\,\mathrm{cm}}=16.7\,\mathrm{cm^{-1}}$ . Thus $\alpha_m=16.7\times2.34=39\,\mathrm{cm^{-1}}$ .

Step 2. Threshold modal gain is the total loss: $g_{th}=\alpha_i+\alpha_m=10+39=49\,\mathrm{cm^{-1}}$ . Notice the mirror loss dominates — a short cavity leaks strongly through its ends, which is the price of the compactness that gives GHz modulation.

Step 3. The mode overlaps the active layer only with weight $\Gamma$ , so the material gain must be larger: $g=g_{th}/\Gamma=49/0.30=163\,\mathrm{cm^{-1}}$ . The active layer must supply about $1.6\times10^{2}\,\mathrm{cm^{-1}}$ of material gain — hundreds of $\mathrm{cm^{-1}}$ , which is why an uncoated 31% facet is a perfectly adequate mirror while an atomic gas (gain of order $1\,\mathrm{cm^{-1}}$ ) could never tolerate such lossy mirrors.

Step 4. Finish with the output power. With internal efficiency $\eta_i=0.80$ , the slope efficiency is $\eta_d=\eta_i\,\alpha_m/(\alpha_i+\alpha_m)=0.80\times39/49=0.64$ , and at $1.55\,\mu\mathrm{m}$ the power slope is $\eta_d(h\nu/e)=0.64\times0.80\,\mathrm{V}=0.51\,\mathrm{W\,A^{-1}}$ . Driving $100\,\mathrm{mA}$ above threshold then gives $P_{out}=0.51\times0.100=0.051\,\mathrm{W}$ . The same laser delivers about 51 mW at a slope of 0.51 W/A — matching the light–current figure, and halving $\alpha_i$ would lift $\eta_d$ to $0.71$ , a direct measure of what each $\mathrm{cm^{-1}}$ of internal loss costs.

Physics of the portrait

The diode laser is the rare device whose decisive idea was theoretical before it was buildable: the double heterostructure was proposed as a way to co-confine carriers and light years before anyone could grow the atomically abrupt interfaces it required. Two physicists share that story, and one of them turned the proposal into the first laser that ran at room temperature.

Origins · 1 minZhores Alferov and how this was discovered

Zhores Alferov

1930–2019 · double-heterostructure semiconductor lasers

Zhores Alferov, at the Ioffe Institute in Leningrad, is the physicist who made the heterostructure laser real. The idea that a thin low-gap layer between wide-gap cladding could trap both carriers and photons was proposed around 1963 by Herbert Kroemer and, independently, by Alferov; Kroemer's paper was famously rejected as having no practical application, and the epitaxial growth it needed did not yet exist. Through the late 1960s Alferov's group mastered liquid-phase epitaxy of the lattice-matched AlGaAs/GaAs system, and in 1970 — essentially simultaneously with Izuo Hayashi and Morton Panish at Bell Labs — they demonstrated the first diode laser to run continuously at room temperature, driving the threshold current density down by two orders of magnitude from the cryogenic homojunction devices of 1962. That single advance put diode lasers into fibre transmitters, compact-disc players, and laser pointers. Alferov shared the 2000 Nobel Prize in Physics with Kroemer "for developing semiconductor heterostructures used in high-speed and opto-electronics" (the other half went to Jack Kilby for the integrated circuit). The heterostructure he built underlies not only every diode laser but the high-electron-mobility transistors in every mobile phone.

History

1962

— first GaAs homojunction diode lasers operate, pulsed and only at 77 K

1963

— Kroemer and Alferov independently propose the double-heterostructure laser

1970

— Alferov, and Hayashi and Panish, achieve room-temperature continuous-wave operation

1996

— Nakamura demonstrates the first InGaN blue-violet laser diode

2000

— Alferov and Kroemer share the Nobel Prize for semiconductor heterostructures

Going further

Texts. Saleh and Teich, Fundamentals of Photonics, 3rd ed., ch. 17–18 · Coldren, Corzine and Mašanović, Diode Lasers and Photonic Integrated Circuits, 2nd ed. (the definitive rate-equation and threshold treatment) · Hecht, Optics, 5th ed., §13.1 · Foot, Atomic Physics, §1.7 and ch. 7 (diode lasers for atomic physics) · Loudon, The Quantum Theory of Light, 3rd ed. · Demtröder, Laser Spectroscopy, Vol. 1, ch. 5.

Provenance. The equations are standard semiconductor-laser results (Saleh and Teich ch. 17; Coldren, Corzine and Mašanović ch. 2–5): the Bernard–Duraffourg condition and interband gain, the threshold modal gain and light–current law from the rate equations, the Fresnel facet reflectivity, the free spectral range, and the Bragg condition. A verifier should recompute: $R=(2.5/4.5)^2\approx0.31$ for $n=3.5$ ; $\Delta\nu_L\approx125\,\mathrm{GHz}$ and mode spacing $\approx1.0\,\mathrm{nm}$ at $1.55\,\mu\mathrm{m}$ for $L=300\,\mu\mathrm{m},\,n_g=4$ (this corrects the base article's "0.3 nm," a factor-of-three error); $\lambda=1240/E_g$ giving $873\,\mathrm{nm}$ for GaAs and $E_g=0.80\,\mathrm{eV}$ for $1.55\,\mu\mathrm{m}$ ; and the Example 2–3 chain $\alpha_m=39$ , $g_{th}=49$ , $g=163\,\mathrm{cm^{-1}}$ , $\eta_d=0.64$ , $P=51\,\mathrm{mW}$ . The $J_{th}$ ranges (homojunction $\sim10^{4}$ – $10^{5}$ , DH $\sim10^{3}\,\mathrm{A\,cm^{-2}}$ ), $T_0$ values ( $\sim120\,\mathrm{K}$ GaAs, $50$ – $70\,\mathrm{K}$ InGaAsP), and Nobel attributions (2000 heterostructures; 2014 blue nitride) are standard.

Prerequisites. Population Inversion and Optical Gain · Stimulated Emission and the Gain Coefficient · Laser Threshold Condition · Optical Resonators and Cavity Modes. Leads to. Longitudinal Modes and Free Spectral Range · Cavity Q Factor and Photon Lifetime · Laser Rate Equations · Gaussian Beams.

Ask the tutor: Ask me to walk through why thinning the active layer lowers threshold but eventually fights the confinement factor, or to work a full threshold-current and slope-efficiency estimate for a laser you specify.

Why does a diode laser need "two" Fermi levels — isn't there only one Fermi level in a solid?

In thermal equilibrium a semiconductor has a single Fermi level and no net gain. A diode laser is driven far from equilibrium: forward bias floods the conduction band with electrons and the valence band with holes faster than they recombine. Within each band the carriers thermalize in femtoseconds, so each band reaches its own quasi-equilibrium with its own quasi-Fermi level — $E_{Fc}$ for electrons, $E_{Fv}$ for holes — while the two bands are not in equilibrium with each other. The split $E_{Fc}-E_{Fv}$ measures how hard you are pumping, and gain appears only when it exceeds the photon energy. That split is the semiconductor's version of population inversion.

If the gain clamps at threshold, how can I get more power out — doesn't more current just do nothing?

More current does plenty — it just cannot raise the gain. Above threshold the modal gain is pinned at the value that balances the loss, because any excess would make the photon number run away and deplete the carriers straight back to threshold; so the carrier density, gain, and spontaneous-recombination rate all clamp. What does not clamp is stimulated emission: every extra electron above $I_{th}$ recombines by stimulating a photon into the lasing mode, so the photon number and output power climb linearly with $(I-I_{th})$ . The clamp is on gain, not power; power is limited instead by heating and catastrophic optical damage at the facet.

How is population inversion in a semiconductor different from a helium-neon or ruby laser?

The bookkeeping is analogous but the physics is broader. An atomic laser inverts a pair of discrete levels; a semiconductor inverts two continuous bands, so the condition must hold at each photon energy separately and takes the form of the Bernard–Duraffourg inequality on quasi-Fermi levels. The consequences differ sharply: the gain spectrum is broad (tens of nm) rather than a sharp line, the wavelength is set by the bandgap and tunes with composition, temperature, and current, and the required carrier densities ( $\sim10^{18}\,\mathrm{cm^{-3}}$ ) and gains (hundreds of $\mathrm{cm^{-1}}$ ) dwarf those of a dilute gas. The upside is that the pump is just an electric current.

What stops a plain Fabry–Pérot diode from running single-mode, and how does a DFB laser fix it?

The gain spectrum is broad — tens of nanometres — while the Fabry–Pérot mode spacing for a $300\,\mu\mathrm{m}$ cavity is only about $1\,\mathrm{nm}$ , so a dozen or more longitudinal modes sit under the gain curve with nearly equal net gain; the laser picks whichever is momentarily favoured and hops between them as current and temperature drift. A distributed-feedback laser replaces the flat facet feedback with a Bragg grating along the waveguide, which reflects strongly only at $\lambda_B=2n_{eff}\Lambda$ . That wavelength-selective loss hands one mode a decisive advantage, giving single-mode emission with side-mode suppression above 40 dB — essential so that fibre dispersion does not smear high-bit-rate pulses.

Why do cleaved facets work as mirrors when they reflect only about 30 percent?

Because the gain is colossal. Fresnel gives $R=((n-1)/(n+1))^2\approx0.31$ for $n\approx3.5$ , so each facet returns under a third of the light — very lossy by the standards of a gas laser (dielectric mirrors above 99.9%). But the material gain of an inverted semiconductor runs to $10^{2}$ – $10^{3}\,\mathrm{cm^{-1}}$ , so even over a $300\,\mu\mathrm{m}$ cavity the single-pass gain easily overcomes the mirror loss $\alpha_m\approx39\,\mathrm{cm^{-1}}$ . The high loss is even useful: most intracavity light is coupled out as usable power rather than trapped, part of why diode lasers reach such high wall-plug efficiency. Coatings mainly tailor the output — a high reflector on the back, an anti-reflection facet for external-cavity tuning.

Why does raising the temperature make a diode laser worse, and what is

T_0

Heating hurts on several fronts. Higher temperature smears the Fermi–Dirac occupation over a wider energy range, so a given carrier density gives less peak gain; carriers leak more easily over the barriers; and in narrow-gap telecom materials, Auger recombination — non-radiative and steeply temperature-dependent — becomes a major carrier sink. The net effect is an empirical rise $I_{th}(T)=I_0\,e^{T/T_0}$ , where the characteristic temperature $T_0$ measures tolerance: larger $T_0$ is a flatter, better device. GaAs has $T_0\approx120\,\mathrm{K}$ but InGaAsP telecom lasers only $T_0\approx50$ – $70\,\mathrm{K}$ , which is why long-haul transmitters are temperature-stabilized with a thermoelectric cooler.