Notes de Cours Technique des Lasers

Sylvain Barbay

June 1, 2023
Contents
1 Properties of laser light 3
1.1 Coherence function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Quasi-monochromatic light . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Spectral linewidth of lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.1 Statistical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Exercise: degree of coherence of an idealized multi-longitudinal
mode laser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 Basics of semiconductor lasers 14

2.1 Semiconductors as laser material . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.0.1 Optical gain in semiconductor materials . . . . . . . . . . 14
2.1.0.2 Quantum confined semiconductor materials . . . . . . . . 16
2.1.0.3 Alpha factor . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.0.4 Optical confinement in semiconductor lasers . . . . . . . 18
2.1.0.5 Optical and Electrical injection is semiconductor lasers . 19
2.1.1 Semiconductor laser rate equations . . . . . . . . . . . . . . . . . . 20
2.1.1.1 Single-mode laser rate equations . . . . . . . . . . . . . . 20
2.1.1.2 Steady-state response . . . . . . . . . . . . . . . . . . . . . 22
2.1.1.3 Multi-mode laser rate equations . . . . . . . . . . . . . . . 23
2.1.1.4 Complex laser rate equations . . . . . . . . . . . . . . . . . 23
2.1.1.5 Small signal analysis . . . . . . . . . . . . . . . . . . . . . 24

2
1 Properties of laser light
In this chapter we review some important properties of laser light. Broadly speaking, a
laser is a system that emits coherent light in a well defined spatial mode. Lasers usu-
ally emit a high-spectral purity electromagnetic field in a confined beam. The spectral
purity ensures that it is possible to have interference fringes interfering an optical
beam following two different paths. This is the notion of temporal coherence. Another
notion of coherence refers to spatial coherence, i.e. the ability to have interference
fringes from different point sources of a laser beam. These two notions are encoded
in the coherence function Γ which we will define in this chapter. We will see how
this function can help in clarifying the relation between the spectral components of a
laser beam and its coherence properties. We will examine what impacts the coherence
properties of lasers.

1.1 Coherence function

Suppose we want to analyze the coherence of a light field characterized by its scalar
complex electric field given in each point is space and time by E(x, y, z, t) ≡ E(r, t).
The physical field (measurable field) is simply defined as the real part of the complex
field Er = Re[E]. We introduce the first order coherence function Γ(1)

Γ(1) (r1 , t1 ; r2 , t2 ) = ⟨E ∗ (r1 , t1 )E(r2 , t2 )⟩

where ⟨. . .⟩ stands for some kind of averaging procedure that we will specify later.
If we are only interested in the temporal coherence we may take r1 = r2 and discard
the spatial dependence such that
(1)
Γ12 (t1 , t2 ) = ⟨E1∗ (t1 )E2 (t2 )⟩
Now in real physical situations we can make two more assumptions that will sim-
plify the above definition. First, we will consider that the averaging procedure intro-
duced (it can be a statistical average for electric fields modeled by a random variable,
or a temporal average if the field is well defined by a simple function of time) does
not depend on the initial choices of t1 or t2 , but only depends on the time difference
τ ≡ t2 − t1 . This is the stationary assumption. Second, we suppose we deal with
ergodic fields, in that the statistical (ensemble) averaging is the same as the temporal
averaging. These allow us to rewrite
Z T /2
(1) 1
Γ12 (τ = t2 − t1 ) = lim E1 (t + τ )E2∗ (t)dt
T →∞ T −T /2

where we have introduced a time average over a time T .

A photodetector used to measure the intensity of a light field in the laboratory is
only sensitive to the average over many periods of the modulus square of the electric
field. It can thus be related to the first order coherence function and we have
1 1 (1)
I(r) = ϵ0 c⟨E1∗ (t)E1 (t)⟩ = ϵ0 cΓ11 (0).
2 2

3
 Figure 1.1: Michelson interferometer

In other words, the intensity is just the self-coherence function for zero time delay τ .
By dimensional analysis, we see that the first order coherence function has dimension
of V 2 L−2 , and since [ϵ0 ] = F L−1 = QV −1 L−1 and [c] = LT −1 we have [I] = QV L−2 T −1 =
JT −1 L−2 . The intensity is thus the power density of the incident beam measured in
W/m2 .
As an application, let us analyze the Michelson interferometer depicted in Figure
1.1. We want the intensity at the photodetector PD at point P. Incident light has
intensity I0 , and Michelson arms have a lengths l1 and l2 respectively. The beamsplit-
ter BS splits the incident energy equally between the two arms. Then we can write
E(P, t) = 21 E(P0 , t1 ) + 21 E(P0 , t2 ): the amplitude of the light impinging on the photode-
tector at time t is the sum of the amplitudes entering in the Michelson interferometer
and traveling in each arm, thus leaving P0 at two different instants in the past. The
factor 1/2 comes from the fact that the light intensity going through one arm is re-
flected and transmitted by the beamsplitter and thus divided by 4, hence a factor 2 in
amplitude. The corresponding intensity in P is just I(P ) = 21 ϵ0 c⟨E(P, t)E ∗ (P, t)⟩. Using
the previous definitions we write
1 1 1
I(P ) = ϵ0 c⟨ (E(P0 , t − τ1 ) + E(P0 , t − τ2 )) (E(P0 , t − τ1 ) + E(P0 , t − τ2 ))∗ ⟩
2 2 2
where we have introduced τ1,2 = 2l1,2 /c. By expanding the previous formula we have

1
I(P ) = ϵ0 c×
 2
1 1
⟨E(P0 , t − τ1 )E ∗ (P0 , t − τ1 )⟩ + ⟨E(P0 , t − τ2 )E ∗ (P0 , t − τ2 )⟩+
4 4

1 ∗ 1 ∗
⟨E(P0 , t − τ1 )E (P0 , t − τ2 )⟩ + ⟨E (P0 , t − τ2 )E(P0 , t − τ2 )⟩
4 4
 
1 1 1 1 ∗ 1 ∗
I(P ) = I0 + I0 + ϵ0 c ⟨E(t)E (t + τ )⟩ + ⟨E (t)E(t + τ )⟩
4 4 2 4 4
with τ = τ2 − τ1 .  
1 1 1 ∗
I(P ) = I0 + ϵ0 c Re⟨E(t + τ )E (t)⟩
2 2 2

4
Figure 1.2: Interferogram in a Michelson interferometer for monochromatic light. Left:
power spectral density of the source.

and
!!
(1)
I0 Γ11 (τ )
I(P ) = 1 + Re (1)
2 Γ11 (0)
(1)
Γ11 (τ )
We introduce the degree of self-coherence γ(τ ) = (1) . It can be shown that |γ(τ )| ≤
Γ11 (0)
1. Hence

I0
I(P ) = (1 + Re (γ(τ )))
2
For monochromatic light, E(t) = E0 ei(k·r−ωt+Φ0 ) and γ(τ ) = e−iωτ = e−i2πντ . Therefore,
I(P) is an oscillating function of time difference τ .

1.1.1 Quasi-monochromatic light

Monochromatic light is an idealization which does not exist in nature. Real light fields
have a non-zero bandwidth and consist of many frequencies. If the frequencies are
closely packed then we speak of a quasi-monochromatic light field (see Figure 1.3).
Then we can describe the field in terms of a carrier at THz frequencies modulated by
a "slow" amplitude such that we can write

E(t) = A(t)eiΦ(t) e−i2πν̄t

with a carrier frequency ν̄ and a complex amplitude A(t)eiΦ(t) . The function A(t) and
Φ(t) are real and "slow" functions of time.
The complex field can be decomposed into Fourier components such that
Z
E(t) = E(ν)e−i2πνt dν

and Z
A(t)eiΦ(t)
= E(ν)e−i2π(ν−ν̄)t dν

Let us calculate the first order coherence function for such a light field. We have

5
 Figure 1.3: Spectral content of a quasi-monochromatic field.

Z T /2
(1) 1
Γ11 (τ ) = lim E(t + τ )E ∗ (t)dt
T →∞ T −T /2
Z T /2 Z ∞ Z ∞
1 −i2πν(t+τ ) ′
= lim dt E(ν)e dν E ∗ (ν ′ )ei2πν t dν ′
T →∞ T −T /2 −∞ −∞
ZZ Z T /2
1 ′
= lim E(ν)E ∗ (ν ′ )e−i2πντ dνdν ′ ei2π(ν−ν )t dt
T →∞ T −T /2

The second integral on the right hand side can be readily evaluated to yield

sin (π(ν − ν ′ )T )
ZZ
(1)
Γ11 (τ ) = lim E(ν)E ∗ (ν ′ )e−i2πντ dνdν ′
T →∞ π(ν − ν ′ )T
The cardinal sine function will shrink to Dirac delta function when T → ∞ (this can
be proved rigorously) therefore

ZZ
(1)
Γ11 (τ ) = E(ν)E ∗ (ν ′ )e−i2πντ δ(ν − ν ′ )dνdν ′
Z
= E(ν)E ∗ (ν)e−i2πντ dν

If we introduce the power spectral density W (ν) = |E(ν)|2 we find the Wiener-
Kinchine theorem
Z +∞
(1)
Γ11 (τ ) = W (ν)e−i2πντ dν
−∞
The first order correlation function is the Fourier transform of the power spectral
density.
By inversion we also get
Z +∞
(1)
W (ν) = Γ11 (τ )ei2πντ dτ.
−∞

This theorem is valid for a general stationary ergodic field. It is very important here
since it relates a quantity that measures the temporal coherence of a light source to
its spectral content. If one knows the spectral content, then the correlation function
is known. We can also introduce the normalized spectral density (as plotted in Figure
1.3)
W (ν)
g(ν) = R ∞
0 W (ν)dν

6
 Figure 1.4: Gaussian light pulse: amplitude of the pulse in blue, full field in red.

(1)
We re going to use this theorem to compute Γ11 for a Gaussian pulse. Let us
consider a short optical pulse of width σ (see Figure 1.4)

E0 −t2 /2σ2 −i2πν̄t

E(t) = √ e e
2πσ
First we need to compute the spectral density of this pulse W (ν) = |E(ν)|2 with
Z +∞
E 2 2
E(ν) = √ 0 e−t /2σ e−i2πν̄t ei2πνt dt
−∞ 2πσ
We re-arrange
Z +∞
E 2 2
E(ν) = √ 0 e−t /2σ ei2π(ν−ν̄)t dt
−∞ 2πσ
We can make appear in the exponential a square (because we know how to integrate
Gaussians), though here the term is complex
Z +∞ √ √
E0 2 2 2 2
E(ν) = √ e−(t/ 2σ−i 2σπ(ν−ν̄)) e−2σ π (ν−ν̄) dt (1.1)
2πσ −∞
One can show that the result of the integration is
2 π 2 (ν−ν̄)2
√
2 /(1/ 2σπ)2
E(ν) = E0 e−2σ = E0 e−(ν−ν̄)
The spectral density W (ν) for a Gaussian wavepacket consists of a Gaussian cen-
tered at the carrier frequency ν̄, whose width directly depends on the pulse width:
the larger the pulse (σ large), the smaller the Gaussian wavepacket width (∝ 1/σ), and
vice-versa.
We use this result to compute the first-order self coherence function

Z ∞
(1)
Γ11 (τ ) = E(ν)E ∗ (ν)e−i2πντ dν
Z−∞
∞
= W (ν)e−i2πντ dν
−∞
Z ∞
(1) 2 π 2 (ν−ν̄)2
Γ11 (τ ) = E02 e−4σ e−i2πντ dν
−∞

7
 Figure 1.5: Degree of coherence of a Gaussian pulse. We plot I(τ ) ≡ 1 + Re[γ(τ )].

And the final result is (by using standard results on the Fourier transform1 ).

(1) E2 2 2
Γ11 (τ ) = √ 0 e−τ /4σ e−2iπν̄τ
2 πσ
2 2
γ(τ ) = e−τ /4σ e−2iπν̄τ
The function is plotted in Figure 1.5.
We see that the degree of coherence is an oscillating function of decreasing ampli-
tude with the delay τ . The fast oscillations occur with a period 1/ν0 ≡ 1/ν̄, i.e. they
testify of the central wavelength of the wavepacket. The characteristic length of the
decay of the coherence is linked to the spectral width of the source ∆ν and we have
∆ν ∝ σ. The larger the spectral content, the faster the degree of coherence decays,
i.e. the less coherent is the light. For a spectral bandwidth tending towards zero, we
recover the case of a monochromatic light.

1.1.1.0.1 Note on the calculus of the FT of a Gaussian

• The integral (Eq. 1.1) to be evaluated can be rewritten with a change of variable
E0 −2σ2 π2 (ν−ν̄)2 √
Z +∞ √ 2
E(ν) = √ e 2σ e−(u−i 2σπ(ν−ν̄)) du
2πσ −∞
−(u−ia)2 √
We can write the complex exponential e by introducing a = 2σπ(ν − ν̄).

2 2 2
1
F T [e−at ] = e−π ν /a
pπ
a
and F T [g(t − a)] = exp(−i2πνa)G(ν)

8
 Figure 1.6: Integration contours.

We will make use of the residue theorem to evaluate the integral. R −z 2 We know that for a
holomorphic function without poles, the contour integration γ e dz = 0 for a closed
contour γ = Γ1 ∪ Γ2 ∪ Γ3 ∪ Γ4 defined R in −z
Figure 1.6.
Rt
2 2
On the bottom contour we have Γ1 e dz = −t e−t dt. On the upper contour we get
−z 2 dz = −t e−(t−ia)2 dt since z = t − ia. This is the integral to be evaluated when
R R
Γ3 e t
t → ∞, with a minus sign because of the integration path on decreasing abscissa. On
2 2 2 2
the left and right contours Γ2 and Γ4 we R have e−z R= e−(±t+iy) =R e−(t −y )∓2ity . Hence,
2 2 2 2
for a contour that extends from ±∞: Γ2 e−z dz ≤ Γ2 |e−z |dz = Γ2 e−(t −y ) dy −−−−→ 0
t→+∞
because y is bounded. The same is true for the other contour.
Therefore
E0 −2σ2 π2 (ν−ν̄)2 +∞ −u2
Z
E(ν) = √ e e du
π −∞
√
The last integral gives π so we get the result.

• Another method to make the integration is based on differential equations. Let

us consider again
Z +∞
E 2 2
E(ν) = √ 0 e−t /2σ ei2π(ν−ν̄)t dt
−∞ 2πσ
By differentiating with respect to ν and by integration by parts we get

Z +∞
dE(ν) E 2 2
= √ 0 i2πte−t /2σ ei2π(ν−ν̄)t dt
dν −∞ 2πσ
 Z +∞ 
i2πE0 2
h 2 2
−t /2σ i2π(ν−ν̄)t
i+∞
2 −t2 /2σ 2 i2π(ν−ν̄)t
= √ −σ e e + σ i2π(ν − ν̄) e e dt
2πσ −∞ −∞
−4π 2 σ(ν − ν̄)E0 +∞ −t2 /2σ2 i2π(ν−ν̄)t
Z
= √ e e dt
2πσ −∞
= −4π 2 σ 2 (ν − ν̄)E(ν)

We thus get a first order differential equation for E(ν) that we can easily solve
2 2 2
to getR E(ν) = E(ν̄)e−2π σ (ν−ν̄) . The integration constant is evaluated using E(ν̄) =
+∞ 2 2
√E0 e−t /2σ dt = E0 .
2πσ −∞

1.2 Spectral linewidth of lasers

Up to now we have introduced tools to characterize the temporal coherence of laser
light, be we have not analyzed the coherence properties of laser light from first prin-
ciples. In the following we are going to review the classical theory of laser coherence
and discuss the various physical and technical phenomena that affect the temporal
coherence of lasers.

9
 Figure 1.7: Coherence of light train.

1.2.1 Statistical approach

We consider here a laser emitting a quasi-monochromatic field represented by the
complex amplitude E(t) = A(t)e−i(ω0 t+Φ(t)) where we consider the slow envelope and
phase A(t) and Φ(t) respectively. As usual, the slowness is appreciated by comparing
to the carrier angular frequency in the hundreds of THz range for visible light.
If we look at the wave train emitted by the laser source, we might observe, as
depicted in Figure 1.7, a regular modulation at the carrier frequency interrupted at
random instant of times by phase jumps. In a laser, an emitter undergoes random
phase changes due to various processes, e.g. collisions of atoms in a gas, spontaneous
emission, ... We are going to show how these random phase jumps affect the linewidth
of lasers. By using the previous definition of the first order coherence function we can
write
Γ(1) (τ ) = ⟨A∗ (t)A(t + τ )e−i(ω0 τ +∆ϕ) ⟩
where the phase difference is defined as ∆ϕ = ϕ(t + τ ) − ϕ(t). We assume that
the amplitude fluctuations of the laser light are not correlated with and independent
of the phase fluctuations. We also assume that the amplitude follows a stationary
process. We can thus factorize Γ(τ ) ≃ ⟨|A|2 ⟩e−iω0 τ ⟨e−i∆ϕ ⟩. We now have to evaluate the
last term, which we can rewrite
Z
−i∆ϕ
⟨e ⟩ = e−i∆ϕ p(∆ϕ)d(∆ϕ) (1.2)

by introducing a probability density p for the phase difference ∆ϕ. In order to

compute this average, we need to know the probability density p. We can assume
that the phase difference fluctuations follow a random process similar to a Brownian
motion, i.e. the phase jumps are independent. This process is thus a Wiener process2 .
Therefore, the probability density function p(∆ϕ) for the stochastic process ∆ϕ is a
normal distribution function such that:
1 2 2
p(∆ϕ) = p e−(∆ϕ) /2σ |τ | .
2π|τ |σ
The second order moment of ∆ϕ writes ⟨(∆ϕ)2 ⟩ = σ 2 τ . So the variance of the fluctua-
tions grows linearly with time, as in a standard Brownian motion. The average phase
2
fluctuations as defined in Eq. 1.2 can thus be computed to give ⟨e−i∆ϕ ⟩ = e−|τ |σ /2 .
This allows us to compute the first order coherence function. As previously, the co-
herence function decays exponentially with delay τ and have oscillations at the carrier
frequency. It is interesting to compute the power spectral density of the source, by
invoking the Wiener-Kinchine theorem. We thus have

2
Wiener process W: Wt+τ − Wt follows a normal distribution with a variance proportional to τ

10
 Z ∞
W (ν) = Γ(1) (τ )ei2πντ dτ
Z−∞
∞
2 /2
= ⟨|A|2 ⟩e−i2πν0 τ e−|τ |σ ei2πντ dτ
−∞

We finally obtain after performing the integral

4⟨|A|2 ⟩σ 2
W (ν) =
σ 4 + 16π 2 (ν − ν0 )2
The power spectral density takes the form of the well known Lorentzian lineshape
with FWHM σ 2 /4π. As a consequence, since the observation of laser lineshapes of-
ten yields Lorentzian lineshapes, we can argue that our physical statistical model of
random phase fluctuations can adequately account for the observations. In our dis-
cussion we have attributed these fluctuations to various processes (atomic collisions
for gas lasers, spontaneous emission, ...). The case of spontaneous emission is partic-
ularly important because it is a fundamental quantum process which is unavoidable.
In fact one can show that the fundamental quantum limit of laser lineshape comes
from the spontaneous emission processes. We will see that later on when we intro-
duce the Schallow-Townes theory of laser linewidth. Before going to this point, and
as an exercise, let us compute the coherence degree of a multimode laser.

1.2.2 Exercise: degree of coherence of an idealized multi-longitudinal mode

laser
For this exercise we consider an idealized multi-longitudinal mode laser. Let us sup-
pose that the laser emits in five equispaced modes that we model by Dirac delta
functions. Here the idealization comes from the fact that we neglect any spontaneous
emission process and the individual laser linewidth have zero width. The laser field
can thus reads

2
X
E(ν) = an δ(ν − νn ) (1.3)
n=−2
2
X
= an δ(ν − ν0 − n∆ν)
n=−2

r (t) = Re[E(t)] = Re[ E(ν)e−i2πνt dν] =

R
where each laser line has frequency. The real field is E
E (ν)e−i2πνt dν. We know from the fact the the field is real
R r
p πthat necessarily we must
have E r (−ν) = E r (ν)∗ . We
pπalso know that F T [cos (2πν 0 t)] = 2 (δ(ν − ν0 ) + δ(ν + ν0 )) and
that F T [sin (2πν0 t)] = i 2 (δ(ν −p ν0 ) − δ(ν + ν0 )). Therefore p by linearity of the Fourier
π π
Transform F T [exp (−i2πν0 t)] = (δ(ν − ν0 ) + δ(ν + ν 0 )) + 2 (δ(ν − ν0 ) − δ(ν + ν0 )) =
√ 2
2πδ(ν − ν0 ). This shows that Eq. 1.3 effectively describes a real electric field com-
posed of several different frequencies. We call ν0 the central frequency of the comb
and each frequency is separated from its neighboring mode by ∆ν. The problem is to
compute Γ(1) in that case (we may assume for simplicity that all the an = 1).

• Answer

We compute Γ(1) :

11
 Z ∞
Γ(1) (τ ) = E(ν)E ∗ (ν)e−i2πντ dν
−∞
Z ∞ 2
X 2
X
= e−i2πντ dν δ(ν − νn )δ(ν − νm )
−∞ n=−2 m=−2
2
X Z ∞
= e−i2πντ δ(ν − νn )δ(ν − νm )dν
n,m=−2 −∞
2
X
= e−i2πνn τ
n=−2
2
X
n
= e−i2πν0 τ e−i2π∆ντ
n=−2

We have thus to evaluate a partial sum of powers. We will make use of a standard
procedure to compute this finite series, by expressing it in terms of difference of
infinite series which we know to compute. To make things clearer we evaluate
M
X
n
S(x) = e−ix
n=−M

We can start by rearranging the summation such that it starts at n = 0

2M
X
n
S(x) = eiM x e−ix
n=0
Then we may express the summation in terms of infinite series
∞ ∞
!
−ix n −ix n
X X
iM x



S(x) = e e − e
n=0 n=2M +1

By playing the same trick as before we can rearrange

∞
X
n  
S(x) = eiM x e−ix 1 − e−i(2M +1)x
n=0
1 − e−i(2M +1)x


iM x
=e
1 − e−ix
by summing the series. Then S(x) reads finally
−i(2M +1)x/2
ei(2M +1)x/2 − e−i(2M +1)x/2


iM x e
S(x) = e
e−ix/2 (eix/2 − e−ix/2 )
sin ((2M + 1)x/2)
=
sin(x/2)
Coming back to our calculation, setting x = 2π∆ντ and M = 2 we have
sin (5π∆ντ )
Γ(1) (τ ) = e−i2πν0 τ
sin(π∆ντ )
A generalization to M modes would have given

12
Figure 1.8: Plot of the real part of Γ(1) for: left, ν0 = 10 and ∆ν = 0.1 and right, ν0 = 10
and ∆ν = 0.2.

Figure 1.9: Coherence length and typical linewidth of some semiconductor lasers.

sin(N π∆ντ ) sin(π∆νBW τ )

Γ(1) (τ ) = e−i2πν0 τ = e−i2πν0 τ
sin(π∆ντ ) sin(π∆ντ )
where ∆νBW is the total bandwidth of the laser modes. The first order correla-
tion function thus consists of a rapidly oscillating function (the complex exponential)
multiplied by an envelope which oscillates less rapidly as can be seen on Figure 1.8.
The envelope decays rapidly and gets to 0 at τ = 1/∆νBW = (N ∆ν)−1 . It oscillates
and then for still larger delays there is a revival since the real part of the coherence
function reaches again 1. In this model, there are an infinite number of revivals
because of the zero linewidth of the individual laser lines. It could be shown that for
non-zero linewidths, the coherence would fall off at infinity. Just looking at the initial
decay close to τ = 0, the typical timescale for coherence loss is given by the inverse
of the bandwidth of the source, and this is a general result: the larger the source
bandwidth, the smaller is the coherence time. For a Lorentzian lineshape, we can
express the coherence length Lc as
c
Lc = cτcoh =
π∆νFWHM
Typical semiconductor laser coherence length and linewidths are plotted on Figure
1.9. For these lasers, the typical linewidth is of the order of several GHz which cor-
responds to less than one meter of coherence length. We see by this example that it
is important to have a these numbers in mind, in particular if one wants to use the
coherence property of lasers for some measurement.

13
2 Basics of semiconductor lasers
Semiconductor lasers constitute about half of the lasers sold today worldwide. They
are ubiquitous in the telecommunication industry, but are also present in everyday
consumer products such as cashier desks, optical storage readers, cars, gaming, and
even face recognition on some high-end smartphones. Semiconductor materials differ
in a number of way of two-level systems which are used for introductory laser courses.
They have their own specificity which is important to bear in mind when one wants
to design a laser system. In this chapter we introduce some basic laser theory as well
as discuss some important characteristics of semiconductor laser gain materials.

2.1 Semiconductors as laser material

Semiconductor lasers are very commonly used because they are compact and energy-
efficient laser sources. This comes from the fact that laser gain materials have intrin-
sically a very high gain, of the order or 103 cm−1 . First we review some properties of
laser gain in semiconductor materials, and then we will introduce a basic semicon-
ductor laser theory.

2.1.0.1 Optical gain in semiconductor materials

Semiconductor materials considered for optical gain usually belong to compounds
of the III-V columns of the periodic table of elements. They are usually found in
the AlGaAs, InGaAs, InP, GaP material families. Note that group IV semiconductor
materials such as the ubiquitous Si or Ge are not standard optical gain materials
because they are indirect bandgap materials: the bottom of the conduction band and
the peak of the valence band do not coincide. Therefore, an optical transition where a
hole in the valence band and an electron in the conduction band are involved in the
creation or destruction of a photon will not be favored because it does not respect the
momentum conservation. Allowed optical transitions in semiconductors are vertical
in the energy-wavevector band diagram.

Figure 2.1: Band structure of a highly-doped or pumped semiconductor material.

Empty circles: holes. Full circles: electrons.

14
Figure 2.2: Typical gain vs wavelength/energy in a (GaIn)(NAs)/GaAs quantum well
ridge waveguide laser structure. Negative gain means absorption. Source
Wikipedia.

A pumped or highly-doped semiconductor material band structure is depicted on

Figure 2.1. It is characterized by a non-equilibrium situation with the simultaneous
presence of a high density of electrons and holes in the same space. In that case,
we can nevertheless define quasi-equilibrium quasi-Fermi levels, describing the oc-
cupancy of states fc,v in the valence and conduction bands respectively, such that
fc,v (E) = (E−EFc,v1 )/kB T . The transition from a → b where a photon of energy h̄ω(k)
e +1
is emitted has a probability proportional to Ra→b = fc (Ea ) (1 − fv (Eb )): it is propor-
tional to the probability of occupancy by an electron of a state of energy Ea times the
probability of occupancy by a hole (or electron vacancy) of the lower state (1 − fv (Eb )).
Population inversion ∆N for a given k can be described in this framework as the
difference of transition probabilities times the density of states ρ(k) available at this
wavevector ∆N = ρ(k)dkV {Ra→b − Rb→a } = ρ(k)dkV {fc (Ea ) − fv (Eb )}. This latter equation
is the equivalent definition of population inversion for Fermi-Dirac statistics as com-
pared to Boltzmann statistics in standard two-level systems. Population inversion
requires then than fc (Ea ) − fv (Eb ) > 0 or in other words
1 1
>
e(Ea −EFc )/kB T +1 e(Eb −EFv )/kB T +1
and hence h̄ω0 = Ea −Eb < EFc −EFv . This condition is called the Bernard and Duraf-
fourg condition. It shows that only photons whose energies are below the quasi-Fermi
energy level difference are amplified. Higher energy photons are mostly absorbed, and
photons whose energy are just at the quasi-Fermi energy level difference undergo nei-
ther gain nor absorption. This occurs for a definite pump and is called transparency.
Note also that there cannot be gain for photon energy less than the bandgap energy
Eg since transitions are then not allowed.
An experimental gain curve for a semiconductor material emitting around 1.3µm is
shown in Figure 2.2.Depending on the current (pump) injected into the laser, different
gain curves are measures. For zero and 5mA injected current there is no gain and only
absorption. For gain above 10mA, a zone of wavelength shows positive gain. Note that
the gain decreases and becomes negative for higher energy as expected by the Bernard
and Durafourg condition. We can nevertheless observe that non zero (but small) gain
can extend below the bandgap: there is a slow decrease of gain for low energy which
is unexpected. This is due to the presence of phonons in the semiconductor. In fact, a

15
complete theory of gain in semiconductor material is very complex and needs to take
into account various processes due in particular to interactions of electrons, and is
beyond the scope of this lecture. It is worth noting also that the maximum gain shifts
towards higher energy: the maximum emission wavelength shifts with the injected
current. The theory used here considered 300K for low currents and 312K for higher
currents to have a good agreement between the curves.

Question:
Can you explain why it is needed to change the model temperature to account for
the experimental data ? What can you say on how temperature affects also the gain
maximum ?

2.1.0.2 Quantum confined semiconductor materials

One of the advantages of semiconductor materials relies on the compactness of the
lasers. This compactness has another consequence: when a layer of semiconductor
material is made sufficiently small, then quantum confinement effects can occur. A
typical example is the quantum well material system (see Figure 2.3). It consists of
a this slice of a semiconductor material (generally of the order of 10nm width) sand-
wiched between two other larger semiconductor materials (the barriers). If the gap
of the thin slice material is less than the one of the barriers, then the electrons in
the quantum well are trapped and confined. Then the motion of the electrons (and
holes) in the small direction will be quantized into discrete energy levels, while keeping
the 2D character in the transverse directions. There is a confinement energy (which
increases with the smallness of the well width) such that the energy bandgap of a
quantum well is of higher energy than the bulk, 3D material counterpart. This can
be seen on the absorption curves on Figure 2.3). The gain is positive for wavelengths
corresponding to a sufficiently high energy of incoming photons, corresponding ap-
proximately to the bandgap energy. The bulk material has a lower bandgap than the
quantum well. If the quantum well is very deep, then it can be considered as an infi-
nite barrier quantum well and the confinement energy, i.e. the difference Eg2D − Eg is
large and depends on a−2 where a is the well width. Note that the confinement energy
is less for holes than for electrons, due to the higher effective mass of holes. The gain
is also larger close to bandgap for quantum wells than for bulk structures.
It is possible to create artificial semiconductor structures which are confined in all
2 or 3 directions. We speak of quantum wires and quantum dots. While quantum
wires are relatively rare, quantum dot materials can be found as laser gain materials
with their specific advantages. On further advantage of quantum confined structures
is the ability to tune the bandgap with the confined dimension size, without touching
the composition of the semiconductor materials.

2.1.0.3 Alpha factor

An important additional parameter characterizing semiconductor gain materials is
the alpha-factor or linewidth enhancement factor. This parameter is unique to semi-
conductor materials and is zero in atomic lasers.
In semiconductor materials, one can obtain the complex susceptibility χ = χ′ + iχ′′
√
(or complex refractive index n̄, through n̄ = 1 + χ) either experimentally or numeri-
cally. Let us consider a general electric field written E(z, t) = 21 E(z)ei(Kz−ωt−ϕ(z)) + c.c.
with K = n0 ω/c is the wavevector, n0 being the background refractive index, ω the
angular frequency, ϕ(z) a propagation dependent phase shift and E the real electric

16
Figure 2.3: Semiconductor quantum well structure and comparison of absorption in
a bulk and quantum-well structure.

Figure 2.4: Gain (left), carrier induced phase shift (middle) and alpha factor (right) for
a 5nm InGaAs/AlGaAs strained quantum well and different carrier densi-
ties 2, 3 and 4×1018 cm−3 . Sold line: full theory. Dashed line: free-carrier
theory. After [Chow, Koch Sargent III, Semiconductor-laser Physics].

17
field. The Bert-Lambert law gives for the gain experienced by a light field at a given
frequency propagating along z: I(z) = I(0)eGz where I(z) = |E(z)|2 . The intensity gain
G, in units of inverse length (L−1 ) and the phase shift derivative are plotted on Fig-
ure 2.4. The phase shift derivative is equivalent to introducing a space dependent
wavevector, and hence additional refractive index δn such that K − dϕ dz = (n0 + δn)K0 ,
with K0 = ω/c the vacuum wavevector. We can see from the previous figure that when
an electric wave propagates in a semiconductor medium, it experiences gain (or ab-
sorption, i.e. negative gain, depending on its wavelength) and an additional phase
shift. The gain and phase shift both depend on the wavelength and on the carrier
density. Since K = n0 K0 we deduce − dϕdz = δnK0 .
We can now introduce the alpha factor defined as
 ′   ′′     
∂χ ∂χ K ∂δn ∂g
α≡ / =− / .
∂N ∂N n0 ∂N ∂N
The alpha factor is defined as the rate of range of the real part χ′ of the complex
susceptibility χ with carrier density N with respect to the rate of change of the imagi-
nary part χ′′ of the complex susceptibility with the carrier density. Equivalently, since
the intensity gain is G = −Kχ′′ and since the phase shift is − dϕ K ′
dz = 2 χ we can relate
the alpha parameter to the rate of change of the amplitude gain g = G/2 and of the
carrier induced refractive index δn, using δn/n0 = χ′ /2. We can introduce the complex
amplitude Beer-Lambert law dA/dz = ḡA with ḡ = g − i dϕ dz and A = Ee
−iϕ(z) . Using

the relations introduced earlier, we can rewrite ḡ = g + i nK0 δn. If we expand the re-
∂g
+ i nK0 δn(Ntr ) + (N − Ntr ) ∂δn


lation about Ntr we get ḡ = g(Ntr ) + (N − Ntr ) ∂N ∂N . Since
∂g
g(Ntr ) = 0 and using the expression for α we get ḡ = i nK0 δn(Ntr ) + (N − Ntr ) ∂N (1 − iα).
∂g 2
The differential gain a ≡ ∂N has units of a cross-section (L ). We can also redefine
the phase of the complex field to incorporate the first constant complex phase shift
into the definition of the complex amplitude A such that a general expression for the
growth of the complex electric field can be written

dE
= a(N − Ntr )(1 − iα)E
dz

2.1.0.4 Optical confinement in semiconductor lasers

There are mainly two types of semiconductor laser structures: the edge-emitting laser
(EEL) structure and the vertical-cavity surface emitting laser (VCSEL) structure.
In an edge-emitting laser, the cavity is perpendicular to the semiconductor growth
axis. Different layers of semiconductor materials are grown by epitaxy on a semicon-
ductor substrate. Optical confinement in the growth direction is done by growing a
high refractive index material between two low index material layers. The gain mate-
rial (the high index material) is composed of quantum wells. Highly doped layers are
inserted to facilitate current injection in the structure. A top electrode in metal (e.g.
gold) is deposited and defines the transverse structure of the mode. In that case we
speak of gain guiding, since optical gain will only occur where the system is pumped
and light will only be guided in these regions. Sometimes, a waveguide structure is
etched on top (ridge waveguide structure): the optical mode is therefore confined also
by the index contrast brought by the ridge, we speak of index guiding. The cavity is
defined by the too facets consisting a either raw interface or a optical coating. Because
of the high index contrast between semiconductor material, whose index of refraction
depends on many parameters but lies usually around n=3.5, and air (n=1), there is
a reflectivity of the facet of about 30%. Light is emitted by both facets on the edge of

18
 Figure 2.5: Vertical-cavity laser and Fabry-Pérot edge emitting laser.

the structure1 , hence the name. The mirrors are technologically defined by cleavage
along a crystallographic axis. The relatively low reflection coefficient of the facets is
compensated by the high gain of semiconductor material and by the length of the laser
cavity which can be up to several millimeters long. Transversely, the mode is defined
by the injected current and/or the ridge structure, and by the optical confinement
along the growth axis. It results that the mode structure is usually very flat and non
circular, with an elliptic far-field.
In a VCSEL, the philosophy is revered: the cavity is along the growth axis. It
consists of two multilayer Bragg mirrors with high reflectivities (> 99%). The cavity
height is very small, of the order of the wavelength: the laser is single longitudinal
mode. The laser is pumped by an annular electrode which defines the transverse
mode structure. The gain region is usually composed of quantum wells. The relatively
small gain region length experienced by the laser cavity light is compensated by the
high reflectivity of the Bragg mirrors. The mode structure is defined by the current
aperture and can be circular, which is convenient for e.g. efficient coupling to an
optical fiber. Also, since the laser does not need cleavage to work, it can be tested
just after fabrication which is an advantage with respect to EEL. The power emitted
by VCSELs is generally smaller than the one emitted by EELs (tens of mW against
hundreds of mW).

2.1.0.5 Optical and Electrical injection is semiconductor lasers

Most semiconductor lasers are injected electrically. This ensures a high wall-plug
efficiency, i.e. a high light throughput compared to the current injected. Some semi-
conductor laser are optically injected and necessitate an additionnal high power laser
source as pump. This is the case e.g. for VECSELs (Vertical External Cavity Surface
Emitting Lasers), a hybrid laser structure composed of a half-VCSEL (a VCSEL with-
out top mirror or with reduced top mirror) and a distant glass mirror (see Figure 2.6).
These lasers have the advantage of being multi-longitudinal mode and hence can be
1
Generally one facet is coated with a dielectric mirror to unbalance the emission towards the low
reflectivity facet.

19
Figure 2.6: Optical pumping of lasers. Simplified band diagram picture and schemat-
ics of a VECSEL (see text).

mode-locked to produce ultra-short pulses. Optical injection only relies on pumping

the gain material with at wavelength higher than the bandgap. Electrons (and holes)
injected at higher energy in the bands cascade to the bottom of the band very rapidly
through emission of phonons. This process is very fast. The difference of energy
between the pump laser and the laser emission (called the quantum defect) must be
chosen as small as possible because the energy lost in the cascade is transformed
into heat.
Electrical injection in the laser structure necessitates the use of doped semicon-
ductor layer to inject the charges carriers to form a diode structure. It is common to
use n-doped substrate and p-doping on top of the laser. A n-doped material is made
with donor impurities (which easily gives up an electron) and creates a high density of
free electrons, while a p-doped material is made with accepting impurities (which at-
tracts electrons) and creates a high density of free holes. The p and n-doped materials
sandwich the undoped intrinsic layer I, the gain section of the laser. At equilibrium
(see Figure 2.7), there is no bias applied to the diode. The Fermi energy level lies
above the conduction band in the highly n-doped region and below the valence band
in the highly p-doped region. By applying a positive bias on the p-type layer (forward
polarisation), electrons in the conduction band can cross the potential barrier and
accumulate in the intrinsic region. The same occurs for the holes. As shown on Fig-
ure 2.7, the carriers are thus injected and confined in the potential well created by
the difference of band-gaps between the middle layer and the highly doped ones. This
prevents the carriers to diffuse in the region where they are in minority, leaving them
in the zone where they are more likely to recombine and emit radiation. An advan-
tage of this heterostructure lies in the double confinement obtained: in addition to
the electronic confinement explained above, the optical mode is also restricted to the
middle layer by the difference of refractive indices.

2.1.1 Semiconductor laser rate equations

2.1.1.1 Single-mode laser rate equations
Semiconductor lasers can be described by rate equations for the carrier density N in
the active region (holes or electrons, [N ] = L−3 ) and the photon density P ([N ] = L−3 )

20
 Figure 2.7: Electrical pumping: laser diode schematics of a PIN junction.

inside the cavity mode by the set of equations

dN J N
= − − A(N − Ntr )P (2.1)
dt eV τnr
dP P
= − + AΓ(N − Ntr )P. (2.2)
dt τp
In Eq. 2.1, the first term describes the injection rate of carriers with the current
J (in Amps), and V the mode volume. Second term describes non-radiative recom-
bination processes of carriers, with timescale τnr . In semiconductor materials, this
timescale is typically of the order of 1ns. The last term describes the conversion of
electron-hole pairs into photons because of either absorption or stimulated emission:
Ntr is the carrier transparency density (at a given wavelength considered here for the
laser emission wavelength), and A is the gain coefficient ([A] = L3 T −1 ). The gain coef-
ficient can be expressed in terms of the differential gain g = dG/dN ([g] = L2 ) where G
is the gain per unit length ([G] = L−1 ) of the material and of the group velocity of the
mode inside the cavity vg such that A = gvg . When the carrier density is less than the
transparency density, the net effect is to increase the number of carriers since light
will be majorly absorbed. On the contrary, if the carrier density injected is greater
than the transparency density, the effect is a reduction of the density of electron-
hole pairs through the creation of stimulated photons (if photons are already present,
i.e. if P ̸= 0). Note that the non-radiative recombination of carriers includes all ef-
fects giving rise to a destruction of electron-hole pairs not associated to the creation
of photons in the laser mode. In semiconductors, there are many such processes,
due to crystal defects in particular.
 More generally,
 we could write the non-radiation
recombination term as −N τnr + BN + CN with (BN )−1 = τsp describing the spon-
1 2

taneous emission process due to the spontaneous decay of electron-hole pairs into
photons escaping the cavity modes and CN 2 the Auger recombination process. The
constant B is called the bimolecular rate. Spontaneous emission can be quite high in
semiconductor lasers. The rate of spontaneous emission depends on the availability
of a free electron and of a free hole, and is therefore a process proportional to N 2 .
It can be thought of as a stimulated emission process due to the vacuum field in an

21
quantum mechanical point of view. The Auger process is important only at very high
currents and involves 3 particles, hence the cubic dependence. These terms can be
ignored in a first approximation, or taken into account globally into the linear term by
introduction a non-radiative decay −N/τ eff with an effective approximated decay rate.
Eq. 2.2 describes the evolution of the photon density into the cavity mode volume.
The first term describes the escape of photons outside of the cavity at a rate τp−1 with a
photon lifetime τp . The photon lifetime depends on the losses at the laser mirrors and
on any additional scattering of photons inside the cavity. Considering the effect of the
mirrors alone, we can write R1 and R2 the reflectivity in intensity of the back and front
laser cavity mirrors. The round trip loss for cavity photons is thus 1−R1 R2 . The round
trip time is Ln/c with L the round trip length (L = 2L0 for a Fabry-Pérot cavity of length
L0 ), n the mean index of refraction inside the cavity and c the speed of light. We can
then write the photon loss rate τp−1 = c(1 − R1 R2 )/Ln. For a EEL, with R1 = R2 ≃ 0.3
and L = 1mm (n=3.5) we get τp ∼ 10ps. For a VCSEL, with R1 = 1, R2 = 0.99, and
L ∼ 1µm we get τp ∼ 1ps. In both cases, the photon lifetime is at least two orders
of magnitude smaller than the carrier lifetime. This is why we say semiconductor
lasers belong to class B lasers2 . The next term describes the creation or destruction
of photons through light-matter interaction (absorption or stimulated emission). This
is the mirror term to the term present in Eq. 2.1, except for a parameter Γ called the
confinement factor. This term describes the non identity of the active material volume
Va and of the cavity mode volume V . Photons created in the active region (which
is usually smaller than the mode volume, especially when the laser gain consists of
quantum confined structures such as quantum wells) will contribute to the photon
density in a ratio Γ = Va /V . It is common to have Γ ∼ 10−2 − 10−3 . The careful reader
will also notice that there is no mirror term for the spontaneous emission. Indeed,
spontaneous emission is an isotropic process3 and most of the spontaneously emitted
photons will not bounce back into the cavity. Spontaneous photons are also emitted
in a broad spectral range while the photon density considered in 2.2 concerns laser
field photons, i.e. at a well defined frequency. Some models consider explicitly the
coupling of spontaneously emitted photons into the laser mode by introducing a factor
βP in Eq. 2.2 with β ≪ 14 .

2.1.1.2 Steady-state response

The steady-state response is obtained by setting the times derivatives
 to zero in the
1
laser rate equations. The photon density equations leads to P − τp + AΓ(N − Ntr ) =
0. The trivial, non-lasing case, solution P = 0 leads to N = Jτ eV . It corresponds to the
nr

laser below threshold, with no laser output and an increasing carrier density with the
increase of the injected current.  
The lasing solution is obtained by requiring − τ1p + AΓ(N − Ntr ) = 0 so that N =
h  i
1 J 1 1
AΓτp + N tr and P = Γτ p eV − τnr AΓτp + N tr . We clearly see in that case that the
photon density P grows linearly with the injected current  J. The laser pumping at
Jth 1 1
threshold Jth corresponds to P (Jth ) = 0 so that eV = τnr AΓτ p
+ Ntr . The laser equa-

2
A class A laser has τp ≫ τm with τm the typical recombination time of the gain material. In that case
the equation for the population inversion can be eliminated and the system is described by only one
equation. Class A lasers such as visible HeNe lasers do not show relaxation oscillations, whereas
class B lasers do show. There is also a less common class C for the laser dynamics.
3
Spontaneous emission can be controlled in specific systems to occur in favored directions.
4
At maximum, β = 1, and we speak in that case of a thresholdless laser.

22
tion can then be rewritten
Γτp
[J − Jth ] .
P =
eV
The laser threshold is affected by all the losses discussed so far. A big part of the
injected current goes to reaching the gain material transparency. Then, the additional
current goes to the compensation of the cavity losses represented by τp . Interestingly,
1
when the laser threshold is reached, the carrier density N is constant (N = AΓτ p
+ Ntr
does not depend on the current injection J). This is called the gain clamping. This
means that right after the laser threshold, all the additional material excitation is
converted into laser photons.

2.1.1.3 Multi-mode laser rate equations

In the case where the laser is multi-longitudinal mode, i.e. several laser frequencies
can lase simultaneously, the above rate equation must be amended to incorporate the
photon density for each laser mode Pm :

dN J N X
= − − A(N − Ntr ) Pm (2.3)
dt eV τnr m
dPm Pm
=− + AΓm (N − Ntr )Pm . (2.4)
dt τp

These equations are more difficult to solve and can give rise to interesting multi-
mode dynamics such as chaos.

2.1.1.4 Complex laser rate equations

Using the previous results on the alpha factor (or linewidth enhancement factor, or
Henry factor), a laser rate equation for the complex electric field amplitude E writes

dN J N
= − − A(N − Ntr ) |E|2 (2.5)
dt eV τnr
dE E A
=− + Γ(N − Ntr )(1 − iα)E. (2.6)
dt 2τp 2
√  √ 
With the ansatz E = P e−iφ5 Eq. 2.6 can be written dE dt = e −iφ Ṗ
√
2 P
− iφ̇ P =
√ −iφ √
− Ie2τp + A2 Γ(N − Ntr )(1 − iα) P e−iφ . By separating the real and imaginary parts we
get
P
Ṗ = − + AΓ(N − Ntr )P,
τp
which is identical to 2.2 and a phase equation

AΓ
(N − Ntr )α.
φ̇ =
2
If the laser intensity is zero, then the phase equation is singular but is is not a
problem since the phase is not defined in that case. This indeterminacy can also
be interpreted physically: when the injection current crosses the laser threshold, the
laser starts with a non zero electric field and the field acquires a phase φ0 which is

5
We use here a phase e−iφ with a minus sign to comply with the temporal dependence chosen above in
e−iωt .

23
 1 α
random. Once the phase is determined and the laser is on, N −Ntr = AΓτp and φ̇ = 2τp .
The last equation can be easily integrated and gives
α
φ − φ0 = t.
2τp

This phase linearly varying in time gives rise to an instantaneous frequency ωα = 2ταp .
If the frequency of the laser field at threshold is ω, then because of the α parameter it
acquires a frequency shift ωα .

2.1.1.5 Small signal analysis

Let us start from the set of equations Eqs. 2.6,2.5 and consider the steady state
solution above laser threshold: N = Nst , P = Pst and φ = φst . We consider a small
fluctuation (δN, δP, δφ) from this state. Then we can perform a linearization of the
system and write

dδN δN
=− − AδN Pst − A(Nst − Ntr )δP
dt τnr
dδP δP
=− + AΓδN Pst + AΓ(Nst − Ntr )δP
dt τp
dδφ AΓ
= δN α.
dt 2
1 Γτp
By replacing the stationary values Nst − Ntr = AΓτp and Pst = eV [J − Jth ] we get

 
dδN 1 Γτp δP
=− +A [J − Jth ] δN −
dt τnr eV Γτp
dδP Γτp
= AΓδN [J − Jth ]
dt eV
dδφ AΓ
= δN α.
dt 2
From the last equation it can be found that the carrier density fluctuations couple
to the phase fluctuations of the laser field through the alpha parameter. This is
a very important observation. In a semiconductor laser, because of a non-zero α
factor, carrier density fluctuations will induce phase, or equivalently, refractive index
fluctuations. They also give rise to laser intensity fluctuations (through δP ), that in
turn modify the carrier fluctuations δN . There is thus a close connections between
carrier and field (intensity and phase) fluctuations in semiconductor lasers. This will
be important when studying the linewidth of lasers.

24