Electro-Optic and Electro-Absorption Modulators

An optical modulator is the hand that writes electronics onto light. A laser supplies a steady carrier; a modulator takes a voltage swinging at gigahertz rates and stamps it onto one of the wave's parameters — phase, amplitude, or polarization — so a fibre can carry a data stream instead of a beam. Its physics comes from two chapters upstream. From the crystal-optics of anisotropic media it inherits the linear electro-optic (Pockels) effect: in a non-centrosymmetric crystal an applied field distorts the index ellipsoid, so the index shifts in proportion to the field, $\Delta n \propto E$ . That is a pure phase effect — it delays the wave without touching its power — so it must be read out by an interferometer or a polarizer before it becomes brightness. From semiconductor band theory it inherits the electro-absorption effect: a field near the band edge changes how strongly the material absorbs, gating the transmitted power directly. Two workhorses dominate. The lithium niobate Mach–Zehnder modulator splits light into two waveguide arms, drives a differential Pockels phase between them, and recombines so that phase becomes intensity through interference; its master figure of merit is the half-wave voltage $V_\pi$ , the drive that flips constructive interference into destructive. The electro-absorption modulator is a III–V semiconductor waveguide tens of microns long that slides its own absorption edge past a fixed laser line — via the bulk Franz–Keldysh effect or, far more strongly, the quantum-confined Stark effect in quantum wells — and so integrates monolithically with a laser. Those two mechanisms, plus free-carrier plasma dispersion in silicon, power every phase shifter, microring, and coupler-switch in an integrated photonic circuit. Every real device is a negotiation among $V_\pi$ , insertion loss, extinction ratio, chirp, and bandwidth.

Key takeaways

A modulator converts an electrical signal into a change in the refractive index (Pockels, linear in $E$ ) or the absorption (Franz–Keldysh / quantum-confined Stark effect) of a waveguide, and thence into a modulated optical intensity.
The Pockels effect gives only phase, $\Delta\phi=(2\pi/\lambda_0)\Delta n\,L$ ; a Mach–Zehnder interferometer turns that into intensity, $T=\cos^2(\pi V/2V_\pi)$ , with the half-wave voltage $V_\pi$ setting the drive.
The length-normalized product $V_\pi L$ is the true figure of merit: $V_\pi\propto 1/L$ , so longer electrodes buy lower drive but cost bandwidth.
Traveling-wave electrodes let the microwave field co-propagate with the light; velocity mismatch $|n_o-n_m|$ gives a sinc roll-off, $f_{3\text{dB}}\approx c/(\pi L|n_o-n_m|)$ .
Electro-absorption modulators are compact and integrate with lasers; the quantum-confined Stark effect shifts an excitonic peak quadratically in $E$ , an order of magnitude stronger than bulk Franz–Keldysh.
Key idea: A voltage changes either the refractive index (Pockels, linear in $E$ ) or the absorption coefficient (Franz–Keldysh / QCSE) of a waveguide; an interferometer or the direct absorption then translates that into a modulated optical intensity, with $V_\pi$ and bandwidth as the master trade-off.

Where this lives today

The lithium niobate Mach–Zehnder modulator has been the beating heart of long-haul and coherent optical communication for three decades: a z-cut or x-cut crystal with velocity-matched traveling-wave electrodes, driven push–pull for chirp-free operation, delivering deep extinction and bandwidths past 40 GHz. Arranged as nested in-phase and quadrature modulators it writes both amplitude and phase, generating the QPSK and 16-QAM constellations that fill today's transoceanic cables. Alongside it, the electro-absorption modulated laser integrates a quantum-well section onto the same III–V chip as a distributed-feedback laser, a compact low-drive transmitter that dominates 10–100 Gb/s datacenter and metro links where footprint and power matter more than the last decibel of chirp. The frontier is thin-film lithium niobate. By etching sub-micron ridge waveguides into a bonded lithium-niobate film, designers push the optical mode into far tighter overlap with the electrode field — raising $\Gamma$ so hard that CMOS-compatible drive voltages coexist with bandwidths beyond 100 GHz, breaking the old $V_\pi$ -versus-speed deadlock the second worked example diagnoses. In parallel, silicon-photonic modulators built on free-carrier plasma dispersion in a reverse-biased pn junction bring modulation into the same CMOS foundries that make processors — trading the pure phase response of Pockels for the manufacturability of silicon.

The physical picture

Picture a Pockels crystal as a voltage-controlled waveplate. Apply a field along the appropriate crystal axis and the index ellipsoid changes: the refractive index becomes $n+\Delta n$ (with the sign of $\Delta n$ set by the crystal orientation and electro-optic coefficient), so after a length $L$ the beam acquires an extra phase $\Delta\phi=(2\pi/\lambda)\Delta nL$ . Nothing about the power has changed — you have merely delayed the crest. To cash that delay out as brightness you must let the delayed wave interfere with an undelayed reference, which is exactly what the two arms of a Mach–Zehnder do: split, phase-shift one arm, recombine. When the arms are in phase the port is bright; when the differential phase reaches $\pi$ they cancel and it goes dark. The voltage spanning that swing is the half-wave voltage $V_\pi$ — the whole story of drive efficiency.

Figure

Two roads to a modulated beam — left: a Pockels crystal in one arm of a Mach–Zehnder tilts the index ellipsoid so a voltage V writes a differential phase Δφ that interference turns into intensity, with LiNbO₃ arms of order L ≈ 4 cm; right: an electro-absorption waveguide only ≈ 100 μm long slides its band edge with the field so the fixed 1.55 μm laser line is absorbed or passed

Electro-absorption is blunter and more visceral. Near a semiconductor band edge the absorption is a steep function of photon energy, so a field that nudges the band edge — by tilting the bands (Franz–Keldysh) or by pulling the confined electron and hole to opposite well walls (the quantum-confined Stark effect) — drags the whole absorption curve past a fixed laser wavelength and switches transmission on and off over microns. The price of directness is that the light you absorb becomes photocurrent and heat, and the absorption edge drifts with temperature. Both mechanisms share one enemy: bandwidth is a race. The microwave drive and the optical wave must stay in step down a long electrode; if the RF wave outruns or lags the light, the phase written into the front of the crystal is unwritten at the back, and high frequencies wash out. The quantitative face of the Pockels device is its transfer function — the cosine-squared law that turns the linear phase ramp into the smooth switching curve every link designer memorizes.

Figure

The punchline plot — the Mach–Zehnder power transfer T = cos²(πV/2V π) for a device with half-wave voltage V π = 3.5 V: full-on (T = 1) at V = 0, full-off (T = 0) at V = 3.5 V, and the quadrature bias point T = 0.5 at V = 1.75 V where the response is most linear for small signals

The core equations

The linear electro-optic effect is written not as a change in $n$ directly but as a perturbation of the ellipsoid coefficient $1/n^2$ — the natural object in crystal optics, since a field distorts the ellipsoid linearly. Inverting to first order gives the index change every modulator design starts from: small, but over many wavelengths of propagation it becomes a usable phase.

\frac{1}{n^2(E)} = \frac{1}{n^2} + r\,E \;\;\Rightarrow\;\; \Delta n \approx -\tfrac{1}{2}\, n^3 r\, E \qquad (1)

where: $n$ — unperturbed index for the relevant polarization; $r$ — the active electro-optic (Pockels) tensor coefficient, e.g. $r_{33}$ in LiNbO $_3$ ; $E$ — applied field along the chosen crystal axis; $\Delta n$ — field-induced index change. The factor $n^3$ is why high-index crystals are prized: the response scales as the cube of the index. That index change accumulates into phase over the interaction length, and the field is set by the voltage across the electrode gap, $E=V/d$ . Writing the phase in terms of $V$ makes the device parameters explicit and exposes exactly which knobs lower the drive.

\Delta\phi = \frac{2\pi}{\lambda_0}\,\Delta n\, L = \frac{\pi\, n^3 r\, L}{\lambda_0}\,\frac{V}{d} \qquad (2)

where: $\lambda_0$ — free-space wavelength; $L$ — electrode / interaction length; $V$ — applied voltage; $d$ — electrode gap. Phase grows linearly in both $V$ and $L$ : doubling the electrode length halves the voltage needed for a given phase. The half-wave voltage is set by the condition $\Delta\phi=\pi$ — the swing that carries a Mach–Zehnder from bright to dark. A real waveguide also carries an overlap factor $\Gamma\le1$ , the fraction of the mode that sits in the applied field, giving the single most quoted modulator number.

V_\pi = \frac{\lambda_0\, d}{n^3 r\, L\,\Gamma} \qquad\Rightarrow\qquad V_\pi L = \frac{\lambda_0\, d}{n^3 r\,\Gamma} \qquad (3)

where: $V_\pi$ — voltage for a $\pi$ phase shift; $\Gamma$ — field–mode overlap ( $0<\Gamma\le1$ ). Because $V_\pi\propto1/L$ , the product $V_\pi L$ is length-independent — it captures the material and geometry alone and is the honest way to compare modulators. Phase is invisible until it interferes. Splitting the input equally, phase-shifting the arms, and recombining gives a power transfer that is a cosine-squared of the differential phase — the switching curve of the second figure — and a push–pull drive swings the arms oppositely, doubling the phase per volt and cancelling chirp.

T = \frac{I_{\text{out}}}{I_{\text{in}}} = \cos^2\!\left(\frac{\Delta\phi}{2}\right) = \cos^2\!\left(\frac{\pi V}{2 V_\pi}\right) \qquad (4)

where: $T$ — intensity transmission of a balanced lossless Mach–Zehnder; $\Delta\phi$ — differential arm phase. The slope is steepest at the quadrature point $V=V_\pi/2$ (where $T=\tfrac12$ ), which is why small-signal analog links bias there. For gigahertz operation the electrode is a traveling-wave transmission line, so the microwave field runs alongside the light. If their group indices differ the two walk off, and the accumulated response is a sinc whose first null sets the bandwidth — the equation that pits length against speed.

f_{3\text{dB}} \approx \frac{c}{\pi L\,|n_o - n_m|} \qquad (5)

where: $n_o$ — optical group index; $n_m$ — microwave effective index of the electrode; $c$ — vacuum light speed. Bandwidth falls with both length and velocity mismatch, so the same long electrode that lowers $V_\pi$ throttles $f_{3\text{dB}}$ unless $n_m$ is engineered toward $n_o$ . For electro-absorption devices the governing quantity is the field-dependent absorption. In bulk material a static field tilts the bands so the electron and hole wavefunctions leak into the gap and absorb photons below $E_g$ by photon-assisted tunneling — the Franz–Keldysh effect, whose lineshape is built from Airy functions.

\alpha_{FK}(\hbar\omega, E) \propto |Ai'(\beta)|^2 - \beta\,|Ai(\beta)|^2,\qquad \beta = \frac{E_g-\hbar\omega}{\hbar\theta_E} \qquad (6)

where: $Ai$ — Airy function from the tilted-band eigenstates; $E_g$ — bulk bandgap; $\hbar\theta_E=(e^2\hbar^2E^2/2\mu)^{1/3}$ — the electro-optic energy set by field $E$ and reduced mass $\mu$ ; $\hbar\omega$ — photon energy, here just below $E_g$ . Raising $E$ pushes absorption further below the gap, gating a sub-gap laser line. Finally, two device-level figures of merit close the loop. The extinction ratio measures how deep the dark state is, and the chirp parameter measures the parasitic phase modulation that rides along with the amplitude modulation and, through fibre dispersion, limits reach.

\text{ER} = 10\log_{10}\!\frac{T_{\text{on}}}{T_{\text{off}}}\ \text{(dB)},\qquad \alpha_{\text{chirp}} = \frac{2\,\mathrm{d}\phi/\mathrm{d}t}{(1/I)\,\mathrm{d}I/\mathrm{d}t} \qquad (7)

where: $T_{\text{on}},T_{\text{off}}$ — bright- and dark-state transmission; $\alpha_{\text{chirp}}$ — Henry-style chirp, the ratio of phase to fractional-amplitude modulation; $I,\phi$ — instantaneous optical intensity and phase. A push–pull Mach–Zehnder reaches $\alpha_{\text{chirp}}=0$ ; an electro-absorption modulator generally does not.

Derivation AHalf-wave voltage and the Vπ·L figure of merit

The goal is to build $V_\pi$ from the ground up for the standard z-cut lithium niobate geometry and to see why it is the length-normalized product $V_\pi L$ , not $V_\pi$ alone, that ranks one modulator technology against another. We take light polarized along the crystal $z$ -axis, which sees the extraordinary index $n_e$ , apply the field along the same axis so the largest tensor element $r_{33}$ is in play, march the induced index change through propagation, and impose the half-wave condition while tracking the modal overlap $\Gamma$ that a real waveguide forces on us.

Step 1. Apply the Pockels relation of equation (1) with $E_z=V/d$ : $\Delta n_e = -\tfrac12 n_e^3 r_{33}\,(V/d)$ . Why: this is the only place the crystal physics enters, so pinning the right coefficient ( $r_{33}$ , not $r_{13}$ ) and the right index ( $n_e$ ) is the whole battle.

Step 2. Convert index change to accumulated phase over length $L$ , weighting by the overlap: $\Delta\phi = (2\pi/\lambda_0)\,|\Delta n_e|\,L\,\Gamma = (\pi n_e^3 r_{33} L\,\Gamma/\lambda_0)\,(V/d)$ . Why: only the fraction $\Gamma$ of the guided mode overlapping the electrode field is retarded, a penalty a plane-wave estimate misses.

Step 3. Impose the switching condition $\Delta\phi=\pi$ and solve: $V_\pi = \lambda_0 d/(n_e^3 r_{33} L\,\Gamma)$ . Why: $\pi$ is exactly the phase that turns constructive interference into destructive in the Mach–Zehnder, so this is the full on-to-off drive.

Step 4. Multiply by $L$ to isolate the invariant: $V_\pi L = \lambda_0 d/(n_e^3 r_{33}\,\Gamma)$ , independent of length. Why: since $V_\pi\propto1/L$ , quoting $V_\pi$ alone is meaningless; the product is what physics fixes and what you trade against bandwidth. For LiNbO $_3$ at $1.55\,\mu$ m ( $n_e\approx2.14$ , $r_{33}\approx30.8$ pm/V, $d\approx10\,\mu$ m, $\Gamma\approx0.5$ ) it comes out near $10$ V·cm — recomputed below.

Go deeper · Derivation B · 1 minTraveling-wave bandwidth from velocity mismatch

Derivation BTraveling-wave bandwidth from velocity mismatch

The aim is to show how a finite electrode length turns a velocity mismatch between the microwave drive and the optical wave into a bandwidth ceiling, and to recover the sinc roll-off and its 3-dB point. The key fact is that the optical wave at position $z$ does not see the drive as it is now — it sees the field launched earlier and propagating with its own index $n_m$ , while the light travels with group index $n_o$ . Integrating that running phase mismatch over the electrode is the whole calculation.

Step 1. Write the phase increment as an integral of the RF field the light meets, $\Delta\Phi = \int_0^L E_{RF}(t - n_o z/c)\,\mathrm{d}z$ with $E_{RF}\propto e^{i(\omega_m t-\omega_m n_m z/c)}$ . Why: the light is a moving probe of the microwave field, so what matters is the field in the frame co-moving with the crest, which mixes both indices.

Step 2. Collect the $z$ -dependence and integrate: $\Delta\Phi\propto\int_0^L e^{-i\omega_m(n_m-n_o)z/c}\,\mathrm{d}z\propto L\,\mathrm{sinc}\!\big(\omega_m L(n_m-n_o)/2c\big)$ . Why: a phase winding linearly along the electrode integrates to a sinc, flat with $L$ only when $n_m=n_o$ — otherwise the tail subtracts what the front added.

Step 3. Set the response to $1/\sqrt2$ for the 3-dB frequency, $m(f_{3\text{dB}})=1/\sqrt2\Rightarrow f_{3\text{dB}}\approx c/(\pi L|n_o-n_m|)$ . Why: bandwidth falls inversely with both length and mismatch — the long electrode that gave a low $V_\pi$ is the one that caps the speed unless $n_m$ is dragged toward $n_o$ .

Step 4. Add the second limiter: microwave conductor loss grows like $\sqrt{f}$ , so the true response is the product of walk-off, RF attenuation, and impedance mismatch. Why: in bulk LiNbO $_3$ the microwave index ( $\approx4$ ) far exceeds the optical index ( $\approx2.2$ ), so designs use thick buffer layers and thin or ridge electrodes to lower $n_m$ — and even then loss eventually wins.

Go deeper · Derivation C · 1 minQuantum-confined Stark effect as a quadratic exciton shift

Derivation CQuantum-confined Stark effect as a quadratic exciton shift

The purpose is to estimate why squeezing electro-absorption physics into a quantum well can make it much stronger than in bulk, and why the resulting excitonic shift is quadratic in field. In bulk, a field tilts the bands and can ionize or dissociate excitons; in a quantum well of width $L_w$ , confinement suppresses ionization over a range of fields, while the electron and hole are displaced toward opposite sides of the well and their bound-state energy shifts. Second-order perturbation theory in the field captures the leading effect and its startling sensitivity to well width.

Step 1. Take the field as a perturbation coupling to carrier position, $H'=eEz$ . Why: the field adds a linear potential ramp across the confinement direction, and $z$ is bounded by $L_w$ — which keeps the effect finite instead of ionizing.

Step 2. The confined ground state has definite parity, so $\langle0|z|0\rangle=0$ and the first-order shift vanishes; the leading term is second order, $\Delta E^{(2)}=\sum_{k\neq0}|\langle0|eEz|k\rangle|^2/(E_0-E_k)$ . Why: no linear shift by symmetry is exactly why the QCSE edge moves as $E^2$ , not $E$ .

Step 3. Scaling: $z$ matrix elements scale as $L_w$ and denominators as $\hbar^2/(m^*L_w^2)$ , giving $\Delta E^{(2)}\propto -\,e^2E^2L_w^4 m^*/\hbar^2$ . Why: the fourth power of well width is the punchline — wider wells shift far more for the same field, trading shift against exciton binding.

Step 4. The device consequence: shifting this sharp, field-tunable excitonic absorption edge past a fixed laser wavelength can produce high extinction in a short interaction length — the basis of electro-absorption modulators and integrated modulated lasers. Why: the strong, sharp QCSE lets a $\sim100\,\mu$ m III–V waveguide do what a centimetres-long LiNbO $_3$ arm does, at the cost of temperature and wavelength sensitivity.

Limiting cases

Case	Result	Meaning
Pockels (linear EO) limit	$\Delta(1/n^2)_i=\sum_j r_{ij}E_j$	Non-centrosymmetric crystals (LiNbO $_3$ , GaAs, KDP): index change linear in field — fast and lossless, but needs interferometric readout.
Kerr (quadratic EO) limit	$\Delta n=\tfrac12 n^3 s E^2$	Centrosymmetric media (liquids, glasses): the linear term vanishes by symmetry, so the index change is quadratic in $E$ — weaker and higher-field.
Franz–Keldysh regime	$\alpha(E)$ enhanced for $\hbar\omega<E_g$	Bulk semiconductor: field-assisted tunneling gives a broad, moderate sub-gap absorption tail with low chirp.
Quantum-confined Stark effect	$\Delta E_{\text{exciton}}\propto -E^2 L_w^4$	Quantum wells: confinement saves the exciton to high field, giving a much stronger, sharper absorption shift than bulk FK.
Velocity-matched electrode	$n_m\to n_o$ , $f_{3\text{dB}}\to$ RF-loss limit	With walk-off eliminated the sinc stays flat and bandwidth is capped only by conductor loss and impedance — the goal of traveling-wave design.

Common trapPhase modulation is not intensity modulation, and a scalar r is not a Pockels coefficient

Two traps sit at the heart of this topic. First, a bare Pockels cell does not modulate power — it only writes phase (or rotates polarization). Left alone it chirps the carrier but leaves the intensity flat; it becomes an amplitude modulator only inside a Mach–Zehnder, a resonator, or between crossed polarizers that convert the phase or polarization change into brightness. Second, the Pockels effect is a tensor. The index change depends jointly on crystal orientation, field direction, and light polarization through $r_{ijk}$ , so quoting a scalar " $r$ " without naming the coefficient and the geometry is meaningless — in LiNbO $_3$ the working element $r_{33}\approx30.8$ pm/V is nearly triple $r_{13}$ , and a design that accidentally uses the wrong one gets $V_\pi$ wrong by the same factor.

Worked examples

1 · Half-wave voltage of a lithium niobate Mach–Zehnder (the number every design starts from)

A z-cut LiNbO $_3$ modulator at $\lambda_0=1.55\,\mu$ m uses the $r_{33}$ geometry: extraordinary index $n_e=2.14$ , $r_{33}=30.8$ pm/V, electrode gap $d=10\,\mu$ m, overlap $\Gamma=0.5$ , electrode length $L=4$ cm. Find $V_\pi L$ and $V_\pi$ .

Step 1. Assemble the cube of the index, the dominant material lever: $n_e^3=2.14^3=\mathbf{9.80}$ — a 10% higher index would buy a 33% larger response.

Step 2. Compute the figure of merit from equation (3): $V_\pi L = \lambda_0 d/(n_e^3 r_{33}\Gamma) = (1.55\times10^{-6})(10\times10^{-6})/(9.80\cdot30.8\times10^{-12}\cdot0.5) = 1.55\times10^{-11}/1.509\times10^{-10}=0.103\ \text{V}\,\cdot\,\text{m}=\mathbf{10.3\ \text{V}\,\cdot\,\text{cm}}$ — the honest ranking number, independent of length.

Step 3. Divide by the actual length: $V_\pi = (V_\pi L)/L = 10.3/4=\mathbf{2.6\ V}$ . A 2.6 V full swing suits modest RF amplifiers, which is why classic LiNbO $_3$ transmitters run centimetres-long electrodes — and then fight the bandwidth penalty of the next example. V_π L ≈ 10.3 V·cm, so V_π ≈ 2.6 V for a 4 cm electrode.

2 · When the electrode is too long: velocity-mismatch bandwidth

Take the same $L=4$ cm electrode. In unmatched bulk LiNbO $_3$ the optical group index is $n_o\approx2.2$ and the microwave index $n_m\approx4.2$ . What is the 3-dB bandwidth, and what does electrode engineering to $|n_o-n_m|=0.3$ buy?

Step 1. Unmatched walk-off, from equation (5): $|n_o-n_m|=2.0$ , so $f_{3\text{dB}}\approx c/(\pi L|n_o-n_m|)=(3.0\times10^8)/(\pi\cdot0.04\cdot2.0)=\mathbf{1.2\ \text{GHz}}$ . A raw long crystal is stuck near a gigahertz — useless for modern links.

Step 2. Velocity-match with thick buffer layers and shaped electrodes to $|n_o-n_m|=0.3$ : $f_{3\text{dB}}\approx(3.0\times10^8)/(\pi\cdot0.04\cdot0.3)=\mathbf{8.0\ \text{GHz}}$ . Same length, same low $V_\pi$ , nearly seven times the bandwidth — the payoff of traveling-wave design.

Step 3. The trade-off: $f_{3\text{dB}}\propto1/(L|n_o-n_m|)$ while $V_\pi\propto1/L$ , so pushing past tens of gigahertz means shortening $L$ (raising $V_\pi$ ) and matching harder. ≈ 1.2 GHz unmatched, rising to ≈ 8 GHz once the mismatch is engineered down to 0.3.

3 · What caps the extinction ratio of a real Mach–Zehnder

A Mach–Zehnder has a 10% amplitude imbalance between its arms (field ratio $b/a=0.9$ ) from imperfect splitters. Even with a perfect $\pi$ dark-state phase, what extinction ratio can it reach?

Step 1. The dark state subtracts the arms, so residual power $\propto(a-b)^2$ and bright power $\propto(a+b)^2$ : $T_{\text{on}}/T_{\text{off}}=((a+b)/(a-b))^2=(1.9/0.1)^2=\mathbf{361}$ .

Step 2. In decibels (equation 7): $\text{ER}=10\log_{10}(361)=\mathbf{25.6\ \text{dB}}$ . A small 10% imbalance caps extinction at about 26 dB however perfectly the phase is set — so deep extinction (35–40 dB) demands split ratios balanced to about 1%, which is why real modulators add trim electrodes. A 10% field imbalance limits the extinction ratio to ≈ 25.6 dB.

Physics of the portrait

Every modulator on Earth runs on an effect one man tabulated before the electron was understood. The linear electro-optic effect carries his name because he measured it, cataloged it, and wrote the book on the crystal optics that frames it.

Origins · 1 minFriedrich Pockels and how this was discovered

Friedrich Pockels

1865–1913, German physicist, Göttingen and Heidelberg · electro-optics and crystal optics

Friedrich Carl Alwin Pockels studied under Woldemar Voigt at Göttingen, the great systematizer of crystal physics, and inherited Voigt's tensor way of thinking. In the early 1890s he made the careful measurements that established the linear electro-optic effect: in crystals lacking a centre of symmetry — quartz, tourmaline, Rochelle salt, and the ones we now build modulators from — an applied field changes the birefringence in direct proportion to the field, distorting the index ellipsoid through the tensor $r_{ijk}$ that bears his name. This linear response is fundamentally different from the quadratic effect John Kerr had found in liquids in 1875, and Pockels showed it is the fingerprint of broken inversion symmetry. His 1906 Lehrbuch der Kristalloptik became the standard crystal-optics reference for a generation, and he held the chair of theoretical physics at Heidelberg. It is a quiet irony that his effect sat as a curiosity for six decades until the laser and the fibre made a fast, low-loss phase shifter the most valuable device in photonics — and that his sister Agnes Pockels, working without a university position, became a founder of surface science in the same years. The lithium niobate Mach–Zehnder that anchors the modern internet is Pockels' tensor, energized.

History

1875

— Kerr discovers the quadratic electro-optic effect in liquids and glasses

1893

— Pockels characterizes the linear electro-optic effect in non-centrosymmetric crystals

1958

— Franz and Keldysh independently predict field-assisted sub-gap absorption

1984

— Miller and co-workers report the quantum-confined Stark effect in quantum wells

2018

— thin-film lithium niobate Mach–Zehnder modulators reach CMOS-level drive at 100 GHz

Going further

Texts. Saleh & Teich, Fundamentals of Photonics, 3rd ed., ch. 21 (Electro-Optics) and ch. 24 (Photonic devices) · Yariv & Yeh, Optical Waves in Crystals and Photonics (the electro-optic tensor and LiNbO $_3$ geometries) · Hecht, Optics, 5th ed., §8.11 (electro-optic effects) · Miller et al., Phys. Rev. Lett. 53, 2173 (1984), the founding QCSE paper. The key relations used throughout are: the Pockels index change $\Delta n=-\tfrac12 n^3 rE$ , the Mach–Zehnder transfer $T=\cos^2(\pi V/2V_\pi)$ , and the traveling-wave sinc response. The LiNbO $_3$ constants ( $n_e\approx2.14$ , $r_{33}\approx30.8$ pm/V at $1.55\,\mu$ m) are standard material values; from them one obtains $V_\pi L\approx10$ V·cm and, for a 4 cm electrode, $V_\pi\approx2.6$ V. The velocity-mismatch bandwidths work out to $1.2$ GHz unmatched and $8$ GHz at $|n_o-n_m|=0.3$ , and a 10% arm-amplitude imbalance caps extinction at $25.6$ dB. The QCSE quadratic $E^2$ shift and $L_w^4$ scaling originate with Miller et al. (1984); the Franz–Keldysh Airy-function lineshape with Franz (1958) and Keldysh (1958); and the silicon plasma-dispersion index change with the Soref–Bennett free-carrier model.

Prerequisites. Waveguide Modes and the Wave Equation · Integrated Photonic Waveguides and Circuits · Polarization-Mode Dispersion and Birefringent Fibers. Leads to. Microring Resonators and Optical Cavities on Chip · Coupled-Mode Theory and Directional Couplers · Dispersion Management and Compensation.

Ask the tutor: Ask me to work through the $V_\pi L$ of a specific modulator, or to compare where a Mach–Zehnder, an electro-absorption modulated laser, and a silicon microring each win.

Why does the Pockels effect give only phase, not intensity?

Because it changes the refractive index, and the index sets how fast a crest moves — the phase — not how much power the wave carries. Applying a voltage to a Pockels crystal simply delays the light, so a photodetector at the output sees constant brightness with an invisible, ever-shifting phase. To make the phase visible you must compare the delayed wave against a reference: split the light, delay one path, and recombine, so interference translates phase into brightness. That is exactly what a Mach–Zehnder does, and why every Pockels intensity modulator wraps the crystal inside one — or places it between crossed polarizers, which does the same conversion for polarization.

What is the half-wave voltage Vπ, and why quote Vπ·L instead?

The half-wave voltage produces a $\pi$ differential phase between the two arms of a Mach–Zehnder — the swing that carries the output from fully bright to fully dark — and it tells you how hard the electronics must push. But $V_\pi$ alone is ambiguous, because it falls inversely with electrode length: a longer crystal reaches $\pi$ at a lower voltage. The comparable figure of merit is the product $V_\pi L$ , independent of length and set only by the material ( $n^3 r$ ), geometry (gap $d$ ), and mode overlap $\Gamma$ . When someone reports a modulator's efficiency, ask for $V_\pi L$ ; for lithium niobate it sits near 10 V·cm, and thin-film designs have driven it several times lower.

Why does making the electrode longer help and hurt at once?

Length is the central trade-off. Because phase accumulates as the light travels through the field, a longer electrode reaches the switching phase at a lower voltage, $V_\pi\propto1/L$ — that is the help. The hurt is that a longer electrode gives the microwave drive and the optical wave more distance to walk out of step when their velocities differ, so bandwidth falls as $f_{3\text{dB}}\propto1/(L|n_o-n_m|)$ . The same length that lowers the drive throttles the speed. Traveling-wave design escapes the bind by matching the microwave and optical velocities — thick buffer layers, ridge or thin electrodes — so a long, low- $V_\pi$ electrode can still run fast.

Franz–Keldysh or quantum-confined Stark effect — which does an EAM use?

Modern high-performance electro-absorption modulators use the quantum-confined Stark effect; bulk Franz–Keldysh is the older, weaker cousin. In bulk a field tilts the bands and lets photons just below the gap tunnel-absorb, but the effect is broad and modest and the field ionizes any exciton. Grow the active region as multiple quantum wells and the barriers forbid that ionization: the electron and hole push to opposite walls but stay bound, and their sharp absorption peak slides to lower energy quadratically in the field. The QCSE shift is roughly an order of magnitude stronger and far sharper, so a device only about 100 μm long gates the light deeply. The price is a temperature- and wavelength-sensitive excitonic edge.

What is chirp, and why does it decide how far a signal reaches?

Chirp is the parasitic phase (instantaneous-frequency) modulation that rides along with the intended amplitude modulation. When a modulator turns the light on and off, if it also swings the optical phase the pulse edges are frequency-shifted — the pulse is "chirped." On its own that is harmless, but fibre is dispersive: different frequencies travel at different speeds, so a chirped pulse smears as it propagates, closing the receiver's eye and capping reach. A push–pull Mach–Zehnder runs chirp-free ( $\alpha_{\text{chirp}}=0$ ) because the arms swing oppositely and the net phase stays put, which is why it dominates long-haul. Electro-absorption modulators carry residual chirp, to be budgeted — or deliberately signed to pre-compensate the fibre's dispersion.

How is a silicon modulator different from a lithium niobate one?

Silicon has a centre of inversion symmetry, so it has no Pockels effect at all — the linear response is forbidden. Silicon modulators instead exploit free-carrier plasma dispersion: injecting or depleting electrons and holes in a pn junction changes the index (and absorption) because free carriers alter the polarizability. The change is smaller and comes with loss, so silicon needs either longer Mach–Zehnder arms or the resonant enhancement of a microring to reach useful extinction. The compensating prize is manufacturability: silicon modulators are built in the same CMOS foundries as processors. Lithium niobate gives the cleanest, lowest-chirp phase response; silicon gives scale and integration. Both now coexist, chosen by application.