Chapter 19Electro-Optic Effects Across Platforms
To modulate at tens of gigahertz you need a faster knob than heat, and each platform offers a different one.
- use the Soref–Bennett relations for plasma dispersion in silicon;
- compute depletion widths and the phase shift of a PN junction in a waveguide;
- compare carrier depletion, injection and accumulation;
- describe the Pockels effect in LN and BTO, including crystal orientation and field overlap;
- describe electro-absorption, and compare platforms by VπL, loss and bandwidth.
Which effect gives the best VπL for a given loss and bandwidth?
19.1Free-carrier plasma dispersion
Silicon’s crystal is centrosymmetric, so it has no linear electro-optic (Pockels) effect (Chapter 1), and its Kerr and Franz–Keldysh effects are weak at telecom wavelengths. What it does have is free carriers. Electrons and holes respond to the optical field like a small plasma: they lower the refractive index and absorb a little light. Soref and Bennett fitted measured absorption spectra and derived, through the Kramers–Kronig relations, the changes at 1550 nm [40, 36]:
At \(10^{18}\) cm⁻³, holes lower the index by \(2.1 \times 10^{-3}\) and absorb 6 cm⁻¹ (26 dB/cm); electrons lower it by \(0.9 \times 10^{-3}\) and absorb 8.5 cm⁻¹. Holes give about three times more index change for the same loss, which is why modulator junctions put more of the mode in the p-type side. The changes are small, a thousandth of the index at most, so a silicon modulator needs either a long phase shifter or a resonator.
19.2PN, PIN and MOS structures
Three junctions move carriers in and out of the mode:
- Depletion. A reverse-biased PN junction across the waveguide. The bias widens the depletion region and sweeps carriers out of it, raising the index. Only majority carriers move, so it is fast, limited by the junction’s RC time to tens of gigahertz, but the change is small.
- Injection. A forward-biased PIN diode fills the intrinsic waveguide with electrons and holes. The index change is large, but the carriers take their lifetime, about a nanosecond, to recombine, limiting the bandwidth to around a gigahertz without electrical pre-emphasis. It suits variable attenuators and slow switches.
- Accumulation. A thin oxide inside the waveguide forms a MOS capacitor. A voltage piles up charge on both sides of the oxide, exactly where the mode is strongest. This gives a large change per volt but a high capacitance.
For an abrupt junction with doping \(N_A\) and \(N_D\), the depletion width under a reverse bias \(V\) is
and the phase shifter’s effective-index change is the overlap of the mode with the carriers removed. To compute it, the TE mode of a 500 × 220 nm rib with a 90 nm slab was solved with femwell, and its sensitivity to the silicon index across the width, \(\partial n_\mathrm{eff}/\partial n(x)\), was taken from first-order perturbation theory (figsrc/ch19_pn.py). Summing that weight times the Soref–Bennett change in each doped region gives \(\Delta n_\mathrm{eff}(V)\), within 0.2 % of a direct mode solve with the junction in place.
With \(N_A = N_D = 5\times10^{17}\) cm⁻³ and the junction in the middle of the rib, \(V_{bi} = 0.92\) V and the depletion region is 69 nm wide at 0 V and 123 nm at 2 V reverse. Between the two, femwell gives \(\Delta n_\mathrm{eff} = 1.13\times10^{-4}\), so
\[V_\pi L = \frac{\lambda\,V}{2\,\Delta n_\mathrm{eff}} = \frac{1.55\times10^{-4}\ \mathrm{cm} \times 2\ \mathrm{V}}{2 \times 1.13\times10^{-4}} = 1.37\ \mathrm{V\cdot cm}.\]The carriers left in the waveguide absorb 13 dB/cm at 0 V (11 dB/cm at 2 V), so a 3 mm phase shifter with \(V_\pi \approx 4.6\) V loses about 3.5 dB. Its junction capacitance is about 0.19 fF/µm, or 0.6 pF for 3 mm.
Table 19.1. Lateral PN phase shifters in a 500 × 220 nm rib at 1550 nm (femwell mode, Soref–Bennett, abrupt symmetric junction).
| Doping (each side) | Depletion width at 0 V | VπL at 2 V | Loss at 0 V | Loss × VπL |
|---|---|---|---|---|
| 2 × 10¹⁷ cm⁻³ | 106 nm | 2.05 V·cm | 4.7 dB/cm | 10 dB·V |
| 5 × 10¹⁷ cm⁻³ | 69 nm | 1.37 V·cm | 13 dB/cm | 18 dB·V |
| 1 × 10¹⁸ cm⁻³ | 50 nm | 1.13 V·cm | 28 dB/cm | 31 dB·V |
Higher doping gives a better \(V_\pi L\) but more loss and capacitance; lighter doping gives a better loss–efficiency product but a longer device. The \(V_\pi L\) also grows with the bias, because the depletion width grows only as \(\sqrt{V_{bi} + V}\). Real designs tune the profile: offset junctions that favour holes, L- and U-shaped junctions that wrap the depletion region around the mode, and interleaved junctions along the waveguide. These raise the efficiency roughly twofold [36].
Try it. Change the doping, the bias and the junction position. The cross-section shows the depletion region moving through the mode, and the plots give the index change and the phase over the whole length.
19.3The Pockels effect
In a non-centrosymmetric crystal an electric field changes the index in proportion to the field, through the electro-optic tensor \(r_{ij}\). For the strongest coefficient of lithium niobate, with the optical and the applied fields both along the crystal’s z axis,
for a single arm with electrode gap \(g\) and electro-optic overlap \(\Gamma\), the fraction of the applied field that reaches the optical mode. The change is instantaneous: the bandwidth is set by the electrodes, not the material, and there is no added optical loss.
For x-cut TFLN with a TE mode, \(n_e = 2.14\) (\(n_e^3 = 9.8\)) and \(r_{33} = 30.9\) pm/V. With electrodes 5 µm apart and \(\Gamma = 0.5\),
\[V_\pi L = \frac{1.55\times10^{-6} \times 5\times10^{-6}}{9.8 \times 30.9\times10^{-12} \times 0.5} = 0.051\ \mathrm{V\cdot m} = 5.1\ \mathrm{V\cdot cm}\]for one arm, and 2.6 V·cm for a push–pull Mach–Zehnder whose two arms see opposite fields. This is worse than silicon’s 1.4 V·cm, but with a propagation loss of a few tenths of a decibel per centimetre instead of 13 dB/cm, so the device can be several centimetres long. Bulk LN modulators, with wide electrode gaps and a weakly confined mode, need well over 10 V·cm; thin films brought it down by confining the mode to a sub-micrometre ridge between closer electrodes [49, 32].
Crystal orientation matters. In x-cut LN the z axis lies in the plane of the film, so electrodes on either side of the ridge drive \(r_{33}\) with a TE mode; in z-cut films the field must be vertical, with electrodes above and below. Barium titanate has a different tensor: its largest coefficient, \(r_{42}\), couples fields at an angle to the crystal axes, so BTO waveguides are oriented at an angle to the crystal and designed for that mixed overlap. Thin BTO films on silicon have shown Pockels coefficients of hundreds of picometres per volt, an order of magnitude more than LN [33]. Organic electro-optic polymers in slot waveguides, and silicon strained by a stressed cladding, are two other ways to give silicon a Pockels response.
19.4Electro-absorption
Instead of changing the index, a field can move a material’s absorption edge:
- Franz–Keldysh effect. In bulk germanium or GeSi near its band edge, a field lets electrons tunnel into the gap, smearing the absorption edge to longer wavelengths. Choosing the GeSi composition places the edge just above the operating wavelength, so a few volts switch the absorption.
- Quantum-confined Stark effect. In quantum wells (InGaAsP or InGaAlAs on InP, or Ge/SiGe on silicon), the excitonic absorption peak shifts strongly to longer wavelengths with a perpendicular field.
Electro-absorption modulators are short (tens of micrometres), so their capacitance is small and they need little voltage. But they work only near the absorption edge, over a narrow band of wavelengths that drifts with temperature. They modulate amplitude, not phase, with some chirp (Chapter 20).
19.5Figures of merit
Three numbers compare phase shifters across platforms: the voltage–length product \(V_\pi L\); the loss–efficiency product \(\alpha V_\pi L\) in dB·V, the loss of a device long enough for a π shift at 1 V; and the bandwidth, with the capacitance per length that sets the drive energy. Table 19.2 gives typical values.
Table 19.2. Phase shifters compared (typical values at 1550 nm). The silicon depletion row is from Table 19.1; the others are representative of published devices.
| Effect | Platform | VπL | Optical loss | Bandwidth | Notes |
|---|---|---|---|---|---|
| Thermo-optic | Si heater | (26 mW for π) | none | ~20 kHz | Chapter 18 |
| Carrier depletion | Si PN | 1–2.5 V·cm | 5–30 dB/cm | 30–60 GHz | CMOS-compatible |
| Carrier injection | Si PIN | < 0.05 V·cm | grows with current | ~1 GHz | slow, lossy |
| Accumulation | Si MOS | 0.2–0.5 V·cm | ~10 dB/cm | 10–40 GHz | high capacitance |
| Pockels | TFLN | 2–3 V·cm (push–pull) | < 0.5 dB/cm | > 100 GHz | linear, no chirp |
| Pockels | BTO on Si | < 1 V·cm | 1–10 dB/cm | tens of GHz | film quality |
| Electro-absorption | GeSi, InP QW | amplitude only | at the band edge | > 50 GHz | narrow band |
No single effect wins. Silicon depletion is the standard because it is free in a CMOS-style process. TFLN wins on linearity, loss and bandwidth for coherent and analogue links. BTO promises the strongest effect on silicon. Electro-absorption gives the smallest devices.
Silicon carrier depletion for the ring modulators, with a doping near 10¹⁸ cm⁻³ and an L- or U-shaped junction for about twice the efficiency of Table 19.1. The resonance shift per volt this gives is designed in Chapter 20.
Reading
- [40] Richard A. Soref and Brian R. Bennett. “Electrooptical effects in silicon.” IEEE Journal of Quantum Electronics, 23(1):123–129, 1987.
- [36] Graham T. Reed, Goran Mashanovich, Frederic Y. Gardes and David J. Thomson. “Silicon optical modulators.” Nature Photonics, 4:518–526, 2010.
- [32] Di Zhu et al. “Integrated photonics on thin-film lithium niobate.” Advances in Optics and Photonics, 13(2):242–352, 2021.
- [49] Cheng Wang et al. “Integrated lithium niobate electro-optic modulators operating at CMOS-compatible voltages.” Nature, 562:101–104, 2018.
- [33] Stefan Abel et al. “Large Pockels effect in micro- and nanostructured barium titanate integrated on silicon.” Nature Materials, 18:42–47, 2019.
- [1] Lukas Chrostowski and Michael Hochberg. Silicon Photonics Design: From Devices to Systems. Cambridge University Press, 2015. https://www.cambridge.org/core/books/silicon-photonics-design/BF3CF13E8542BCE67FD2BBC7104ECEAB
- Silicon modulators move free carriers: at 10¹⁸ cm⁻³, holes lower the index by 2.1 × 10⁻³ at 26 dB/cm of loss, three times more index per unit loss than electrons.
- A lateral PN junction (5 × 10¹⁷ cm⁻³) in a 500 × 220 nm rib gives \(V_\pi L\) ≈ 1.4 V·cm with 13 dB/cm of loss; doping trades efficiency against loss and capacitance.
- Injection is efficient but slow (~1 GHz); accumulation is efficient but capacitive.
- The Pockels effect in TFLN gives about 2.6 V·cm in push–pull with almost no loss and > 100 GHz; BTO’s coefficient is an order of magnitude larger.
- Electro-absorption modulators are short and low-voltage but narrow-band; compare platforms by \(V_\pi L\), \(\alpha V_\pi L\) and bandwidth.