PPLN Λ = 18.99 µm · 1550 → 775 nm · 2 px/µm

6.2 Traveling-wave design: velocity, impedance, RF loss

L1 Phenomenon

At 100 GHz the drive signal on the chip has a wavelength of about 1.4 mm (c / (f nRF) with nRF = 2.2 derived), shorter than the electrode. The electrode is therefore a transmission line: the microwave and the light race down the chip together. If they travel at the same speed and the microwave does not fade, every slice of the electrode adds phase in step, and the modulator stays fast however long it is.

EO response, forward + reflected wave, ZS = 50 Ω0 to 300 GHz
0255075100125150175200225250275300−15−12−9−6−303frequency (GHz)EO response (dB)
f3dB > 300.0 GHz|nRF − ng| 0.000 RF loss at f3dB – fvel none floss  306.2 GHz
28 rows
10.0 mm
2.25
2.25
f axis
 3.00 dB/cm
Loss scaling
 50.0 Ω
= Z0
Model derived: transmission line with forward and load-reflected waves, source reflection, light transit at ng (model in L2 below; model class of Alferness 1982 and Chung 1991); response normalized to its DC value. Dashed verticals: velocity-only and loss-only 3 dB limits. Presets load nRF, ng and, where the row has them, Z0, RF loss and length; the marker is the row's reported bandwidth (arrow = lower bound).atlas eb97f96 · 2026-10-20 · fetched 2026-10-25
electrode (Au), RF wave at 100 GHz, n_RF 2.25: λ = 1.33 mm+10voltage seen by one light pulse along the electrode (1 = DC)10 mm
050100150200250300−15−12−9−6−303frequency (GHz)EO response (dB)
3 dB bandwidth> 300.0 GHz
Index mismatch
0.000
Velocity limit
none
Loss limit (6.4 dB)
 306.2 GHz
RF loss at f3dB
–
10.0 mm
2.25
2.25
 3.00 dB/cm
 50.0 Ω
Loss scaling
= Z0
Load
Frequency axis
Model: transmission line with forward and load-reflected waves, source impedance ZS = 50 Ω, light transit at ng; model class of Alferness 1982 and Chung 1991; response normalized to its DC value; dashed verticals: velocity-only and loss-only 3 dB limits; presets load nRF, ng and, where present, Z0, RF loss and length; marker = reported bandwidth (arrow = lower bound). Set nRF = ng and loss 0: the response is flat at every length. The strip shows the RF voltage that one optical pulse sees as it crosses the electrode at 100 GHz; the 1550 nm path is drawn in false color. derivedatlas eb97f96 · 2026-10-20 · fetched 2026-10-25

L2 Line model, limits, levers

Model

Voltage on a line with propagation constant γ = α + j ω nRF / c, load reflection ΓL = (ZL - Z0)/(ZL + Z0), source reflection ΓS = (ZS - Z0)/(ZS + Z0):

V(z)=V+[e−γz+ΓLe−2γLeγz],V+=VSZ0/(Z0+ZS)1−ΓSΓLe−2γL

The light sees the voltage at the time it reaches z, t = t0 + ng z / c. The modulation it accumulates, normalized to the electrode length:

m(f)=V+[1−e−u+Lu+L+ΓLe−2γLeu−L−1u−L],u±=α+jωnRF∓ngc

EO response (dB) = 20 log10 |m(f) / m(0)|, with m(0) = VS ZL / (ZS + ZL). derived from transmission-line theory; the model class is the one in Chung, Chang, Adler 1991 [Chung et al. 1991] and Alferness 1982 [Alferness 1982].

Single-effect limits (derived here from the matched-load form, ZS = ZL = Z0):

LimitCondition3 dB point
Velocity mismatch only (α = 0)|sin x / x| = 1/√(2) at x = π f L |nRF - ng| / c = 1.392f3dB L = 0.443 c / |nRF - ng| derived
RF loss only (nRF = ng)(1 - exp(-a)) / a = 1/√(2) at a = α L = 0.738 Nptotal electrode RF loss at f3dB = 6.4 dB derived

Worked numbers derived, using the Chen 2022 (c) device (capacitively loaded, nRF 2.35, ng 2.2, Z0 50.5 ohm, RF loss 2.947 dB/cm at 67 GHz, L = 10 mm) reported:

  • Velocity term alone: f3dB = 0.443 c / (0.15 × 10 mm) = 88.5 GHz.
  • Loss term alone, loss scaling as √(f): 6.4 dB is reached at 67 GHz × (6.4 / 2.947)2 = 317 GHz.
  • Reported 3 dB bandwidth: > 67 GHz (bound) reported.

Design levers:

  • Velocity. In a quasi-TEM line nRF = c √(L' C') and Z0 = √(L' / C'); adding capacitance per length (T-rail loading) raises nRF and lowers Z0 together. Of 28 TFLN/TFLT device rows that report both indices, 23 have |nRF - ng| ≤ 0.10; the others are Chen 2022 (c) (0.15), Renaud 2022 (3um-738) (0.16), Mao 2022 (a) (0.28), Lin 2026 (a) (0.50, lumped), Chen 2023 (a) (ng 4.0, approx.) derived.
  • RF loss. Narrow gaps raise conductor loss; T-rail segments keep the rails wide while bringing the field close to the waveguide reported. Low-k underfill was used to reduce RF loss and reach velocity matching with a narrow gap reported.
  • Impedance. A source at 50 ohm into a lower-Z0 line divides the voltage; load mismatch adds a counter-propagating wave whose contribution averages out once ω (nRF + ng) L / c >> 1.
Velocity matching: nRF vs ng, TFLN and TFLT device rows28 rows; 23 within |Δn| ≤ 0.10
22.22.42.62.833.23.43.63.844.222.533.54optical group index ngRF index nRFnRF = ng
GSGCPWGSSGT-railslow wavelumpedother / nonesize: length 1, 5, 10, 20 mmband: |nRF − ng| ≤ 0.10
Rows that report both indices reported; click a mark to load it into the EO response plot above.atlas eb97f96 · 2026-10-20 · fetched 2026-10-25
22.22.42.62.833.23.43.63.844.222.533.54optical group index ngRF index nRFnRF = ng
GSGCPWGSSGT-railslow wavelumpedother / none
28 published traveling-wave LN and LT modulators that report both indices reported; band: |nRF − ng| ≤ 0.10; size: length 1, 5, 10, 20 mm; 23 sit within 0.10 of the diagonal, where the microwave keeps pace with the light. Marker size grows with electrode length; click one to load it into the response model above.atlas eb97f96 · 2026-10-20 · fetched 2026-10-25

Slow light and loaded electrodes

In a slow-light waveguide the phase from a given index change grows with the optical group index, which lowers VπL; the electrode must then slow the microwave to the same group velocity. A 1 mm SiN-loaded TFLN MZM with slow-light waveguides and capacitively loaded slow-wave electrodes reached 0.21 V cm with bandwidth > 110 GHz reported.

L3 Atlas RF columns and measurement

  • Atlas RF fields: RF index, optical group index, Z0, and RF loss in power dB/cm at a stated frequency (amplitude coefficients converted). A cross-section's RF index and impedance are not the effective values of a periodically loaded line, so cross-section values do not give the loaded-line nRF or Z0.
  • Measured RF loss is usually quoted at one frequency; the √(f) scaling (conductor skin effect) is a model assumption, and dielectric loss scales as f.
  • Instrument roll-off, probe and cable de-embedding, and the photodiode response enter measured EO S21 and are removed by the authors with stated or unstated calibrations.

Sources

5 cited in 6.2
RefTitleLink
Alferness 1982Waveguide Electrooptic Modulatorsdoi:10.1109/tmtt.1982.1131213
Chung et al. 1991Modeling and optimization of traveling-wave LiNbO/sub 3/ interferometric modulatorsdoi:10.1109/3.81370
Kharel et al. 2021Breaking voltage–bandwidth limits in integrated lithium niobate modulators using micro-structured electrodesdoi:10.1364/OPTICA.416155
Liu et al. 2025Ultrahigh‐Efficiency Dual‐Band Thin‐Film Lithium Niobate Modulator Incorporating Low‐k Underfill for 390 Gbit s −1 PAM8 Transmissiondoi:10.1002/lpor.202402053
Shen et al. 2024Highly Efficient Slow‐Light Mach–Zehnder Modulator Achieving 0.21 V cm Efficiency with Bandwidth Surpassing 110 GHzdoi:10.1002/lpor.202401092