Potassium niobate thin-film electro-optic Mach-Zehnder modulator for silicon photonics

Potassium niobate (KNbO3) is a perovskite ferroelectric whose electro-optic tensor coefficients substantially exceed those of lithium niobate — a material that has dominated optical modulation for two decades. The silicon photonics industry is in the middle of a forced substitution cycle: the incumbent thin-film lithium niobate (TFLN) platform is increasingly supply-constrained, technically crowded, and approaching fundamental limits in modulation efficiency at the bandwidth operating points demanded by 800G and 1.6T datacom. KNbO3 on a lattice-matched or index-engineered buffer (MgO or silicon nitride) offers a practical path to breaking the Vπ·L barrier that TFLN cannot clear without sacrificing bandwidth or propagation loss. This asset occupies a genuine lead position in the "Potassium-niobate thin-film electro-optic modulator" patent family. The device concept — KNbO3 thin film integrated into a Mach-Zehnder traveling-wave modulator (MZM-TWM) architecture on either MgO or SiN buffer layers — is structured to exploit KNbO3's superior electro-optic tensor and dielectric properties at the specific film geometry, electrode pitch, and frequency windows where the efficiency figure of merit (Vπ·L) can reach approximately 0.59 V·cm at bandwidths approaching 82 GHz. That combination of efficiency and bandwidth simultaneously outperforms commercially deployed TFLN and emerging thin-film lithium tantalate (TFLT) in the modulation figure-of-merit space most relevant to high-speed intensity modulation. The filing is squarely aimed at datacom and silicon photonics — markets where switching speed, energy per bit, and chip-scale integration density are decisive. The timing aligns with a technology transition: large hyperscalers and optical transceiver vendors are actively evaluating materials alternatives to LiNbO3 as they push beyond 112 Gbaud per lane, and no mature commercial KNbO3 modulator product yet exists, leaving clear whitespace in both the market and the patent landscape.

KNbO3 crystallizes in the orthorhombic Amm2 space group at room temperature and is a classic displacive ferroelectric. Its electro-optic r33 coefficient is among the highest of any perovskite oxide — generally cited in the literature as 64 pm/V or above, compared with roughly 30 pm/V for LiNbO3 — which directly translates to a lower half-wave voltage-length product when the material is incorporated into a traveling-wave Mach-Zehnder architecture. The device geometry targeted here places the KNbO3 layer on either an MgO buffer (for near-epitaxial quality with controlled lattice mismatch) or a silicon nitride buffer (for CMOS-back-end-of-line compatibility), each choice influencing the mode-overlap integral between the optical field and the electro-optic layer as well as the velocity-matching conditions between the RF traveling wave and the guided optical mode. The computational figure-of-merit sweep is the core quantitative output. Lattice Graph executed a sweep across 243 distinct electrode-geometry and film-parameter combinations and a broader 500-geometry parametric space. Of these, 21 out of 243 (8.6 percent) and 141 out of 500 (28.2 percent) configurations satisfied all three simultaneous constraints: Vπ·L below 1 V·cm, 3-dB modulation bandwidth above 70 GHz, and propagation loss below 1 dB/cm at 1550 nm. The best-performing geometry in the sweep achieves Vπ·L approximately 0.59 V·cm at approximately 82 GHz — a figure that, if confirmed experimentally, would sit below anything publicly demonstrated on TFLN platforms at equivalent bandwidth without heroic electrode engineering. The dielectric characterization underpinning the RF velocity-match and electrode design is grounded in two independent DFT sources. The in-house Gamma-point DFPT calculation (pipeline designation BB-1b) yields an electronic (high-frequency) dielectric constant ε∞ of approximately 5.41 for the KNbO3 layer modeled here. This is combined with the experimentally established and literature-corroborated total dielectric constant ε_total of approximately 30.3 drawn from the published corpus, which governs the RF traveling-wave phase velocity. Accurate values for both limits are essential: ε∞ sets the optical refractive index and mode effective index, while ε_total determines the microwave effective index of the coplanar waveguide electrode. Velocity mismatch between these two values is a principal bandwidth limiter in traveling-wave modulators, and the combination places KNbO3 in a favorable range relative to LiNbO3 (where ε_total in the relevant crystallographic direction is substantially larger, worsening velocity mismatch at high frequency). It should be noted that the standard multi-MLIP consensus workflow applied elsewhere in the Lattice Graph portfolio — cross-validation by MACE, CHGNet, MatterSim, and ORB for phonon stability — was not the primary analytical pathway here. KNbO3 is a well-established, experimentally synthesized ferroelectric with extensive published literature confirming its Amm2 phase stability; the bulk of the computational work therefore focused on device-level simulation (the dielectric and geometry sweeps) rather than on phase-stability screening of an unknown compound. This is appropriate for the material class and does not diminish the rigor of the device simulations, but it means the main open validation gate is fabrication-level: a physical modulator must demonstrate the predicted Vπ·L, 3-dB bandwidth, and propagation loss numbers before the asset transitions from a computationally grounded prediction to an experimentally confirmed device.

The silicon photonics optical modulator market sits within the broader datacom transceiver and photonic-integrated-circuit industry. Driven by hyperscaler capital expenditure on AI training clusters and front-haul/back-haul optical interconnects, demand for 400G, 800G, and emerging 1.6T coherent and intensity-modulated links is growing rapidly. Electro-optic modulators are a critical component in every optical transceiver, and device performance — particularly the Vπ·L figure of merit, which directly sets driver voltage requirements and energy per bit — is a primary technical differentiator. The addressable market for high-speed EO modulators and the IP that governs them is estimated in the range of one to five billion dollars annually, with the higher end of that range reflecting the full scope of photonic transceiver silicon that depends on modulator IP. This estimate is Lattice Graph's own commercial-context assessment and should be treated as an order-of-magnitude sizing, not a market-research forecast. Buyers of this IP are primarily photonic-integrated-circuit developers and optical transceiver vendors who are already evaluating or integrating thin-film EO platforms. The licensing logic is straightforward: whoever fabrics a KNbO3 MZM on MgO or SiN and sells it into the datacom ecosystem would need to clear a license. Royalty structures for foundational device patents in photonics typically run in the range of low single-digit percentage of module revenue or a flat per-unit rate, though the exact model would depend on whether the acquirer seeks an exclusive field-of-use license or a portfolio cross-license. The claim structure covering the specific buffer/film/pitch windows — tied expressly to KNbO3's index and EO tensor, not carried over from LiNbO3 or LiTaO3 platforms — gives the patent a degree of specificity that makes it harder to design around without exiting the high-performance regime entirely.

Vpi*L<1 V*cm modulator outperforming TFLN/TFLT in the FOM window

KNbO3 + MZ/traveling-wave topology + buffer chemistry + film/pitch windows tied to KNbO3 index/EO tensor

The natural acquirers and licensees for this asset are organizations already active in thin-film EO modulator development or silicon photonics platform integration. That includes vertically integrated optical transceiver companies building their own PIC platforms (such as those supplying hyperscalers with 800G/1.6T modules), foundry platforms that offer EO modulator process design kits and are evaluating materials beyond LiNbO3, and fabless photonics IP companies that aggregate modulator-related patents for licensing programs. Defense and government photonics programs investing in compact EO modulators for sensing, LiDAR, or RF-photonics applications represent a secondary acquisition channel, particularly given KNbO3's favorable EO properties in the near-infrared. Any organization that is today shipping or developing TFLN or TFLT products and is monitoring the next generation of EO material platforms would have direct strategic interest in controlling or at minimum clearing this IP before committing to a KNbO3 development program. The asset is also relevant to university and national-laboratory groups that may seek to experimentally validate the computational predictions — establishing a collaboration or sponsored-research agreement early would both accelerate the fabrication validation gate and provide an additional layer of evidence supporting issued claims. A licensing-first strategy (non-exclusive to stimulate market development, with an option to convert to exclusive) may be optimal given the pre-commercial state of the technology, but an outright assignment to a strategic buyer who intends to develop a KNbO3 platform is equally viable.

The primary technical risk is deposition-quality uncertainty. KNbO3 thin-film deposition by sputtering, pulsed-laser deposition, or MOCVD is significantly less mature than LiNbO3 on lithium niobate substrates, and achieving the low propagation-loss values (below 1 dB/cm) required to pass the three-constraint filter will demand careful control of film crystallinity, surface roughness, and ferroelectric domain orientation. If achievable propagation loss in practical KNbO3 films is materially higher than modeled, a significant fraction of the predicted passing geometries may fall out of the feasible window, narrowing the design space and weakening the case for the efficiency advantage over TFLN. The secondary risk is freedom-to-operate: the assessment is currently narrow, and any continuation or broadened reissue of relevant prior art, or a broad EO-material claim from an active TFLN platform filer, could create clearance work. The roadmap to de-risk both vectors is the same: a fabricated KNbO3 MZM device that demonstrates measured performance within the claimed windows simultaneously validates the simulation, anchors the claims to a demonstrated embodiment, and clarifies any FTO exposure by forcing prior art into a concrete technical comparison. A targeted collaboration with an academic thin-film deposition group — likely six to eighteen months to first optical characterization — is the most efficient path to that gate.