scienceBuilt for Marvell silicon-photonics

800G → 1.6T → 3.2T optical. MMI + ring + grating signoff on one /v1/photonics surface.

Marvell’s 800G PAM4 + 1.6T coherent + co-packaged optics roadmap turns on MMI splitter uniformity, ring resonator Q + tuning range, and grating coupler bandwidth. ChipletOS Photonic Signoff ships a research-preview analytical physics engine(~21% mean error vs FDFD-grade references; trained surrogates pending the FDTD corpus) — MMI / ring / grating / MZI / photonic crystal all served today.

Research preview: analytical engine, all 6 primitivesConformal CI: lands with the trained surrogates✓ Lumerical / OmniSim / ANSYS export
Your specific photonic challenges

What Marvell SiPh hits on every design iteration — and how ChipletOS Photonic Signoff collapses the loop.

MMI splitter uniformity

800G PAM4 16-channel WDM

Insertion-loss imbalance < 0.3 dB across 16 outputs is the spec. ChipletOS MMI research-preview analytical engine (trained surrogate pending FDTD corpus) returns {insertion_loss_dB, imbalance_dB, L_pi_um} in < 10 ms per call — drop into the Marvell layout-iteration loop directly.

Ring resonator tuning

WDM tunability + Q ≥ 10k

Ring surrogate v1 returns {Q_loaded, FSR_nm, FWHM_nm} + a 95% conformal prediction interval sourced from a held-out test residual quantile — bounded predictions for every (radius, gap, width, λ) Marvell wants to evaluate.

Grating coupler BW

3-dB BW ≥ 40 nm for CWDM

Grating surrogate v1 returns {coupling_efficiency, peak_wavelength_nm, bandwidth_3dB_nm} — driver-friendly outputs that match the Marvell CWDM band plans. Pareto-explore (period / duty / etch / angle) against your target.

Try the live demos

Three Modal endpoints. Public. No auth.

  • GET /v1/photonics/signoff-health — reports research-preview status: analytical engine on all 6 primitives; trained surrogates pending FDTD corpus.
  • POST /v1/photonics/predict-waveguide-mode — closed-form analytical model (canonical for waveguide).
  • POST /v1/photonics/validate-against-ieee — published-paper cross-check across 5 SOI Neff papers, within the analytical-model expected band vs published silicon-photonic references.
Try the playground →Pilot SOW ($150K, 6 weeks) →
See Trust & Validation for the full validation-suite methodology and our published-paper cross-check numbers.