The HBM4 packaging stack you’re shipping past Blackwell needs glass-substrate impedance signoff that the rest of the EDA industry hasn’t built. ChipletOS HBM4 cal head is the first-deployed BEM-vs-Palace μ-correction surface for this regime — 90.3% gap closure, v2-clean retrain (2026-05-05). ULTRA_HIGH_FREQ now ties HBM4 at 90.5% — Genesis is the only platform with calibrated full-wave-grade accuracy at >100 GHz.
d ∈ [25, 50] µm × p ∈ [40, 80] µm × t ∈ [300, 600] µm. The HBM4 corner Genesis trains on is the same corner Nvidia’s Blackwell+ packaging team designs into. Live envelope gate; runs only when input is in-distribution.
Raw BEM-vs-Palace gap on test slice: 4.47 Ω (13.06% mean abs err vs Palace truth). After v2-clean retrain: 1.42% mean abs err. 90.3% of the gap closed.
/v1/glass-pdk/predict-impedance + /v1/glass-pdk/geometry-from-target + /v1/coupons/export-fab. Activate via CHIPLETOS_PALACE_RESIDUAL_HEAD=1.
Per-regime witness JSONs ship with the cal head, including filters_applied with per-criterion drop counts (HBM4 9.9% of rows filtered for Palace solver failures on hard meshes). Scope: simulator-vs-simulator calibration; VNA truth pending.
Defensive acquisition because Nvidia/Marvell were ahead on photonic switch. Same pattern: substrate-layer IP that locks competitors out. Genesis owns the EDA-adjacent workflow lock on glass.
Same defensive pattern. Glass-substrate workflow lock without ChipletOS = your HBM4+ packaging team integrates 3-5 separate point tools. With ChipletOS = single workflow lock across 5 EDA vendors + 5 multi-solver-verified golden kits.
Up to $5.5B PRE-REVENUE. Glass substrate is the integration substrate. Genesis ships the cross-EDA workflow today.
$700M, ~10× rev, workflow lock-in premium. Genesis ships the same posture: cross-EDA workflow lock + Genesis-side cert + customer-side native signoff.