Intellectual Property

909 Filed Subclaims. 9 Technology Areas.

280 utility filed Jan 2026 · 629 provisional · 146 independent · 0 issued to date · deadline 2027-01-31

Comprehensive IP covering glass interposer impedance, hybrid bonding yield prediction, isolation synthesis, wafer control, and thermal intelligence for advanced heterogeneous integration.

11 file-ready patent provisional bundles

Inverse design, Pareto optimization, fab-coupon export, per-regime calibration heads, lab readiness scoring, route-backed signoff, cross-solver consensus, adversarial verification, MCP agent pack, adjoint-BEM correlation, and provenance lineage — each with code-to-claim trace, §112 enablement evidence, retraction check, and counsel-ready cover memo.

Cross-physics calibration — 6 of 6 regimes deployed

Per-regime BEM-vs-Palace μ-correction cal heads close the gap between fast quasistatic and full-wave solvers. Latest retrain (2026-05-05) closes 72.5% of the pooled gap (17.10% → 4.71%) at an average 74.0% per-regime closure: ULTRA_HIGH_FREQ 90.5% (29.72% → 2.96%), HBM4 90.3% (13.06% → 1.42%), WIDE_PITCH 87.6% (10.26% → 1.52%), UCIE 81.6% (12.07% → 2.17%), EXTREME_TIGHT 50.6% (16.84% → 7.23%), MMWAVE 43.4% (20.64% → 12.95%). 4/6 regimes elite (>80% closure). Pattern reproduces across all regimes — tighter geometries calibrate cleaner; honest disclosure on the wide-distribution and mmWave regimes. Palace is full-wave FEM cross-physics truth, not VNA measurement.

Productization — 261 live API endpoints

Multi-objective Pareto inverse design · IPC/SEMI DRC validation · cross-solver matrix (100-geometry × 5-solver) · fab-coupon export (GDS + 12-layer stack-up + per-foundry SOW) · measurement validation with honest sim-vs-VNA disclosure. All endpoints pytest-covered with 1121 tests + buyer-runnable adversarial harness (100/100 pass + 30/30 OOD recall).

gavel146Independent Filed Claims
hub9 + QUALTechnology Areas + Quality Plane
scheduleJan 2026Priority Date (filed claims)
description4 New DraftsCross-discipline integration, Glass SI/PI, Bonding Yield

Highest-Value Independent Claims

Top 10 Claims

Full claim text, evidence, and buyer-specific analysis available under NDA.

1

Overlay Metrology → Yield via Spatial Autocorrelation

Yield Prediction Engine

Bridges overlay registration measurements to hybrid bonding yield prediction via pairwise autocorrelation and spatial defect correlation length.

Relevant to: KLA, ASML

2

End-to-End GDS-to-Yield Pipeline for Hybrid Bonding

Yield Prediction Engine

Seven-stage physics chain from CMP recess to contact mechanics to void nucleation to thermal stress to Monte Carlo yield, with mandatory data contracts between stages.

Relevant to: Corning, KLA, TSMC

3

Multiconductor BEM for Through-Glass Via Impedance

Glass Interposer PDK

Boundary-element multiconductor capacitance extraction for TGV arrays, capturing discrete ground-via placement effects that coaxial formulas cannot.

Relevant to: Corning, Intel, Cadence, Synopsys

4

Bayesian Sequential Calibration of Yield Model Parameters

Yield Prediction Engine

Conjugate Normal-Normal sequential updates over correlation length, CMP recess variance, and anneal stress, narrowing confidence intervals by 90% within 10 measurements.

Relevant to: KLA, Corning

5

Closed-Loop Adjoint Workflow for Isolation Synthesis

Isolation Synthesis Engine

Parameterizes isolation corridors as a 2D permittivity field, runs forward FDTD with adjoint gradient computation, and exports DRC-verified GDSII in a single closed loop.

Relevant to: Cadence, Synopsys, Qualcomm, Intel

6

Chirped LFI Dielectric Shield

RF & Photonics Suite

Chirped TiO2/SiO2 dielectric stack providing broadband laser fault injection rejection above 40 dB at 532/808/1064 nm simultaneously with visible transmittance above 90%.

Relevant to: Defense, Qualcomm, Apple

7

Multi-Dk Inverse Impedance Design

Glass Interposer PDK

Analytical-seed plus BEM-verify iterative search for target characteristic impedance across glass substrates of varying dielectric constant.

Relevant to: Corning, Intel, Cadence

8

Differentiable Yield-Aware Inverse Layout

Yield Prediction Engine

Fourier Neural Operator backpropagation through density, density gradient, and CMP recess channels to optimize layout density patterns for predicted yield.

Relevant to: KLA, TSMC

9

Spectral FFT Contact with Smoothed Dugdale CZM

Yield Prediction Engine

Fourier-space elastic contact combined with smoothed Dugdale cohesive zone optimization to compute wafer bonding gap and void distribution.

Relevant to: Infrastructure for hybrid bonding yield prediction

10

Zernike-Decomposed Iterative Learning Control

Wafer Control Suite

Noll-normalized Zernike mode decomposition with order-dependent gain scheduling for iterative wafer flatness control in lithography.

Relevant to: ASML, Tokyo Electron

11

Lot Intelligence Report — 4-Signal Release Verdict

Quality / Inspection Plane

System and method for generating a manufacturing lot release recommendation by aggregating (a) IPC-A-610G / SEMI E10 / ISO 9000-3 defect taxonomy resolution, (b) WM-811K-derived yield-pattern hypotheses by process step, (c) ranked CAPA recommendations keyed by regulatory reference, and (d) a queryable provenance graph binding each output field to its source solver, dataset, and validation method.

Relevant to: Absolics, DNP, Rapidus, Corning QA, Intel Foveros, KLA

12

Provenance Graph Binding — Solver / Dataset / Validation Audit Trail

Quality / Inspection Plane

Persistent provenance graph linking each manufacturing output field to a tuple of source solver, source dataset, validation method, and evidence path. Exposed via API for buyer audit.

Relevant to: All QA / compliance teams

13

Hand-Engineered Explainable BEM Fingerprint (128-dim pgvector, NEW, April 13)

Glass Interposer PDK

128-dimensional fingerprint built from glass/metal/via one-hots + Z₀/IL/RL/L/C/R/G frequency sweep + derived physics scalars. Every bucket has human-readable meaning, unlike opaque ViT/CNN embeddings. Enables sub-10 ms ANN lookup across the 15.92M-row long-format BEM corpus.

Relevant to: Corning, EDA houses

14

Shape-Invariant Wafer-Map Fingerprint

Quality / Inspection Plane

Radial + angular + quadrant + zone + line-residual feature extraction producing a 128-dim classifier over wafer maps of arbitrary size (26×26 to 212×204 dies). Pairs with a trained CNN classifier over the same WM-811K corpus (811,457 real fab wafers, 9 failure classes).

Relevant to: KLA, Archer

15

DRC ↔ IPC/SEMI/ISO Taxonomy Mapping

Quality / Inspection Plane

Every geometric DRC violation carries an optional IPC-A-610G / SEMI E10 / ISO 9000-3 `defect_code`, making the entire design-rule system speakable in buyer-QA-team language. Continuation claim of Rank 11.

Relevant to: All QA / compliance teams

lock

Full claim text under NDA

Complete independent claim language, prosecution status, filing dates, evidence packages, and valuation analysis are available in the NDA data room.

Moat Depth

Design-Around Evidence

Each claim is backed by quantitative evidence that competitors cannot trivially replicate or circumvent the protected methods.

Wafer Control Suite

982/1000 wins (simulation)
Test

ILC vs 5 classical controllers (PID, LQR, MPC, SM, Fixed) — synthetic Monte Carlo

What it proves

Control law non-trivially superior in 1,000-case analytical-plant simulation; hardware/FEM-in-loop validation future work

Glass Interposer PDK

0/41,700 designs escape BEM
Test

BEM vs coaxial analytical across 41,700 TGV designs

What it proves

Coaxial formula is 55–225% wrong on glass TGV geometries

Isolation Synthesis Engine

r = 1.0000 (energy objective), r = 0.9863 (spectral objective)
Test

Adjoint gradient vs finite-difference validation (EM Isolation Compiler)

What it proves

Adjoint gradient computation correct to numerical precision for energy FoM; spectral FoM within r>0.8 pass threshold.

Yield Prediction Engine

96% CI reduction in ≤10 wafers
Test

Overlay autocorrelation → yield calibration

What it proves

Correlation length converges from 294µm to 12.4µm with no competing published method

Glass Interposer PDK

4.0% MAE
Test

BEM vs 5 IEEE-published HFSS-coaxial reference points (not VNA)

What it proves

Matches Eagle XG, AF32, Borosilicate, EN-A1, and Quartz HFSS-coaxial reference values within ±40-60% published-tolerance bands. measurements.json has 13 Z₀ entries (10 source_type=simulation + 1 measurement + 2 other) after Sprint 41 added 5 new papers. Real VNA campaign queued (~$200-500K wet lab).

Glass Interposer PDK

180 valid cross-validations
Test

BEM vs FastHenry2 (LGPL, built from source)

What it proves

20 geometries × 3 glasses × 3 frequencies. Z₀ 31.8–66.0 Ω, L 0.128–0.277 nH. Independent LGPL solver agrees.

Glass Interposer PDK

50/50 transient sims; Z₀ at 10 GHz (excitation freq) shows 20.7% median |Δ| vs BEM (was 55.3% at wrong 28 GHz extraction)
Test

AWS Palace FEM (Apache-2.0) at correct excitation frequency

What it proves

Root-caused extraction bug: source spectral power near zero at prior 28 GHz extraction frequency. Corrected at 10 GHz. Residual 20.7% is a factor-of-2 lumped-port convention.

Glass Interposer PDK

20 geometries × 3 refinement levels; max 0.039% |Δ| across 4× refinement
Test

BEM mesh-convergence study (standalone physics claim)

What it proves

BEM solver converges to a single answer independently of discretization — no cross-solver dependency. Standalone defensibility claim for due diligence.

Glass Interposer PDK

18 frequency-points across 5 peer-reviewed papers; mean bias −1.6%, RMSE 5.8%, max |err| 12.7% at 77 GHz
Test

IEEE validation corpus extended

What it proves

Tightens the headline 4.0% MAE figure with frequency-resolved data. Clean quasi-static BEM regime boundary at ~60 GHz.

Glass Interposer PDK

2/3 witness geometries pass at 28 / 77 GHz under the finite-conductivity model; tighter mmWave/HBM4/UCIe regimes remain outside the promotion gate.
Test

Local source-built OpenEMS high-band named matrix

What it proves

OpenEMS is a real independent full-wave witness on a bounded named set. golden_50ohm and golden_mmwave_77ghz pass; tighter mmWave/HBM4/UCIe regimes await extraction sanity.

Quality / Inspection Plane

HOLD_FOR_REVIEW returned correctly with 2 defects (PATTERN_CTR, VOID_WIRE), 2 yield patterns, 1 CAPA, 2 provenance entries
Test

Lot Intelligence Report endpoint verified live

What it proves

4-signal → 1-verdict aggregation works end-to-end against live production database.

Portfolio Coverage

9 Technology Areas

Each area combines production solvers, validated benchmarks, and independent claims filed with a January–February 2026 priority date.

Wafer Control Suite

Iterative learning control, wafer flatness, Zernike decomposition

Package Mechanics Engine

Kirchhoff-von Karman FEM, panel warpage, structural coupling

Thermal Intelligence Engine

Binary mixture thermal, Rohsenow boiling, 2D resolved thermal

RF & Photonics Suite

TMM wideband isolation, chirped LFI shields, photonic thermal

Critical Materials Engine

PFAS remediation cascades, rare-earth selectivity

Battery Physics Engine

Gyroid tortuosity, Monroe-Newman CCD, dendrite prediction

Glass Interposer PDK

BEM multiconductor impedance surrogate — R²=0.9999966 on 6.75M-row corpus, 23 commercial glasses, 4-solver cross-validation, per-prediction confidence intervals, 5/5 cross-EDA export

Isolation Synthesis Engine

Adjoint topology optimization, GDSII export, DRC integration

Yield Prediction Engine

GDS-to-yield pipeline, Bayesian calibration, KLA Archer integration

Quality / Inspection Plane

33-type IPC/SEMI/ISO defect taxonomy, 12-pattern CAPA library, WM-811K wafer-map corpus (811K real fab wafers), provenance graph, MCP tool server

For strategic partners and acquirers

Build-vs-buy infrastructure for AI-native EDA

909 filed patent claims spanning chiplet packaging, photonic ICs (ChipletOS Photonic Signoff sub-brand), multi-physics signoff, and bondability screening. AI-EDA distribution via ChipletOS-MCP (Anthropic gallery submission in flight). 39-entry retraction registry with root-cause documentation for every withdrawn claim. 17/17 validation suite gates passing at every commit.

One platform, two markets: chiplet packaging (today) and silicon photonics (build-out under the ChipletOS Photonic Signoff sub-brand). Synopsys-Lumerical-class photonic-EDA story without the silo — same surrogate stack, same audit gates, same cross-vendor exporter discipline.

For strategic discussion: nick@chipletos.com

Outside the chiplet stack

Sister Portfolio: FluxZero

Same ownership. Separately branded. FluxZero is a PFAS-free, Marangoni-active, self-pumping fluid built to replace 3M Novec in two-phase immersion cooling before EPA and EU restrictions take the incumbent off the market.

scheduleEnd of 20253M Novec shutdown
wavesMarangoni-activeSelf-pumping mechanism
device_thermostatGPU · Edge · OrbitalTwo-phase immersion stack

The two-phase immersion stack that cools GPU clusters, edge compute, and orbital electronics is about to lose its incumbent PFAS dielectric. EPA and EU restrictions are closing the door behind 3M's Novec discontinuation. FluxZero is the PFAS-free replacement: a fluid whose surface-tension gradient pumps itself across the boiling boundary without external work. Owned by ChipletOS; commercialized under its own brand.

Visit fluxzerofluid.com →
lock

Full Portfolio Under NDA

Request data room access for complete claim text, prosecution status, filing packages, evidence bundles, valuation analysis, and buyer-specific technical deep-dives.

Request NDA + Data Room Access