System Architecture
The architecture behind the Glass Package Signoff suite: Glass PDK RF, EM Isolation Compiler, Bondability Pipeline, route-backed bundles, and generated diligence artifacts from the same stack.
Monolithic Core,
Modular Execution.
ChipletOS is organized around one flagship path rather than a loose collection of modules. The architecture turns package requirements into RF signoff, isolation signoff, bondability screening, Touchstone content, report markdown, bundle hashes, and provenance from the same typed API surface.
- hub
Domain Modules
Versioned corpora, buyer-regime releases, S2P registry assets, and schema-complete model lanes.
- terminal
Unified Core
Shared signoff logic, bundle generation, OOD handling, provenance, and decision-summary assembly.
- router
FastAPI Gateway
Route-backed buyer workflows for TGV signoff, package signoff, registry retrieval, and calibration updates.
The photonics/ sibling tree
The Photonic Signoff sub-brand extends the Genesis monorepo as photonics/ sibling to glass_pdk/. Same validation suite, same retraction registry, same buyer-DD harness — applied to silicon-photonic IC signoff. Six primitives, 4 solvers, 3 optimizers, 5 exporters, AIM-Photonics-class DRC, published-paper validation.
- account_tree
4 Solvers
Full-wave reference solver (process-isolated truth), cross-physics solver, mode solver, closed-form analytical hot path.
- widgets
6 Primitives
Waveguide · MZI · MMI · ring resonator · grating coupler · photonic crystal. Per-primitive trained AI surrogate.
- tune
3 Optimizers + 5 Exporters + DRC + Validation
Inverse design (target → geometry) + Pareto (multi-objective) + closed-loop synthesis. Lumerical .lsf / Photon Design Omnisim XML / ANSYS Lumerical-FDTD INI / KLA Kandela XML / GDSII. AIM-Photonics-class DRC + published-paper cross-check.
Genesis/
├── glass_pdk/ # chiplet (live, 6.75M-row corpus)
│ ├── optimization/
│ ├── exporters/ # 5 EDA vendors
│ ├── drc/ # IPC-A-610G / SEMI E10
│ └── validation/
└── photonics/ # SUB-BRAND v1
├── solvers/
│ ├── reference_runner.py # full-wave reference (process-isolated)
│ ├── cross_physics.py # cross-physics solver
│ ├── mode_solver.py # mode solver
│ └── analytical.py # closed-form hot path (~0.04 ms)
├── primitives/
│ ├── waveguide.py # closed-form analytical model
│ ├── mzi.py
│ ├── mmi.py
│ ├── ring.py
│ ├── grating.py
│ └── photonic_crystal.py
├── optimization/
│ ├── inverse_design.py
│ ├── pareto.py
│ └── closed_loop.py
├── exporters/
│ ├── lumerical_lsf.py # .lsf / .ldf
│ ├── omnisim_xml.py
│ ├── ansys_lumerical.py
│ ├── kla_kandela.py
│ └── gdsii.py
├── drc/ # AIM-Photonics-class
├── validation/ # published-paper cross-check
└── ai/inference/
└── waveguide_surrogate.py # AI surrogate (analytical fallback today)API Gateway & Router Mapping
Every domain is mapped via lazy-loaded routers to ensure minimal cold-start latency and strict physical boundary isolation.
| Buyer-Facing Category | Flagship Surface | Key Routes | Status |
|---|---|---|---|
| Glass RF Signoff | schema-complete inference + signoff | `/v1/glass-pdk/predict-impedance-schema-complete`, `/v1/glass-pdk/tgv-signoff` | Stable |
| Inverse Design | target Z₀ → recovered (d, p, t); surrogate + adjoint-BEM | `/v1/glass-pdk/geometry-from-target` (?refine=adjoint optional) | Live |
| Package Signoff | integrated flagship route | `/v1/chiplet-suite/package-signoff` → decision summary, report, manifest, provenance | Stable |
| Isolation + Bondability | buyer-facing signoff layer | `/v1/isocompiler/isolation-signoff`, `/v1/bondability/bondability-signoff` | Stable |
| Registry + Evidence | library + diligence surface | `/v1/library/s2p`, claim trace, canon facts, one-command evidence pack | Stable |
Solver Hierarchy & Neural Acceleration
32+ production solvers orchestrated through a unified dispatch layer. Top solvers ranked by composite maturity score.
BEM TGV Impedance
Boundary Element Method for glass through-via impedance extraction. Validated against 6 IEEE publications.
TMM RF Isolation
Transfer Matrix Method for RF isolation verification with machine-epsilon energy conservation.
ILC Zernike Controller
Iterative Learning Controller for wafer-level lithography correction using Zernike decomposition.
IsoCompiler Adjoint
Adjoint topology optimizer for EM isolation synthesis. 0/20 alternative approaches beat the closed-loop pipeline.
LBM Solver
Lattice Boltzmann Method for high-fidelity fluid and thermal simulations within chip packaging.
PINN Layers
Physics-Informed Neural Networks providing millisecond-level surrogate approximations.
FDTD Engine
Finite-Difference Time-Domain solvers for sub-nanometer signal integrity analysis.
Cross-Domain Orchestration
Multi-physics coupling is managed via our Dispatch Layer. We handle “Cross-Pollination Inventions” where thermal data directly feeds electromagnetics solvers in a single unified pipeline.
Yield-aware isolation ensures that solver failures in one domain do not contaminate the manufacturing dataset of another.
SDF Engine & EDA Export
The core geometry engine generates Signed Distance Functions (SDF) for pixel-perfect physical modeling. This translates seamlessly into industry-standard GDSII exports, verified for structural validity in tools like KLayout and HFSS.
Enterprise Deployment Architecture
Stateless Scaling
API targets are purely stateless, allowing for infinite horizontal scaling on Kubernetes clusters.
IP Boundaries
Zero-trust networking ensures that physical process data remains within domain-isolated namespaces.
Docker Targets
Optimized container images for both standard x86 compute and dedicated GPU-accelerated nodes.
Telemetric Visibility
Real-time observability across the entire solver stack. We treat simulation latency as a first-class citizen.
{"status": "processing", "request_id": "8f3a-921c", "node": "gpu-04"}
{"solver": "LBM-Thermal", "iteration": 450, "residual": 1.2e-7}
{"memory_usage": "14.2GB", "latency": "142ms", "coupling": true}
{"event": "EXPORT_TRIGGERED", "target": "GDSII_VERIFIED"}
_