2026-10-11 17:13 UTC

PCBSchemaGen’s authors claim their released 227-task benchmark and deterministic five-layer verifier can score generated PCB schematics against real-IC pin and topology constraints without an LLM or functional simulation, enabling reproducible structural evaluation of schematic-synthesis systems.

state: watchingheat: lowuncertainty: mediumconvergesscott: lowllm-code-synthesis electronic-design-automation verifiable-generationHuanghaohe ZouPCBSchemaGen authors

What is this?

PCBSchemaGen (Zou et al., arXiv 2602.00510, 2026) is a training-free, inference-time framework that uses a frozen LLM to generate PCB schematics as SKiDL/KiCad-compatible programs, grounded by a knowledge graph induced from IC datasheets and scored by a deterministic five-layer verifier (structural checks against real-IC pin and topology constraints — no golden references, no functional simulation), with candidate refinement via a Thompson Sampling bandit. The authors released two benchmark suites totaling 227 real-IC tasks across 22 circuit domains (including a held-out public-schematic-derived suite) plus the verifier itself (github.com/HZou9/PCBSchemaGen_v2). The paper reports cited-by 3 and a separate, independently developed PCB-Bench (Li et al.) exists for evaluating LLMs on PCB design, indicating a small emerging area; the HN discussion saw essentially no traction (1 point), and the snippets establish the artifacts' existence and design claims but not independent reproduction or functional (as opposed to structural) circuit correctness. A separate Show HN project, CircuitPilot (LLM plus rules-based Critic for PCB workflows), suggests verifier-gated PCB generation is being pursued by multiple teams independently.

Why it matters to Scott

Converges squarely with Scott's mechanically-different-verifiers and validation-gated LLM extraction canon — a deterministic, datasheet-grounded five-layer checker used as reward and repair gate for LLM code synthesis is exactly the pattern his wikis argue for — but the CircuitPilot corroboration only adds a second instance of a pattern he already holds, and the load-bearing open question (does structural verification predict functional circuit correctness?) mirrors the ModelRift silent-geometry finding without resolving it. PCB/EDA is not a territory he builds in, nothing here challenges or extends a load-bearing claim, and no independent adoption of the benchmark has materialized: the world agreeing with him again in an adjacent domain is not news for him.
ip:concept.mechanically-different-verifiersdev:concept.validation-gated-llm-extractionradar:modelrift-cad-agent-silent-geometryradar:contract-verifier-llm-gpu-kernelsradar:flare-milp-lean-verificationradar:nomoreda-agent-facing-eda
queries asked of Scott's wikis
  • mechanically different verifiers deterministic reward verifier LLM synthesis
  • validation-gated extraction structural vs functional correctness gap
  • LLM code synthesis benchmark verifiable reward no golden reference
  • domain-specific constraint checker as reward signal for code generation
  • training-free inference-time framework bandit candidate refinement repair loop
  • EDA / electronics domain LLM tooling prior work or projects

Measured heat

now 0 pts/hpeak 0 pts/hcomments 0/hpeers p14momentum: steady2 platformsage 605h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion

How the heat travelled

09-16 11:22 (minted)⭐ origin echo-reconstructedReleases two benchmark suites totaling 227 PCB schematic-synthesis tasks and a deterministic five-layer structural verifier for candidate SK
HZou9 / PCBSchemaGen authors on github (echo) · attributed from hn.story.49724579 · published time unknown
—
09-16 10:48first on hacker news · published · lag ?PCBSchemaGen: Reward-Guided LLM Code Synthesis for PCB Schematic Design
teleforce
—
09-16 10:48amplified on hacker newshn.story.49724579
teleforce
peak 1 · 0 comments · 26% of case engagement
09-22 14:55amplified on hacker news 👑hn.story.49802378
Dota-12
peak 3 · 0 comments · 76% of case engagement
09-16 11:20our radar first saw it · lag ?discovery anchor: hn.story.49724579—
pace: p36 vs 1032 stories at the 336h mark (now 605h old) — ahead of agentgate-signed-agent-receipts (1.3x), behind agent-memory-add-search-evaluation (0.8x)

Evidence (3) — ⭐ canonical anchor

sourceobjectauthorscorecomments
🟧 hnPCBSchemaGen: Reward-Guided LLM Code Synthesis for PCB Schematic Design
Retrieved article excerpt

Open article · Retrieved 2026-09-16T11:22:14.886320+00:00

# PCBSchemaGen — Benchmarks & Verifier

This repository provides the **two benchmark suites** and the **deterministic 5-layer verifier** from the paper
*PCBSchemaGen: Reward-Guided LLM Code Synthesis for Printed Circuit Board (PCB) Schematic Design with Structured Verification*.

It lets anyone (i) load the **227 PCB schematic-synthesis tasks**, and (ii) run the **deterministic structural verifier**
that scores a candidate SKiDL design against real-IC pin- and topology-level constraints — no LLM, no API key,
and no cached results required.

## What's included

```
PCBSchemaGen_v2/
├── benchmarks/
│   ├── pcbbench/benchmark.tsv                   # 62 hand-authored tasks (Easy/Medium/Hard)
│   └── open_schematics/ose_165_task_specs.json  # 165 tasks from public schematics
├── framework/
│   ├── topo/                                    # the deterministic 5-layer verifier (self-contained)
│   └── task_config.yaml                         # per-task verification parameters
├── kg/kg_open_schematics.json                   # schema-induced KG for Open-Schematics-Eval
├── component.json / kg_component.json / kg.json # schema-induced KG for PCBBench
├── requirements.txt
└── LICENSE
```

## Benchmarks at a glance

| Suite | Tasks | Difficulty split | Real ICs | Domains |
| --- | --- | --- | --- | --- |
| PCBBench (`benchmarks/pcbbench/benchmark.tsv`) | 62 | 17 Easy / 28 Medium / 17 Hard | 41 commercial | 22 |
| Open-Schematics-Eval (`benchmarks/open_schematics/ose_165_task_specs.json`) | 165 | 40 Easy / 48 Medium / 77 Hard | 439 commercial | 22 |
| **Total** | **227** | 57 / 76 / 94 | 480 | 22 |

Each PCBBench task row specifies: task id, difficulty level, circuit type, the natural-language task,
input/output nodes, input/output voltages, and the required-component set.

## The deterministic 5-layer verifier

The verifier scores a candidate schematic against structural constraints derived from real IC datasheets —
fully deterministic, with no golden reference and no functional simulation:

1. **L1 ERC** — electrical invariants (VDD/GND disjointness, power reachability, ground integrity)
2. **L1b Role** — pin-role compatibility under the 32-role schema
3. **L2 Template** — per-IC subcategory connection templates
4. **L3 Topology** — subgraph isomorphism vs. canonical motifs (half-bridge, sync-buck, three-phase inverter, …)
5. **L4 Power** — domain-specific power rules (Kelvin source, decoupling, isolation, gate-resistor)

See the paper appendix (verification details) for the full predicate set and reward ladder.

## Install

```
python3 -m pip install -r requirements.txt   # networkx, pyyaml
```

## Using the verifier

The verifier is self-contained in `framework/topo/` (only depends on `networkx`). It exposes, among others:

```
from framework.topo import (
    KGStore, index_snapshot,
    check_system_topology, is_complex_task,
    validate_complex_task, get_validation_feedback_for_llm,
)

# 1. Load the schema-induced KG (component.json + kg_component.json)
kg = KGStore(base_dir=".")

# 2. Index a candidate design snapshot (nets / parts extracted from a SKiDL build)
snapshot = index_snapshot(...)

# 3. Run the layered checks
#    - system-level validation for Hard / complex tasks:
report = validate_complex_task(...)
#    - or per-layer checks: check_system_topology(...), run_phase2_checks(...)
```

Per-task verification parameters (match mode, enabled checks, skipped rules) live in
`framework/task_config.yaml`. Exact function signatures are documented in each module's
docstring under `framework/topo/` and in the paper appendix.

## License

Released under the MIT License (see `LICENSE`). Open-Schematics-Eval tasks are derived from the
public `open-schematics` dataset (MIT). Vendor datasheets referenced during KG construction are
cited, not redistributed.

## Citation

```
@inproceedings{pcbschemagen2026,
  title     = {PCBSchemaGen: Reward-Guided LLM Code Synthesis for Printed Circuit Board (PCB)
               Schematic Design with Structured Verification},
  author    = {Zou, Huanghaohe and Han, Peng and Nazerian, Emad and
               Zhang, Mafu and Guo, Zhicheng and Huang, Alex Q.},
  year      = {2026},
}
```
teleforce10
🟧 echo.github ⭐Releases two benchmark suites totaling 227 PCB schematic-synthesis tasks and a deterministic five-layer structural verifier for candidate SKHZou9 / PCBSchemaGen authors——
🟧 hnShow HN: CircuitPilot – We built a physics engine to auto-route PCBs using LLMsDota-1230

Interpretation history

Decision trace