I open-sourced an ML model that predicts chip design outcomes — 92% accuracy on real OpenLane runs

[n57d30top]

Hey everyone,

If you've ever designed a chip with OpenLane, you know this loop:

Draw a layout → Run OpenLane (30 min) → It fails → Repeat

Most runs fail. And you only find out after waiting 30 minutes. Worse — the knowledge from that failure disappears. Nobody else learns from it.

I built Open Silicon Triage — a tool that adds a 2-second ML check. It looks at your run metrics and tells you reject (will fail), review (uncertain), or run (go for it). Trained on 76 real OpenLane runs, 92% accuracy, 100% on held-out test data.

🛠️ How to use it for your own designs

1. Train a model (takes seconds):

python cli/train.py --features corpus/seed-corpus.json --out-model my-model.joblib --out-summary results.json
2. Check your design after an OpenLane run:

python cli/ingest.py --openlane-dir --variant "my-design" --out my-run.json
python cli/predict.py --model my-model.joblib --features my-run.json --out-summary verdict.json
Replace with the path to your OpenLane output (the folder containing reports/ and logs/), and "my-design" with whatever you want to call your design.

The output tells you whether your metrics look like past successes or past failures — before you commit to another 30-minute run.

🤝 The community part
Here's the real idea: every run you share makes the predictor smarter for everyone. Think Waze, but for chip design.

After your OpenLane run, one command submits your data:

python cli/submit.py --variant "my-design"
The script automatically extracts the metrics and strips all local paths — your design files stay completely private. Only anonymous performance numbers are shared.

At 76 records it already works. At 1,000 it'll be very accurate. At 10,000 it could become the standard pre-check for the whole OpenLane community. Failed runs are just as useful as successes — they teach the model what to avoid.

GitHub: https://github.com/n57d30top/open-silicon-triage

Even one run from you helps everyone. 🙏