Hill-Climbing Chip Design with a Self-Improving Loop
Our loop beat the best frontier coding agent on all six design variants, and every design was bit-exact.
Research
ChipBench
Our benchmark for AI on chip design, across CPU, accelerator and network design.
Science in Broad Daylight
Every “timing closed” an agent claims should point to the tool report that proves it.
The Goal of Science Is Not to Win
Using cheap lint, coverage, and early timing signals to choose the next design candidate.
AI Should Build Its Own Research World Model
A timing-closure campaign that remembers which paths fail, which fixes were tried, and how the slack changed.
Agent-Native Research Artifacts
A finished block hands off its failed floorplans and rejected experiments too, ready for the next team to rerun.
Beyond Natural Language: An Agent-Native Language for Autonomous Science
Agents write each hypothesis in a formal language that a checker can verify.
Curie: Toward Rigorous and Automated Scientific Experimentation with AI Agents
Controls, reruns, and audit logs, built into every experiment an agent runs.
EXP-Bench: Can AI Conduct AI Research Experiments?
We tested whether AI agents can take a real research experiment from setup to result.