AI That Reads Chip Rules and Writes Code for Designers

Designing a computer chip is like building a miniature city with millions of rules: wires can't be too close, layers have to line up, and materials have specific limits. Engineers typically read those rules and manually rewrite them into code that design software can understand. That step is slow, expensive, and prone to human error, especially as chips get more complex.

AutoDRI, a new system from researchers, automates this translation using multiple AI agents that work together like a team. One agent reads the rule, another converts it into a mathematical format, and a third double-checks the result. It builds a library of known shapes and uses a clever model for tricky multi-patterning rules, where chips are made in multiple passes. The system turns plain-English rules into instructions for a CP-SAT solver, a kind of logical problem-solving engine.

In experiments on 41 chip cells with over 10 complex rules, AutoDRI achieved near-perfect accuracy: 33 out of 33 correct integrations using one advanced AI, and 32 out of 33 with another. Crucially, the generated code passed industrial-quality verification checks called DRC and LVS, meaning it's not just theoretically correct — it works in practice. The system also runs just as fast as manually written code, so it doesn't sacrifice speed.

Why should you care? Chips are the brains inside everything from phones to medical devices. Speeding up chip design could lower costs, shorten product development cycles, and let smaller companies innovate without a huge team of specialists. The catch: real chips have thousands of rules, and AI can still make mistakes. So this is a promising leap forward, not yet a replacement for humans.