Rule2DRC: Benchmarking LLM Agents for DRC Script Synthesis with Execution-Guided Test Generation

Researchers introduced Rule2DRC, a benchmark for LLM-based DRC script synthesis from natural-language chip design rules.

The benchmark includes 1,000 rule-to-script tasks and 13,921 evaluation layouts, making evaluation much more realistic and scalable.

They also released SplitTester, which uses execution feedback to generate discriminative tests and better choose among candidate scripts.

The work tackles a major bottleneck in semiconductor layout automation by measuring whether generated scripts actually work, not just whether they resemble reference code.