The semiconductor industry has hit a wall: human intelligence can no longer keep pace with the complexity of silicon. For decades, translating thousands of Design Rule Check (DRC) instructions from text manuals into machine code has been grueling manual labor. Top-tier engineers spent years laboring over SVRF or KLayout scripts, only to have to start from scratch every time a foundry moved to a new process node. Rule2DRC, a project presented by Professor Sung-Hyun Oh’s team at Seoul National University in collaboration with the Samsung AI Center, aims to clear this exact bottleneck.
From Code Comparison to Execution-Based Validation
Previous attempts to automate DRC failed because they relied on superficial metrics like code similarity. Rule2DRC shifts the paradigm: it is not just a text generator, but an Action Model. The system doesn't guess whether the code "looks right"; it executes it. The solution is built on a massive benchmark of 1,000 instruction pairs and 13,921 chip layouts. According to materials prepared for ICML 2026, Rule2DRC runs generated scripts through actual verification engines to test functional viability rather than literary style.
Rule2DRC goes beyond simple code similarity assessment by directly executing generated algorithms within a verification environment to confirm their functionality.
This execution-guided approach closes a critical vulnerability—the risk of translation error. As detailed in the arXiv preprint, the goal is to create a reliable bridge between human language and hardware architecture. This is no longer about playing with chatbots; it is an industrial-grade circuit where AI understands the logic of verification scripts as deeply as their syntax.
Integration into the Fabrication Workflow
Professor Oh’s team worked in close coordination with the Samsung AI Center to ensure the agent could survive in a real-world factory environment. The resulting application interprets natural language and immediately outputs production-ready code. For Samsung, this is a matter of economics: automating the routine allows for a radical reduction in Time-to-Market and, more importantly, reduces dependence on the chronic shortage of verification engineers—a talent pool that is always smaller than the ambitions of chipmakers.
In the modern arms race, design verification speed is becoming a decisive competitive advantage. The Rule2DRC case is a significant signal to the market: AI is moving from generating essays to managing physical architecture. We are looking at a future where transitions to new process nodes will be limited by the laws of physics and power consumption, rather than how many engineers managed to rewrite the documentation. However, one question remains: how does Samsung plan to verify the "verifier"? When script complexity exceeds the capacity for human audit, hunting for logical hallucinations in code will become a billion-dollar task with zero margin for error.