Back to AI Research

AI Research

EvoDRC: A Self-Evolving Agentic Framework for Autom... | AI Research

Key Takeaways

  • EvoDRC is an automated framework designed to solve a major bottleneck in semiconductor manufacturing: Design Rule Check (DRC) closure.
  • Design rule check (DRC) closure remains a major bottleneck in advanced-node physical design.
  • Although detailed routers are rule-aware, residual design rule violations (DRVs) often require manual engineering change order iterations.
  • Automating this process is challenging because repairs must account for complex geometric interactions, preserve circuit connectivity, and avoid introducing new violations.
  • We present EvoDRC, a skill-evolution framework for agentic block-level DRC repair.
Paper AbstractExpand

Design rule check (DRC) closure remains a major bottleneck in advanced-node physical design. Although detailed routers are rule-aware, residual design rule violations (DRVs) often require manual engineering change order iterations. Automating this process is challenging because repairs must account for complex geometric interactions, preserve circuit connectivity, and avoid introducing new violations. We present EvoDRC, a skill-evolution framework for agentic block-level DRC repair. EvoDRC initializes layer-specific repair skills using knowledge distilled from an unrelated reference design and continuously evolves these skills using traceable repair experience collected from the target design. EvoDRC decomposes the layout into bounded repair regions and assigns an LLM repair agent to each region. Local DRC analysis, connectivity-checking, and impact-preview tools provide feedback on proposed modifications. Repair operations and their resulting DRV changes are stored in a knowledge database and used to evolve the repair skills. Experiments on seven block-level designs from the DAC26 DRC Benchmark show that EvoDRC achieves a 73.5\% overall reduction compared to the reported baseline.

EvoDRC is an automated framework designed to solve a major bottleneck in semiconductor manufacturing: Design Rule Check (DRC) closure. In advanced chip design, even after automated routing, layouts often contain residual design rule violations (DRVs)—geometric errors that must be fixed before a chip can be produced. Traditionally, this requires engineers to manually inspect and repair these errors, a time-consuming and iterative process. EvoDRC automates this by using a team of AI agents that learn from experience, allowing them to repair complex block-level layouts without human intervention.

How EvoDRC Works

The framework operates through a three-stage process. First, it decomposes a large, complex chip layout into smaller, manageable "repair crops." This allows the system to focus on specific problem areas rather than trying to process the entire design at once. Second, it assigns an LLM-based repair agent to each crop. These agents are equipped with specialized tools that allow them to preview the impact of their changes, check for connectivity issues, and run local DRC analysis to ensure their fixes don't create new problems. Finally, the system records every successful repair in a "Knowledge Database," which is then used to evolve and improve the agents' skills for future iterations.

The Power of Skill Evolution

A unique aspect of EvoDRC is its ability to "learn" as it works. It initializes its repair capabilities by distilling knowledge from a reference design and then continuously refines those skills based on the specific challenges found in the target design. By storing both the repair operations and their outcomes, the framework creates a traceable history of what works and what does not. Two specialized agents—a "Skill Refiner" and a "DB Summarizer"—constantly update the system's repair knowledge, ensuring that the agents become more effective at resolving violations over time.

Scalability and Performance

To ensure the system remains reliable, EvoDRC enforces strict rules for its agents. For example, it categorizes objects in the layout based on their editability, preventing agents from making changes that could break the circuit's logic or damage standard cells. It also uses a "Skill Judge" to verify that any new repair strategies are backed by actual evidence from previous successful fixes. When tested on seven block-level designs from the DAC26 DRC Benchmark, which included up to 765 violations per block, EvoDRC achieved a 73.5% reduction in total DRVs compared to the reported baseline. This demonstrates that the framework can effectively scale to handle the complexity of modern, advanced-node physical design.

Comments (0)

No comments yet

Be the first to share your thoughts!