The semiconductor industry is undergoing a fundamental transformation. Advances in open-source Electronic Design Automation (EDA) toolchains, open process design kits (PDKs), and accessible fabrication through multi-project wafer shuttles and chiplet integration are lowering the barriers to custom silicon – ushering in what may be called the Open Silicon Era. At the same time, the research landscape is being reshaped by breakthroughs in artificial intelligence (AI) and machine learning (ML) and by a renewed interest in principled combinatorial optimization. Yet the three communities most relevant to this moment – EDA researchers, AI/ML researchers, and combinatorial algorithms researchers – remain largely disconnected, despite deeply complementary expertise.

This Dagstuhl Seminar is designed to bring these communities into sustained, structured dialogue. The EDA community brings deep domain knowledge and practical experience with chip design at scale, but many of its core methods remain heuristic-driven and evolve in isolation from broader algorithmic progress. The combinatorial algorithms community, spanning theoretical algorithmics, algorithm engineering, and high-performance computing, offers exactly the tools to address this gap: many core EDA problems are open algorithmic challenges with rich structure, independent of any ML consideration. The AI/ML community, meanwhile, has produced techniques including reinforcement learning, graph neural networks, and generative models that are directly applicable across the EDA stack; in turn, chip design offers both communities high-stakes, structurally rich problems with clear correctness criteria and domain-specific data that general benchmarks rarely provide.

The seminar will address the full hardware design stack, from architectural exploration and high-level synthesis (HLS) through logic and physical design to verification and test, asking at each level where combinatorial methods provide indispensable guarantees, where discriminative ML can accelerate without sacrificing correctness, and how far generative AI can go in transforming design entry itself. Large language models and diffusion models are beginning to generate RTL descriptions, suggest design optimizations, and offer conversational interfaces to non-expert users; whether and how these capabilities can be made reliable, verifiable, and practically useful across the design flow is one of the central open questions the seminar will address. A related focus is the interplay between openness and manufacturability: open toolchains and accessible fabrication programs let researchers prototype and tape out designs, turning algorithmic and AI innovations into silicon and closing the loop between theory and practice.

Open-source platforms are central to realizing this vision. Transparent toolchains and open PDKs provide shared infrastructure for data collection, algorithmic benchmarking, and AI training. Crucially, they also close the loop from research to silicon: innovations can be prototyped, validated on real design problems, and taped out through accessible fabrication programs before any transition to commercial flows. This combination of openness and manufacturability is what makes the current moment distinctive. The seminar will also engage design practitioners and early adopters of AI-based design techniques as active participants, grounding the research agenda in the realities of production environments and ensuring that the resulting roadmap, benchmark suites, and open infrastructure have a clear path to impact.

Creative Commons BY 4.0

Antonino Tumeo, Lana Josipovic, Nesreen Ahmed, and Robert Wille