Seminar: ACRC Webinar
From Assistants to Autonomous Agents: The Evolution of AI in Chip Design
From Assistants to Autonomous Agents: The Evolution of AI in Chip Design
Abstract:
In 2017, DeepMind’s AlphaZero learned to play chess, Go, and shogi from scratch with no human game data, no hand-crafted heuristics, just the rules and three days of self-play. In Go, it beat AlphaGo, the system that had just defeated the world champion, 100 games to zero, discovering strategies that 3,000 years of human play had never produced. That moment was not just a milestone in game-playing AI. It was a preview of what is coming in every complex engineering domain.
This talk traces that same three-phase arc inside electronic design automation. Phase one is AI as optimizer: reinforcement-learning systems like DSO.ai and VSO.ai that search the implementation space more efficiently than human teams, delivering 25% power reduction and 4.5% frequency improvement on production designs. Phase two is AI as assistant and co-creator: the Synopsys.ai Copilot, which answers tool questions in natural language, generates Tcl and Python scripts on demand, creates RTL and testbenches from design specifications, and produces SVA assertions for formal verification. Phase three, where we are today, is AI as autonomous agent: systems that accept a goal (“autonomously improve this RTL for power and performance”), then plan, orchestrate, and execute multi-step EDA workflows with minimal human intervention.
Important: Registration is required.
Zoom link will be sent after registration.