ML for Hardware: A New Frontier in Design Automation
Arash Ardakani - UC Berkeley
Feb. 7, 2025, 2:30 p.m. - Feb. 7, 2025, 3:30 p.m.
MAASS 217
Hosted by: Hsiu-Chin Lin
In the field of electronic design automation (EDA), integrating machine learning (ML) techniques has the potential to transform traditional design workflows, especially in verification. Conventional EDA methods, primarily based on heuristic optimization, often struggle to scale with the increasing complexity of modern digital circuits. By introducing ML-driven methodologies, this presentation focuses on enhancing the verification process through advanced sampling techniques for circuit satisfiability (CircuitSAT), a computationally challenging task central to digital circuit testing and verification. Our approach redefines the CircuitSAT problem by applying a differentiable, gradient-based sampler that models the task as a multi-output regression problem. This reframing enables efficient and diverse solution generation directly from the circuit structure, bypassing traditional CNF representations. Through parallelized, element-wise operations executed on GPUs, the proposed method achieves substantial runtime improvements over heuristic-based samplers, delivering orders-of-magnitude acceleration. Extensive testing across benchmark circuits confirms the effectiveness of this ML-driven sampling method, marking a significant advancement in the verification capabilities of EDA tools and laying the groundwork for scalable, reliable hardware design automation.
Dr. Arash Ardakani (he/him/his) is a postdoctoral researcher in the Department of Electrical Engineering and Computer Sciences (EECS) at the University of California, Berkeley, where he is mentored by Professor John Wawrzynek. He is also affiliated with the Berkeley Institute for Data Science (BIDS) and Berkeley Wireless Research Center (BWRC). Before joining UC Berkeley, he served as a senior researcher at Huawei Technologies' Noah’s Ark Research Lab in Montreal, Canada. Dr. Ardakani earned his Ph.D. in Electrical and Computer Engineering from McGill University and his M.Sc. in Electrical Engineering from Sharif University of Technology. His research spans ML-driven electronic design automation (EDA), hardware accelerator design for ML models, scientific computing methods for efficient neural computation, and stochastic computing. Dr. Ardakani has authored or co-authored over 30 peer-reviewed articles published in leading conferences and journals. His work combines theoretical insights with practical applications to advance intelligent and efficient computing systems.
