Unlocking the Future: Designing 3D AI Chips with AI Algorithms and Limited Data

SystemX Affiliates: login to view related content.

Unlocking the Future: Designing 3D AI Chips with AI Algorithms and Limited Data
Thursday, May 9, 2024 - 4:30pm to 5:30pm
Shriram 104
Sung Kyu Lim - DARPA & Georgia Institute of Technology
Abstract / Description: 

*To receive email announcements and live stream information for upcoming seminars, please subscribe to the SystemX Seminar/EE310 Mailing list here.

Data, chips, and algorithms form the essential trio at the heart of the AI revolution. As technology evolves and market dynamics shift, the demand for advanced AI chips, especially for model training, underscores the critical importance of hardware in AI progress. Training AI models demands hardware capable of handling complex tasks, much like coordinating a bustling city, where GPUs and high-bandwidth memory chips play crucial roles. However, efficiently moving data within this "city" remains a challenge, often consuming more resources than computation itself.

3D Integrated Circuit (IC) architecture presents a solution by stacking chips to provide massive direct connections, thereby improving energy efficiency and reducing latency. Despite the promise shown by advances in foundry technology and domain-specific architecture innovations, progress is restrained by the limitations of current Electronic Design Automation (EDA) tools.

In this context, my research group at Georgia Tech has developed integrated solutions over the past two decades, harnessing both AI and non-AI algorithms. Our aim is to forge next-generation 3D AI chips that set new benchmarks for energy efficiency and data bandwidth, driving forward the capabilities of AI hardware.


Dr. Sung Kyu Lim has been a Program Manager at DARPA's Microsystems Technology Office since August 2022, focusing on electronic design automation (EDA) programs. Concurrently, he has been serving as Professor at Georgia Tech's School of Electrical and Computer Engineering since 2001. Dr. Lim earned his degrees from UCLA's Computer Science Department between 1994 and 2000. His research centers on architecture, design, and EDA for 2.5D and 3D integrated circuits, resulting in over 400 papers. He has received multiple awards, including Best Paper Awards from IEEE Transactions on CAD in 2022 and ACM Design Automation Conference in 2023, the NSF CAREER Award in 2006, and was featured in the Communications of the ACM in 2014. Dr. Lim is an IEEE Fellow, recognized for his contributions to EDA and 3D integrated circuits. A sample of his talk presented at the DARPA Electronics Resurgence Initiative 2.0 Summit 2023 is available at: https://www.youtube.com/watch?v=VpvkWUju-mk&list=PL6wMum5UsYvaychdWCfbmaGSqQ7YqQE4d&index=18&t=199s