Jun 5 (Wed) @ 11:00am: "Realizing Practical LLM-assisted AI Assistant in the Semiconductor Domain,” Yueling Jenny Zeng, ECE PhD Defense

Abstract

The emergence of Large Language Models (LLMs) offers new opportunities for applying Machine Learning (ML) and Artificial Intelligence (AI) in semiconductor chip design and test (D&T). Realizing these opportunities requires a fundamentally different thinking from the past. For more than two decades, the semiconductor industry has been exploring applications of ML in D&T. Despite many promises, notable challenges remain in many contexts. In the first part of the talk, I will review key observations from selected past works and highlight the challenges faced in those works. From there I will elaborate a view to differentiate ML from applying ML in D&T, where the latter is called Decision-Support ML (DSML). This DSML view implies that in order to implement an AI assistant for enabling human decision making in a D&T application, one needs to answer two essential questions: What is domain knowledge? And how domain knowledge is used? Following these two questions, I will present the design principles of Intelligent Engineering Assistant (IEA) and show its first practical realization, the IEA-Plot, in an industrial environment. Through a tangible product like IEA-Plot, I will explain why LLM plays an essential role in IEA and how the two fundamental questions can be answered. In the last part of the talk, I will introduce a new approach called Oracle-Checker (OC) scheme, inspired by the theoretical ideas of Interactive Proofs, and designed to enable effective utilization of a generative LLM like GPT. I will conclude this talk by pointing out future directions from our IEA design and the OC scheme.

Bio

Yueling Jenny Zeng is a PhD candidate in Computer Engineering at Electrical and Computer Engineering department, University of California, Santa Barbara (UCSB). She received B.S. and M.S. both in Electrical and Computer Engineering from the same department in 2018 and 2020, respectively. Zeng received the Best Paper Award from the IEEE International Test Conference (ITC), twice, in 2020 and in 2022, where she was the presenter for both works. ITC is the oldest, largest and premier conference in the field of electronic test. Zeng is the recipient of the 2023 IEEE/TTTC G.W. Gordon Student Service Award. This award is designed to honor students for credible service to IEEE Test Technology related activities, programs and organizations. During her graduate study at UCSB, she was honored as an Outstanding Teaching Assistant three times. Zeng worked for NXP Semiconductor in Austin Texas for two summers as a data scientist intern. Then, she worked for LinkedIn Corporation, Sunnyvale, CA for two summers as a ML research intern. Her research interests broadly include various aspects of building an ML application, to serve the needs of a semiconductor chip design company or an Internet service company. In recent years, her interests focus on LLMs and how to utilize LLMs to build a practical AI system.

Hosted by: Professor Li-C. Wang

Submitted by: Yueling Zeng <yuelingzeng@ucsb.edu>