Qualcomm awards grad students for innovative framework to compiler design

Oct. 3, 2022

Two Princeton ECE graduate students — Yi Li and Yu Zeng — have won a 2022 Qualcomm Innovation Fellowship recognizing their work to increase the efficiency and performance of modern computing systems.

The fellowship program recognizes innovative Ph.D. students across a broad range of technical research areas. Li and Zeng were one of 19 teams selected. Their one-year fellowship includes $50,000 each in financial support as well as mentorship from Qualcomm engineers. Since the program’s inception in 2009, Qualcomm has awarded more than $15 million in funding.

Both Li and Zeng are fifth-year Ph.D. students working jointly with Sharad Malik, the George Van Ness Lothrop Professor in Engineering, and Aarti Gupta, professor of computer science. Their project focuses on a compiler framework for the systems that power today’s most advanced AI and augmented reality applications. In these domains, traditional chips don’t meet computational demands, and traditional programming languages require too much overhead. To speed things up, engineers have turned to specialized hardware accelerators and high-level domain-specific languages (such as TensorFlow and PyTorch). While both approaches have proven highly effective, the growing mix of systems has created new inefficiencies — and demands great engineering effort — in the compiler layer of the stack, which allows the hardware and software to work together.

Li and Zeng, with their advisers and other collaborators, have identified an end-to-end approach that automates and standardizes this layer to ease production headaches and improve system performance. The Qualcomm fellowship will fund their research and, through its mentorship aspect, allow them to develop this framework in the context of real-world engineering problems.

Li earned his bachelor's degree from Peking University. Zeng earned a bachelor's degree from Fudan University and a master's degree from the University of Michigan. Their winning proposal, “Bridging the Gap between Domain-Specific Languages and Hardware Accelerators with Unified Software/Hardware Abstractions,” was selected from more than 100 applicants from top North American programs.

Sitewide Category