Yifan Sun wants to help microchip designers better understand how computer chips work and why they fail. His proposal to study what he calls a “human-centered” approach to chip performance has now been recognized by the National Science Foundation with a prestigious CAREER award.

“My CAREER proposal advances a new paradigm in computer architecture by integrating human-centered methods and generative AI into simulation and performance analysis,” said Sun, an assistant professor in William & Mary’s department of computer science. “As chips grow more complex, traditional tools fall short in helping architects understand, diagnose and improve designs.” 

Sun is the latest – and 19th – William & Mary computer scientist to receive a CAREER award. The award recognizes early-career faculty leadership in both teaching and research. The value of the award is close to $600,000 in grant funding.

Sun’s research focuses on chip performance by developing a method for figuring out why chips fail by improving the visualization of hardware behavior. Traditional computer architecture research mainly helps researcher evaluate factors like how fast a program can finish, if a certain type of attack can break the system, or how much energy it consumes.

“The goal of human-centered method is to care about how the user interprets the changes,” Sun said. “I mainly care about humans as chip designers. I especially care about if a chip designer can interpret the performance changes of a chip and understand the complex internal mechanisms of the chip.”

Sun says every facet of modern technology now relies on chips to do more and more heavy lifting. “The more we ask chips to do, the more complex they become, and the more vulnerable they are to slight imperfections or bugs,” he said. “We want higher performance, we want chips to calculate faster and faster, but there are problems – security issues and bugs – caused by imperfections in design. The chip is getting so complex that we need to help designers understand how to make better chips.”

Current standards that assess chips, Sun said, prioritize technical metrics like accuracy and speed. He wants to make those tools human-centered, intuitive and enhanced by the use of AI. His research builds intuitive visualizations, collaborative workflows and AI-powered insight generation into an assessment model that will help chip designers reduce costly design flaws.

Sun said improving the performance of chips could lead to breakthroughs in myriad fields, since so many aspects of modern life now rely on computer technology. 

“Say you need a CT scan. There’s actually a lot of computing power necessary to read medical imagery and what is providing the power is the chips,” Sun said. “Perhaps a chip can now see down to one millimeter of an image. If you had a better chip maybe you could see the image with finer granularity. If you could see things more clearly, perhaps you could diagnose a problem earlier.”    

Helping designers understand chip problems will result in better chips and less failure, Sun said. “Some bugs will cause a chip to generate the wrong result. It can be very expensive to fix those problems.”

The grant funding will be used to support doctoral candidates who will assist with the research. Sun has five full-time doctoral students and one part-time student in his lab.

Sun expects the project to take up to five years to complete although work has already begun.