Xiong recognized for contributions to semiconductor manufacturing process

By Peter Murphy

Published February 27, 2024

Jinjun Xiong, SUNY Empire Innovation Professor in the Department of Computer Science and Engineering, has been elevated to fellow in the Institute of Electrical and Electronics Engineers (IEEE). The organization cited his contributions to process variation modeling, circuit yield optimization and their applications in industry. 

“I model the data while dealing with a lot of uncertainty. The essence of my work is about how to optimize under the uncertainties. This has been my body of expertise, and it can be translated to AI research as well. ”
Jinjun Xiong, SUNY Empire Innovation Professor
Department of Computer Science and Engineering

IEEE is the world’s largest technical professional organization. It has over 427,000 members across 190 countries who are engineers, scientists and allied professionals. The organization produces over 30% of the world's literature in electrical, electronics and computer engineering.

According to Xiong, the contributions he was recognized for are the culmination of work he has been conducting for over a decade to create the software and tools to help designers build better semiconductor chips.

A typical semiconductor chip contains billions of transistors. When transistors shrink in size, more transistors can be packed into the chip while requiring less power to function, allowing the chips in mobile phones, medical devices, automobiles and other devices to have enhanced performance over previous versions. At least, this is what should happen, says Xiong. Technology scaling, or Moore’s law, could also cause deviations in the manufacture of semiconductor chips, leading to varying performance and power.

Process variation, in this instance, makes semiconductor chip designers’ jobs to produce optimal chips for different devices arduous. To tackle these challenges, Xiong and his team model, optimize and test chips.

“Every two or three years, technology will shrink the size of transistors—on and off switches in almost all of our chips—which means the performance of the technology will double. You can see the performance getting better and better,” Xiong says. “But the resulting process variation and its impact on chip design and testing will become more difficult to handle. A lot of my work is about how to address process variation impacts,” Xiong says.

Jinjun Xiong.

Transistors equal in size perform differently and generate varied levels of power when they are placed at different locations throughout the chip. Transistors placed close together are more likely to have similar performance compared to those that are placed farther apart. This phenomenon is called spatial variation and is one of the factors Xiong and his team model.

“An important factor to consider is spatial correlation. It’s the same chip, but you have transistors at different locations. Think of it like building a house and that all your beams vary in their length or strength, and you put one sized beam on the second floor and a different sized beam on the third floor. The house’s stability will be unpredictable if not modeled correctly,” Xiong says.

Xiong began this work conceptually as a PhD student at the University of California, Los Angeles. He was able to put these concepts into practice when he joined IBM in 2006. By the time he left IBM in 2021, many of his research results and technologies had been integrated into the company’s design and test flows, helping IBM to ship multiple generations of its high-performance server chips and American Semiconductor Innovation Coalition (ASIC) chips.

Xiong has nearly 45 patents related to process variations, and some of his patents have also been licensed to external companies for commercialization. 

A career built on data

Xiong’s current research is in artificial intelligence. He was recently named director of the University at Buffalo’s Institute for Artificial Intelligence and Data Science (IAD) and is the scientific director and co-director of the $20 million National AI Institute for Exceptional Education at UB. The shift from modeling process variation to AI is not a stark change. According to Xiong, the work overlaps.

“Dealing with process variation requires modeling the process. You need to make sure that you know the process data and manufacturing data. From the data, you build your process variation model,” Xiong says. “I model the data while dealing with a lot of uncertainty. The essence of my work is about how to optimize under the uncertainties. This has been my body of expertise, and it can be translated to AI research as well.”

During his career, Xiong has also examined smarter and renewable energy and cognitive computing. In all of his research areas he has worked with data, including big data analytics. He continues to work within those areas today in his various roles, including in a new role as an affiliated faculty member with UB’s Center for Advanced Semiconductor Technologies.

“UB has established a new semiconductor microelectronics center,” Xiong says. “Given the recent CHIPS Act and UB’s central role as part of the federally dedicated tech hub, I would love to use my background and new capacity as director of IAD to see how we can really establish UB at the forefront of semiconductor research with AI.”