LAWRENCE — Turning powerful graphics cards that increase the speed and performance of video games into reliable processors for high-performance computing has earned a University of Kansas researcher a prestigious National Science Foundation (NSF) award.
Xin Fu, an assistant professor in electrical engineering and computer science (EECS), has received a $430,000 NSF Early Career Development (CAREER) Award. The five-year grant supports junior faculty members who have shown exceptional promise in teaching and research.
“I am honored to receive this award,” Fu said. “This grant will help me develop tools to assess the reliability of next-generation throughput processors integrated with emerging technologies. That assessment will then lead to predicting, detecting and, finally, tolerating various types of errors. The end result will be that a wide range of disciplines will be able to use throughput processors for their data processing needs.”
Complex 3D images were overwhelming computer resources, leading to decreased overall performance. To ease the strain, researchers developed a Graphic Processing Unit (GPU) to build images and manage other large efforts. GPUs consist of smaller, more efficient cores designed to perform multiple tasks simultaneously, known as parallel processing. Further advancements were needed, however, to improve performance and energy efficiency, so new technologies were added to GPU systems that were not necessarily reliable.
With her CAREER Award, Fu is developing tools to assess reliability in GPUs integrated with emerging technologies, such as non-volatile memory (NVM). Leaked power can account for nearly 50 percent of a chip’s power consumption, according to EE Times. To create more energy-efficient computers, researchers started adding NVM to GPUs for its low rate of unintended power leaks. Fu’s work will start with an assessment of the reliability of the data that these new integrated GPUs produce.
Since each new technology has positive and negative impacts on reliability, Fu will develop vulnerability models to asses for three kinds of reliability errors: particle-strike soft errors, aging-effect hard errors and manufacturing process variations. These models will lead to the prediction of errors in new GPU systems and the creation of lightweight error detections techniques. Finally, the error detection techniques will lead to designing systems that are fault-tolerant, or more reliable, at a lower cost.
“It is critical to harness novel technologies’ benefits and overcome their shortcomings on reliability to develop robust, high-performance and energy-efficient GPU processors,” Fu said.
Once Fu’s work on this proposal is done, she hopes to benefit numerous real life applications by leveraging these advanced processing technologies. The varied fields of finance, medicine, biology, aerospace and geology could all benefit from faster and more reliable data computation. On a more general level, reliable chips are critical to large-scale growth in supercomputing.
Fu came to KU in 2010 after graduating with her doctorate in computer engineering from the University of Florida. She also is a recipient of Kansas NSF EPSCoR First Award and NSF Computing Innovation Fellow (CIFellow). She has taught courses in digital design, computer architecture and computer organization. She conducts research at KU’s Information and Telecommunication Technology Center.
Her CAREER Award marks the second for the department this spring. EECS Assistant Professor Andy Gill also received a CAREER Award for his functional programming.