A new AI-based technology developed by Cornell researchers will help gain new insights into how our brains respond to external stimuli. Read more about Movies, music and pictures can train synthetic brain
Cornell ECE researchers receive best paper award
Cornell ECE Assistant Professor Zhiru Zhang and graduate students from his group recently received the Best Paper Award for the Short Paper Category at the 26th IEEE International Symposium on Field-Programmable Custom Computing Machines (FCCM).
Cornell ECE Assistant Professor Zhiru Zhang and graduate students from his group, in collaboration with Professor Evangeline F.Y. Young from The Chinese University of Hong Kong, recently received the Best Paper Award for the Short Paper Category at the 26th IEEE International Symposium on Field-Programmable Custom Computing Machines (FCCM) held in Boulder, C.O., April 29 through May 1. The paper, “Fast and Accurate Estimation of Quality of Results in High-Level Synthesis with Machine Learning” is authored by Steve Dai, Yuan Zhou, Hang Zhang, Ecenur Ustun, Evangeline F.Y. Young and Zhiru Zhang.
While high-level synthesis (HLS) is becoming the go-to tool for implementing high-performance applications on FPGA, inaccurate quality of results estimates of the final hardware, such as how large the circuit will be and what clock frequency is achievable, remain a major pain point of the HLS technology. Because HLS lacks complete information on the subsequent FPGA implementation process, area and timing estimates reported by HLS often deviate significantly from final results. To get an accurate understanding of the final design, FPGA designers must spend hours, if not days, to run through the entire FPGA implementation process.
“This paper applies machine learning techniques to generate much more accurate area and timing estimates for HLS,” said Steve Dai, lead author on the paper and Ph.D. student in the Zhang Research Group. “Machine learning models allow us to holistically capture the multitude of factors affecting estimation accuracy. These automatically-learned models also teach us (the tool developers) the important features to consider for getting accurate estimates.”
Many FPGA designers relate to the inaccurate HLS estimation problem and believe that the Zhang Group’s proposed approach constitutes a promising direction in closing the accuracy gap. “While we should definitely leverage FPGA to accelerate machine learning, it is also important to take full advantage of machine learning to improve the FPGA design automation flow. Our paper represents an early attempt on the latter direction,” said Dai.
The IEEE Symposium on Field-Programmable Custom Computing Machines is the original and premier forum for presenting and discussing new research related to computing that exploits the unique features and capabilities of FPGAs and other reconfigurable hardware. Over the past two decades, FCCM has been the place to present papers on architectures, tools, and programming models for field-programmable custom computing machines as well as applications that use such systems. Find out more at http://fccm.org/2018/.
The Zhang Research Group in the School of Electrical and Computer Engineering at Cornell University investigates new applications, programming models and computer-aided design (CAD) algorithms and tools to enable productive design and implementation of highly efficient application- and domain-specific computer systems. The group’s cross-cutting research intersects CAD, compilers and computer architecture at multiple scales, from circuit-level building blocks, to chip-level processor and co-processor cores, as well as system-level heterogeneous compute nodes. Learn more about the group and their work at http://zhang.ece.cornell.edu.