Cornell University received a three-year Project Grant award of $598,449 from the National Science Foundation Division of Computing and Communication Foundations under the Computer and Information Science and Engineering federal grant program (CFDA 47.070).
The award will support research and development of specialized computer chips and software programs to accelerate artificial intelligence workloads using fine-grained unstructured sparsity and mixed precision techniques. Cornell University will optimize unrolled deep neural network circuits for implementation on field-programmable gate arrays and investigate hardware architecture enhancements like time-multiplexing and in-memory computing. The university will also codesign deep learning sparsification algorithms including pruning, quantization, and parameter sharing to extract maximum efficiency from the hardware. Outcomes are expected to include new bit-programmable hardware architectures, deep neural network sparsification methods, and a research framework to jointly optimize sparse neural networks and hardware.