Ph.D. Dissertation Research
Dissertation Title:
- Optimizing FPGA Architecture for Deep Learning Workloads
Highlights:
- Adding matrix multiplier blocks to FPGAs (“Tensor Slices”)
- Impact: The FPGA industry has developed similar blocks recently like Intel AI Tensor Block and Achronix Machine Learning Processor. Although these blocks do not support matrix multiplication natively, the fact that industry is finding similar solutions demonstrates that our research is on the right track.
- Incorporating compute-in-memory in FPGAs (“CoMeFa RAMs”)
- Impact: Our paper proposing these memory blocks at FCCM’22 won the Best Paper Award, and companies like Rapid Silicon have expressed interest in this technology.
- Benchmarks for FPGA architecture research (“Koios Benchmarks”)
- Impact: Koios is being used by many researchers worldwide and also by companies like Efinix in their regression suite.
