ECE Seminar: Towards High-Efficiency and High-Usability Hardware Acceleration

Abstract: The exploding complexity and computation efficiency requirements of modern applications are stimulating a strong demand for hardware acceleration with heterogeneous platforms such as FPGAs. The stringent requirements of these modern applications also pose several challenges for hardware acceleration systems. In particular, achieving high-efficiency, high-usability hardware acceleration targeting highly heterogeneous systems is one major challenge. In this talk, I will first give an overview of the challenges that modern hardware acceleration systems are facing, as well as my previous works in tackling these challenges. I will then introduce my recent representative works in detail, including PyLog language and compiler for Python-based high-level programming for FPGAs, Tangram language and compiler for highly efficient GPU code generation, Chai-FPGA for enabling CPU-FPGA collaborative computing, etc. Evaluation shows the promising performance and flexibility of these works when solving the hardware acceleration challenges. I will conclude my talk with my envision of future hardware acceleration research.