• Develop Deep Learning Compilers for performance and programmability.

• Led a Software team specialized in Heterogenous Compute (with GPU/DSP) and Deep Learning

• Developed compiler/runtime for Compute APIs such as OpenCL

• Led design/implementation of an OpenCL1.2-conformant software stack on ARM/NEON, which is 50% faster than competitors’ counterpart

• Architected a unified OpenCL runtime, which is shared by various in-house processors such as CPU, GPU, and DSP and is 20X faster than competitors’ counterpart

• Optimized Renderscript on GPUs; the resulted software stack is at least 50% faster than competitor’s counterpart according to popular benchmarks

• Developed world’s first production-quality device fusion technology, which can execute workload on CPU and GPU simultaneously with dynamic scheduling capability

• Conducted research and published well-cited papers on prestigious conferences/journals such as TECS about system-level modeling and simulation

• Received 3 awards for my doctoral research as listed below in the Honors section

• Explored advanced AMBA bus architecture by automatically generating models 20X faster than counterparts in the commercial tools in terms of simulation speed with the same timing accuracy

• Constructed an OpenRISC based processor simulator that can run 50X faster than the corresponding cycle-accurate models while providing the same cycle count information