Previous Job
ML Acceleration Engineer ( sw-hw Codesign )
Ref No.: 20-02587
Location: Menlo Park, California
Client Acceleration Engineer (S/W-H/W Codesign)

Job Summary

We are looking for an Engineer to implement efficient h/w acceleration of Machine Learning algorithms for Computer Vision in embedded domain, with an emphasis on performance and power. We are looking for someone with strong software development skills, familiarity with Client algorithms like CNN's and hands-on experience in s/w-h/w codesign, especially in the context of Client.

Work include :
  1. Collaborate with computer architects, software, Client and silicon engineers, to map and optimize Client workloads on various backend targets including CPU's, DSP's and Deep Learning Accelerators.
  2. Run analysis/profiling, identify performance and power bottlenecks on the actual h/w, virtual platforms, simulators or emulators and provide feedback for optimizations across the entire stack.
  3. Work with Deep Learning compilers; identify the correct knobs for best efficiency and influence new feature additions.
  4. Develop optimized kernels and port various Client libraries to backends like DSP's with custom ISA;
  5. Ensure high quality by creating tests and automation infrastructure.
  6. Partner with productization teams and driver/firmware teams to integrate Client acceleration into shipping software and create any new tools as necessary.

  • BS in EE/Computer with 5+ industry experience. MS or PhD with industry experience is preferable.
  • Strong coding skills in C/C++ or Python.
  • Experience with h/w acceleration on GPU's/CPU's/DSP's/custom h/w.
  • Familiarity of Client algorithms like CNN's and frameworks like Tensorflow/Pytorch.
  • Familiarity with profiling and debug tool; Tools in context of Client is a plus.
  • Familiarity with Deep learning compilers like tensor-rt, XLA is a plus.
  • Understanding of Client algorithm optimizations for low power like quantization, pruning etc. is a plus.
  • Comfortable with reading others code, tracing them, and code refactoring