A hardware–software blueprint for flexible deep learning specialization


This article describes the Versatile Tensor Accelerator (VTA), a programmable DL architecture designed to be extensible in the face of evolving workloads. VTA achieves “flexible specialization” via a parameterizable architecture, two-level Instruction Set Architecture (ISA), and a Just in Time (JIT) compiler.

IEEE Micro
Ziheng Jiang
Ziheng Jiang
Ph.D. Student

His research center around co-designing efficient algorithms and systems for machine learning.