1. Introduction
This paper is published in 2019. Due to the compact or sparse models are different from the traditional large ones in that there is much more variation in their layer shapes and sizes, the researchs design a flexible accelerator, called eyeriss v2, exploits sparsity for performance improvement.
2. Innovation points
The following is the Eyeriss v2 top-level architecture.
1. Propose a novel NoC, called hierarchical mesh, that is designed to adapt to a wide range of bandwidth requirements.
2. Design a new PE architecture that exploits the sparsity in weights and activations simultaneously
3. Summary
The Eyeriss v2 specification is shown below.