BitSystolic: A 26.7 TOPS/W 2b∼8b NPU With Configurable Data Flows for Edge Devices

Efficient deployment of deep neural networks (DNNs) emerges with the exploding demand for artificial intelligence on edge devices. Mixed-precision inference with both compressed model and reduced computation cost enlightens a way for accurate and efficient DNN deployments. Despite obtaining mixed-precision DNN models at the algorithmic level, there still lacks sufficient hardware support. In this work, we propose BitSystolic, a neural processing unit based on a systolic array structure. In BitSystolic, the numerical precision of both weights and activations can be configured in the range of 2- 8 bits, fulfilling different requirements across mixed-precision models and tasks. Moreover, BitSystolic can support various data flows presented in different types of neural layers (e.g., convolution, fully-connected, and recurrent neural layers) and adaptive optimization of data reuse by switching between the matrix-matrix mode and vector-matrix mode. We designed and fabricated the proposed BitSystolic composed of a 16x16 systolic array. Our measurement results show that BitSystolic features the unified power efficiency of up to 26.7 TOPS/W with 17.8 mW peak power consumption across various layer types.

Automated summary

The article introduces a neural processing unit based on a systolic array structure called BitSystolic, which can support numerical precision configuration in the range of 2-8 bits, fulfilling different requirements across mixed-precision models and tasks. The unit can also flexibly support data flows in various types of neural layers and achieve adaptive optimization of data reuse by switching between different modes.