SCNN: An Accelerator for Compressed-sparse Convolutional Neural Networks

TL;DR

SCNN is a specialized accelerator architecture that leverages sparsity in CNN weights and activations to significantly improve performance and energy efficiency, making it suitable for mobile and embedded applications.

Contribution

The paper introduces a novel sparse dataflow and hardware design for CNN acceleration that exploits zero-valued weights and activations to enhance efficiency.

Findings

SCNN achieves 2.7x performance improvement over dense accelerators.

SCNN reduces energy consumption by 2.3x compared to dense CNN accelerators.

The architecture effectively maintains and processes sparse data in compressed form.

Abstract

Convolutional Neural Networks (CNNs) have emerged as a fundamental technology for machine learning. High performance and extreme energy efficiency are critical for deployments of CNNs in a wide range of situations, especially mobile platforms such as autonomous vehicles, cameras, and electronic personal assistants. This paper introduces the Sparse CNN (SCNN) accelerator architecture, which improves performance and energy efficiency by exploiting the zero-valued weights that stem from network pruning during training and zero-valued activations that arise from the common ReLU operator applied during inference. Specifically, SCNN employs a novel dataflow that enables maintaining the sparse weights and activations in a compressed encoding, which eliminates unnecessary data transfers and reduces storage requirements. Furthermore, the SCNN dataflow facilitates efficient delivery of those weights and activations to the multiplier array, where they are extensively reused. In addition, the accumulation of multiplication products are performed in a novel accumulator array. Our results show that on contemporary neural networks, SCNN can improve both performance and energy by a factor of 2.7x and 2.3x, respectively, over a comparably provisioned dense CNN accelerator.