SP-CNN: A Scalable and Programmable CNN-based Accelerator

Abstract

Specialized accelerators have become prevalent in mobile computing platforms for their ability to perform certain tasks, such as image processing, at a lower power cost than CPUs or GPUs. In this paper, we focus on using Cellular Neural Networks (CNN) as a specialized accelerator. CNN is a neural computing paradigm that is well suited for image processing applications. However, hardware implementations were originally developed only to handle relatively small image sizes. In this paper, we propose SP-CNN, an architecture as well as a multiplexing algorithm that provides scalability to CNN applications. We demonstrate the proposed multiplexing algorithms over a set of six image processing benchmarks, as well as present a performance and power analysis of SP-CNN.