A Simple Method to Reduce Off-chip Memory Accesses on Convolutional Neural Networks

2019/01/28 by Do Yun Kim, Kim, Doyun, Kyoung-Young Kim +5
Computer Science · Engineering · #Advanced Memory and Neural Computing #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Ferroelectric and Negative Capacitance Devices #Machine Learning (cs.LG) #Neural and Evolutionary Computing (cs.NE)

paper · pdf · doi:10.48550/arxiv.1901.09614

openalex publication_date 2019/01/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

For convolutional neural networks, a simple algorithm to reduce off-chip memory accesses is proposed by maximally utilizing on-chip memory in a neural process unit. Especially, the algorithm provides an effective way to process a module which consists of multiple branches and a merge layer. For Inception-V3 on Samsung's NPU in Exynos, our evaluation shows that the proposed algorithm makes off-chip memory accesses reduced by 1/50, and accordingly achieves 97.59 % reduction in the amount of feature-map data to be transferred from/to off-chip memory.

Citations

• BinaryConnect: Training Deep Neural Networks with binary weights during propagations
• Fixed Point Quantization of Deep Convolutional Networks
• MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

Related

• MRAM Co-designed Processing-in-Memory CNN Accelerator for Mobile and IoT Applications
• Enabling Large Neural Networks on Tiny Microcontrollers with Swapping
• Architectural Implications of Neural Network Inference for High Data-Rate, Low-Latency Scientific Applications
• High Performance Zero-Memory Overhead Direct Convolutions
• Block Convolution: Towards Memory-Efficient Inference of Large-Scale CNNs on FPGA