Skip to main content
Top

2024 | Book

High Energy Efficiency Neural Network Processor with Combined Digital and Computing-in-Memory Architecture

Author: Jinshan Yue

Publisher: Springer Nature Singapore

Book Series : Springer Theses

insite
SEARCH

About this book

Neural network (NN) algorithms are driving the rapid development of modern artificial intelligence (AI). The energy-efficient NN processor has become an urgent requirement for the practical NN applications on widespread low-power AI devices. To address this challenge, this dissertation investigates pure-digital and digital computing-in-memory (digital-CIM) solutions and carries out four major studies.

For pure-digital NN processors, this book analyses the insufficient data reuse in conventional architectures and proposes a kernel-optimized NN processor. This dissertation adopts a structural frequency-domain compression algorithm, named CirCNN. The fabricated processor shows 8.1x/4.2x area/energy efficiency compared to the state-of-the-art NN processor. For digital-CIM NN processors, this dissertation combines the flexibility of digital circuits with the high energy efficiency of CIM. The fabricated CIM processor validates the sparsity improvement of the CIM architecture for the first time. This dissertation further designs a processor that considers the weight updating problem on the CIM architecture for the first time.

This dissertation demonstrates that the combination of digital and CIM circuits is a promising technical route for an energy-efficient NN processor, which can promote the large-scale application of low-power AI devices.

Table of Contents

Frontmatter
Chapter 1. Introduction
Abstract
Artificial Intelligence (AI) has promoted the development of modern society in many aspects [1], and will profoundly change human social life and the world [2].
Jinshan Yue
Chapter 2. Basics and Research Status of Neural Network Processors
Abstract
The NN algorithm has experienced a long history with several times of upsurges and troughs. In the 1890s, the research related to the human brain has arisen.
Jinshan Yue
Chapter 3. Energy-Efficient NN Processor by Optimizing Data Reuse for Specific Convolutional Kernels
Abstract
This chapter studies the data reuse characteristics of digital ASIC NN processors, and proposes a CNN processor, named KOP3, that is optimized for specific convolutional kernel sizes. Compared to previous NN processors that achieve higher flexibility but sacrifice a certain energy efficiency, the proposed KOP3 architecture trade-offs the flexibility for different convolutional kernel sizes with energy efficiency.
Jinshan Yue
Chapter 4. Optimized Neural Network Processor Based on Frequency-Domain Compression Algorithm
Abstract
This chapter introduces a designed neural network processor that improves energy efficiency with a frequency-domain compression algorithm. This chapter first analyzes the significant power and area overhead of NN processors based on irregular sparse compression technology due to the support for sparsity, and then introduces the frequency-domain structural compression algorithm adopted in this work.
Jinshan Yue
Chapter 5. Digital Circuits and CIM Integrated NN Processor
Abstract
This chapter first introduces the advantages of CIM in terms of energy efficiency compared to pure digital circuits based NN processors, and analyzes the deficiencies of CIM system chips, as well as the challenges of data reuse and sparsity optimization at the system level.
Jinshan Yue
Chapter 6. A “Digital+CIM” Processor Supporting Large-Scale NN Models
Abstract
Based on the research work in Chap. 5, this chapter further analyzes the optimization space of “Digital+CIM” NN processors at the system level, and points out the efficiency and accuracy challenges of system-level CIM chips when running large-scale NN applications.
Jinshan Yue
Chapter 7. Summary and Prospect
Abstract
NN algorithms have been applied in various fields such as security, autonomous driving, finance, healthcare, and will continue to penetrate more segmented scenarios, profoundly changing and promoting the development of human society.
Jinshan Yue
Metadata
Title
High Energy Efficiency Neural Network Processor with Combined Digital and Computing-in-Memory Architecture
Author
Jinshan Yue
Copyright Year
2024
Publisher
Springer Nature Singapore
Electronic ISBN
978-981-9734-77-1
Print ISBN
978-981-9734-76-4
DOI
https://doi.org/10.1007/978-981-97-3477-1