Optimized Near-Zero Quantization Method for Flexible Memristor Based Neural Network
Open Access
- 21 May 2018
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 6, 29320-29331
Abstract
Due to controllable conductance and non-volatility, flexible memristors are regarded as a key enabler for building Artificial Neural Network (ANN) based learning algorithms in flexible and wearable systems. However, existing flexible memristors are suffering from limited number of conductance values, issues limiting large-scale integration and insufficient accuracy that cannot support accurate computation of ANN. In this paper, solutions are proposed for the three major challenges of the flexible memristor, the feasibility of a three-layer fully-connected neural network on MNIST and a thirteen-layer Convolutional Neural Network (CNN) on CIFAR-10 using the flexible memristor based on single-walled carbon nanotubes (SWCNTs) network/polymer composite and hydrophilic Al2O3 dielectric are studied. The evaluation result show that in the fully-connected neural network system, it is able to recognize MNIST with an accuracy above 90% after 4-bit quantization, 52.05% decrease in interconnection numbers in the circuit and up to 40% random error introduced, and in the CNN on CIFAR-10, the system can retain an accuracy above 86% with less than 4% accuracy loss after 5-bit quantization, 59.34% decrease in interconnection numbers in the circuit and up to 40% random error injected.Keywords
Funding Information
- National Natural Science Foundation of China (61571137)
- Shanghai Pujiang Program (17PJ1400800)
- Shanghai Science and Technology Innovation Program (17JC1401400)
- JSSF project (20150164)
This publication has 40 references indexed in Scilit:
- Face classification using electronic synapsesNature Communications, 2017
- Memristive Model for Synaptic CircuitsIEEE Transactions on Circuits and Systems II: Express Briefs, 2016
- Feature Extraction Using Memristor NetworksIEEE Transactions on Neural Networks and Learning Systems, 2015
- Memory and Information Processing in Neuromorphic SystemsProceedings of the IEEE, 2015
- Interaction of bipolaron with the H2O/O2 redox couple causes current hysteresis in organic thin-film transistorsNature Communications, 2014
- Memristor Crossbar-Based Neuromorphic Computing System: A Case StudyIEEE Transactions on Neural Networks and Learning Systems, 2014
- Statistical parametric speech synthesis using deep neural networks2013
- Charge Transfer Equilibria Between Diamond and an Aqueous Oxygen Electrochemical Redox CoupleScience, 2007
- Object Recognition with Features Inspired by Visual Cortex2005
- Gradient-based learning applied to document recognitionProceedings of the IEEE, 1998