Please use this identifier to cite or link to this item:
Title: Fast prototyping of neural network application
Authors: Liu, Yingmin
Keywords: Engineering::Electrical and electronic engineering
Issue Date: 2022
Publisher: Nanyang Technological University
Source: Liu, Y. (2022). Fast prototyping of neural network application. Master's thesis, Nanyang Technological University, Singapore.
Abstract: SRAM has become the most important memory structure in storage due to its small size and high speed. Many new structures have been developed like 8T, 9T, 10T, 11T SRAM based on 6T SRAM. This work focused on the study of 8T SRAM due to its stability. Then we discussed about the current mode and voltage mode to find a good control method. Current mode has greater accuracy than voltage mode however it is not stable. We choose the voltage mode because it is faster and stabler when the circuit is small. To solve the problem of fast training and classification, hardware accelerator is proposed. There are many new platforms appearing and they can develop the accuracy and emulate the circuit. The most suitable platform to estimate the system-level performance quickly is NeuroSim+MLP. This method uses multilayer perceptron (MLP) to test the benchmark and classify MNIST dataset. Then, we use Virtuoso to simulate writing and reading “1” operation with umc401p library to emulate 8T SRAM and the memory cell has 32 units in our test. When changing the number of cells opened, the change rate of RBL voltage will vary. Then we put the data into ADC in NeuroSim to evaluate this accelerator. The accuracy is about 96.32%. To compare the performance of this new method, we also run MNIST dataset by Python. The result shows NeuroSim+MLP occupies small area and little source and has better accuracy than computer only.
Schools: School of Electrical and Electronic Engineering 
Research Centres: A*STAR Institute of Material Research and Engineering 
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Theses

Files in This Item:
File Description SizeFormat 
LIU YINGMIN_dissertation_amended.pdf
  Restricted Access
3.98 MBAdobe PDFView/Open

Page view(s)

Updated on Apr 17, 2024


Updated on Apr 17, 2024

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.