Date of Award
Electrical and Computer Engineering
Miller, W. C.,
Engineering, Electronics and Electrical.
CC BY-NC-ND 4.0
This thesis explores the design and implementation of a multilayer programmable optically coupled neural network in the Northern Telecom 1.2$\mu$ Complementary Metal Oxide Semiconductor (CMOS) process. The motivation for this work originated from the results obtained from the fabricated implementation of fixed weight optically coupled neural networks in the Northern Telecom 3$\mu$ CMOS process. Previous designs were fabricated and tested with remarkable results. The new design of the optically coupled neural network is a translation of the previous designs from 3$\mu$ to 1.2$\mu$ CMOS technology with the improvement of the programmability. The new programmable neural network contains 5 x 5 photosensitive elements as the input devices, 7 neurons, and 112 synaptic weights. Also, the design includes 596 bit memory as the on-chip weight storage. This digital memory, along with other analog circuitries, makes the network a hybrid network. Each synaptic weight is represented by a 5-bit digital signal, and digital to analog (D/A) conversion is followed for the analog computation. The Very Large-Scale Integration (VLSI) implementation mask layouts of this network are completely verified. Functionality of the building blocks for the network is proved by the SPICE simulations. The main usage of this network is for the pattern recognition which is currently involved in another industrial research project at the University of Windsor. The programmability of the network provides the feature of multi-set of patterns. It also develops the features of a training loop network. Source: Masters Abstracts International, Volume: 34-02, page: 0832. Advisers: W. C. Miller; G. A. Jullien. Thesis (M.A.Sc.)--University of Windsor (Canada), 1994.
Lei, Ka Wa., "A 1.2 micron neural network design." (1994). Electronic Theses and Dissertations. 512.