Development of Neural Modules Based on Si/PLZT Technology for Opto- Electronic Implementations of Neural Networks
Abstract
The objective of the research program was to design opto-electronic neuron modules communicating via free-space optical interconnects, and to develop Si/PLZT opto-electronic integrated circuit technology in order to implement these designs. First, device and system requirements for artificial neural networks were studied. Minimum performance requirements for artificial neural networks implementations were extracted. Next, two opto-electronic neural network architectures were developed. The first achieves reconfigurable optical interconnects using photorefractive crystals. This system was theoretically and experimentally investigated. The second reconfigurable weights based on Si/PLZT technology. A prototype of this system was successfully built and tested. The performance of both systems exceed the minimum requirements. The first year effort involved the design of the optoelectronic architectures, the further development of the Si/PLZT process, the development of optimal neural networks data-encoding methods, and analysis of the system performances. The remaining period entailed the experimental demonstration of the CMTM system, an the development, characterization, and application of the D-STOP prototype system.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 31, 1992
- Accession Number
- ADA257937
Entities
People
- Sadik Esener
- Sing H. Lee
Organizations
- University of California, San Diego