Electrically Modifiable Nonvolatile SONOS Synapses for Electronic Neural Networks.
Abstract
This research addresses the implementation of an electronic element, which emulates the biological synaptic interconnection, in an artificial electronic neural system. The basic interconnection, or the weight, consists of an electrically reprogrammable, nonvolatile, analog conductance which programs at 5V levels. In addition, the fabrication technology for this synaptic interconnection is compatible with existing CMOS VLSI processes. The attractive features of this synaptic weight will be discussed in this report. Furthermore, this report examines the material needs, the-device structures, the use of the synaptic weights in a two-tap weight linear adaptive neural-like circuit and the issue of integrating both the synaptic weight elements and the peripheral circuit onto a single silicon wafer.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 30, 1992
- Accession Number
- ADA258318
Entities
People
- Amit Banerjee
- Chun-yu M. Chen
- Margaret French
- Marvin H. White
Organizations
- Lehigh University