A capacitive neural network for associative memory
(1989) Proceedings of the Tenth Symposium on Information Theory in the Benelux — Location: Houthalen, Belgium (25.May.1989)
Files
5-Acapacitiveneuralnetworkforassociativememory.pdf
Restricted Access - Adobe PDF
- 1.73 MB
Details
- Authors
- Abstract
- Neural networks show interesting properties in vision problems, such as pattern or character recognition. However, the advantages of such networks (speed, convergence, . . .) are only fully exploited with dedicated processors. This paper presents a VLSI architecture for Hopfield-like fully interconnected networks. Its particularity is to use capacitors as synaptic connections in place of resistors or current sources. This network is programmed by a learning rule adapted to such networks where the dynamics of the connection weights is restricted to some discrete values. The feasibility of this architecture has been proved by an 8-neurons network built with discrete components.
- Affiliations
Citations
Verleysen, M., Jespers, P., & Martin, D. (1989). A capacitive neural network for associative memory. In Barbe, A.M.; (ed.), Proceedings of the Tenth Symposium on Information Theory in the Benelux (p. p. 73-79). Werkgemeenschap voor inf.- & communicatietheorie. https://hdl.handle.net/2078.5/253758