Design for a Manufacturing Method for Memristor-Based Neuromorphic Computing Processors
Abstract
It is well received that conventional CMOS technology is approaching its physical limitations. Researchers have started to explore the potential replacement by leveraging the advances of nanotechnology. Very recently, memristor attracted growing attentions since the first physical realization reported by HP Labs in 2008. Unique characteristics like non-volatility, re-configurability, and analog state storage make memristor become a very promising candidate for the realization of artificial neural systems. In this project, we developed a SPICE-compatible model of memristor and designed CMOS-mimicked memristor cells for system development. Then we proposed a memristor-based design of bidirectional transmission excitation/inhibition synapses and implemented a neuromorphic computing system based on our proposed synapse designs. The robustness of our system is also evaluated by considering the actual manufacturing variability with the emphasis on process variations. After that, we discussed memristor-based crossbar neuromorphic architecture. Finally, we compared the designs of synapse network-based and crossbar-based neuromorphic computing systems.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2013
- Accession Number
- ADA581795
Entities
People
- Beiye Liu
- Yiran Chen
Organizations
- University of Pittsburgh