Neuromorphic CMOL circuits

This is a brief review of the recent work on the development of neuromorphic architectures for future hybrid CMOS/nanowire/MOLecular ("CMOL") circuits. Such circuits may provide the first chance for the implementation of advanced information processing systems with areal density of (beyond 10¹² active functions per cm²) comparable to that of the human cerebral cortex, while operating at much higher speed (up to 10²⁰ operations per second per cm²), at acceptable power consumption. Our group has suggested a family of distributed crosspoint networks ("CrossNets") that are natural for implementation in CMOL technology, and has shown that such networks may be trained to perform at least the effective pattern recognition in the Hopfield mode. Work on CrossNet training to perform more complex tasks in under way.