An Integrated CMOS/Memristor Bio-Processor for Re-configurable Neural Signal Processing

This paper proposes a bio-processor for neural signal analysis. The device architecture features an analogue Front-End and a Process Element, the latter can be scaled as an array. Rather than a single dedicated algorithm, the Process Element supports multiple analysis modes, utilising the analogue behaviour of memristors. When used as part of an array structure, each Process Element can be programmed independently and furthermore, the array elements can be electrically interconnected in an arbitrary manner. The device facilitates an inter-network of in-memory computation units, i.e. an inter-network of functions. This supports construction of a system that is highly scalable, re-configurable and thus adaptive. The device enables multi-functional neural recording and processing, for early stage signal exploration. The device has been implemented using a standard 180nm CMOS process with the addition of back-end-of-line (BEOL) memristor deposition.
¹
Although targeted at neural signal analysis, the device and the architecture described is considered general purpose and may find application within other disciplines.