Fingerprinting Memristors for Neural Computing

An Artificial Neural Network (ANN) is a computational model that is inspired by the way biological neural networks in the human brain process information as it is a possible path toward ultra-fast real-time Artificial Intelligence (AI).The memristor technology has been reported as an excellent solution to implement basic the computational elements that are called nodes or units. It receives input from some other units, or perhaps from an external source. Each input has an associated weight which can be modified so as to model synaptic learning, response times below a milli second are needed.

We are proposing to use the natural variations in manufacturing of each memristor-based node as “fingerprint”. The injection of extremely small electric currents (a few nano Amperes) in each node forces the resistance to drop abruptly by several orders of magnitudes through the formation of a conductive path between the two electrodes. These conductive paths dissolve as soon as the current injection stops as the nodes return to their initial state. A repeat injection of currents into the same node results in a predictable effect in resistance drop. Different, stable resistance values in each node can be achieved by injecting different current values, thereby yielding large entropy. With no changes to existing memristor manufacturing technologies, we suggest the software and computing architecture to enhance resilience, reduce power losses, lower latencies, and enhance cybersecurity.