Strategies to Finding Optimal Parameters for Plasticity Changes in Memristor-Based Systems for Neuromorphic Data Computing

Abstract Current publication highlights exemplary commercially available memristive chips, capable of replicating processing nodes found in artificial neural networks and reservoirs. These hardware components can become foundational elements of neuromorphic computing. To achieve this, several pre-defined node-to-node connection strength modulation effects must be integrated into a potential system-on-a-chip. KNOWM memristors exhibit many of these effects, including potential-dependent potentiation-depression behavior. The results from our preliminary tests, which identify optimal electric parameters for achieving strong and consistent synaptic responses, serve as a basis for research into more complex phenomena, such as spike-rate-dependent plasticity (SRDP) and spike-time-dependent plasticity (STDP). To advance the development of information processing systems that incorporate novel materials, it is essential to establish universal testing and benchmarking protocols. As material-based neural networks are anticipated to play a significant role in future computational tasks, these protocols will ensure consistent and comparable performance assessments across various memristive systems. The proposed methodology offers a cost-effective and accessible approach to benchmarking in this emerging field.