Modeling and simulation of neuromorphic units based on resistive memory devices

This paper details the development and simulation of a neuromorphic unit model based on memristors, designed to replicate key synaptic characteristics essential for neuromorphic computing. These characteristics include Long-Term Potentiation (LTP), Long-Term Depression (LTD), Short-Term Potentiation (STP), Short-Term Depression (STD), and Spike-Timing-Dependent Plasticity (STDP), all of which are critical for learning and memory processes in biological systems. The model employs the Piecewise Linear (PWL) method, chosen for its simplicity, continuous differentiability, and high computational efficiency. These properties enable accurate and resource-efficient simulation of memristor dynamics, crucial for large-scale neuromorphic systems. To ensure broad applicability, the model is implemented in multiple formats: Verilog-A for hardware simulation, C language for general-purpose programming, and macro models compatible with popular simulation tools such as HSPICE, Cadence, MATLAB, and LTSPICE. This versatility allows seamless integration into various research and development workflows. The paper provides a thorough analysis of the model's resistive variation equations, IV relationships, and resistance drift phenomena, which are fundamental to understanding memristor behavior and synaptic plasticity. It also demonstrates the construction of macro models in simulation software, offering practical guidance for implementation. Extensive simulations confirm the model's ability to accurately simulate LTP, STP, LTD, STD, and STDP, validating its effectiveness in capturing the complex dynamics of biological synapses. This makes the model a valuable tool for advancing neuromorphic computing research. In conclusion, the neuromorphic unit model presented here offers a robust and efficient framework for simulating synaptic characteristics in memristor-based systems. Future work may focus on further refining the model and exploring its applications in more complex neuromorphic architectures.