CNN-Based Automated Parameter Extraction Framework for Modeling Memristive Devices

The introduction of an automated parameter extraction framework for memristive devices represents a significant leap in the modeling of resistive random access memory (RRAM), which is crucial for next-generation nonvolatile memory (NVM) and in-memory computing applications. Traditional methods often rely on numerous fitting parameters that necessitate extensive manual tuning, making them time-consuming and less adaptable. The new framework employs a convolutional neural network (CNN) trained on synthetic datasets to generate initial parameter estimates directly from device I-V characteristics. This approach is refined through heuristic optimization blocks that minimize errors, achieving low error rates across diverse device characteristics. Evaluated against key NVM metrics, including set voltage and reset voltage, the framework demonstrates its potential to enhance the efficiency and accuracy of RRAM modeling, paving the way for advancements in memory technology.