PROMPT: Prediction Model for RRAM Programming Optimized by Adaptive Mechanism and Progressive Continuous Transformation

Resistive Random-Access Memory (RRAM) has emerged as a promising candidate for next-generation non-volatile memory due to its fast access speed, low power consumption, and compatibility with in-memory computing. However, the inherent variability and nonlinearity in resistance evolution pose significant challenges for precise and efficient programming, particularly under conventional schemes such as Incremental Step Pulse Programming (ISPP), which suffers from high latency and device wear due to excessive pulse applications.To improve the efficiency of RRAM programming, we propose PROMPT, an AI-assisted framework designed to optimize the programming process. Specifically, PROMPT integrates: (1) a MLP-based adaptive classification module that dynamically evaluates the current resistance state of the memory cell, and (2) an LSTM-based sequence prediction model that accurately forecasts the optimal programming voltage required to reach the target resistance state. The predicted voltage serves as the initial programming pulse, serving for following step-by-step programming, thus reducing redundant operations. Based on our evaluation, PROMPT achieves a voltage prediction accuracy of 92.04% and reduces programming complexity by 54.75%.