Changhao Wang, S. Yuan, Nicolò Bellarmino, Danyang Chen, Nima Kolahimahmoudi, Honghao Wang, Hanzhi Xun, Xiuyan Li, Lin Wang, Chujun Yin, Chaobo Li, Mottaqiallah Taouil, Moritz Fieback, Said Hamdioui, Giovanni Squillero, Riccardo Cantoro
Physics-based compact models for emerging non-volatile memories (NVMs) are often limited by the com plex interactions of microscopic domains and defects that are difficult to capture analytically, resulting in reduced ac curacy and simulation efficiency. To address this challenge, a machine learning (ML)-based approach is proposed using artificial neural networks (ANNs) trained entirely on device measurement data, enabling a direct translation of fabrica tion characteristics into SPICE-compatible circuit models. The resulting models achieve high accuracy (MSE: 0.724, adjusted R2: 0.998), significantly outperforming physics- based baselines with an 18× lower MSE for polarization and a two-order-of-magnitude precision improvement in FeFET current simulation, while accurately capturing the wake-up process. Furthermore, the model demonstrates robust out- of-distribution (OOD) extrapolation to unseen ferroelectric thicknesses and a 33.7% improvement in simulation speed. These results validate the ML-based approach as a highly efficient, SPICE-compatible solution for next-generation memory.