查看文章

A two-tier multivariate-kriging interpolation method is proposed as a computationally efficient approach to model hafnia-based FeFET devices. We investigate how well this data-driven approach captures device-to-device variabilities when applied to realistic datasets sampled from a physics-based compact model with artificial variance. The framework provides methods for future analysis of experimental data, as well as selection of device operating conditions with the aim of unveiling FeFET strengths and weaknesses and optimizing these synaptic devices for neuromorphic circuit integration.