A Variability-Aware Data Preprocessing Method for Data-Driven Memristive Device Inverse Modeling

Data-driven learning methods have great potential in optimizing memristive devices, yet its application in memristor inverse modeling is still facing big challenge due to the scarcity of experimental data and device randomness. Here, we proposes a novel data preprocessing method including data generation, data augmentation and statistical preprocessing. With the aid of above techniques, we manage to achieve less than 5 % relative errors in the test regression problem just need a few original I-V curves. Comparing with relevant studies, this work just needs a few of original experimental or simulated I-V curves which is achievable under laboratory conditions.