Macro modeling of liquid crystal cell using machine learning method: reservoir computing approach

Abstract

A macro model of liquid crystal cells including electrical and optical behaviors has been developed using a machine learning framework called reservoir computing and implemented into a circuit simulator. Assuming the arbitrary time steps given from the circuit simulator, we confirmed that our model in which the time-continuous reservoir update equation is discretized by a fourth-order Runge–Kutta method shows high prediction accuracy even at the different time steps from that in the training phase. The director distribution of liquid crystals, which is the microscopic state that realizes the specific macroscopic characteristic, capacitance, and transmittance, is not uniquely determined. Therefore, it is essential to utilize the reservoir’s ability to memorize history to improve prediction accuracy. We found it effective to adjust the parameters that control memory length and update speed according to the response time of each capacitance and transmittance.