Reservoir Computing Model for Liquid Crystal Cell Responsive to Design Parameter and Its Augmentation With Novel Transfer Learning Framework

Macro models of liquid crystal cells that prioritize input-output responses over physical accuracy are high-speed and valuable for large-scale circuit simulations. However, the simplicity of the macro model makes it challenging to achieve accurate responses when design parameters, such as the electrode width, change. In this study, we have developed a macro model of liquid crystal cells that follows changes in design parameters using “reservoir computing (RC),” one of the machine-learning (ML) methods. We have obtained high accuracy with trained and untrained design values by adding a port to provide a design value and by adjusting the hyperparameters that significantly affect prediction accuracy. In addition, we demonstrated that connecting a well-trained model with a specific design value in series or parallel to the reservoir has achieved equivalent prediction ability even when trained with a smaller amount of data. This new transfer learning (TL) approach in RC can be used even when the dimension of the output weight matrix changes, which has not been covered in previous studies. Display designers can use our model in circuit simulators to optimize design values while estimating the impact of design values on optical properties.