Fig. 6: Demonstration of waveform classification and time-series prediction.
From: Analog reservoir computing via ferroelectric mixed phase boundary transistors
a Schematic of the MPBTFT-based ARC system utilizing the masking process. Following the masking process, the virtual node states created by each volatile DG MPBTFT are fed into the readout network and generate outputs. The time interval (τ) indicates the total duration of each input pulse train consisting of M timeframes. The weights (wout) of the readout network are trained through a linear regression method. b Inputs and c classification results of sine and square waveforms obtained from the MPBTFT-based ARC system. Each waveform is composed of eight-time steps. The input data with the initial 300-time steps are used for training, while the rest are used for testing. d Time-series prediction results obtained from the MPBTFT-based ARC system. The initial 300 data points are used for training, while the rest are used for testing. The system successfully predicts the time-series dataset, achieving a low NRMSE of 0.035. e Two-dimensional (2D) representation of the Hénon map. The results exhibit excellent consistency between the target and predicted values. f Results of complex real-world predictive task: the number of confirmed COVID-19 cases prediction. The black and red lines represent the actual and predicted number of confirmed cases, respectively. The data from January 2020 to February 2022 are used for training, while the rest are used for testing. The MPBTFT-based ARC system demonstrates an ability to forecast the future number of confirmed COVID-19 cases by leveraging historical data on previously confirmed cases.
