Recognition and Prediction of Nonlinear Dynamic States of Microelectronic Devices Based on Machine Learning Algorithms

Abstract

The accurate identification and prediction of nonlinear dynamical states of microelectronic devices, especially Duffing vibrator systems, have become particularly important with the development of microelectronics technology. The complex dynamical behaviors of these systems pose a challenge to traditional analysis methods, and machine learning-based approaches provide an efficient and accurate new way to solve this problem. This paper presents a new state recognition system for the Duffing oscillator, designed using the extreme learning machine algorithm. The aim is to address the issues of large computation and limited detection accuracy associated with traditional recognition and detection methods. This paper also constructs a new detection system based on the noise precursor phenomenon of Josephson junctions. The experiment showed that the constructed system had a recognition accuracy of 93.3% for the training set samples, a running time of 70 seconds, and better computational performance than traditional detection methods. The average accuracy of the Josephson junction bifurcation prediction system in detecting multiple bifurcation points was over 91%, with a maximum of 95%, which was more than 10 percentage points higher than traditional methods. The results of this paper have certain value in the field of nonlinear dynamic state detection of microelectronic devices, and can provide technical reference for the study of nonlinear dynamic equations of other devices.