Inverse Neural Network Modeling for System-Level Electromagnetic Radiation Prediction in Nanoelectronic Platforms

In older to solve the problem of system-level electromagnetic compatibility in multi-equipment stack, a novel method for electromagnetic interference (EMI) prediction based on neural network (NN) inverse model was proposed. First of all, the single device emitters are measured, and the system-level emitters are simulated by the principle of noise source emitters equivalence, and the training sample set is obtained. Secondly, the radiation intensity, frequency and coordinates of each equipment are selected as the input parameters, and the system-level electric field radiation emission is selected as the output parameters, based on the levenberg-marquardt (LM) algorithm, the reverse model of the three-layer back propagation (BP) neural network is established, finally, the output of the neural network is calculated by combining numerical algorithms such as the trial and conjugate gradient method. The results show that the validation error of the model is obviously improved compared with the traditional three-layer LM-BP neural network, the verification error is reduced from 0.4159% to 0.0997% by using the neural network inverse model with conjugate gradient method solution. This method not only does not depend on the complex neural network structure, but also improves the model precision under the limited training data scale, it provides a new and efficient solution for the EMC evaluation of electronic information platforms such as chips, system integration, EMI challenges in compact nanoelectronic circuits.