Near-Field Radiation Prediction for Neuromorphic Chip Based on Deep Learning

Abstract

This article proposes an enhanced equivalent dipole model, integrating convolutional neural networks and fully connected networks, to predict near-field magnetic fields in neuromorphic chips. Traditional dipole models struggle with complex electromagnetic interactions, such as multiple reflections and diffractions, which occur in dense neuromorphic circuits. To address these challenges, the proposed method constructs a coefficient matrix based on the spatial relationships between scanning points and dipoles. This matrix is used as input to the neural network, which predicts the radiation field magnitude as output. The model is trained to accurately predict magnetic fields beyond the scanned area. The approach is validated through both numerical simulations and experimental measurements, showing a relative error of approximately 5% between predicted and measured values, indicating high accuracy.