J. Senthilnath, C. Hao, N. Lettsome, F. Cawthorne, T. Venkatesan, and F. C. Wellstood, "Inferring wire length and depth from magnetic field images via deep-spatial physics informed model," Proc. 27th IEEE Electronics Packaging Technology Conf. (EPTC’25), 2025
Abstract:
In this paper, we propose a novel method called the Deep-Spatial Physics Informed Model (D-SPIM), which combines a deep learning-based convolutional neural network (CNN) with a spatial physics model to convert magnetic images of current-carrying circuits into current density images. The D-SPIM framework consists of two key components: i) CNN models that classify wire length and predict the ratio of wire length to wire depth (ℓ/z), and ii) a spatial physics model that determines the depth z and employs the Biot-Savart law with FFT, using the predicted z value for accurate image conversion. The resultant current density images provide valuable insights into the spatial distribution of current.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the Agency for Science, Technology and Research - Industry Alignment Fund – Pre-Positioning
Grant Reference no. : M23K8a0050