Flash-CNNCap: Capacitance Extraction via Image Mapping

Researchers proposed Flash-CNNCap, which reframes parasitic capacitance prediction as an image-to-image regression task over contribution maps.

The method reconstructs full capacitance matrices in O(n) passes instead of O(n^2), significantly improving inference efficiency.

A U-Net variant matched or outperformed baselines on CapBench, showing strong accuracy for full-matrix capacitance extraction.

Integrated into a DEF-to-SPEF flow, the approach delivered notable speedups over OpenRCX for EDA workflows.