Computational lithography optimizes complex, curved geometric patterns on photomasks for full-chip UV semiconductor manufacturing

https://www.eurekalert.org/news-releases/1136380

"curvilinear masks essential for pushing chip manufacturing down to advanced angstrom-era nodes, but limited by excessive computing costs/ memory... overcome using streamlined deep-learning surrogate model (tunable U-Net) trained on structural data predicting wafer-imaging patterns rapidly... pairs AI with slice-based gradient calculation eliminating need to store massive 3D data fields... validated on sub-20-nm target features... high -yield/ -NA EUV lithography for high-performance logic/ memory silicon chips"