Training procedure for scanning electron microscope 3D surface reconstruction using unsupervised domain adaptation with simulated data

Accurate metrology techniques for semiconductor devices are indispensable controlling the manufacturing process. For instance, dimensions of a transistor’s current channel (fin) an important indicator device’s performance regarding switching voltages and parasitic capacities. We expand upon traditional 2D analysis by utilizing computer vision full-surface reconstruction. propose data-driven approach that predicts dimensions, height width (CD) values, fin-like structures. During operation, method solely requires experimental images from scanning electron microscope patterns concerned. introduce unsupervised domain adaptation step to overcome gap between simulated data. Our model is further fine-tuned with measurement second scatterometry sensor optimized through tailored training scheme optimal performance. The proposed results in accurate depth predictions, namely 100% interwafer classification root-mean-squared error 0.67 nm. R2 intrawafer on 0.59 0.70. Qualitative also indicate detailed surface features, such as corners, accurately predicted. study shows z-metrology can be viable high-volume manufacturing.