Misalignment measurement is a fundamental but challenging issue in optical lithography. Unfortunately, even with research efforts spanning a few decades, sub-nanoscale misalignment measurement remains a challenge. Through as a promising strategy, the high-accuracy moiré-based misalignment measurement is influenced by unfavorable factors and can only achieve accuracy on the nanoscale with regularly-manufactured micron-scale linear alignment marks. Herein, we propose a one-step deep learning-based misalignment regression measurement strategy. With a trained deep neural network, our scheme achieved sub-nanoscale accuracy with micron-scale linear alignment marks. Compared with existing measurement strategies, it was also robust to fabrication deficiencies, environmental noise, alignment ability, and systematic errors. This scheme holds promise for improving the accuracy of existing misalignment measurements in proximity, x-ray, extreme ultraviolet (EUV), projective and nanoimprint lithographies.

失准测量是光学光刻中的一个基本但具有挑战性的问题。不幸的是，即使经过数十年的研究努力，亚纳米级失准测量仍然是一个挑战。通过作为一种有前途的策略，高精度的基于莫尔条纹的失准测量受到不利因素的影响，并且只能使用常规制造的微米级线性对准标记才能达到纳米级的精度。在本文中，我们提出了一种基于深度学习的失步回归测量策略。通过训练有素的深度神经网络，我们的方案通过微米级线性对准标记实现了亚纳米级的精度。与现有的测量策略相比，它在制造缺陷，环境噪声，对准能力和系统误差方面也很强大。