Answers are grounded in this paper and cite their sources.
Iterative process optimization in two-photon lithography (TPL) is fundamentally constrained by a critical metrology bottleneck: the reliance on slow, resource-intensive, and destructive scanning electron microscopy (SEM) for structural verification. To overcome this bottleneck, we introduce a resource-efficient inspection framework that provides high-fidelity fabrication feedback using only rapid, in-situ optical imaging. By leveraging cross-modal machine learning to transfer high-resolution metrological knowledge from SEM to optical microscopy, our approach significantly reduces the need for high-resolution microscopy and the time-consuming post-illumination processing steps during the optimization loop. We implement a dual-scale inspection strategy applied to grayscale projection TPL, which is a high-throughput layer-by-layer printing technique. At the global scale, a structure-level model provides highly reliable screening of complete samples, achieving 95% grouped accuracy and 100% within-one-score accuracy across an eight-level ordinal scale. Complementing this, an area-level model enables the granular detection of localized polymerization deviations within specific geometric sectors with 88% accuracy. We demonstrate the framework’s adaptability through transfer learning, where pre-training on cylindrical structures reduced the mean absolute error on unseen 3D woodpiles by 79%. Ultimately, this rapid cross-modal inspection approach enables high-fidelity structural assessment using only optical microscopy, eliminating routine SEM inspection and providing a scalable pathway for quality assessment in nanoscale 3D printing.
No takes yet. Share an insight, caveat, or question.
Ghosh et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: