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

Abstract

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

An In-Depth Analysis of Domain Adaptation in Computer and Robotic Vision

“…[175] Presents a model that works upon generative adversarial networks for semi-supervised learning Can improve performance by learning from both labeled and unlabeled data [176] Introduces a new method for learning invariant features for domain adaptation Can improve performance by learning representations that are invariant to domain shift [177] Introduces a new method for domain adaptation that combines invariant representations and self-supervised learning Can improve performance by learning representations that are invariant to domain shift and using self-supervised learning to learn features that are transferable to new domains [178] Presents a model that works upon meta-learning for transferable features Can improve performance by learning features that are transferable to new domains [179] Presents a model that works upon multi-task learning and attention Can improve performance by using multi-task learning and attention to learn representations that are invariant to domain shift [180] Presents a model that works upon multi-task learning for few-shot image classification Can improve performance by learning multiple tasks with few examples [181] Presents a model that works upon patch-level self-supervised learning Can improve performance by using patch-level self-supervised learning to learn features that are invariant to domain shift [182] Presents a model that works upon self-supervised contrastive learning Can improve performance by learning representations that are invariant to domain shift [183] Presents a model that works upon self-supervised contrastive learning Can improve performance by learning representations that are invariant to domain shift [184] Presents a model that works upon self-supervised learning and synthetic data Can improve performance by using self-supervised learning to learn features that are invariant to domain shift and by generating synthetic data that are like the target domain [185] Presents a model that works upon synthetic data and domain-invariant feature aggregation…”