INFORMATION TECHNOLOGY

Machine learning accelerates nanoscale 3D X-ray imaging


Guide

Researchers at MIT and Argonne National Laboratory have developed a machine learning-based method (APT) that can accelerate non-invasive imaging of nanoscale 3D objects using synchrotron X-rays. APT utilizes regularization priors and X-ray propagation principles to produce accurate IC reconstructions with fewer scan angles, reducing data acquisition and calculation time by nearly 100 times without compromising quality. The technology has potential applications in materials science and biology.

Research background

Three-dimensional X-ray imaging enables non-invasive monitoring of the interior of objects with nanoscale resolution, and integrated circuits are important to check their manufacturing integrity due to their highly regular and diverse geometric properties. The high penetration depth of X-rays helps to recover information inside the sample under angular sampling schemes. The Ptychographic X-ray Computed Tomography (PXCT) protocol proposed by Iksung Kang et al. combines this property with translational scanning for lensless high spatial resolution to determine nanoscale details inside the volume of biological samples. Using this technique, the researchers demonstrated nondestructive imaging of integrated circuits manufactured using 22nm technology at 14.6nm resolution. PXCT provides non-destructive inspection of manufactured samples, eliminating the restrictive need for limited depth access caused by electron scattering in STEM, such as destructive measures such as delayering. This allows the plant to connect to a synchrotron radiation X-ray source and perform quality assurance without the need for destructive measures. However, the limitation of this technique is that it requires two types of scanning: angle and translation, and it does not expand well as the volume of the object increases. A new X-ray microscope called Velociprobe uses the flight-scanning ptychography to significantly reduce data acquisition time. However, for a typical 100×100×5 mm³ IC, the total data acquisition and reconstruction time is estimated to be more than two months. This work proposes a machine learning framework designed to reduce data acquisition and computation time for IC reconstruction under X-ray ptycho-tomographic geometry (Figure 1), thereby providing a non-invasive and efficient solution for inspection purposes.

Research innovation

This work mainly discusses the effectiveness of training APT using gold-standard reconstruction of a single sample. The authors focus on two issues: 1) how to ensure the accuracy of the gold standard, and how to deal with the gap with physical samples; 2) Whether the APT is overtrained on a specific sample. Here, a machine learning framework is proposed that can provide reliable IC reconstruction with sampling angle density N∗ ∼ 29 and total sampling angle range q∗ ∼ ±17?, with a large reduction in number and total angle range, ×12 and ×4.2, respectively. For the reconstruction of integrated circuit chips in the test volume (4.48×93.2×3.92mm3), data acquisition and reconstruction based on this machine learning framework takes only 38 minutes. These improvements equate to an overall (acquisition plus computation) time reduction of nearly ×108 compared to the current latest iterative reconstruction methods. This method can be applied to a variety of physical systems that mathematically represent forward models. In addition, this method is not limited to specific sample geometries and can be extended to other types of samples.

Figure 1 Implementation of X-ray ptycho-tomography and APT

The article was recently published in Light: Science & Applications in the top international academic journal “Attentional Ptycho-Tomography (APT) for three-dimensional nanoscale X-ray imaging with minimal data acquisition and computation.” Iksung Kang is the first author and Professor George Barbastathis is the corresponding author. (Source: LightScience Applications WeChat public account)

Related paper information:https://www.nature.com/articles/s41377‍-023-0‍1181-8

Special statement: This article is reproduced only for the need to disseminate information, and does not mean to represent the views of this website or confirm the authenticity of its content; If other media, websites or individuals reprint and use from this website, they must retain the “source” indicated on this website and bear their own legal responsibilities such as copyright; If the author does not wish to be reprinted or contact the reprint fee, please contact us.



Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button