Machine learning could revolutionize the way we manufacture and test electronics, according to researchers from MIT and Argonne National Laboratory. Integrated circuits, or microchips, are the building blocks of modern electronics, and their continued miniaturization has led to increasingly complex and powerful devices. However, traditional imaging techniques are no longer sufficient for inspecting and testing these small components. One promising method for imaging nanoscale components is synchrotron X-ray ptychographic tomography, but this requires hours or even days to get a single reconstruction. To speed up this process, researchers trained a neural network called APT (Attentional Ptycho-Tomography) to predict accurate reconstructions of the objects in a fraction of the time it would normally take. The network uses regularizing priors in the form of typical patterns found in integrated circuit interiors and the physics of X-ray propagation through the object. The researchers tested their technique on an integrated circuit and were able to capture detailed images in just a few minutes.
The lead author of the research is Iksung Kang, who works at MIT and Argonne National Laboratory.
MIT is a world-renowned research university in Cambridge, Massachusetts, USA. Founded in 1861, it is dedicated to advancing knowledge and educating students in science, technology, and other areas of scholarship that will best serve humanity in the twenty-first century. Argonne National Laboratory is a multidisciplinary science and engineering research center that explores solutions to pressing national problems in science and technology. Located in Lemont, Illinois, it aims to solve the largest scientific and engineering challenges in energy, the environment, and national security.
