TSOM method transforms conventional optical microscopes to truly 3D metrology tools for nanoscale to microscale dimensional analysis with potentially sub-nanometer scale resolution. The method can be used in both reflection and transmission modes of microscopes. It is applicable to a wide variety of target materials ranging from transparent to opaque, and shapes ranging from simple nanoparticles to complex semiconductor memory structures, including buried structures under transparent films.
We use TSOM images to inspect defects on semiconductors through comparing images with various methods. The following methods are required to achieve better results. First, do registration all images which gained from one semiconductor with through-focus scanned with optical microscopy. Second, denoise registered images to get fine images and it prevents misunderstanding noise as defects. We applied other techniques to develop process by reducing cost for compute. In 2-D image, we can find defects more easily by kernel based defect detection method. Then, we take a template around defects and find the similar templates in other images using fast normalized cross correlation. It means we don't have to inspect whole images which costs a lot. To compare TSOM images, we also need to do align images since we extract images in different 3-D cubes, which stacked with through-focus scanned images.
Using upon methods, we compare different semiconductors with given through-focus scanned images. Finally, difference between TSOM images can be achieved by subtracting each other.
In these days, we are trying to use deep learning methods for original TSOM images or 3D cubes.