[2025] Design custom chips to accelerate scanning electron microscope (SEM) machine image processing (with ASML)

Next-generation scanning electron microscope (SEM) machines will be generating enormous amounts of data (petabytes per day). This data has to be processed in an efficient manner (both in terms of cost as well as energy).

The SEM machines are used for defect inspection, i.e. finding defects that occur during the chip manufacturing process. Any such defect can make a chip worthless, and wafers need to be checked during the production process so that errors can be detected, allowing to improve or stop the process.

This defect inspection works by comparing SEM images to reference data. Algorithms range from simple denoising with difference detection to small convolutional neural networks (more algorithms can be explored). Pre-processing is needed to align different images, typically including an FFT.

In this project, you will design an efficient chip that does these steps, and is still reprogrammable to support long-term updates. The processing can be parallelized across many different beams, so small chips that process only part of the data are preferred.

This project is a collaboration with ASML and is supported by a stipend.

Contact

Last modified: 2025-02-10