Researchers at MIT and the Chinese University of Hong Kong have developed a machine learning-powered digital simulator for the photolithography process, frequently used in the manufacture of computer chips and optical devices. The team has built a digital simulator that can model the photolithography system based on real-world data, allowing for a greater level of accuracy in spectral modelling. The simulator is then integrated with another digital system that emulates performance in down-stream processes.
This pioneering method has the potential to enable scientists and engineers to create more advanced and efficient optical devices for sectors including mobile technology, augmented reality, medical imaging, entertainment and telecommunications. Previously, aspects such as the high cost of real-world data and technical difficulties in integrating the software meant that the method was seldom explored, as explained by co-lead author of the paper, Cheng Zheng.
This method, known as neural lithography, has been dubbed a “learned simulator”, because it uses a machine learning model to rectify specific pivotal deviations in a system. By using real experimental data from a user’s photolithography system, the necessary deviations can be generated.
The digital simulator consists of two segments – the optical model which captures how light is projected onto a device’s surface, and the resist model which charts how the photochemical reaction creates characteristics on the surface. Adjoined, the researchers believe that these two simulators can guide users to a design that matches their desired performance objectives.
The technique underwent testing with the successful fabrication of a holographic component that creates a butterfly image when illuminated. Other techniques were unable to recreate the near-perfection of this experiment. With consumer camera tests and complex device model tests planned in the near future, the researchers also hope to expand their approach to include other types of photolithography systems, such as those using deep or extreme ultraviolet light.
The research was financially supported by a range of institutions including Fujikura Limited, the U.S. National Institutes of Health, and the Hong Kong Innovation and Technology Fund.
