Neural Lithography: How Machine Learning Can Improve Photolithography
Researchers from MIT and the Chinese University of Hong Kong are using machine learning to close the gap between design and manufacturing of optical devices. By utilizing real data from a photolithography system, they have developed a digital simulator that accurately models the manufacturing process and enables users to produce optical devices that better match their designs and reach better task performance.
The applications for this technology are vast, ranging from mobile cameras and medical imaging to entertainment and telecommunications. This new technique can help scientists and engineers create more accurate and efficient optical devices by utilizing real-world data, and the research promises to have a significant impact on the industry.
Photolithography involves projecting light onto a surface to etch features. However, tiny deviations during the manufacturing process can cause the fabricated devices to fall short of their intended design. Many existing design approaches rely on equations derived from physics to model the photolithography process, but these equations cannot capture all the unique deviations specific to a photolithography system.
The researchers’ technique, called neural lithography, involves using physics-based equations as a foundation and incorporating a neural network trained on real, experimental data from a user’s photolithography system. This neural network learns to compensate for many of the system’s specific deviations and increase the accuracy of the digital simulator.
The digital lithography simulator consists of two separate components: an optics model and a resist model. These two simulators work together within a larger framework to help users achieve the best performance from their fabricated objects on downstream tasks. The researchers have already tested their technique by fabricating a holographic element and a multilevel diffraction lens, both of which exceeded the performance of devices designed using other techniques.
Moving forward, the researchers aim to enhance their algorithms to model more complicated devices and test the system using consumer cameras. They also plan to expand their approach to be used with different types of photolithography systems. This promising research is supported in part by the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund. Overall, it shows the potential for machine learning to revolutionize the design and manufacturing of optical devices.