Researchers at MIT and the Chinese University of Hong Kong have developed a machine learning model to close the gap between design and manufacturing in the field of photolithography. The technique, which involves manipulating light to etch onto surfaces, sees use in the creation of computer chips and optical devices but often falls short of intentions due to minor deviations during the process. Utilising real data from the photolithography system, the team have created a digital simulator to more accurately predict the fabrication of a design.
By incorporating this simulator into the design framework alongside another digital emulator, users can produce an optical device that effectively fulfils its intended purpose. The potential areas of benefit from these advancements range from mobile cameras and augmented reality to medical imaging and telecommunications. And given the model employs real-world data, it could feasibly be applied to various photolithography systems.
However, the process is not without its challenges. Building a high-fidelity dataset capable of effective coordination between software and hardware required extensive exploration and the development of tailor-made tools and data-exploration methods. Despite the substantial resource investment and initial difficulty, the results far outweighed those obtained from simulators based on analytical equations.
Named “neural lithography,” this method combines physics-based equations into a neural network trained on real-world, experimental data from a user’s photolithography system. It then uses this to compensate for specific deviations of the system. By creating broad design ranges, the system can learn to tweak the discrepancies between the final structure and the design intent.
Further, two separate components make up the digital lithography simulator, an optics model and a photochemical reaction model. The system then ties these into a larger framework that presents the user with a design pathway to reach their targeted results.
Upon testing, the system proved successful in the fabrication of a holographic element that generated a near-perfect butterfly image when illuminated. Additionally, it can enhance image quality in multilevel diffraction lenses. Future plans include refining the algorithms to model more complex devices while testing it on consumer cameras and other photolithography systems.
The research, funded in part by the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund, will be presented at the SIGGRAPH Asia Conference.
