Photolithography, the technique used to create computer chips and optical devices, often results in minuscule deviations from design intentions. With the goal of closing the gap between design and manufacturing, a team of researchers from MIT and the Chinese University of Hong Kong, led by mechanical engineering graduate student Cheng Zheng, used machine learning to create a simulator for a specific photolithography process.
The digital simulator was developed using real data from the photolithography system and can more accurately mimic how the system would build a design. It was integrated into a design framework alongside another simulator emulating the performance of the fabricated device. The pairing allows for more accurate creation of optical devices that better match their design and perform optimally when used.
Despite the challenges of costliness and developing an effective way to build a high-fidelity dataset, the team proved that real data works more efficiently and accurately than simulated data created by analytical equations. The method, named “neural lithography”, uses the physics-based equations paired with a neural network trained on experimental data, learning to compensate for the specific deviations of the photolithography system.
The photolithography simulator consists of an optics model and a resist model, which captures the projection of light and the resulting photochemical reactions on the device’s surface, respectively. This simulator is then connected to another that predicts the device’s performance on specified tasks.
When compared to devices made with other techniques, objects manufactured using their simulator, like a holographic element producing a butterfly image when lit, closer resembled the initial design and showed better image quality.
The researchers plan on improving their algorithms to model more complex devices, and test this technique with consumer cameras and different photolithography system types. The team’s research was supported by the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund, and some of the work was conducted using MIT.nano’s facilities. The research findings will be presented at the SIGGRAPH Asia Conference.
