Photolithography is a complex process often used in making computer chips and lenses where light is expertly etched onto a surface to create features. However, tiny deviations that occur during the manufacturing process often result in the final product not meeting the initially intended design.
To rectify this, a team of researchers from MIT and the Chinese University of Hong Kong have utilized machine learning to develop a digital simulator that emulates a specific photolithography manufacturing process. The simulator utilizes genuine data gathered from the photolithography system, allowing for a more accurate simulation of the system’s fabrication of a design.
This simulation is incorporated into a design framework, together with another simulator that mimics the performance of the device produced in downstream tasks, such as making images with computational cameras. These connected simulators enable the user to generate an optical device that aligns better with its design and reaches optimal task performance.
By using real-world data, this technique can be utilized in numerous photolithography systems and can aid scientists and engineers in creating more precise and efficient optical devices for uses in numerous industries, including cameras, augmented reality, medical imaging, entertainment, and telecommunications. The improved efficiency of using real-world data over data produced by simulators was underlined by Cheng Zheng, one of the co-lead authors of the research.
Dual simulators are integrated into the technique, with an optics model representing how light is projected onto the device’s surface and a resist model illustrating how the photochemical reaction creates features on the surface. During a downstream task, this digital lithography simulator is connected to a physics-based simulator to predict the fabricated device’s performance. A larger framework allows these systems to demonstrate to the user how to create a design that will meet performance goals.
Their model was tested by producing a holographic element that generated a butterfly image when light shone on it. Compared to previous techniques used, their holographic element produced a more exact result that was closer to the original design. Other devices such as a multilevel diffraction lens also produced a superior image quality.
In future, the team intends to improve their algorithms so that they can model more intricate devices. The system will be tested with consumer cameras, and they aim to expand the approach to use with different types of photolithography systems, including those using deep or extreme ultraviolet light. The research, supported in part by the U.S. National Institutes of Health, was carried out using facilities at MIT.nano.
