Photolithography, the process of manipulating light to etch features on to a surface, is crucial in making computer chips and optical devices. However, the performance of devices made using this process often falls short of their original designs due to minute deviations during manufacturing. To address this design-to-manufacturing gap, researchers from MIT and the Chinese University of Hong Kong have used machine learning to develop a digital simulator that can predict and correct these deviations.
The digital simulator, which the researchers call “neural lithography,” mimics the photolithography manufacturing process and takes into account real-world data from the system, enabling it to better simulate how the system fabricates a design. It is then incorporated into a design framework along with another simulator, portraying the performance of the fabricated device in downstream tasks, such as taking images with computational cameras. This method allows the creation of optical devices that more closely match their design and deliver optimal task performance.
The importance of using real-world data was emphasized by Cheng Zheng, a mechanical engineering graduate student at MIT. Zheng clarified that real data works more efficiently and precisely than data generated by simulators composed of analytical equations. These findings demonstrate this model’s potential for fabricating more accurate and efficient optical devices.
The digital simulator consists of two main components – an optics model, which shows how light is projected onto a device, and a resist model, which illustrates how the light-triggered photochemical reaction produces features on a surface. In usage, these models work together within a framework to guide the user in creating a design that meets their performance objectives.
The researchers tested this technique by manufacturing a holographic element that generated a butterfly image when illuminated. Their holographic element yielded a nearly flawless butterfly image that was closely aligned with the original design, compared to devices designed using previous methods.
The researchers plan to further refine their algorithms to model more complex devices and to apply the system to consumer cameras. They also aim to expand their approach to different types of photolithography systems, such as those utilizing deep or extreme ultraviolet light.
This project has been supported by the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund, and partially conducted using MIT.nano’s facilities. The research will be presented at the SIGGRAPH Asia Conference.
