Researchers from MIT and the Chinese University of Hong Kong are using machine learning to close the gap between design and manufacturing processes in photolithography – a method used in the creation of computer chips and optical devices. Photolithography involves using light to etch features onto a surface. However, tiny variations during production often lead to the final product not matching up to the original design.
The team created a digital simulator, using real data from the photolithography process, allowing it to more precisely predict how a design would be fabricated. The simulator is part of a larger design framework, which includes another simulator that imitates the performance of the end product, providing a tool for designers to create an optical device that closely aligns with its original design and performs optimally.
This process, named “neural lithography”, could improve the accuracy and efficiency of optical devices, such as mobile cameras and medical imaging equipment. Furthermore, the method can be applied to a range of photolithography systems due to its use of real-world data.
Despite the significant costs and challenges involved in obtaining real data, MIT graduate student Cheng Zheng argues it is vital to future success. The research team also created an optics model showing how light is projected on the surface of the device, and a resist model outlining how the photochemical reaction occurs to fabricate features on the surface.
The two models function simultaneously within a large framework that instructs the user on how to create a design reaching maximum performance levels. Graduate student Guangyuan Zhao adds that devices fabricated in this way have the best possible performance level, even when compared to those produced by other methods.
The team used their technique to construct a holographic element that creates a butterfly image when exposed to light. The result was a near-perfect butterfly more closely aligning with the original design, highlighting the success of their approach. Similarly, their multilevel diffraction lens displayed superior image quality.
Looking ahead, the team wants to refine their technology to model more intricate devices and trial the system using consumer cameras. The research was supported by the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund, and utilized MIT.nano’s facilities.
