Photolithography, a technique used to etch precise features onto surfaces for the creation of computer chips and optical devices, is often inaccurately executed due to tiny deviations during manufacturing. In an attempt to bridge this gap between design and production, a team of researchers from MIT and the Chinese University of Hong Kong have developed a digital simulator using machine learning, which emulates the photolithography manufacturing process. The simulator uses real data from photolithography systems to render a more accurate model of the system’s fabrication design.
Integrated into a design framework alongside another simulator that mimics the performance of the manufactured device in activities such as generating images with computational cameras, these linked simulators permit the user to create an optical device that better aligns with its design, thus improving task performance. This could potentially result in more precise and efficient optical devices for a multitude of applications.
Additionally, given that the learning pipeline for the simulator uses real-world data, it can be applied across various photolithography systems. Despite the expense and complexity of gathering real data and integrating software and hardware successfully, the researchers found that real data worked more efficiently and accurately than data produced by simulators built from analytical equations.
The technique developed by the researchers, termed “neural lithography,” employs physics-based equations as a foundation, and then integrates a neural network that is trained on real, experimental data from the user’s photolithography system. This network consequently learns to account for many of the system’s specific discrepancies.
The simulator has two primary components: an optics model that represents how light is projected onto a device’s surface, and a resist model which describes how the photochemical reaction results in features on this surface. Within a larger framework, the user outlines their desired outcomes and the two simulators work together to show how a design can meet these performance goals.
In future, the researchers aim to enhance their algorithms to better model more complex devices and to also test the system with consumer cameras. They also plan to expand the adaptability of their approach for different types of photolithography systems. The research, backed by the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund, was conducted partly using MIT.nano’s facilities.
