Photolithography, the process of using light to precisely etch designs onto a surface, is a primary method for creating computer chips and optical devices. However, it’s common for slight deviations during manufacturing to cause the final product to diverge from the intended design. To bridge this gap, researchers from MIT and the Chinese University of Hong Kong employed machine learning to construct a digital photolithography simulator based on real data from the process, promising higher accuracy in replicating how a design would be fabricated.
The simulator has been integrated into a design framework along with another that simulates the performance of the final device in various tasks, such as image production with computational cameras. Users may use these interconnected simulators to create devices more closely aligned with the original design goals and optimal task performance.
Applications span mobile cameras, augmented reality, medical imaging, entertainment, and telecommunications. Given its foundation on real-world data, the technology can potentially be adapted to various photolithography systems. However, collecting this data can be challenging and costly, and it’s been uncharted territory when it comes to balancing software and hardware for high-quality dataset building, according to Cheng Zheng, co-lead author of the paper detailing the research.
The researchers’ system, dubbed neural lithography, combines physics-based equations with a neural network trained on actual data from the photolithography process. This combination helps to compensate for the system-specific discrepancies that physics equations alone can’t account for. They compiled this data by fabricating numerous designs varying in feature sizes and shapes, comparing the end structures with the planned specifications, and feeding these pairings into their neural network.
Their digital simulater integrates an optics model, illustrating how light projects on the device’s surface, and a resist model, showing how photochemical reactions create surface features. Users can specify desired outcomes, and the simulators will show how to achieve it in the design phase. The method was successfully tested by producing a holographic element that created a near-perfect butterfly image when light was shone upon it.
Looking ahead, the researchers aim to finetune their algorithms to facilitate modeling of more complex devices, test the system with consumer cameras, and adapt their approach to different types of photolithography systems that use deep or extreme ultraviolet light. Funding for the research was provided in part by the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund. Some work was conducted using facilities at MIT.nano.
