Photolithography, a technique for fabricating computer chips and optical devices, frequently encounters problems due to minute deviations during the manufacturing process. To address this, scientists from MIT and the Chinese University of Hong Kong have successfully used machine learning to build a digital simulator that effectively mimics certain photolithography manufacturing processes. The simulator, which utilizes real-life data from the photolithography system, more accurately models how a system would construct a product.
The team combined this simulator with another that emulates the performance of a device’s downstream tasks – for instance, generating images via computational cameras. Together, these connected simulators allow users to develop an optical device that adheres closely to its design outline and performs tasks to an optimal level. This new approach could significantly enhance the design and creation of optical devices for several applications, such as mobile cameras, augmented reality, entertainment, and telecommunications.
Regardless of this, getting real-world data can be costly and there is no set precedent for coordinating hardware and software to build a high-fidelity data set. However, the benefits of real data yield precise and efficient results, surpassing those generated by simulators that primarily use analytical equations.
This research, described as neural lithography, forms the photolithography simulator from physics-based equations, incorporating a neural network trained on empirical data from a user’s photolithography system. It can compensate for many of the system’s specific deviations. This new simulator comprises two separate parts: an optics model and a resist model.
The user provides desired outcomes for the device, and the simulator guides the user to create a design that meets those objectives. One example of its effectiveness showed when they fabricated a holographic element that produced a near-perfect butterfly image. Future plans involve refining their algorithms to model more complex appliances and testing the system with consumer cameras.
The research received support from the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund, and was conducted at MIT.nano’s facilities. They are due to present the research at the SIGGRAPH Asia Conference.
