The researchers integrate this simulator into a design framework, along with another digital simulator that emulates the performance of the fabricated device in subsequent tasks, such as producing images with computer cameras. These interconnected simulators allow the user to fabricate an optical device that better matches their design and achieves optimal performance in the task.
This technique could help scientists and engineers create more accurate and efficient optical devices for applications such as mobile cameras, augmented reality, medical imaging, entertainment, and telecommunications. And because the digital simulator's learning process uses real-world data, it can be applied to a wide range of photolithographic systems.
“This idea seems simple, but the reasons it hasn’t been attempted before are that real-world data can be expensive and there’s no precedent for how to effectively coordinate software and hardware to build a high-fidelity dataset,” explains Cheng Zheng, a mechanical engineering graduate student and co-lead author of an open-access paper describing the work. “We took risks and did extensive exploration—for example, developing and testing characterization tools and data exploration strategies—to determine a scheme that works. The result is surprisingly good, demonstrating that real-world data performs much more effectively and accurately than data generated by simulators composed of analytical equations. While it can be expensive and you might feel a bit lost at first, it’s well worth the effort.”.
Zheng co-authored the article with Guangyuan Zhao, a lead co-author and graduate student at the Chinese University of Hong Kong, and their advisor, Peter T. So, a professor of mechanical and biological engineering at MIT. The research will be presented at the SIGGRAPH Asia Conference.
Print with light
Photolithography involves projecting a light pattern onto a surface, which triggers a chemical reaction that etches the features into the substrate. However, the finished device ends up with a slightly different pattern due to minute deviations in light diffraction and small variations in the chemical reaction.
Because photolithography is complex and difficult to model, many existing design approaches rely on equations derived from physics. These general equations provide some insight into the manufacturing process, but they cannot capture all the specific variations of a photolithographic system. This can lead to devices underperforming in real-world applications.
For their technique, which they call neural lithography, the MIT researchers build their photolithography simulator using physics-based equations as a foundation, and then incorporate a neural network trained with real experimental data from a user's photolithography system. This neural network, a type of machine learning model loosely based on the human brain, learns to compensate for many of the system's specific deviations.
The researchers collect data for their method by generating many designs that cover a wide range of feature sizes and shapes, which they then manufacture using the photolithography system. They measure the final structures and compare them to the design specifications, matching that data and using it to train a neural network for their digital simulator.
"The performance of learned simulators depends on the data input, and data generated artificially from equations cannot cover real-world deviations, which is why it is important to have real-world data," Zheng says.
Dual simulators
The digital lithography simulator consists of two separate components: an optical model that captures how light is projected onto the surface of the device, and a resistance model that shows how the photochemical reaction occurs to produce features on the surface.
In a subsequent task, they connect this learned photolithography simulator to a physics-based simulator that predicts how the fabricated device will behave—for example, how a diffractive lens will diffract the light falling on it.
The user specifies the desired results for the device. These two simulators then work together within a broader framework that shows the user how to create a design that achieves those performance goals.
"With our simulator, the manufactured object can achieve the best possible performance in a subsequent task, such as computational cameras, a promising technology for making future cameras miniaturized and more powerful. We demonstrated that even if post-calibration is used to try to obtain a better result, it will still not be as good as having our photolithographic model in the loop," Zhao adds.
They tested this technique by fabricating a holographic element that generates a butterfly image when light strikes it. Compared to devices designed using other techniques, their holographic element produced a nearly perfect butterfly that better matched the design. They also fabricated a multilevel diffraction lens with better image quality than other devices.
In the future, the researchers want to refine their algorithms to model more complex devices and test the system with consumer cameras. They also want to broaden its scope so it can be used with different types of photolithographic systems, such as those employing deep or extreme ultraviolet light.
This research is funded, in part, by the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund.
###
