Photolithography is a process used to precisely etch features onto a surface, commonly employed in the fabrication of computer chips and optical devices. However, tiny deviations during manufacturing can lead to these devices falling short of their intended design. To address this, researchers from MIT and the Chinese University of Hong Kong have developed a machine learning-based digital simulator that mimics the photolithography manufacturing process, incorporating real-world data to more accurately model fabrication.
This simulator is integrated into a design framework alongside another digital simulator that emulates the performance of the fabricated device in downstream tasks, such as producing images with computational cameras. By connecting these simulators, users can create optical devices that better match their designs and achieve optimal performance.
The application of this technique extends to various fields including mobile cameras, augmented reality, medical imaging, entertainment, and telecommunications, potentially allowing for the creation of more accurate and efficient optical devices. The use of real-world data in the digital simulator's learning pipeline makes it applicable to a wide range of photolithography systems.
The researchers' approach, known as neural lithography, involves using physics-based equations as a foundation for their photolithography simulator and incorporating a neural network trained on real experimental data from a user's photolithography system to compensate for specific deviations. The method involves generating various designs covering a range of feature sizes and shapes, fabricating them using the photolithography system, and using the resulting data to train a neural network for their digital simulator.
The digital simulator consists of two components: an optics model that captures how light is projected onto the device's surface, and a resist model that depicts the photochemical reaction that produces features on the surface. These simulators are part of a larger framework that aids users in designing devices that meet specific performance goals.
Experimental tests demonstrate the effectiveness of this technique, as evidenced by the fabrication of a holographic element and a multilevel diffraction lens, both exhibiting improved performance compared to devices designed using other methods. Future research aims to enhance algorithms to model more complex devices, test the system using consumer cameras, and expand its applicability to different types of photolithography systems.
Funding for this research was provided by the U.S. National Institutes of Health, Fujikura Limited, and the Hong Kong Innovation and Technology Fund, with operations conducted at MIT.nano’s facilities.
