MIE’s Xian Du and Students Awarded U.S. Patent for Breakthrough in High-speed Autofocus Control

Content

The quality control of microelectronic pattern production is based largely on the real-time inline inspection. However, current state-of-the-art autofocus technologies often create manufacturing bottlenecks due to inefficient and computationally heavy methods. Addressing this crucial bottleneck, Associate Professor Xian Du of the UMass Amherst Mechanical and Industrial Engineering (MIE) Department, along with two student collaborators, has been awarded U.S. Patent Number 12242169 for a trailblazing, high-performance, autofocus technology. 

The newly patented  autofocus method dramatically reduces the number of required input images and reduces the computational complexity, significantly enabling faster image-processing. This method efficiently improves the speed and precision of inline inspection in advanced manufacturing.

In addition to Du, the patent credits two co-inventors: Yingyang Yan, an MIE postdoctoral associate and lab manager in Du’s Intelligent Sensing Lab; and Peter DiMeo, a former undergraduate researcher in Du’s lab and currently a research and development software engineer at MathWorks in Natick, Massachusetts.

Du’s current research focuses on health monitoring and quality control of advanced manufacturing processes through the integration of physics-informed machine learning, in-situ metrology, and multi-modal sensing. 

The backstory of the new patent is that the inline inspection of manufacturing processes – meaning real-time monitoring and metrology for the roll-to-roll continuous printing of micro- and nano-scale flexible electronics – often relies on very speedy and accurate image-acquisition and image-processing techniques to facilitate the measurement of microelectronic pattern production. The problem is that many of these techniques are not sufficient for industrial inspection due to their lack of fast autofocus for real-time application.

According to the patent proposal, “These techniques are either too slow for real-time imaging applications, are extremely computationally expensive, or require training a model that is specific to the imaging configuration.” 

The solution to these issues involves two aspects. One is the “fast and accurate autofocus control using Gaussian standard deviation and gradient-based binning.” The second aspect is to “determine a control input for the piezoelectric-controlled-motion stage using a long short-term memory (LSTM) backpropagation network trained to minimize a cost function over a defined prediction horizon.” This twofold solution is the basis for Du’s new patent.

According to the patent application, “Instead of iteratively searching for the optimal focus using an optimization process, the proposed algorithm can directly calculate the mean of the Gaussian-shaped, focus-measure curve to find the optimal focus location and uses the focus-measure-curve standard deviation to adapt the motion step size.” 

In another aspect, the solution “uses LSTM to directly link the open-loop, polyelectrolyte-membrane, input voltage to the response variable and subsequently uses a single-model-predictive controller for complete system control,” which avoids the separate, time-consuming, internal, closed-loop control of the precision-motion stage.

The technology is currently being commercialized through IntelliRes LLC, a UMass spin-off founded by Du, with the goal of delivering intelligent sensing and imaging solutions to industry and the broader market.

The three inventors conclude that the resulting technology is “a highly efficient method for image-based autofocus,” and its utility extends beyond flexible microelectronics. The technology is also ideal for applications that include the imaging and metrology of biological samples or scenarios in which non-image-based autofocus methods pose risks to damaging heat- or photo-sensitive biological samples. 

Whether used in manufacturing, biomedical imaging, or precision metrology, this autofocus technology stands to make a major impact. Its ability to provide real-time adaptive focusing opens new doors for high-throughput, high-accuracy, inspection systems in both research and industry. See videos about autofocus technology: https://youtu.be/JsnkZCogaYE?feature=shared; https://youtu.be/mCBgKQHWmFc?feature=shared. (April 2025)