Smarter sensors save time and energy

Imagine a surgical robot that could detect the boundary between a tumor and healthy tissue during an operation; not by sending images offsite for testing, but by quickly analyzing subtle differences fast enough to guide the surgeon’s next move.

Modern optical sensors passively collect data, which is then exported for analysis. Researchers at Texas A&M University are developing optical sensors that can compress and analyze their own data, saving time, money and energy.

Led by Drs. Yuxuan Cosmi Lin and Linda Katehi, the team recently published the framework of their novel sensors, which they call electrochromic hyperspectral embedding (ECHSE). The system shifts intelligence from back-end computer analysis into the sensor itself, allowing for reduction in data transfer while maintaining strong vision performance. 

This project involved several Texas A&M researchers, including postdoctoral researcher Ran Li, recent electrical engineering graduate Dr. Chaoyi He, and Ph.D. student Enzi Zhai.

For everyday users, this technology could lead to smaller, faster and more energy-efficient intelligent sensors. These systems could enable real-time decision-making without reliance on cloud computing or large AI chips. 

Another potential application of the in-sensor technology is space exploration. Currently, lunar rovers are used to collect images of moon rock samples, which are sent back to Earth for composition analysis. Using smart sensors, astronauts could analyze a sample on the moon in real time, determining if an area contains ice or rare minerals.

The team hopes to develop a general working principle for in-sensor intelligence, in which researchers would build sensors that can adaptively track useful information in real time. They also plan to expand the type of data the smart sensors can capture by expanding the device to differentiate motion and wavelength ranges. 

“We are interested in technology translation. We want to explore simple fabrication approaches that could make these systems low-cost, scalable and compatible with industrial manufacturing,” said Lin, an assistant professor of materials science and engineering. “The long-term goal is to build task-specific intelligent sensing systems that are not only powerful in the lab, but also practical for real-world deployment.”

Developing this type of sensor requires engineers from across many disciplines. With contributors from Texas A&M’s Department of Materials Science and Engineering and Department of Electrical and Computer Engineering, the team was able to benefit from the College of Engineering’s resources and opportunities that allowed these different fields to come together. 

“This project opened a new world for me. It pushed me to learn across many different fields, from electrochromic materials and photodetectors to machine learning and hardware systems,” said Li. “More importantly, Dr. Lin taught me what it really means to understand something; not just to get a result, but to connect the mechanism, the evidence, and the meaning behind it.”

External collaborators on this project include Dr. Yi Huang and Professor Qiangfei Xia from the University of Massachusetts Amherst, Professor Vinod K. Sangwan and Professor Mark C. Hersam from Northwestern University, Professor Xi Ling from Boston University, Professor Xu Zhang from Carnegie Mellon University, and Dr. Helen Xie from Google X.

By Alyssa Schaechinger, Texas A&M University College of Engineering

###