"Reading the Direction of Light to Reconfigure Operation" Overcoming the Limitations of Autonomous Vehicle 'Vision'

A next-generation polarization sensor capable of reading the direction of light and autonomously reconfiguring its operation has been developed. Conventional image sensors rely solely on light intensity (detection) and have limitations in precisely identifying the orientation and surface structure of objects. This is also why autonomous vehicles have struggled to distinguish between water and asphalt on nighttime roads using current sensors. In contrast, the newly developed sensor stands out by utilizing polarization information—an intrinsic property of light that vibrates in a specific direction—to autonomously find optimal operating conditions and adjust its function accordingly.

Polarized Artificial Intelligence (AI) Sensor Platform Experimental Image (AI Generated). KAIST

KAIST announced on May 12 that the research team led by Professor Junki Suh from the Department of Biological Chemistry and Engineering has developed a self-reconfigurable polarization sensor array technology based on this principle.

To overcome the limitations of existing sensors, the research team devised a polarization-based sensor technology capable of simultaneously recognizing the vibration direction of light. They also implemented a heterostructure combining tellurium (Te) and rhenium disulfide (ReS₂), two different materials, effectively leveraging the property that their response to light varies depending on the crystal orientation.

For the precise stacking of these two materials in an intersecting manner, epitaxial atomic layer deposition was employed. The core of this deposition method is a process that precisely stacks materials one atomic layer at a time to control the crystal structure. According to the research team, this allowed the crystal structures of the two materials to align exactly, resulting in higher reproducibility and stable performance compared to conventional approaches.

This structure enables the sensor’s operating state to be freely adjusted using only light, without the need for external electrical signals. Additionally, when light is irradiated, charge transfer and trapping phenomena occur at the material boundaries, resulting in a bipolar photoresponse—a phenomenon in which the current direction reverses depending on conditions such as light intensity, wavelength, and direction.

(From left) Hanbin Cho, PhD candidate; Wenxuan Zhu, postdoctoral researcher; Jungki Seo, professor; Changhwan Kim, integrated master's and doctoral program. KAIST

Most notably, the technology developed by the research team can be applied to an in-sensor computing structure, in which the sensor itself processes data. This makes it possible to efficiently handle multidimensional optical information that changes over time without the need for complex computational processes.

In actual experiments, the research team confirmed that the technology achieves an object recognition accuracy of over 95% for moving objects, demonstrating its potential for broad applications in fields such as autonomous driving and medical diagnostics.

Professor Suh commented, "This research is significant in that it provides a new foundation for artificial intelligence (AI) vision technology that can acquire richer visual information than conventional sensors by utilizing polarization data. We expect that this achievement will also contribute to the realization of low-power, high-efficiency AI systems in the future."

Meanwhile, Wenxuan Zhu (postdoctoral researcher) and Changhwan Kim (doctoral candidate) are listed as co-first authors, with Professor Junki Suh as the corresponding author. The results of this research were published in the international journal Nature Sensors on April 14.

This content was produced with the assistance of AI translation services.