Symposium X—Frontiers of Materials Research

Benjamin C.K. Tee, National University of Singapore

AI Skins for Good

Written by Don Monroe

In his Symposium X presentation, Benjamin C.K. Tee of the National University of Singapore discussed applications of sensor devices that use soft materials, including electromechanical sensors, self-healing devices, stretchable electronics, and neuro-inspired devices.

The widespread adoption of artificial intelligence (AI) is driven by the recent explosion in the amount of data available, much of which comes from sensors, Tee said. Thus, making sensors more ubiquitous, versatile, and robust can provide data for an even wider range of applications for the benefit of humankind.

Materials science and engineering are critical for enabling these novel devices, and there have been many good devices and systems proposed, Tee said. The resulting electronic skin integrates and processes many sensory inputs in parallel,” he noted. He emphasized the importance of a multi-disciplinary approach that extends beyond device design to include a systems perspective, including AI and even biological science.

The human sense of touch augments vision for many tasks, such as manipulating objects and navigating the world. Skin-like structures incorporating force-sensitive devices could therefore be very powerful in robotics, for example grasping delicate items, and, Tee said, perhaps other applications that we haven’t thought of yet. In one example, he described a heads-up display for a surgeon (or trainee) that tracks in real time the forces on a scalpel from healthy or tumor tissue.

Traditional electronic materials have stiffnesses of many gigapascals, much greater than that of skin, which is of order megapascals, or of tissues such as brains, below a kilopascal. “One of the biggest challenges to developing soft, flexible materials that can perform functions is the ability of human skin to maintain functionality and robustness despite mechanical strain,” Tee said. “Generally, you can stretch 10–15%.”

He showed one enabling technology in which helical metal wires embedded in elastomer suffer no change in conductivity even with very large strain. His team has developed a platform that accurately predicts the behavior of well-characterized materials in novel devices.

Unfortunately, soft materials tend to be viscoelastic, resulting in large hysteresis in force sensing. Although machine learning can partially compensate for this behavior, Tee said, “Despite the best AI, you need good sensors to get good accuracy,” as illustrated by an improved, almost hysteresis-free device his team developed.

Large-area electronic-skin systems also struggle to convey the information from many sensors. “It can be quite impractical to do this with normal wires,” Tee said. To address this issue, his team developed ACES, or asynchronous coded E-skins, which multiplexes the outputs of many sensors, with orders of magnitude faster response times—almost 10 megahertz instead of around a kilohertz. “Such systems might make it more scalable,” Tee noted.

Another powerful communication strategy is the use of action potentials like those used in human brains. This biology-inspired strategy allows more efficient transmission of sensory data. Indeed, companies including Intel have developed “neuromorphic” chips that perform powerful computations with much less power than traditional graphics processing units.

Large arrays of sensors also need to be robust against failures of individual devices. Tee showed an example system design that has multiple wires for each device to provide redundancy, as well as reconfigurability.

At the device level of robustness, he described research on self-healing materials, which can repair themselves even under water. In addition to being more convenient, such long-lived devices could reduce the significant environmental impact of electronic waste.

Symposium X—Frontiers of Materials Research features lectures aimed at a broad audience to provide meeting attendees with an overview of leading-edge topics.