Robotic Probe Speeds Materials Measurement

Scientists are working to discover new semiconductor materials for more efficient electronics. The speed at which researchers can manually measure material properties slows innovation. MIT researchers have developed a fully autonomous robotic system that could make the process faster.

The system uses a robotic probe to measure photoconductance, an electrical property describing a material's responsiveness to light. The machine-learning model guiding the robot incorporates materials-science knowledge from experts. This helps the robot select the best contact points for the probe to maximize information about photoconductance, while a specialized planning procedure optimizes movement between contact points.

In a 24-hour test, the robotic probe took over 125 unique measurements per hour, showing greater precision and reliability than other AI-based methods. This could accelerate the development of solar panels.

Tonio Buonassisi, professor of mechanical engineering, finds the system exciting because it enables autonomous, contact-based characterization methods. Contact is needed for some material properties, so speed and information maximization are important. Alexander (Aleks) Siemenn, Basita Das, Kangyu Ji, and Fang Sheng are co-authors of the paper in Science Advances.

Making Contact

Since 2018, Buonassisi’s lab has been developing a fully autonomous materials discovery laboratory. They have been focusing on perovskites. Previously, they created methods to synthesize and print combinations of perovskite material and imaging-based methods to determine material properties. However, a probe placed on the material, combined with light and electrical response measurement, is the most accurate way to characterize photoconductance.

Siemenn said that to allow the experimental laboratory to operate quickly and accurately, they needed a solution that would produce the best measurements while minimizing the time it takes to run the whole procedure. This required integrating machine learning, robotics, and material science into an autonomous system.

The robotic system uses its camera to image a slide with perovskite material. Computer vision divides the image into segments that are fed into a neural network model incorporating domain expertise from chemists and materials scientists.

Siemenn adds that including human experts is important because the robots can improve repeatability and precision. The model uses domain knowledge to select optimal contact points based on sample shape and material composition. A path planner then finds the most efficient route for the probe to reach these points.

The machine-learning approach is adaptable because the printed samples have unique shapes. Buonassisi notes that the samples are almost like snowflakes because it is difficult to get two that are identical.

Once the path planner identifies the shortest path, it signals the robot’s motors to manipulate the probe and take measurements. The neural network model's self-supervised nature is key to the approach's speed because it determines optimal contact points directly on a sample image without labeled training data.

The researchers also made the system faster by enhancing the path planning procedure. They found that adding randomness to the algorithm helped it find the shortest path.

Buonassisi says that hardware building, software, and understanding materials science must come together to innovate quickly.

Rich Data, Rapid Results

The researchers tested each component of the system. The neural network model located better contact points with less computation time than seven other AI-based methods. The path planning algorithm also found shorter path plans than other methods. During a 24-hour autonomous experiment, the robotic system took over 3,000 unique photoconductance measurements at over 125 per hour.

The precise measurements enabled the researchers to identify hotspots with higher photoconductance and material degradation areas.

Siemenn says that the ability to gather rich data at fast rates without human guidance opens opportunities to discover and develop new high-performance semiconductors for sustainability applications like solar panels.

The researchers want to continue developing this robotic system to create a fully autonomous lab for materials discovery.

This work is supported by First Solar, Eni through the MIT Energy Initiative, MathWorks, the University of Toronto’s Acceleration Consortium, the U.S. Department of Energy, and the U.S. National Science Foundation.