Solar cells, transistors, LEDs, and batteries with boosted performance require better electronic materials which are often discovered from novel compositions. Scientists have turned to AI tools to identify potential materials from millions of chemical formulations, with engineers developing machines that can print hundreds of samples at a time, based on compositions identified by AI algorithms. Verifying the performance of these printed materials, however, has been a major hurdle.
Now, a new computer vision technique developed by MIT engineers can quickly analyze images of printed semiconducting samples, significantly speeding up the characterization of new electronic materials. This technique provides quick estimates of two key electronic properties: band gap (electron activation energy), and stability (longevity). The new technique offers an 85-times faster characterization of electronic materials compared to the standard benchmark approach.
The researchers aim to use the technique to expedite the search for viable solar cell materials and to incorporate it into an entirely automated materials screening system. “The application space for these techniques ranges from improving solar energy to transparent electronics and transistors,” says MIT graduate student, Alexander Siemenn.
Typically, when a new electronic material is synthesized, its properties are characterized by a “domain expert” using a benchtop tool called a UV-Vis. This process is accurate but time-consuming, with experts characterizing roughly 20 material samples per hour.
To increase the speed of this process and clear this bottleneck in materials screening, the researchers used computer vision to analyze optical features in an image quickly and automatically. This led them to develop two computer vision algorithms to interpret images of electronic materials, one for estimating band gap and the other for determining stability.
The team applied these two new algorithms to characterize the band gap and stability for about 70 printed semiconducting samples. This entire band gap extraction process took about six minutes, compared to the several days a domain expert would typically take. The researchers also figured out that changes in the material’s color could indicate the rate of degradation in the material system.
Tests for stability involved placing the sample in a chamber where environmental conditions could be varied. Images were taken every 30 seconds over two hours, with the second algorithm applied to the images of each sample over time to estimate the degree to which each droplet changed color, or degraded under various environmental conditions.
The team’s band gap and stability results, compared to those obtained through manual measurements by a domain expert, were found to be 98.5% and 96.9% as accurate, respectively, and 85 times faster.
