But to date, there hasn't been an equally rapid way to confirm that these printed materials actually perform as expected. This final step in materials characterization has been a significant bottleneck in the advanced materials selection process.

Now, a new computer vision technique developed by MIT engineers is significantly accelerating the characterization of newly synthesized electronic materials. The technique automatically analyzes images of printed semiconductor samples and quickly calculates two key electronic properties for each sample: the band gap (a measure of electron activation energy) and stability (a measure of longevity).

The new technique accurately characterizes electronic materials 85 times faster than the standard reference method.

The researchers intend to use the technique to accelerate the search for promising materials for solar cells. They also plan to incorporate the technique into a fully automated materials screening system.

"Ultimately, we envision incorporating this technique into a self-contained laboratory of the future," says Eunice Aissi, a graduate student at MIT. "The complete system would allow us to pose a materials problem to a computer, have it predict potential compounds, and then run it 24/7, fabricating and characterizing those predicted materials until we arrive at the desired solution.".

"The field of application for these techniques ranges from improving solar energy to transparent electronics and transistors," adds Alexander (Aleks) Siemenn, a graduate student at MIT. "It really covers the whole range of areas where semiconductor materials can benefit society.".

Aissi and Siemenn detail the new technique in a new study appearing today in Nature Communications. Their co-authors at MIT are graduate student Fang Sheng, postdoctoral fellow Basita Das, and mechanical engineering professor Tonio Buonassisi, along with former visiting professor Hamide Kavak of Cukurova University and visiting postdoctoral fellow Armi Tiihonen of Aalto University.

Optical power

Once a new electronic material has been synthesized, the characterization of its properties is usually carried out by a "subject matter expert" who examines one sample at a time using a benchtop tool called a UV-Vis spectrometer, which scans different colors of light to determine where the semiconductor begins to absorb more intensely. This manual process is precise, but also slow: A subject matter expert typically characterizes about 20 material samples per hour, a snail's pace compared to some printing tools that can print 10,000 different combinations of materials per hour.

"The manual characterization process is very slow," says Buonassisi. "It gives you a high degree of confidence in the measurement, but it doesn't keep up with the speed at which material can be applied to a substrate today.".

To speed up the characterization process and eliminate one of the biggest bottlenecks in materials screening, Buonassisi and his colleagues turned to computer vision, a field that applies computer algorithms to quickly and automatically analyze the optical characteristics of an image.

"Optical characterization methods are very powerful," Buonassisi points out. "Information can be obtained very quickly. There is a wealth of detail in the images, across many pixels and wavelengths, that a human cannot process but that a machine learning computer program can."

The team realized that certain electronic properties—namely, band separation and stability—could be estimated based solely on visual information, if it was captured in sufficient detail and interpreted correctly.

With that goal in mind, the researchers developed two new computer vision algorithms to automatically interpret images of electronic materials: one to estimate the bandwidth and another to determine stability.

The first algorithm is designed to process visual data from highly detailed hyperspectral images.

"Instead of a standard camera image with three channels—red, green, and blue (RBG)—the hyperspectral image has 300 channels," explains Siemenn. "The algorithm takes that data, transforms it, and calculates a bandwidth. We run that process extremely fast.".

The second algorithm analyzes standard RGB images and assesses the stability of a material based on the visual changes in the material's color over time.

"We discovered that the color change can be a good indicator of the degradation rate of the material system we are studying," explains Aissi.

Material compositions

The team applied the two new algorithms to characterize the band gap and stability of about 70 printed semiconductor samples. They used a robotic printer to deposit samples onto a single slide, like cookies on a baking sheet. Each deposit was made with a slightly different combination of semiconductor materials. In this case, the team printed different proportions of perovskites, a type of material that is expected to be a promising solar cell candidate, although it is also known to degrade rapidly.

"People try to change the composition - adding a little of this, a little of that - to try to make [perovskites] more stable and high-performing," Buonassisi explains.

Once 70 different compositions of perovskite samples were printed onto a single slide, the team scanned the slide with a hyperspectral camera. They then applied an algorithm that visually segments the image, automatically isolating the samples from the background. They ran the new band separation algorithm on the isolated samples and automatically calculated the band separation for each sample. The entire band gap extraction process took about six minutes.

"Normally, it would take an expert in the field several days to manually characterize the same number of samples," says Siemenn.

To test stability, the team placed the same slide in a chamber where environmental conditions, such as humidity, temperature, and light exposure, were varied. They used a standard RGB camera to take an image of the samples every 30 seconds for two hours. They then applied the second algorithm to the images of each sample over time to estimate the degree to which each droplet changed color or degraded under the different environmental conditions. Ultimately, the algorithm produced a "stability index," or a measure of each sample's durability.

To verify this, the team compared their results with manual measurements of the same drops taken by a subject matter expert. Compared to the expert's baseline estimates, the team's results for band gap and stability were 98.5% and 96.9% more accurate, respectively, and 85 times faster.

"We were constantly amazed at how these algorithms were able not only to increase the speed of characterization, but also to obtain accurate results," says Siemenn. "We envisioned that this could fit into the current automated materials pipeline we're developing in the lab, so that we could run it in a fully automated way, using machine learning to guide us to where we want to discover these new materials, printing and characterizing them, all with very fast processing."

This work has been partly funded by First Solar.

Author: Jennifer Chu, MIT News