Increasing the efficiency of solar cells, transistors, LEDs and batteries will require better electronic materials made from novel compositions that have yet to be discovered.
To accelerate the search for advanced functional materials, scientists are using artificial intelligence tools to identify promising materials from hundreds of millions of chemical formulations. Together, engineers build machines that can print hundreds of material samples simultaneously based on chemical composition determined by AI search algorithms.
However, until now there has been no equally quick way to check whether printed materials actually perform as expected. This last step of material characterization was the main bottleneck in the advanced material selection process.
Now, a up-to-date computer vision technique developed by MIT engineers significantly accelerates the characterization of newly synthesized electronic materials. This technique automatically analyzes images of printed semiconductor samples and quickly estimates two key electronic properties of each sample: bandgap (a measure of electron activation energy) and stability (a measure of durability).
The up-to-date technique accurately characterizes electronic materials 85 times faster compared to standard benchmarking approaches.
Scientists intend to apply this technique to accelerate the search for promising solar cell materials. They also plan to incorporate this technique into a fully automated materials inspection system.
“Ultimately, we plan to use this technique in an autonomous laboratory of the future,” says MIT graduate Eunice Aissi. “The entire system would allow us to give a computer a materials problem, predict potential compounds, and then work 24/7 to create and characterize the predicted materials until we achieve the desired solution.”
“The application space for these techniques ranges from solar energy enhancement to transparent electronics and transistors,” adds MIT graduate Alexander (Aleks) Siemenn. “It really covers the whole range of applications where semiconductor materials can benefit society.”
Aissi and Siemenn describe the up-to-date technique in detail in: study that comes out today IN . Their MIT co-authors include graduate student Fang Sheng, postdoc Basita Das, and mechanical engineering professor Tonio Buonassisi, as well as former visiting professor Hamide Kavak of Cukurova University and postdoc Armi Tiihonen of Aalto University.
Power in optics
Once a up-to-date electronic material has been synthesized, its properties are typically characterized by a “subject matter expert” who examines a single sample using a laboratory tool called UV-Vis, which scans lightweight of different colors to determine where the electronic material is located. the semiconductor begins to absorb more strongly. This manual process is precise, but also time-consuming: a specialist in the field typically characterizes about 20 material samples per hour – a snail’s pace compared to some printing tools that can create 10,000 different material combinations per hour.
“The manual characterization process is very slow,” says Buonassisi. “They provide high measurement confidence, but are not matched to the speed at which matter can currently be placed on a substrate.”
To speed up the characterization process and remove one of the biggest bottlenecks in materials inspection, Buonassisi and his colleagues focused on computer vision, a field that uses computer algorithms to quickly and automatically analyze an image’s optical features.
“Optical characterization methods are powerful,” notes Buonassisi. “You can get information very quickly. Images with many pixels and wavelengths contain a richness that a human simply cannot process, but a machine learning computer program can.
The team realized that certain electronic properties – namely bandgap and stability – could be estimated from visual information alone if the information was captured in sufficient detail and interpreted correctly.
With this goal in mind, researchers developed two new computer vision algorithms to automatically interpret images of electronic materials: one to estimate the bandgap and the other to determine stability.
The first algorithm is designed to process visual data from highly detailed hyperspectral images.
“Instead of the standard camera image consisting of three channels – red, green and blue (RBG) – the hyperspectral image has 300 channels,” explains Siemenn. “The algorithm takes this data, transforms it and calculates the bandgap. We are moving through this process extremely quickly.”
The second algorithm analyzes standard RGB images and assesses material stability based on visual changes in the material’s color over time.
“We found that color change can be a good indicator of the rate of degradation in the material system under study,” says Aissi.
Material compositions
The team used two up-to-date algorithms to characterize the bandgap and stability of approximately 70 printed semiconductor samples. They used a robotic printer to place the samples on a single slide, like cookies on a baking tray. Each deposit was made from 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 candidate for solar cells, although it is also known to degrade quickly.
“People try to change the composition – add a little bit of this, a little bit of that – to get it [perovskites] more stable and efficient,” says Buonassisi.
After printing 70 different compositions of perovskite samples on one 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 up-to-date bandgap algorithm on isolated samples and automatically calculated the bandgap for each sample. The entire bandgap extraction process took approximately six minutes.
“It would typically take a domain expert several days to characterize the same number of samples manually,” says Siemenn.
To test stability, the team placed the same preparation in a chamber where it varied environmental conditions such as humidity, temperature and lightweight exposure. They used a standard RGB camera to take a photo of samples every 30 seconds over a two-hour period. They then applied a second algorithm to images of each sample over time to estimate the degree to which each droplet changed color or degraded under different environmental conditions. Ultimately, the algorithm created a “stability index,” a measure of each sample’s durability.
As a follow-up, the team compared their results with manual measurements of the same droplets taken by an expert in the field. Compared to expert benchmark estimates, the team’s bandgap and stability results were 98.5% and 96.9% correct, respectively, and 85 times faster.
“We were continually amazed at how these algorithms were able to not only increase the speed of characterization, but also produce accurate results,” says Siemenn. “We see this being incorporated into the current automated materials pipeline that we are developing in the lab, so that we can run it in a fully automated way, using machine learning to pinpoint where we want to discover these new materials, print them, and then properly characterize them, and all this with very fast processing.”
This work was supported in part by First Solar.