In a recent article published in the journal Nature Communications, researchers introduced a suite of automated characterization (autocharacterization) tools that leverage adaptive computer vision to measure the key properties of semiconductor materials rapidly and accurately. They demonstrated the application of these tools on a high-throughput synthesis platform, which produces unique perovskite semiconductors in one hour.
The band gap computations shown are for N = 201 unique perovskite samples across three independent trials. The solid black line is the line of best fit to the band gap data and the dashed black line is the y = x line. Histogram distributions of both autocharacterization and domain expert band gaps are shown on the right and top of the plot area, respectively. The color of the scatter points corresponds to the proportion of MA, x, in the formamidinium (FA) and methylammonium (MA) mixed-cation composition FA1−xMAxPbI3. Image Credit: https://www.nature.com/articles/s41467-024-48768-2
Background
Semiconductor materials are widely used in various fields such as electronics, optoelectronics, solar cells, and sensors. However, discovering and optimizing new semiconductor materials is a challenging task, as it requires exploring a large and complex material search space and characterizing the material properties that affect the performance and stability of devices.
High-throughput synthesis methods have been developed to accelerate the production of diverse material samples, but they face a bottleneck in the characterization process, which is often slow, manual, and rigid. Therefore, there is a need for automated and scalable characterization tools that can keep pace with high-throughput synthesis and provide rapid and accurate feedback on material properties.
About the Research
In this paper, the authors aimed to address the challenge of characterizing high-throughput synthesized semiconductor materials by developing auto-characterization tools that use scalable computer vision techniques to extract key information from image data. They focused on the perovskite semiconductor system FA1-xMAxPbI3, 0 ≤ x ≤ 1, which has shown promising potential for solar cell applications but also exhibits complex composition-dependent properties such as band gap and degradation.
The study used a high-throughput inkjet printing platform to deposit 200 unique perovskite samples on a glass substrate in one hour, creating a compositional gradient of formamidinium (FA) and methylammonium (MA) cations. The samples had variable morphologies and were not compatible with existing characterization tools.
Therefore, a hyperspectral imager and a standard red-green-blue (RGB) camera were used to capture the reflectance spectra and the color change of the samples, respectively. The image data was then processed by the developed autocharacterization tools, which consisted of a computer vision segmentation tool, a composition mapping tool, a band gap autocharacterization tool, and a degradation autocharacterization tool.
Firstly, the computer vision segmentation tool identified and indexed each sample and its corresponding reflectance spectra from the hyperspectral datacube, enabling parallel measurements across many samples. Secondly, the composition mapping tool determined the proportion of FA and MA cations in each sample by integrating the pump speeds of the inkjet printer over time and spatially mapping them onto the segmented samples.
Thirdly, the band gap autocharacterization tool computed the direct band gap of each sample by transforming the reflectance spectra to Tauc curves and finding the best-fit linear regression line between the Tauc peaks using a recursive segmentation and iterative fitting algorithm. Finally, the degradation autocharacterization tool quantified the degree of degradation of each sample by integrating the color change over time, using a calibrated RGB color space and a degradation intensity metric.
Research Findings
The performance of the autocharacterization tools was validated by comparing the results obtained with conventional methods and domain expert evaluation. The outcomes showed that the autocharacterization tools achieved high accuracy and speed in measuring the composition, band gap, and degradation of the perovskite samples. Additionally, the composition mapping tool was verified by X-Ray diffraction and X-Ray photoelectron spectroscopy, which confirmed the expected shifts in the crystal structure and the elemental composition of the samples.
Specifically, the band gap autocharacterization tool demonstrated a strong linear fit with an R² of 0.975 when compared to expert-calculated band gaps, with 98.5 % accuracy within a 0.02 eV range. The degradation autocharacterization tool exhibited a precision-recall area under the curve of 0.853 and a maximal accuracy of 96.9 % relative to the ground truth degradation determined by the pre- and post-band gap deviation.
Furthermore, the autocharacterization tools significantly sped up the characterization process, taking only six minutes to compute the band gap and 20 minutes to detect the degradation of 200 samples, compared to 510 minutes and several hours or days, respectively, for conventional methods.
Applications
The proposed tool has significant implications for the discovery and optimization of semiconductor materials. It enables rapid, accurate feedback on material properties, allowing for faster screening and selection of optimal compositions and conditions.
Additionally, it facilitates the exploration of larger and more complex material search spaces, aiding in the discovery of novel and high-performance materials. The tool can also be adapted to other material systems and properties, such as organic photovoltaics, nanomaterials, and porous media, by modifying computer vision algorithms and data analysis methods.
Conclusion
In summary, the novel tool demonstrated effectiveness in extracting composition, band gap, and degradation information from hyperspectral and RGB image data, achieving throughputs up to 85 times faster than conventional methods. Future work should focus on extending these autocharacterization tools to multi-phase materials, which may feature multiple band gaps, and to other material systems and properties. Additionally, integrating these tools with machine learning and artificial intelligence techniques can further enable data-driven and intelligent material design.
Journal Reference
Siemenn, A.E., Aissi, E., Sheng, F. et al. Using scalable computer vision to automate high-throughput semiconductor characterization. Nat Commun 15, 4654 (2024). https://doi.org/10.1038/s41467-024-48768-2, https://www.nature.com/articles/s41467-024-48768-2
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.