[Press Release] Professor Jeonghoon Kwak's research team develops a performance optimization technology for organic thermoelectric devices using machine learning
From left: Professor Jeonghoon Kwak(corresponding author), Jeehyun Jung, PhD candidate (co-first author)
The College of Engineering at Seoul National University announced that Professor Jeonghoon Kwak's research team from the Department of Electrical and Computer Engineering has developed a machine learning-based Design of Experiments (DOE) method for efficiently optimizing the performance and manufacturing conditions of organic thermoelectric devices.
Organic thermoelectric devices convert low temperature waste heat, such as that generated by human skin or electronic devices, into electrical energy. This research conveys the first instance of utilizing machine learning in the field of organic thermoelectric devices. The newly developed experimental design method is being recognized as an innovative approach for effectively optimizing the performance of these devices, which was previously challenging due to the multitude of variables involved.
The study was led by PhD candidates Jeehyun Jung and Sooyeon Park from the Department of Electrical and Computer Engineering at Seoul National University. The findings were published on November 26 in the prestigious international journal Advanced Energy Materials (Impact Factor: 24.4), which focuses on energy and materials science.
Organic thermoelectric devices are gaining attention as energy harvesting solutions for next-generation wearable devices due to their excellent mechanical flexibility, large-area manufacturability, and potential for mass production. They are also attracting interest in the field of temperature sensors. However, unlike conventional thermoelectric technologies that rely on crystalline inorganic materials to interconvert heat and electricity, organic thermoelectric devices utilize doped semicrystalline polymer thin films, making it challenging to find optimal performance conditions. The use of these polymer thin films creates complex interactions between manufacturing variables (such as doping concentration, film formation methods, and annealing temperature) and thermoelectric performance factors (such as electrical conductivity and Seebeck coefficient). As a result, optimizing the performance of organic thermoelectric devices has traditionally required significant time, effort, and iterative experimentation.
To address this inefficiency, Professor Jeonghoon Kwak's research team introduced a machine learning-based Design of Experiments (DOE) methodology. The team began by identifying four key process variables affecting the performance of organic thermoelectric devices: spin speed, doping solution concentration, doping time, and annealing temperature. They then established four distinct conditions for each variable. Using traditional methods, this would have necessitated fabricating at least 256 (4 to the 4th power) thermoelectric devices to evaluate all possible combinations of process conditions.
However, leveraging their AI-based experimental design method, the research team successfully reduced this number to just 16 (4 × 4). This streamlined approach enabled them to identify the significance of each process variable and determine the optimal conditions for enhancing device performance with remarkable efficiency.
The machine learning-based Design of Experiments (DOE) method, which successfully predicts the optimal performance of organic thermoelectric devices while minimizing repetitive experiments, is expected to significantly contribute to improving device performance. Moreover, it provides valuable insights for the development of materials and processes. These advanced organic thermoelectric devices are anticipated to be widely used as power sources for wearable devices and small electronic gadgets.
Jeehyun Jung, the first author of the paper, commented, "This research is a successful example of AI application, efficiently deriving optimal thermoelectric performance with minimal experiments using machine learning-based techniques. The results are particularly meaningful because they demonstrate a shift from traditional iterative experimentation to data-driven scientific design."
Professor Jeonghoon Kwak, who supervized the research, stated, "The AI-based experimental design significantly reduced research time and costs while enabling a more systematic understanding of the high-dimensional interactions that were previously difficult to investigate."
Currently leading the Advanced Opto & Nano Electronics Laboratoryat Seoul National University, Professor Kwak plans to continue researching the development of organic thermoelectric devices, as well as the fabrication processes and performance optimization of various electronic devices using organic semiconductors. Jeehyun Jung is also pursuing research to further enhance the performance of organic thermoelectric devices, focusing on fabrication processes and device designs essential for advancing clean energy technologies that utilize waste heat.
Source: http://www.mediadale.com/news/articleView.html?idxno=224078
Translated by: Dohyung Kim, English Editor of the Department of Electrical and Computer Engineering, kimdohyung@snu.ac.kr
