Organic semiconductors are attracting attention as materials for low-temperature waste-heat recovery owing to their low cost, mechanical flexibility, solution processability, and chemical stability. In particular, semiconducting polymers have emerged ...
Organic semiconductors are attracting attention as materials for low-temperature waste-heat recovery owing to their low cost, mechanical flexibility, solution processability, and chemical stability. In particular, semiconducting polymers have emerged as promising thermoelectric (TE) materials for diverse energy harvesting and sensing applications, because their solution processability, mechanical flexibility, structural tunability, and intrinsically low thermal conductivity make them well suited for TE applications. Optimizing the processing conditions of conjugated-polymer-based organic electronic devices is essential for improving their performance. However, the structural and energetic disorders inherent in conjugated polymers, together with process-dependent variations in their properties, make it difficult to establish universal principles for process optimization. As a result, extensive trial-and-error experiments are still often required to maximize TE performance, and the absence of systematic guidelines remains a major bottleneck.
In this study, systematic process optimization studies on two representative organic thermoelectric systems were conducted to establish a process-level understanding of how doping and post-treatment routes govern microstructure, charge transport, and thermoelectric performance. Building on these insights, a machine learning (ML)-based design of experiments (DOE) strategy was established to quantitatively analyze the effects of processing parameters and efficiently identify optimal conditions over a wide process-parameter space. By comparing and analyzing various molecular doping methods and multistep post treatment, the effects of each doping route on microstructure and charge transport were clarified, and it was shown how the trade-off between carrier concentration and the Seebeck coefficient can be controlled by tuning the processing conditions. Through these two systems, a process-level understanding was obtained of how the choice of doping and post-treatment routes controls the linkage between processing, microstructure, and TE characteristics.
Based on these process optimization insights, the ML-based DOE approach enabled quantitative analysis of how each processing parameter influences the TE properties of the polymers and made it possible to predict optimal processing conditions with a minimal number of experiments over a wide process-parameter space. In particular, DOE uses statistical designs such as orthogonal arrays to sample intrinsically correlated process factors as nearly uncorrelated independent variables, thereby allowing efficient separation and evaluation of main effects and interaction effects. An ML regression model trained on the experimental data obtained from DOE was then used to fill the unmeasured regions through interpolation and limited extrapolation, reconstructing a continuous response surface over the entire process-parameter space. Furthermore, by analyzing the morphology and electronic states of PBTTT-based doped polymer films, the physical origin of the enhanced power factor under the ML-suggested optimal conditions was identified, and the validity of the predictions was verified. The ML-based process optimization method was also extended to various combinations of host polymers and molecular dopants, and was found to be consistently effective across different organic TE systems, demonstrating the generality and robustness of this approach. The proposed methodology is broadly applicable to conjugated polymer systems for organic thermoelectric energy conversion and provides practical guidelines for determining optimal processing conditions to efficiently realize high-performance organic TE devices.