Charge Mobility Maximization in Organic Field-Effect Transistors via Design of Experiments and Machine Learning

The field of flexible, printable, and biocompatible electronics is being boosted by the synthesis of novel high-mobility organic semiconductors. At the same time, the successful exploitation of organic semiconductors in real-life applications relies equally on advances in processing techniques, understanding of structure-property relations in large-area thin-films and device engineering. Thus, the complexity of this innovation process requires optimization tasks which are time- and cost-demanding, and often not sustainable, especially in the context of academic research.<br/>In this work, we adopt design of experiments and machine learning techniques to maximize the field-effect mobility in organic field-effect transistors (OFETs), focusing on a multi-factor optimization of the processing of the semiconducting layer.^[1] This approach allows to test and optimize several variables simultaneously and converge to an optimal combination, requiring a much lower number of experiments than in a standard one-variable-at-a-time optimization process.^[2]<br/>Firstly, the method is validated with OFETs based on P(NDI2OD-T2), a well-known high-mobility n-type polymer, also known with its commercial name “N2200”. We investigate the interplay of 4 parameters affecting the polymer deposition by off-center spin-coating (solvent, concentration, spin-coating speed and crystallization temperature) and, by performing less than 50 experiments, we achieve optimized OFETs with state-of-the-art field-effect-mobilities (&gt; 1 cm²/Vs).<br/>Subsequently, we apply this optimization procedure to blends of novel promising small-molecules, namely cumulenic sp-carbon atom wires^[3-5], with insulating polymers. We deposit thin-films of the active material by wire-bar coating, a scalable printing technique, and achieve a fast enhancement of the OFETs performance by tuning the blend composition and the processing parameters.<br/>Finally, we discuss on the physical motivations underpinning the choice of the parameters to optimize, with the aim to provide general insights on machine learning-driven optimization in OFETs.<br/><br/>[1] S. Pecorario, S. Reale, et. al., manuscript in preparation.<br/>[2] B. Cao, L. A. Adutwum, A. O. Oliynyk, E. J. Luber, B. C. Olsen, A. Mar, J. M. Buriak, ACS Nano 2018, 12, 7434.<br/>[3] A. D. Scaccabarozzi, A. Milani, S. Peggiani, S. Pecorario, B. Sun, R. R. Tykwinski, M. Caironi, and C. S. Casari, J. Phys. Chem. Lett., 11, 1970−1974, (2020).<br/>[4] C. S. Casari, M. Tommasini, R. R. Tykwinski and A. Milani, Nanoscale 8 (8), 4414−4435, (2016).<br/>[5] S. Pecorario, et. al., Stable and Solution Processable Cumulenic sp-Carbon Wires: A New Paradigm for Organic Electronics, submitted.