(346c) Machine Learning Applications in Prediction of Propagation Rate Coefficients of Acrylate-Based Polymers for Electronics Applications

Acrylate-based polymers are commonly employed in electronics applications, such as antireflective coatings and photoresists. Microelectronics specialty chemicals have evolved largely and become more sophisticated, as the technology node in semiconductor manufacturing processes has shrunk to the smaller feature sizes. Resist and anti-reflective formulations generally contain several components, including polymer resin (film forming resin), solvent, sensitizer, either photo-initiator or photoacid generator, and additives such as surfactants. The polymer resin is the major component in the formulation; hence, the precise manufacturing and quality control of both chemical and physical properties of the polymer resins are essential. A mathematical model is an indispensable tool to predict and control the properties of polymers, and it can be implemented in manufacturing systems for real-time control of reactions. Along these lines, kinetic modeling is central to developing optimal reaction conditions. Among the different reactions steps, the propagation rate plays an essential role in determining molecular weight (MW), molecular weight distribution (MWD), and composition for co-polymers, thus influencing the performance of products. Therefore, determining the propagation rate accurately is critical in modeling of polymerization processes, product design, and manufacturing scale-up. As Machine Learning (ML) has become popular due to its ability in adapting complex mathematical calculations to big data, analyzing, and decision-making, this tool is aggressively utilized in chemistry (e.g., Chemception [1]) as well as in the case of complex non-linear processes for which phenomenological understanding is limited [2]. In this work, we developed a kinetic modelling using Deep Learning (DL), such as Long Short-term Memory (LSTM), to study the reaction kinetics of Poly(Methyl Methacrylate-co-Glycidyl Methacrylate), determining the propagation rate coefficients and the reactivity ratios. The reactivity ratios predicted using the ML techniques were compared against the experimental data.

[1] Garrett B. Goh et. al., Chemception: A Deep Neural Network with Minimal Chemistry Knowledge Matches the Performance of Expert-developed QSAR/QSPR Models., 2016. (arXiv:1706.06689).

[2] S. Curteanu and F. Leon, Polym.-Plast. Technol. Eng., 2006,45, 1013.