Machine Learning-Based Multiscale Modeling and Control of Quantum Dot Manufacturing and Their Applications
Abstract
In the past few years, there has been a major impetus in the search for quantum dots (QDs), which are a type of semiconducting nanocrystals (NCs) with tunable optical and optoelectronic properties for next-generation photonic devices. This can be attributed to their relatively high photoluminescence quantum yield, wide color gamut, tunable optoelectronic properties, and cost-effective solution processibilities. Furthermore, the rising market share of these applications has led to an increased demand for fast and scalable production of QDs and the associated optoelectronics devices. However, there are major commercialization challenges associated with the manufacturing of QDs: (a) lack of mechanistic understanding of the crystallization kinetics of various QD systems, which hinders the predictive control of the QD size distribution; (b) absence of a well-established paradigm for high-fidelity modeling and scale-up of various QD manufacturing processes (e.g., crystallization and thin-film deposition); and (c) no presence of computationally efficient solutions for control and optimization of QD processes.
To address these knowledge gaps, in this work, we develop different models to describe the mechanism of QD crystal growth, enable fast-scalable manufacturing of QDs and the associated optoelectronic devices, and develop an appropriate control framework for various QD processes. First, a first-principled kinetic Monte Carlo (kMC) was developed and experimentally validated to describe the crystallization kinetics of QDs. Second, to resolve the various issues associated with the batch synthesis of QDs, continuous manufacturing of QDs using a plug flow crystallizer (PFC) was demonstrated using a multiscale modeling approach. Further, this approach was extended to two-phase slug flow crystallizers (SFCs) by combining the construction of a CFD-based multiscale model. Also, a highly efficient data-driven optimal control framework was formulated using a deep neural network (DNN) to control QD crystal size and distribution. Third, modeling of thin-film deposition required for manufacturing of solar cells and high-resolution displays was performed. Specifically, a multiscale model that combines the surface-level discrete element method (DEM) model of QD aggregation and macroscopic mass and energy balance equations was developed for describing the spray coating of QDs. Further, given the computational expense of this model, a surrogate DNN model was developed, which was integrated with a model predictive controller (MPC) to control the film characteristics (i.e., thickness and roughness). Next, although the resulting thin-films are of the desired quality, they are chemically labile and cannot withstand subsequent downstream processing during the manufacturing of LEDs or solar cells. Thus, a kMC model was developed to describe crosslinking of QD thin-films for increased chemical robustness resulting in the manufacturing of high-resolution displays. Moreover, it is important to note that all of the above-developed models were experimentally validated using appropriate experimental observations.
Lastly, although the above-developed model accurately describes various processes related to QD manufacturing, these models are very system-specific, and cannot be easily extrapolated to other QD systems. To provide a concrete direction for addressing this issue in the future, we propose a transformer-based hybrid model, which can leverage the remarkable transfer learning properties of transformers for better generalization across different QD systems. Overall, the proposed work addresses three major challenges in the QD field (i.e., control of QD kinetics, continuous production of QDs, and designing manufacturing processes for fast scale-up of QD-based devices) by developing various experimentally validated multiscale models and combining them with an appropriate control framework.
Subject
quantum dotsmachine learning
multiscale modeling
crystallization
thin-film deposition
optimal control
MPC
Citation
Sitapure, Niranjan Arvind (2023). Machine Learning-Based Multiscale Modeling and Control of Quantum Dot Manufacturing and Their Applications. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /199931.