Date of Award
Fall 2023
Document Type
Open Access Thesis
Department
Mechanical Engineering
First Advisor
Ming Hu
Abstract
Machine learning has demonstrated superior performance in predicting vast materials properties. However, predicting a continuous material property such as phonon density of states (DOS) is more challenging for machine learning due to the inherent issues of data smoothing and sensitivity to peak positions. In this work, phonon DOS of 2,931 inorganic cubic structures with 63 unique elements from the Open Quantum Materials Database are calculated by high precision density functional theory (DFT). With these training data, we build an equivariant graph neural network (GNN) for total phonon DOS of crystalline materials that utilizes site positions and atomic species as input features. The computational cost of training the GNN model is several orders of magnitude cheaper than full DFT calculations. More interestingly, the trained GNN model can predict partial DOS of the constituent atomic species even if such data were not included in the training, which demonstrates GNN’s capability in predicting the species contributions (node-level) of partial DOS from the total DOS predictions without additional computational cost. We then deploy the trained GNN model to predict phonon DOS of 4,626 cubic materials with band gap >0.2 eV to search for thermally conductive substrates for cooling a few representative high electron mobility transistors (HEMT) in terms of high interfacial thermal conductance (ITC). Our results show that high vibrational similarity or phonon DOS overlap is not a necessary requirement to obtain high ITC as evidenced in BN/MgO interface with ITC of 1,044 MW/m2K despite of phonon DOS overlap of only 0.22. Moreover, we highlight that the LTC of substrates does not always play a significant positive role in determining ITC when cooling HEMT devices. However, higher LTC substrates indeed implies a higher magnitude of heat flux that can be transferred from the interfacial region. This work demonstrates the power of GNN models and paves the way for high-throughput screening of novel crystalline materials with desirable high ITC for phonon-mediated thermal management of wide bandgap electronics.
Rights
© 2024, Mohammed Saif Ali Al-Fahdi
Recommended Citation
Al-Fahdi, M. S.(2023). Rapid Prediction of Phonon Density of States by Graph Neural Network and High-Throughput Screening of Candidate Substrates for Wide Bandgap Electronic Cooling. (Master's thesis). Retrieved from https://scholarcommons.sc.edu/etd/7532