Thermal conductivity modeling using machine learning potentials: application to crystalline and amorphous silicon

Abstract

First-principles based modeling on phonon dynamics and transport using density functional theory and Boltzmann transport equation has proven powerful in predicting thermal conductivity of crystalline materials, but it remains unfeasible for modeling complex crystals and disordered solids due to the prohibitive computational cost to capture the disordered structure, especially when the quasiparticle "phonon" model breaks down. Recently, machine-learning regression algorithms show great promises for building high-accuracy potential fields for atomistic modeling with length and time scales far beyond those achievable by first-principles calculations. In this work, using both crystalline and amorphous silicon as examples, we develop machine learning based potential fields for predicting thermal conductivity. The machine learning based interatomic potential is derived from density functional theory calculations by stochastically sampling the potential energy surface in the configurational space. The thermal conductivities of both amorphous and crystalline silicon are then calculated using equilibrium molecular dynamics, which agree well with experimental measurements. This work documents the procedure for training the machine-learning based potentials for modeling thermal conductivity, and demonstrates that machine-learning based potential can be a promising tool for modeling thermal conductivity of both crystalline and amorphous materials with strong disorder.

Neuroevolution machine learning potentials: Combining high accuracy and low cost in atomistic simulations and application to heat transport

“…However, we note that the quantum effects can be very strong at low temperatures. Our results here agree well with previous studies using the GAP-SOAP potential [66] and the DP potential [67]. A more systematical study considering different temperatures, strains (stresses), cell sizes, and quenching rates is beyond the scope of this paper.…”

“…MD simulations with ML potentials have been applied to study heat transport properties of a number of materials, including, e.g., GeTe and MnGe compounds [61][62][63][64], diamond and amorphous silicon [65][66][67], multilayer graphene [68], monolayer silicene [69], CoSb 3 [70], monolayer MoS 2 and MoSe 2 and their alloys [71], C 3 N [72], α-Ag 2 Se [73,74], β-Ga 2 O 3 [75], Tl 3 VSe 4 [59], PbTe [59], and SnSe [76]. There are also works that exclusively used the Boaltzmann transport equation (BTE) approach to calculate thermal conductivity based on force constants determined from ML potentials [77][78][79][80][81][82].…”

Phonon-engineered extreme thermal conductivity materials

“…Effects of defects, interfaces, and nanostructure distributions on heat conduction cannot be dealt with confidence, partly owing to the lack of knowledge in the details of such imperfections in practical materials. Machine learning of microscopic structures combined with predictive modeling of heat conduction may be a fruitful direction to pursue, especially for high-fidelity modeling of phonon transport at extreme conditions, 126 amorphous materials 127 , as well as the structural design of maximized/minimized thermal resistance of superlattices 128 and even polymers. 129 For fast screening of materials, convenient and high-throughput thermal measurement tools are necessary.…”

Nanostructured and Heterostructured 2D Materials for Thermoelectrics

“…For the materials genome, a highthroughput calculation method is required and this can be achieved with machine learning. Successful examples for machine learning in materials search and design can be found for interfacial thermal conductance, [175], [255], [256] bandgap, [257] and interatomic force constants [258], [259], [260] used in MD simulations. Machine learning driven by experimental data is desired for thermoelectric studies but is still lacking due to the challenge of high-throughput measurements at the nanoscale.…”