Source-linked AI summary
Materials discovery and properties prediction in thermal transport via materials informatics: a mini-review
Xiao Wan, Wentao Feng, Yunpeng Wang, Chengcheng Deng, Nuo Yang
TL;DR
Materials with special thermal properties are sought for applications such as heat dissipation and thermoelectric power generation, but discovery is costly and time-consuming. This mini-review surveys materials informatics methods and applications for thermal transport, including thermal conductivity, interfacial conductance, and thermoelectricity. It summarizes representative findings and identifies challenges including descriptor selection and physical interpretation.
Problem
Materials with suitable thermal properties are in demand, while their discovery is constrained by cost, time, and hardware limitations.
Source: §1 · p. 3Method
The mini-review summarizes materials-informatics procedures, machine-learning algorithms, and applications to thermal conductivity, interfacial thermal conductance, and thermoelectric conversion efficiency.
Source: §1 · p. 4, §6 · p. 14Results
The review presents representative materials-informatics findings across thermal conductivity, interfacial thermal conductance, and thermoelectricity, including reported predictions and experimentally confirmed thermoelectric candidates.
Source: §3 · p. 8, §4 · p. 10, §5 · pp. 12–13Takeaways & Limitations
The review highlights efficient algorithms, effective descriptors, transfer of data into physical knowledge, and physical interpretation as important challenges for materials informatics in thermal transport.
Source: §6 · p. 14Machine learning alone cannot fully reveal underlying physical mechanisms, which require support from other theoretical or simulation methods.
Source: §6 · p. 14
Abstract
from arXiv · showhide
There has been an increasing demand for materials with special thermal properties, whereas experimental discovery is high-cost and time-consuming. The emerging discipline `Materials Informatics' is an effective approach that can accelerate materials development by combining material science and big data technique. Recently materials informatics has been applied to the design of novel materials such as thermal interface materials for heat-dissipation, and thermoelectric materials for power generation. This mini-review summarized the research progress on the applications of materials informatics for the thermal transport properties prediction and discovery of materials with special thermal properties, including optimal thermal conductivity, interfacial thermal conductance and thermoelectricity efficiency. In addition, some perspectives are given for the outlook of materials informatics in the field of thermal transport.
1. Introduction
Thermal transport properties matter across electronics and thermoelectric applications, but discovering materials with suitable properties is constrained by cost, time, and hardware. The review surveys materials informatics and its applications to thermal conductivity, interfacial thermal conductance, and thermoelectric conversion efficiency.
1. Introduction
- High thermal conductivity supports heat dissipation in electronic devices, while interfacial thermal conductance dominates heat transfer across microscale and nanoscale interfaces.Low thermal conductivity is also essential for thermoelectric power generation from low-grade waste heat.Source: §1 · p. 3
- Discovery of materials with adaptive thermal properties remains challenging in experiments and simulations because of cost, time, and hardware limitations.Materials informatics combines materials-science simulations or experiments with machine-learning algorithms to accelerate property prediction and discovery.Source: §1 · p. 3
- The review summarizes materials-informatics research on thermal conductivity, interfacial thermal conductance, and thermoelectric conversion efficiency, and discusses remaining challenges and future directions.Source: §1 · p. 4
2. Materials informatics
Materials informatics links materials data to property prediction and discovery through data procurement, representation, and mining. The review describes data sources and descriptors alongside supervised-learning approaches, including Bayesian optimization, random forests, and neural networks.
2. Materials informatics
- The materials-informatics framework comprises data procurement, data representation, and data mining for finding materials, predicting properties, or gaining physical insights.Source: §2 · p. 5
2.1 Data procurement
- Data procurement collects materials' physical properties and structural information from calculations, experiments, and online libraries to construct training databases.Source: §2.1 · p. 5
2.2 Data representation
- Data representation stores material structure and property information using descriptors that map to target thermal-transport properties; descriptor choice is key to prediction and search.Identified descriptors can support intrinsic or extrinsic searches for optimal materials or property prediction within databases.Source: §2.2 · p. 5
2.3 Data mining
- Data mining commonly uses supervised learning; Bayesian optimization updates a Gaussian-process model by sampling promising observations, while random forests combine decision trees and can assess descriptor importance.Random forests are described as robust to outliers and high-dimensional data; the review also describes artificial neural networks trained and tested on input-output relationships.Source: §2.3 · pp. 6–7
3. Thermal conductivity
Materials informatics has been applied to thermal-conductivity discovery and prediction in solids, composites, and liquids. The reviewed work highlights descriptor choice and combines machine-learning models with calculations or experimental data.
3. Thermal conductivity
- Among 79,000 database entries, Carrete et al. proposed three half-Heusler compounds; larger average atomic radii at A and B positions tended to accompany lower thermal conductivity.Source: §3 · p. 8
- Random-forest regression combined with first-principles calculations was introduced to estimate κω for compounds, and descriptor selection substantially affected lattice thermal-conductivity prediction.Tanaka’s group generated compound descriptors from simple atomic representations; their Bayesian-optimization application to lattice thermal-conductivity data showed good predictive performance.Source: §3 · p. 8
- Machine-learning models for composite materials, paired with effective-medium theory, produced effective thermal-conductivity predictions that matched experimental data well.The review concludes that different descriptors should be investigated and compared to reduce prediction deviation.Source: §3 · pp. 8–9
- An artificial neural network predicted ethylene-glycol–water solution thermal conductivity with high accuracy, while an MLP-ANN modeled nanofluid effective thermal conductivity.The reported advantage over practical experiments was lower time and cost.Source: §3 · p. 9
4. Interfacial thermal transport
The review describes machine-learning prediction of interfacial thermal transport and optimization of interfaces for extreme interfacial thermal conductance (ITC).
4. Interfacial thermal transport
- Interfacial thermal transport matters for high-power micro- and opto-electronic devices, where interfaces are numerous and thermal boundary resistance can adjust whole-system thermal conductivity.The review identifies prediction of interfaces with very low or very high thermal boundary resistance as useful for guiding interface discovery.Source: §4 · p. 10
- Artificial neural networks performed best among the tested methods for predicting graphene–h-BN ITC, while machine-learning methods predicted thermal boundary resistance more accurately than AMM and DMM.The thermal boundary resistance predictions were compared with experimental values; the graphene–h-BN methods used temperature, coupling strength, and tensile strain as inputs.Source: §4 · p. 10
- Combining atomistic Green’s function calculations with Bayesian optimization identified Si/Ge superlattice interfaces with the highest and lowest ITC using only a few interface calculations.The review says these results deepen understanding of interfacial thermal-transport mechanisms and indicate cost-effectiveness for designing nanostructures with special thermal properties.Source: §4 · pp. 10–11
5. Thermoelectricity
The review surveys materials-informatics approaches to thermoelectric materials, including property prediction, high-ZT searches, and experimental recommendations. Reported studies identify performance-related features, screen for low thermal conductivity, and optimize or validate materials for thermoelectric applications.
5. Thermoelectricity
- Bayesian optimization identified 221 low-thermal-conductivity materials from 54,779 compounds, while separate work optimized graphene for higher ZT and identified half-Heusler features associated with high ZT [49].The half-Heusler features were a large lattice parameter and either a wide gap at high temperatures or a large effective mass of holes at room temperature; the low-conductivity search alone did not identify high-ZT materials.Source: §5 · p. 12
- An open machine-learning recommendation engine proposed materials with low thermal conductivity and high electrical conductivity, and tested RE12Co5Bi (RE = Gd, Er) compounds were confirmed experimentally to exhibit surprising thermoelectric performance.The review suggests this recommendation-engine paradigm could promote discovery of good thermoelectric materials.Source: §5 · p. 13
- Random forests were reported to predict the Seebeck coefficient accurately, reducing the need to synthesize or calculate materials to obtain that property.The review also identifies prediction of thermoelectric properties as important alongside searching for high-ZT materials.Source: §5 · p. 13
- High-throughput calculations have also been applied successfully to the discovery of new thermoelectric materials [57].Source: §5 · p. 13
- The review describes materials-informatics methods for searching for high-ZT materials and predicting thermoelectric properties, including machine-learning-derived rules and an open recommendation engine for experimental researchers.High ZT requires low thermal conductivity, high electrical conductivity, and a high Seebeck coefficient, despite these properties being closely correlated.Source: §5 · pp. 12–13
6. Summary and perspectives
The review summarizes materials-informatics applications to thermal transport and discusses factors affecting discovery efficiency and predictive efficacy. It identifies unresolved challenges in algorithms, descriptors, physical interpretation, and preparing data for complex nanostructures.
6. Summary and perspectives
- The review presents a framework and conclusions for discovering materials with optimal thermal conductivity, interfacial thermal conductance, and thermoelectric efficiency using materials informatics.It also describes practical implementation procedures and machine-learning algorithms used in the field.Source: §6 · p. 14
- Efficient algorithms, effective descriptors, transfer of data into practical knowledge, and physical interpretation remain challenges for materials informatics.Machine learning alone cannot fully reveal underlying physical mechanisms, so other theoretical or simulation methods are needed.Source: §6 · p. 14
- Preparing data for complex nanomaterials and nanostructures is time-consuming because their larger simulation cells require lengthy calculations.Source: §6 · p. 14
Figure
The figures cover materials relationships, thermal-transport prediction and optimization, and thermoelectric characterization.
Figure 1
- Fig. 1 depicts the processing-structure-property-performance relationships of materials.Source: Figure 1 · p. 16
Figure 2
- Fig. 2 presents a prototype Half-Heusler structure with primitive vectors and a conventional cell, alongside the elements considered in the study.Source: Figure 2 · p. 16, Figure · p. 16
Figure 3
- Fig. 3 concerns correlation between experimental values and values predicted by different methods.Source: Figure 3 · p. 17
Figure 4
- Fig. 4 depicts optimized Si-Si and Si-Ge interface structures for maximum and minimum interfacial thermal conductance.The figure also shows 10 optimization runs with different initial candidate choices and probability distributions of calculated ITC.Source: Figure · p. 17, Figure 4 · p. 17
Figure 5
- Fig. 5 shows crystal structures of K2CdPb and Cs2[PdCl4]I2 predicted to have low lattice thermal conductivity.The stated target is below 0.5 W/mK at 300 K with a band gap below 1 eV.Source: Figure 5 · p. 18, Figure · p. 18
Figure 6
- Fig. 6 characterizes RE12Co5Bi compounds with RE = Gd or Er across temperature.The plotted properties are electrical resistivity, Seebeck coefficient, thermal conductivity, and thermoelectric figure of merit zT.Source: Figure · p. 18, Figure 6 · p. 18