Exploring diamondlike lattice thermal conductivity crystals via feature-based transfer learning

At a Glance

Metadata	Details
Publication Date	2021-05-10
Journal	Physical Review Materials
Authors	Shenghong Ju, Ryo Yoshida, Chang Liu, Stephen Wu, Kenta Hongo
Institutions	Japan Advanced Institute of Science and Technology, The University of Tokyo
Citations	50
Analysis	Full AI Review Included

Exploring Diamond-Like Lattice Thermal Conductivity Crystals via Feature-Based Transfer Learning

Executive Summary

This technical analysis summarizes the findings of the research paper concerning the identification of ultrahigh lattice thermal conductivity ($\kappa_L$) materials using advanced machine learning (ML) techniques, directly correlating these findings with 6CCVD’s MPCVD diamond capabilities for thermal management applications.

Novel Screening Methodology: A feature-based transfer learning approach successfully bridged the gap between “big data” (harmonic phonon properties, 320 crystals) and “small data” (thermal conductivity, 45 crystals) to enable extrapolative prediction.
Diamond Alternatives Identified: Screening over 60,000 inorganic compounds identified 14 novel crystals with ultrahigh $\kappa_L$, serving as potential alternatives to diamond for heat spreaders.
Record Thermal Conductivity: Cubic Boron Arsenide (BAs) was calculated to possess the highest $\kappa_L$ (3411 W/mK), exceeding the benchmark diamond (3048 W/mK).
Critical Calculation Method: The study validates that accurate $\kappa_L$ prediction for materials exceeding 1000 W/mK requires solving the Iterative Boltzmann Transport Equation (IBTE), as the Relaxation Time Approximation (RTA) significantly underestimates these values.
Material Physics Insights: High $\kappa_L$ is correlated with low three-phonon scattering phase space ($P_3$) and low average/maximum dipole polarizability and Van der Waals radius, descriptors qualitatively related to anharmonicity.
Application Focus: The identified materials (C, BAs, BN, BC${2}$N, C${3}$N$_{4}$) are critical for next-generation thermal management solutions in high-power electronic and optical devices.

Technical Specifications

The following table extracts key quantitative data points and material properties relevant to ultrahigh thermal conductivity from the research.

Parameter	Value	Unit	Context
Highest Calculated $\kappa_L$	3411	W/mK	Cubic BAs (mp-ID: 10044), Iterative BTE
Diamond Benchmark $\kappa_L$	3048	W/mK	Fd-3m structure, Iterative BTE
Cubic BN $\kappa_L$	1876	W/mK	F-43m structure, Iterative BTE
Wurtzite BAs $\kappa_L$	2947	W/mK	P6₃mc structure, Iterative BTE
Lowest $P_3$ Value	0.6397	10^-4 cm	Cubic BAs
Diamond $P_3$ Value	1.0005	10^-4 cm	Fd-3m structure
$\kappa_L$ Training Data Range	< 370	W/mK	Small data set (45 materials)
$\kappa_L$ Extrapolated Range	1000 - 3000	W/mK	14 identified crystals
DFT Energy Convergence Threshold	10^-8	eV	Final thermal conductivity calculation
DFT Force Convergence Threshold	< 2	meV/Å	Hellmann-Feynman forces
RTA vs. IBTE Difference	Significant	W/mK	For $\kappa_L$ > 1000 W/mK (RTA underestimates)

Key Methodologies

The research employed a sophisticated computational workflow combining database screening, machine learning, and first-principles calculations to identify high-$\kappa_L$ materials.

Database Screening: Over 60,000 inorganic crystal entries from the Materials Project database were initially screened. Materials with band gaps smaller than 0.1 eV or molecular crystals were excluded.
Feature Property Calculation ($P_3$): The three-phonon scattering phase space ($P_3$) was calculated for 320 materials using harmonic interatomic force constants derived from Density Functional Perturbation Theory (DFPT) via the ALAMODE package. $P_3$ served as the proxy feature property.
Transfer Learning Model Training:
- A fully-connected pyramid neural network was pre-trained using the 320 $P_3$ instances and 290 compositional descriptors.
- The pre-trained subnetwork was then transferred and combined with a Random Forest model, trained on a small dataset of 45 known $\kappa_L$ values.
First-Principles Validation: The top-14 candidates identified by the ML model (those with the lowest predicted $P_3$) were validated using detailed anharmonic lattice dynamics calculations.
Accurate $\kappa_L$ Determination: Lattice thermal conductivity ($\kappa_L$) was calculated by solving the Iterative Boltzmann Transport Equation (IBTE), which is essential for accurate results in ultrahigh conductivity crystals where the Relaxation Time Approximation (RTA) fails.
DFT Parameters: Calculations utilized Quantum ESPRESSO with kinetic energy cutoffs of 80 Ry (wave functions) and 400 Ry (charge density).

6CCVD Solutions & Capabilities

The research confirms that diamond and diamond-like materials (Cubic BAs, Cubic BN) remain the gold standard for ultrahigh thermal conductivity, essential for next-generation thermal management. 6CCVD provides the high-quality MPCVD diamond materials and customization services required to immediately implement or extend this research.

Research Requirement/Finding	6CCVD Solution & Value Proposition
Benchmark Material (Diamond, $\kappa_L$ ~3048 W/mK)	Optical Grade Single Crystal Diamond (SCD): To achieve the intrinsic thermal conductivity limits validated by the IBTE method, high purity and low defect density are paramount. Our MPCVD SCD offers the highest crystalline quality, ensuring minimal phonon scattering and maximum $\kappa_L$.
Need for Large-Area Heat Spreaders	Polycrystalline Diamond (PCD) Wafers up to 125mm: For scaling thermal management solutions in power electronics and large optical systems, 6CCVD provides PCD wafers up to 125mm in diameter, offering high thermal performance across large areas.
Surface Quality for Integration	Precision Polishing (Ra < 1nm SCD, < 5nm PCD): The integration of diamond into devices requires ultra-smooth surfaces. We offer polishing down to Ra < 1nm for SCD and Ra < 5nm for inch-size PCD, minimizing thermal boundary resistance and surface scattering.
Exploration of Diamond Alternatives (e.g., BN, BAs)	Custom Substrates and Boron-Doped Diamond (BDD): While BAs and complex nitrides are promising, 6CCVD provides the foundational materials and custom substrates (up to 10mm thick) necessary for synthesizing and integrating these novel compounds, including Boron-Doped Diamond (BDD) for applications requiring controlled electrical conductivity.
Device Integration and Contacting	In-House Custom Metalization: Successful device integration requires specific contact layers. 6CCVD offers internal metalization capabilities, including Au, Pt, Pd, Ti, W, and Cu, tailored to meet the thermal and electrical requirements of high-power devices.
Extrapolative Prediction & Material Selection	Expert Engineering Support: The complexity of predicting $\kappa_L$ (requiring IBTE and advanced descriptors like $P_3$) necessitates expert material consultation. 6CCVD’s in-house PhD team can assist engineers and scientists in translating theoretical ML findings into practical material specifications for similar ultrahigh thermal conductivity projects.

For custom specifications or material consultation, visit 6ccvd.com or contact our engineering team directly.

View Original Abstract

Ultrahigh lattice thermal conductivity materials hold great importance since\nthey play a critical role in the thermal management of electronic and optical\ndevices. Models using machine learning can search for materials with\noutstanding higher-order properties like thermal conductivity. However, the\nlack of sufficient data to train a model is a serious hurdle. Herein we show\nthat big data can complement small data for accurate predictions when\nlower-order feature properties available in big data are selected properly and\napplied to transfer learning. The connection between the crystal information\nand thermal conductivity is directly built with a neural network by\ntransferring descriptors acquired through a pre-trained model for the feature\nproperty. Successful transfer learning shows the ability of extrapolative\nprediction and reveals descriptors for lattice anharmonicity. Transfer learning\nis employed to screen over 60000 compounds to identify novel crystals that can\nserve as alternatives to diamond.\n

Tech Support

Original Source

DOI: https://doi.org/10.1103/physrevmaterials.5.053801