The use of machine learning in computational molecular design has great potential to accelerate the discovery of innovative materials. However, its practical benefits still remain unproven in real-world applications, particularly in polymer science. We demonstrate the successful discovery of new polymers with high thermal conductivity, inspired by machine-learning-assisted polymer chemistry. This discovery was made by the interplay between machine intelligence trained on a substantially limited amount of polymeric properties data, expertise from laboratory synthesis and advanced technologies for thermophysical property measurements. Using a molecular design algorithm trained to recognize quantitative structure—property relationships with respect to thermal conductivity and other targeted polymeric properties, we identified thousands of promising hypothetical polymers. From these candidates, three were selected for monomer synthesis and polymerization because of their synthetic accessibility and their potential for ease of processing in further applications. The synthesized polymers reached thermal conductivities of 0.18–0.41 W/mK, which are comparable to those of state-of-the-art polymers in non-composite thermo-plastics.

There is a growing demand for the use of machine learning (ML) to derive fast-to-evaluate surrogate models of materials properties. In recent years, a broad array of materials property databases have emerged as part of a digital transformation of materials science. However, recent technological advances in ML are not fully exploited because of the insufficient volume and diversity of materials data. An ML framework called “transfer learning” has considerable potential to overcome the problem of limited amounts of materials data. Transfer learning relies on the concept that various property types, such as physical, chemical, electronic, thermodynamic, and mechanical properties, are physically interrelated. For a given target property to be predicted from a limited supply of training data, models of related proxy properties are pretrained using sufficient data; these models capture common features relevant to the target task. Repurposing of such machine-acquired features on the target task yields outstanding prediction performance even with exceedingly small data sets, as if highly experienced human experts can make rational inferences even for considerably less experienced tasks. In this study, to facilitate widespread use of transfer learning, we develop a pretrained model library called XenonPy.MDL. In this first release, the library comprises more than 140 000 pretrained models for various properties of small molecules, polymers, and inorganic crystalline materials. Along with these pretrained models, we describe some outstanding successes of transfer learning in different scenarios such as building models with only dozens of materials data, increasing the ability of extrapolative prediction through a strategic model transfer, and so on. Remarkably, transfer learning has autonomously identified rather nontrivial transferability across different properties transcending the different disciplines of materials science; for example, our analysis has revealed underlying bridges between small molecules and polymers and between organic and inorganic chemistry.

The heat transport properties of aromatic polyimide thin films have become more important in the use for the electric insulation in the microelectronic devices with highly integrated circuits. The various kinds of measuring methods have been applied to obtain the anisotropic thermal conductivity and thermal diffusivity of thin films, however, if the specimens are soft and transparent, the conventional methodology requires highly advanced technology in preparing the specimens for the measurement and the results obtained vary widely. The purpose of this study is to apply the temperature wave analysis (TWA) method to measure the thermal diffusivity of thin films and spin-coated layers of aromatic polyimide in the thickness direction at various temperatures. The TWA is an absolute method to determine the thermal diffusivity by using the phase shift of temperature wave. We have performed measurements on the five different chemical structures of aromatic polyimide, including polyimide isoindoloquinazolinedione (PIQ), pyromellitic dianhydride and 4,4′-oxydianiline (PMDA/ODA), 3,3′,4,4′,-biphenyltetracarboxylic dianhydride and p-phenylenediamine (BPDA/PPDA), 3,3′,4,4′-biphenyltetracarboxylic dianhydride and 4,4′-oxydianiline (BPDA/ODA), and 3,3′,4,4′-benzophenonetetracarboxylic dianhydride and 3,3′-diaminobenzophenone (BTDA/DAB). As a result, thermal diffusivity of thin films in a thickness range from 0.1 to 300 μm at a temperature range from 10 to 570 K is obtained. The thickness dependence of thermal diffusivity of spin-coated layers of PIQ exhibits a good coincidence with the tendency of molecular anisotropy observed by attenuated total reflection Fourier transform infrared spectroscopy. In the low temperature below 20 K the amorphous PMDA/ODA film exhibits a substantial increase, which can be understood by considering the phonon mean free path of amorphous systems as first noted by Kittel for inorganic glasses. The glass transition of BTDA/DAB is observed as a rapid decrease of thermal diffusivity. The results indicate the availability of TWA for the present needs of accurate but simple measuring technique to determine the thermal diffusivity of thin polymer films, which is strongly dependent on the chemical and anisotropic structures.

Thermal diffusivity and thermal conductivity of single crystals of ${\mathrm{Nd}}^{3+}$ doped $\mathrm{Gd}\mathrm{V}{\mathrm{O}}_4$ , $\mathrm{Y}\mathrm{V}{\mathrm{O}}_4$ , $\mathrm{Lu}\mathrm{V}{\mathrm{O}}_4$ , and ${\mathrm{Y}}_3{\mathrm{Al}}_5{\mathrm{O}}_{12}$ are precisely measured over a wide range of doping concentration from $0.5\text{to}15\mathrm{at.}\%$ by temperature wave analysis. Thermal diffusivity serves as the most sensitive parameter to detect the effect of doping on thermal properties, where ${\mathrm{Nd}}^{3+}$ doped $\mathrm{Gd}\mathrm{V}{\mathrm{O}}_4$ exhibits a decrease in thermal diffusivity (it has changed about 20% in their values in the $c$ axis) but an increase in heat capacity (only 1.7%). It has long been understood that the thermal conductivity of $\mathrm{Y}\mathrm{V}{\mathrm{O}}_4$ is inferior to that of ${\mathrm{Y}}_3{\mathrm{Al}}_5{\mathrm{O}}_{12}$ ; however, the thermal conductivity of $\mathrm{Y}\mathrm{V}{\mathrm{O}}_4$ in the $c$ axis shows the highest value in all four crystals compared at $1\mathrm{at.}\%$ of ${\mathrm{Nd}}^{3+}$ doping concentration. Thermal conductivity exhibits a decrease ($\propto {\epsilon }^{-1∕2}$ , $\epsilon$ : mass variance) with an increase of doping concentration, that is characteristic of Klemens’ point defect model for the phonon scattering. In the numerical fitting, the anisotropic decrease of thermal conductivity in different crystal axes, [100], [001], and [110], is reflected as the different phonon velocities. The simulated and the experimental thermal conductivity show the various decreasing curvatures indicating that the effect of mass difference of the host and the doping cations serves as an important factor, as with the crystallographic anisotropy, on the thermal conductivity of the doped single crystals.