# Fileset

[micromachines-14-00186.pdf](https://mdr.nims.go.jp/filesets/bf8cec5a-1805-48b9-81c7-02a524097369/download)

## Creator

[Yen-Ju Wu](https://orcid.org/0000-0003-2647-3407), [Yibin Xu](https://orcid.org/0000-0001-8600-8748)

## Rights



## Other metadata

[Data-Driven Design of Transparent Thermal Insulating Nanoscale Layered Oxides](https://mdr.nims.go.jp/datasets/c222a45f-0984-474f-9118-a9cb989c15ca)

## Fulltext

Data-Driven Design of Transparent Thermal Insulating Nanoscale Layered OxidesCitation: Wu, Y.-J.; Xu, Y.Data-Driven Design of TransparentThermal Insulating NanoscaleLayered Oxides. Micromachines 2023,14, 186. https://doi.org/10.3390/mi14010186Academic Editor: Rui LiReceived: 13 December 2022Revised: 28 December 2022Accepted: 5 January 2023Published: 11 January 2023Copyright: © 2023 by the authors.Licensee MDPI, Basel, Switzerland.This article is an open access articledistributed under the terms andconditions of the Creative CommonsAttribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).micromachinesArticleData-Driven Design of Transparent Thermal InsulatingNanoscale Layered OxidesYen-Ju Wu 1,2,* and Yibin Xu 11 Research and Service Division of Materials Data and Integrated System (MaDIS), National Institute forMaterials Science (NIMS) 1-1 Namiki, Tsukuba 305-0044, Ibaraki, Japan2 International Center for Young Scientists (ICYS), National Institute for Materials Science (NIMS), 1-2-1 Sengen,Tsukuba 305-0047, Ibaraki, Japan* Correspondence: wu.yenju@nims.go.jpAbstract: Predicting the interfacial thermal resistance (ITR) for various material systems is a time-consuming process. In this study, we applied our previously proposed ITR machine learning modelsto discover the material systems that satisfy both high transparency and low thermal conductivity. Theselected material system of TiO2/SiO2 shows a high ITR of 26.56 m2K/GW, which is in good agree-ment with the predicted value. The nanoscale layered TiO2/SiO2 thin films synthesized by sputteringexhibits ultralow thermal conductivity (0.21 W/mK) and high transparency (>90%, 380–800 nm).The reduction of the thermal conductivity is achieved by the high density of the interfaces with ahigh ITR rather than the change of the intrinsic thermal conductivity. The thermal conductivity ofTiO2 is observed to be 1.56 W/mK with the film thickness in the range of 5–50 nm. Furthermore,the strong substrate dependence is confirmed as the thermal conductivity of the nanoscale layeredTiO2/SiO2 thin films on quartz glass is three times lower than that on Si. The proposed TiO2/SiO2composites have higher transparency and robustness, good adaptivity to electronics, and lower costthan the current transparent thermal insulating materials such as aerogels and polypropylene. Thegood agreement of the experimental ITR with the prediction and the low thermal conductivity ofthe layered thin films promise this strategy has great potential for accelerating the development oftransparent thermal insulators.Keywords: thermal conductivity; thermal insulator; thin film; superlattice; transparency1. IntroductionTransparent thermal insulating materials have been widely used in decreasing heatloss, collecting solar energy, and increasing efficiency of clean energy usage, such as thermalcollectors [1], phase-change memory that relies on self-heating to switch between memorystates [2], thermoelectrics [3], and thermal insulation windows [4]. For such applications,low thermal conductivity and high transmittance are the two essential properties. Theability to reduce heat loss and to provide high transmittance depends on the materialand operating temperature. Porous materials such as aerogels, polymer materials such aspolypropylene, and amorphous SiO2 (a-SiO2) have been conventionally used as transparentthermal insulators. More specifically, aerogels have been frequently used to provide goodthermal insulation. However, there are still issues to be overcome for those materials. Forexample, ensuring good thermal insulation and the high transparency of silica aerogels stillremains to be a big challenge [5]. For porous materials, their low robustness and high costlimit their application. For polymer materials, although they generally show lower thermalconductivity and relatively high transmittance of ~80%, their low melting points reduce itspracticality at high temperatures. As for inorganic a-SiO2, although it is commonly used asa thermal insulator in electronics, its thermal conductivity (1.38 W/mK) [6] is higher thanthat of aerogels or polymers.Micromachines 2023, 14, 186. https://doi.org/10.3390/mi14010186 https://www.mdpi.com/journal/micromachineshttps://doi.org/10.3390/mi14010186https://doi.org/10.3390/mi14010186https://creativecommons.org/https://creativecommons.org/licenses/by/4.0/https://creativecommons.org/licenses/by/4.0/https://www.mdpi.com/journal/micromachineshttps://www.mdpi.comhttps://orcid.org/0000-0003-2647-3407https://doi.org/10.3390/mi14010186https://www.mdpi.com/journal/micromachineshttps://www.mdpi.com/article/10.3390/mi14010186?type=check_update&version=1Micromachines 2023, 14, 186 2 of 12On the other hand, inorganic nanocomposite structures with periodic or aperiodicmultilayers have become new prominent candidates to reduce thermal conductivity toa value that is even lower than that of homogeneous amorphous structures. Cahill et al.found that the Si–Ge superlattice exhibited a large reduction in its thermal conductivitybecause of the interfacial thermal resistance (ITR) [7]. The larger interface density wasproven to contribute to the lower thermal conductivity compared to the continuous filmsin various nanolaminates [8–12]. Hu et al. proposed the aperiodic layered structures ofgraphene and MoS2 to reduce thermal conductivity [13]. In such structures, it is consideredthat the phonon propagation is hindered by scattering into random directions or associatedinterferences when phonons encounter interfaces in nanostructured materials.So far, various methods to describe the ITR at the interface are proposed such as theacoustic mismatch model (AMM) [14], the diffuse mismatch model (DMM) [15], and themolecular dynamics (MD) simulation [16–18]. Although these models and simulationsprovide a theoretical way to evaluate the ITR, it is still difficult to achieve high accuracyor to conduct high-throughput predictions. Although AMM and DMM provide simplepictures on the ITR, many factors are overlooked which may affect the ITR in those models.MD simulations can include those factors, but their computational cost is generally highand predicting the ITR with high accuracy would be a time-consuming process. Anotherapproach to predict the ITR is machine learning and we have shown this approach in ourprevious work [19]. Once we trained a machine learning model, making predictions is asignificantly time-efficient process compared to the conventional methods. By utilizingmachine learning and the databases for the ITR and other materials properties, we canconstruct a model that takes into account various chemical, physical, and synthesis processfactors to predict the ITR with high accuracy [19,20]. In our previous studies, we havesuccessfully realized the nanocomposite thin films with ultralow thermal conductivity bythe interfacial design based on machine learning [19,21]. In this study, we further discusshow we consider the transparent property through the combination of machine learningand experimental optimization to realize the transparent thermal insulators based on theITR machine learning model constructed in our previous works.High optical transmittance is another crucial issue to be addressed in relation totransparent thermal insulating materials. To discover material systems that satisfy bothproperties, the search space for material candidates should be confined to transparentmaterials with large band gaps. We selected transparent material systems with high ITRbased on the prediction by the ITR model. Further, we synthesized those selected materialsystems through the nanocomposite optimization by sputtering and measuring their ITRand optical transmittance. The thermal conductivity and the ITR were measured by thefrequency-domain thermoreflectance (FDTR). The structures of the layered films withvarious periodic thicknesses and substrates were characterized by X-ray diffraction (XRD)and transmission electron microscopy (TEM).2. Experimental Procedure2.1. Thin Film DepositionThe TiO2/SiO2 layered thin film samples were prepared on quartz glass (Qz) or Sisubstrates in the sputtering system (CFS-4EP-LL, Shibaura Mechatronics Corp., Yokohama,Japan) at a pressure of approximately 6 × 10−5 Pa before deposition. The pressure wasmaintained at 0.4 Pa (20 sccm Ar flow) during the deposition process. Ar was used asthe sputtering gas for Au at 20 sccm, whereas both Ar and O2 were applied for TiO2 (Ar:16 sccm; O2: 4 sccm) and SiO2 (Ar: 13 sccm; O2: 13 sccm). The RF power for both TiO2 andSiO2 was set at 200 W, while the DC power was set at 50 W for Au. TiO2:SiO2 in Table 1shows the thicknesses of TiO2 and SiO2 layers corresponding to the quartz crystal resonator.The thicknesses of TiO2 and SiO2 were set to be 1 nm, 5 nm, and 30 nm. The 30 nm sample(total thickness is 60 nm) is used to validate the ITR with the predicted values, and the 1 nmand 5 nm samples (total thickness is 100 nm) are used for analyzing the interface effecton the thermal conductivity. After the deposition of the TiO2/SiO2, an Au layer with theMicromachines 2023, 14, 186 3 of 12thickness of 120 nm was deposited without evacuation at the top as a heat absorber forthe thermal measurement. The total film thickness and the thickness of each layer wereanalyzed through TEM (JEM-ARM200F, JEOL Ltd., Tokyo, Japan). The structural propertiesof the thin film were characterized through XRD (SmartLab, Rigaku Corp., Tokyo, Japan).Table 1. Experimental parameters of the TiO2/SiO2 samples.Sample TiO2:SiO2 [nm] The Number ofInterfaces SubstrateTS-Qz-30 30:30 2 QzTS-Qz-5 5:5 20 QzTS-Qz-1 1:1 100 QzTS-Si-30 30:30 2 SiTS-Si-5 5:5 20 SiTS-Si-1 1:1 100 Si2.2. Heat Conduction Equation for the ITR and Thermal ConductivityThe thermal resistance measurement was performed through FDTR [22]. The thermalresistance was measured along the perpendicular direction (cross-plane) to the Qz or Sisubstrate. Heat conduction was assumed to be one-dimensional because the laser spot wasmuch larger than the film thickness (Equation (1)) [23].T(0)qd0=e−i π4√2ωλ3C3+ R0 +(1− λ2C2λ3C3)d2λ2+(1− λ1C1λ3C3)d1λ1+(12− λ0C0λ3C3)d0λ0(1)where T(0) is the Au temperature; q is heat per unit volume; C is heat capacity per unitvolume; λ is thermal conductivity; R0 is the sum of the ITRs at Au/SiO2, TiO2/SiO2 (itvaries with the interface number), and TiO2/substrate; subscripts 0, 1, 2, and 3 denote Au,TiO2, SiO2, and the substrate, respectively. The temperature on the Au film surface, T(0),was detected through the thermoreflectance method using a probe laser with the appliedalternating current with the frequency ω. If we plot T(0)qd0versus ω−1/2, the intercept Rgives the sum of the last four terms of Equation (1). The second term, R0, can be calculatedwith the known thickness, specific heat, and thermal conductivity of the Au, SiO2, TiO2films, and the substrate. We also conducted the measurement for the TiO2 films withdifferent thicknesses and plotted R as a function of d. The slope of the line was equalto 1λ1− C1λ3C3. The thermal conductivity of TiO2 (λ1) was obtained as 1.56 W/mK at allthe thicknesses we measured: 5 nm, 10 nm, 30 nm, and 50 nm. The value was lowerthan the reported value of bulk polycrystalline TiO2 (8.9 W/mK) [24]. Table 2 presentsthe thermophysical properties of the materials used in the calculation. We subsequentlydetermined the thermal conductivity along the cross-plane by dividing the film thicknessby the total thermal resistance.Table 2. Thermophysical properties of the materials used in the calculation.Material Volumetric HeatCapacity (×106 J/m3K)Thermal Conductivity(W/mK)ITR(m2K/GW)Au 2.509 [24,25] 298 [24,25] -TiO2 2.76 [24,25] 1.56 -Si 1.66 [24,25] 148 [24] -SiO2 1.65 [24,25] 1.38 [6,24] -Au/SiO2 5 [6]Au/TiO2 and TiO2/Qz 23.26Micromachines 2023, 14, 186 4 of 123. Result3.1. Data-Driven Material SelectionFirst, the proposed ITR machine learning model [19] was applied to select the trans-parent interface with high ITR. Figure 1 shows the flow chart of our data-driven schemefor the materials selection. We used ITR dataset, which was collected from 87 publishedpapers, and the descriptor dataset constructed for 298 single-element materials or binarycompounds. The ITR dataset contains the ITR values of various interfaces with temperature,the synthesis method, the thermal measurement method, the sample pretreatment, and itsoriginal references. The descriptor dataset is composed of the physical (e.g., melting point,density), chemical (e.g., electronegativity, binding energy), and process (e.g., film thickness)descriptors. The details of the datasets are available in our previous work [26].Figure 1. Schematics of the data-driven material selection to combine the machine-learning methodand experimental validation. (a) ITR prediction model. Detailed information of the datasets ispresented in our previous work [26]. (b) Thermal conductivity of various transparent materials andtheir predicted ITR with SiO2. (c) Schematic of the experimental validation set up. The images showthe FDTR of ITR measurement. The probe laser of 635 nm and pump laser of 405 nm were applied fordetecting and heating the sample. The bilayer thin film with Au-top layer deposited on the substratewas used for evaluating the ITR value after the prediction with the model. TiO2/SiO2 with a highITR of 26.56 m2K/GW (pink star in (b)), which was close to the predicted values, was selected.According to our previous works, ordinal least-square linear regression showed poorpredictive performance and least absolute shrinkage and selection operator (LASSO) ne-glected important descriptors such as measurement temperature [19,26]. Support vectormachine regression (SVM), Gaussian process regression (GPR), and LSBoost regressionwere used in our previous studies in consideration of the dataset size and their abilityto describe nonlinear effects. SVM and GPR are kernel-based methods and radial basisfunction (RBF) kernel was used for both methods. LSBoost regression performs least-squareboosting which fits regression ensembles to minimize mean-squared error. The coefficientof determination, R2, of SVM, GPR, and LSBoost are 0.879, 0.916, and 0.919, respectively.The initial algorithm settings and additional details on those algorithms are described inour previous papers [19,20].The transparent materials in the searching space were screened by the band gap thatwas >2.8 eV to be transparent in the visible range. Note that some transparent materialsmight be excluded due to the underestimation of the simulated values for band gap [27].The 70 materials, which met the band gap criterion and had all the necessary descriptorsfor the ITR prediction, formed more than 4800 possible candidate interfaces. To reducethe number of candidates, one of the materials in the interface was fixed as SiO2 dueto its easiness to synthesize and its low thermal conductivity. The ITR values predictedby our machine learning models and experimental thermal conductivity values in thepolycrystalline bulk form, which were collected from Thermophysical Properties ResearchMicromachines 2023, 14, 186 5 of 12Center Data Series [24], are shown in Figure 1b. The ITR values predicted by SVM, GPR,and LSBoost models are shown by the blue, purple and green colors, respectively. Generally,the predicted values differ among various models even if the coefficient of determination ofR2 is all higher than 0.85. Therefore, we selected the candidates that were predicted to havehigh ITR by all the three models. The experimental ITR of TiO2/SiO2 (TS-Qz-30 sample),26.56 m2K/GW, was close to the predicted values (Figure 1b). After the experimentalvalidation of the ITR of TiO2/SiO2, we further synthesized TiO2/SiO2 into nanoscalelayered thin films (Figure 1c) by sputtering to increase the interface density and analyzethe ITR effect on the thermal conductivity.3.2. StructureFigure 2 shows the XRD pattern of the TiO2/SiO2 samples. The samples on both theQz and Si substrates showed rutile TiO2(210) and Au phases of the top layer, indicating thatthe films were composed of crystalline TiO2 and amorphous SiO2. The SiO2(100) peak camefrom the Qz substrate instead of the layered thin film. Interestingly, the additional phasesof Ti2O3 (104), (110), and (214) only existed in the samples on the Si substrates, although thesamples with the same layer thickness on Qz and Si were simultaneously deposited in thesame sputtering. The existence of these additional phases may be attributed to the sametetrahedral atomic environment of O in Ti2O3 and of Si in Si and the similar atomic distancebetween O–Ti (0.203 nm) in Ti2O3 and Si–Si (0.235 nm) in Si. Moreover, the peak intensity ofTi2O3 (104) increased with the number of interfaces, implying the strong relation betweenformation of the Ti2O3 phase and the interfacial region.Figure 2. X-ray diffraction (XRD) of TiO2/SiO2 on the quartz glass (Qz) (a) and Si (b) substrates.Figure 3 presents TEM images of the layered samples on the Qz and Si substrates. Theexperimental parameters are shown in Table 1 for samples of (a,e) TS-Qz-1, (b,f) TS-Qz-5,(c,g) TS-Si-1, and (d,h) TS-Si-5. All films were deposited well without porosities. TheMicromachines 2023, 14, 186 6 of 12thickness of each layer matched the sputtering deposition setting of 1 and 5 nm. Theinterface of the thinner-layer samples (Figure 3e,g) was unclear when compared with thethicker-layer samples (Figure 3f,h).Figure 3. TEM images of the TiO2/SiO2 multilayered samples. Samples on quartz glass (Qz) of(a,e) 1 nm and (b,f) 5 nm for each layer and on Si of (c,g) 1 nm and (d,h) 5 nm for each layer.The total thickness of all samples was 100 nm. The scale bar is 50 nm and 10 nm for upper andbottom, respectively.3.3. TransmittanceFigure 4 shows that the nanoscale layered TiO2/SiO2 has high transmittance in thevisible range of 380–800 nm. It shows higher transmittance in the range of 380–500 nmwhile lower transmittance in the range of 550–800 nm compared to Qz. The averagetransmittance in the visible range of the nanoscale layered TiO2/SiO2 reached 92.6% and91.1% for TS_Qz_1 and TS_Qz_5, respectively.Figure 4. The transmittance of the nanoscale layered TiO2/SiO2. Inset: images of the corresponding films.Micromachines 2023, 14, 186 7 of 123.4. ITR and Thermal Conductivity at Room TemperatureTable 3 presents the cross-plane thermal conductivity (λ⊥) of the samples with var-ious number of interface (N). R0*, which is obtained by subtracting the ITR of Au/SiO2(5 m2K/GW) [6] from R0 (the second term in Equation (1)), is the ITR of all the TiO2/SiO2interfaces. The ITR and thermal conductivity were measured by FDTR and evaluated viaEquation (1) with the thermophysical properties in Table 2. Detailed information is shownin Section 2.2.Table 3. Thermal conductivity of the samples. N is the number of the interfaces.Sample N λ⊥(W/mK) R0*/N (m2K/GW)TS-Qz-30 2 0.65 25.64TS-Qz-5 20 0.26 15.71TS-Qz-1 100 0.21 4.08TS-Si-30 2 0.97 10.38TS-Si-5 20 0.96 1.82TS-Si-1 100 0.54 1.18The thermal conductivity of the samples on Qz decreased from 0.26 W/mK to 0.21 W/mKand that of the samples on Si decreased from 0.96 W/mK to 0.54 W/mK with the increasinginterfaces (N: 20 to 100). All samples deposited on Qz showed lower thermal conductivitiesthan those on the Si substrates. The ITR of each interface (R0*/N) decreased as the thicknessof each layer decreased.4. Discussion4.1. ITRThe thermal conductivity and the ITR of all samples with different layer thicknesseswere measured and calculated, as described in Section 3.4. The TiO2/SiO2 samples on Qzand Si both exhibited a decrease in the thermal conductivity compared with the respectivebulk constituents of SiO2 and TiO2. Their thermal conductivities at 300 K decreased by 85%from SiO2 and by 87% from TiO2.If we use the thermal resistor model reported by Böttger et al. [28] in Equation (2) topredict the thermal conductivity (λ⊥), which is in the case of negligible interface resistance,we obtain:λ⊥ = λ2λ1/λ2(1 + d1/d2)d1/d2 + λ1/λ2(2)where λ represents thermal conductivity; d represents thickness; and subscripts 1 and2 represent the material beside the interface. The estimated thermal conductivity of themultilayered TiO2/SiO2 of 1.46 W/mK was higher than our observation in Table 3. Theoverestimation of λ⊥ indicated that the ITR contribution cannot be neglected.4.2. Structural Effect on the ITRThe ITR decreased as the thickness of each layer decreased from 25.64 m2K/W (30 nm,Qz) to 4.08 m2K/W (1 nm, Qz) (Table 3). The structure of the interfacial region approachedthe amorphous state as the thickness of each layer decreased. The phonon modes availablefor heat transfer were then broadened [29]. This may cause more overlapping phononmodes, in which heat can be transported between two materials beside the interfaces,resulting in a lower ITR. Moreover, with the increase of interface density, the layer thicknessbecame thinner than the mean free path of the dominant phonons. The phonons mayhave a high possibility to transport (tunnel) across the interfaces and attribute to the lowerthermal resistance at the interfaces [11]. In short, the thickness-dependent ITR is significantfor the nanoscale layered TiO2/SiO2. All samples deposited on Qz (without Ti2O3 phases)showed lower thermal conductivities relative to those on the Si substrates (with Ti2O3Micromachines 2023, 14, 186 8 of 12phases), corresponding to Figure 2. This may imply the low ITR of Ti2O3/SiO2 relative toTiO2/SiO2.To summarize the structural effect of the nanoscale layered TiO2/SiO2 on the ITR, thefollowing three are preferred to realize low thermal conductivity (high ITR): (1) samplesdeposited on Qz to prevent the Ti2O3 phase formation; (2) sharp interface; and (3) highinterface density. A trade-off existed between the film thickness and the interface densitywhen the total thickness was fixed. The aperiodic stacking order can disrupt the secondaryperiodicity (superlattice phonons) and reduce the phonon lifetimes, resulting in lowerthermal conductivity. Therefore, the stacking order can be further optimized by combiningit with the proposed strategy of material system selection.4.3. Comparison with Other Transparent Layered MaterialsTable 4 and Figure 5 show the thermal conductivities of the reported inorganic pore-free materials ranging from 0.48 W/mK to 18.0 W/mK. The materials were categorized intosingle crystals and composite nanostructures. The composite nanostructures were classifiedby the band gap value (Eg) of 3 eV. If the Eg of at least one material of the compositenanostructure is smaller than 3 eV (composite nanostructure 1), it is considered to benontransparent. If the Eg of both materials are larger than 3 eV (composite nanostructure 2),it is considered to be transparent. Most of the composite nanostructures reach a lowerthermal conductivity than the single crystal (e.g., PbTe [30], Zn4Sb3 [31], SnSe [32], andAgSbTe2 [33]). In our previous work, Bi/Si reached an ultralow thermal conductivity of0.16 W/mK [21], which is as low as that of polymer materials. Hu et al. proposed anaperiodic stacking composite nanostructure of graphene/MoS2 with an even lower thermalconductivity [13]. The abovementioned thermal insulators are unsuitable for transparentapplications because of the smaller Eg of the materials in the nanostructure. In comparisonwith the reported composite nanostructures composed of the materials with Eg > 3.0 eV, thethermal conductivity of the layered TiO2/SiO2 in this study achieves the lowest values of0.21 W/mK as shown in Figure 5. The characteristics of lower cost and higher robustnessthan porous materials (e.g., aerogel), the higher temperature tolerance than polymers, andlower thermal conductivity than the a-SiO2 (common thermal insulators in electronics)promise the nanoscale layered TiO2/SiO2 as transparent thermal insulators for thermalcollectors, thermal insulation windows, or thin film thermoelectric.Table 4. Thermal conductivity at room temperature (R.T.) of the reported inorganic pore-free materials.ID Material Thermal Conductivity(W/mK) Ref.1 PbTe 2 [30]2 Zn4Sb3 5 [31]3 SnSe 0.6 [32]4 AgSbTe2 0.7 [33]5 Si/SiGe 9.6 [34]6 GaAs/AlAs 6.5 [35]7 Si/Ge SLNWs (superlattice nanowires) 1.87 [36]8 Si/SiO2 0.75 [37]9 W/Al2O3 0.53 [38]10 Ta/TaOx 0.37 [39]11 Mo/Si 1.2 [40]12 Ge2Sb2Te5/ZnS:SiO2 0.25 [41]13 Au/Si 0.33 [42]14 Bi2Te3/Sb2Te3 0.33 [43]15 Bi/Si (previous work) 0.16 [21]16 graphene/MoS2 0.03 [13]Micromachines 2023, 14, 186 9 of 12Table 4. Cont.ID Material Thermal Conductivity(W/mK) Ref.17 TiN/SiO2 2 [44]18 TiN/AlCrN 2.4 [28]19 ZrO2/Y2O3 1.4 [7]20 ZrO2:Y/SiO2 1.15 [7]21 Si3N4/SiO2 1.14 [2]22 HfO2/Al2O3 1.02 [9]23 TiO2/Al2O3 0.95 [10]24 Al2O3/SiO2 0.57 [45]25 Cr2O3/SiO2 0.52 [45]26 Y2O3/SiO2 0.5 [45]27 Al2O3/SiO2 0.48 [2]28 TiO2/SiO2 (this work) 0.21Figure 5. Thermal conductivity at R.T. of the reported inorganic pore-free materials. The labelcorresponds to the material systems in Table 4. Eg > 3 eV (<3 eV), the material is transparent(nontransparent) to visible light. In this study, TiO2/SiO2 achieved the lowest thermal conductivitycompared to the other reported transparent layered composites.5. SummaryA data-driven strategy for exploring inorganic material systems with high trans-parency and high ITR for the transparent thermal insulator was demonstrated. We appliedan ITR machine learning model, which can consider the chemical, physical, and processproperties for predicting ITR and achieve higher predictive performance than the con-ventional prediction models (AMM, DMM), to select the transparent material systemsamong more than 4800 candidates. After the experimental validation, TiO2/SiO2 wasselected in terms of easiness of synthesis and the intrinsic low thermal conductivity. Thismaterial system achieved a high ITR of 26.56 m2K/GW, which is in good agreement withthe prediction by our machine learning model. Through the ITR prediction by machinelearning, the exploration of the material systems for transparent thermal insulators canbe accelerated.Micromachines 2023, 14, 186 10 of 12The TiO2/SiO2 nanoscale layered thin films with high interface density were furthersynthesized by sputtering. They showed an ultralow thermal conductivity of 0.21 W/mKand high transmittance to visible light. The low thermal conductivity was attributed to thehigh ITR between the alternating layers and the low intrinsic thermal conductivity of thecomponent materials. The layered thin films on the quartz substrate had lower thermalconductivity than those on Si due to the substrate dependence of the structure.The good agreement of the experimental ITR with the prediction and the low ther-mal conductivity of the layered thin films promise that this strategy has great potentialfor developing transparent thermal insulators in the future. Note that the compoundformation at the interfaces and the substrate dependence, which are yet considered inmaterial searching, also have a substantial effect on the ITR. An improvement made byfurther considering experimental parameters and interfacial reactions would give a morecomprehensive exploration.Author Contributions: Formal analysis, investigation, writing—original draft preparation, Y.-J.W.;resources, writing—review and editing, Y.X. All authors have read and agreed to the publishedversion of the manuscript.Funding: This work was supported by Japan Science and Technology Agency (JST) CREST GrantNumber JPMJCR21O2, and the “Materials research by Information Integration” Initiative (MI2I)project of the Support Program for Starting Up Innovation Hub of JST.Data Availability Statement: The experimental ITR dataset and descriptor dataset can be found inthe file “training dataset for ITR prediction” at https://doi.org/10.5281/zenodo.3564173 (accessedon 12 December 2022), which can be used as a training dataset for predicting ITR directly.Conflicts of Interest: The authors declare no conflict of interest.References1. Freeman, J.; Hellgardt, K.; Markides, C.N. An assessment of solar–Thermal collector designs for small-scale combined heatingand power applications in the United Kingdom. Heat Transf. Eng. 2015, 36, 1332. [CrossRef]2. Fong, S.W.; Sood, A.; Chen, L.; Kumari, N.; Asheghi, M.; Goodson, K.E.; Gibson, G.A.; Wong, H.S.P.J. Thermal conductivitymeasurement of amorphous dielectric multilayers for phase-change memory power reduction. Appl. Phys. 2016, 120, 015103.[CrossRef]3. Yang, C.; Souchay, D.; Kneiss, M.; Bogner, M.; Wei, H.M.; Lorenz, M.; Oeckler, O.; Benstetter, G.; Fu, Y.Q.; Grundmann, M.Transparent flexible thermoelectric material based on non-toxic earth-abundant p-type copper iodide thin film. Nat. Commun.2017, 8, 16076. [CrossRef] [PubMed]4. Hu, F.; An, L.; Li, C.N.; Liu, J.; Ma, G.B.; Hu, Y.; Huang, Y.L.; Liu, Y.Z.; Thundat, T.; Ren, S.Q. Transparent and Flexible ThermalInsulation Window Material. Cell Rep. Phys. Sci. 2020, 1, 100140. [CrossRef]5. Wang, J.Y.; Petit, D.; Ren, S.Q. Transparent thermal insulation silica aerogels. Nanoscale Adv. 2020, 2, 5504. [CrossRef]6. Kato, R.; Hatta, I. Thermal conductivity and interfacial thermal resistance: Measurements of thermally oxidized SiO2 films on asilicon wafer using a thermo-reflectance technique. Int. J. Thermophys. 2008, 29, 2062. [CrossRef]7. Cahill, D.G.; Bullen, A.; Lee, S.M. Interface thermal conductance and the thermal conductivity of multilayer thin films. High Temp.-High Press. 2000, 32, 135. [CrossRef]8. Chiritescu, C.; Cahill, D.G.; Heideman, C.; Lin, Q.Y.; Mortensen, C.; Nguyen, N.T.; Johnson, D.; Rostek, R.; Bottner, H.J. Lowthermal conductivity in nanoscale layered materials synthesized by the method of modulated elemental reactants. Appl. Phys.2008, 104, 033533. [CrossRef]9. Gabriel, N.T.; Talghader, J.J. Thermal conductivity and refractive index of hafnia-alumina nanolaminates. J. Appl. Phys. 2011,110, 043526. [CrossRef]10. Ali, S.; Juntunen, T.; Sintonen, S.; Ylivaara, O.M.; Puurunen, R.L.; Lipsanen, H.; Tittonen, I.; Hannula, S.P. Thermal conductivity ofamorphous Al2O3/TiO2 nanolaminates deposited by atomic layer deposition. Nanotechnology 2016, 27, 445704. [CrossRef]11. Yang, P.; Tang, Y.; Yang, H.; Wu, Y. Thermal conductivity reconstruction at TiO2/ZnO multilayer nanoscale interface structure.Sci. Adv. Mater. 2014, 6, 1986. [CrossRef]12. Venkatasubramanian, R. Lattice thermal conductivity reduction and phonon localizationlike behavior in superlattice structures.Phys. Rev. B 2000, 61, 3091. [CrossRef]13. Hu, S.; Ju, S.; Shao, C.; Guo, J.; Xu, B.; Ohnishi, M.; Shiomi, J. Ultimate impedance of coherent heat conduction in van der Waalsgraphene-MoS2 heterostructures. Mater. Today Phys. 2021, 16, 100324. [CrossRef]14. Little, W.A. The transport of heat between dissimilar solids at low temperatures. Can. J. Phys. 1959, 37, 334–339. [CrossRef]15. Swartz, E.T.; Pohl, R.O. Thermal boundary resistance. Rev. Mod. Phys. 1989, 61, 605. [CrossRef]https://doi.org/10.5281/zenodo.3564173http://doi.org/10.1080/01457632.2015.995037http://doi.org/10.1063/1.4955165http://doi.org/10.1038/ncomms16076http://www.ncbi.nlm.nih.gov/pubmed/28681842http://doi.org/10.1016/j.xcrp.2020.100140http://doi.org/10.1039/D0NA00655Fhttp://doi.org/10.1007/s10765-008-0536-4http://doi.org/10.1068/htwi9http://doi.org/10.1063/1.2967722http://doi.org/10.1063/1.3626462http://doi.org/10.1088/0957-4484/27/44/445704http://doi.org/10.1166/sam.2014.2125http://doi.org/10.1103/PhysRevB.61.3091http://doi.org/10.1016/j.mtphys.2020.100324http://doi.org/10.1139/p59-037http://doi.org/10.1103/RevModPhys.61.605Micromachines 2023, 14, 186 11 of 1216. Zhan, T.Z.; Minamoto, S.; Xu, Y.B.; Tanaka, Y.; Kagawa, Y. Thermal boundary resistance at Si/Ge interfaces by molecular dynamicssimulation. AIP Adv. 2015, 5, 047102. [CrossRef]17. Donadio, D.; Galli, G. Atomistic simulations of heat transport in silicon nanowires. Phys. Rev. Lett. 2009, 102, 195901. [CrossRef]18. Donadio, D.; Galli, G. Thermal conductivity of isolated and interacting carbon nanotubes: Comparing results from moleculardynamics and the Boltzmann transport equation. Phys. Rev. Lett. 2007, 99, 255502. [CrossRef]19. Wu, Y.-J.; Fang, L.; Xu, Y. Predicting interfacial thermal resistance by machine learning. Npj Comput. Mater. 2019, 5, 56. [CrossRef]20. Zhan, T.; Fang, L.; Xu, Y. Prediction of thermal boundary resistance by the machine learning method. Sci. Rep. 2017, 7, 7109.[CrossRef]21. Wu, Y.-J.; Sasaki, M.; Goto, M.; Fang, L.; Xu, Y. Electrically conductive thermally insulating Bi–Si nanocomposites by interfacedesign for thermal management. ACS Appl. Nano Mater. 2018, 1, 3355. [CrossRef]22. Xu, Y.B.; Wang, H.T.; Tanaka, Y.; Shimono, M.; Yamazaki, M. Measurement of interfacial thermal resistance by periodic heatingand a thermo-reflectance technique. Mater. Trans. 2007, 48, 148. [CrossRef]23. Wang, H.; Xu, Y.; Goto, M.; Tanaka, Y.; Yamazaki, M.; Kasahara, A.; Tosa, M. Thermal conductivity measurement of tungstenoxide nanoscale thin films. Mater. Trans. 2006, 47, 1894. [CrossRef]24. Touloukian, Y.S. The TPRC Data Series; Plenum Publishing Corporation: Ann Arbor, MI, USA, 1970.25. National Institute for Materials Science (NIMS). Inorganic Material Database (AtomWork-adv). 2017. Available online: https://atomwork-adv.nims.go.jp/ (accessed on 12 December 2022).26. Wu, Y.-J.; Zhan, T.; Hou, Z.; Fang, L.; Xu, Y. Physical and chemical descriptors for predicting interfacial thermal resistance. Sci.Data 2020, 7, 36. [CrossRef]27. Kim, S.; Lee, M.; Hong, C.; Yoon, Y.; An, H.; Lee, D.; Jeong, W.; Yoo, D.; Kang, Y.; Youn, Y.; et al. A band-gap database forsemiconducting inorganic materials calculated with hybrid functional. Sci. Data 2020, 7, 387. [CrossRef] [PubMed]28. Böttger, P.H.M.; Braginsky, L.; Shklover, V.; Lewin, E.; Patscheider, J.; Cahill, D.G.; Sobiech, M.J. Hard wear-resistant coatings withanisotropic thermal conductivity for high thermal load applications. Appl. Phys. 2014, 116, 13507. [CrossRef]29. Chen, L.; Kumari, N.; Hou, Y. Thermal resistances of crystalline and amorphous few-layer oxide thin films. AIP Adv. 2017,7, 115205. [CrossRef]30. Chen, X.; Weathers, A.; Carrete, J.; Mukhopadhyay, S.; Delaire, O.; Stewart, D.A.; Mingo, N.; Girard, S.N.; Ma, J.; Abernathy, D.L.;et al. Twisting phonons in complex crystals with quasi-one-dimensional substructures. Nat. Commun. 2015, 6, 6723. [CrossRef][PubMed]31. Snyder, G.J.; Christensen, M.; Nishibori, E.; Caillat, T.; Iversen, B.B. Disordered zinc in Zn4Sb3 with phonon-glass and electron-crystal thermoelectric properties. Nat. Mater. 2004, 3, 458. [CrossRef]32. Zhao, L.D.; Lo, S.H.; Zhang, Y.; Sun, H.; Tan, G.; Uher, C.; Wolverton, C.; Dravid, V.P.; Kanatzidis, M.G. Ultralow thermalconductivity and high thermoelectric figure of merit in SnSe crystals. Nature 2014, 508, 373. [CrossRef]33. Morelli, D.T.; Jovovic, V.; Heremans, J.P. Intrinsically minimal thermal conductivity in cubic I—V—VI 2 semiconductors. Phys.Rev. Lett. 2008, 101, 035901. [CrossRef]34. Huxtable, S.T.; Abramson, A.R.; Tien, C.-L.; Majumdar, A.; LaBounty, C.; Fan, X.; Zeng, G.; Bowers, J.E.; Shakouri, A.; Croke, E.T.Thermal conductivity of si/sige and sige/sige superlattices. Appl. Phys. Lett. 2002, 80, 1737. [CrossRef]35. Luckyanova, M.N.; Johnson, J.A.; Maznev, A.A.; Garg, J.; Jandl, A.; Bulsara, M.T.; Fitzgerald, E.A.; Nelson, K.A.; Chen, G.Anisotropy of the thermal conductivity in GaAs/AlAs superlattices. Nano Lett. 2013, 13, 3973. [CrossRef]36. Hu, M.; Poulikakos, D. Si/Ge superlattice nanowires with ultralow thermal conductivity. Nano Lett. 2012, 12, 5487. [CrossRef]37. Liao, Y.X.; Iwamoto, S.; Sasaki, M.; Goto, M.; Shiomi, J. Heat conduction below diffusive limit in amorphous superlattice structures.Nano Energy 2021, 84, 105903. [CrossRef]38. Costescu, R.M.; Cahill, D.G.; Fabreguette, F.H.; Sechrist, Z.A.; George, S.M. Ultra-low thermal conductivity in W/Al2O3nanolaminates. Science 2004, 303, 989. [CrossRef]39. Ju, Y.S.; Hung, M.T.; Carey, M.J.; Cyrille, M.C.; Childress, J.R. Nanoscale heat conduction across tunnel junctions. Appl. Phys. Lett.2005, 86, 203113. [CrossRef]40. Bozorg-Grayeli, E.; Li, Z.; Asheghi, M.; Delgado, G.; Pokrovsky, A.; Panzer, M.; Wack, D.; Goodson, K.E. Thermal conductionproperties of Mo/Si multilayers for extreme ultraviolet optics. J. Appl. Phys. 2012, 112, 083504. [CrossRef]41. Kim, E.K.; Kwun, S.I.; Lee, S.M.; Seo, H.; Yoon, J.G. Thermal boundary resistance at Ge2Sb2Te5/ZnS:SiO2 interface. Appl. Phys.Lett. 2000, 76, 3864–3866. [CrossRef]42. Dechaumphai, E.; Lu, D.; Kan, J.J.; Moon, J.; Fullerton, E.E.; Liu, Z.; Chen, R. Ultralow thermal conductivity of multilayers withhighly dissimilar debye temperatures. Nano Lett. 2014, 14, 2448. [CrossRef]43. Touzelbaev, M.N.; Zhou, P.; Venkatasubramanian, R.; Goodson, K.E. Thermal characterization of Bi2Te3/Sb2Te3 superlattices.J. Appl. Phys. 2001, 90, 763–767. [CrossRef]http://doi.org/10.1063/1.4916974http://doi.org/10.1103/PhysRevLett.102.195901http://doi.org/10.1103/PhysRevLett.99.255502http://doi.org/10.1038/s41524-019-0193-0http://doi.org/10.1038/s41598-017-07150-7http://doi.org/10.1021/acsanm.8b00575http://doi.org/10.2320/matertrans.48.148http://doi.org/10.2320/matertrans.47.1894https://atomwork-adv.nims.go.jp/https://atomwork-adv.nims.go.jp/http://doi.org/10.1038/s41597-020-0373-2http://doi.org/10.1038/s41597-020-00723-8http://www.ncbi.nlm.nih.gov/pubmed/33177500http://doi.org/10.1063/1.4886182http://doi.org/10.1063/1.5007299http://doi.org/10.1038/ncomms7723http://www.ncbi.nlm.nih.gov/pubmed/25872781http://doi.org/10.1038/nmat1154http://doi.org/10.1038/nature13184http://doi.org/10.1103/PhysRevLett.101.035901http://doi.org/10.1063/1.1455693http://doi.org/10.1021/nl4001162http://doi.org/10.1021/nl301971khttp://doi.org/10.1016/j.nanoen.2021.105903http://doi.org/10.1126/science.1093711http://doi.org/10.1063/1.1931827http://doi.org/10.1063/1.4759450http://doi.org/10.1063/1.126852http://doi.org/10.1021/nl500127chttp://doi.org/10.1063/1.1374458Micromachines 2023, 14, 186 12 of 1244. Tkadletz, M.; Lechner, A.; Schalk, N.; Sartory, B.; Winkler, M.; Mitterer, C. Reactively sputtered TiN/SiO2 multilayer coatings withdesigned anisotropic thermal conductivity—From theoretical conceptualization to experimental validation. Surf. Coat. Technol.2020, 393, 125763. [CrossRef]45. Alvarez-Quintana, J.; Peralba-Garcia, L.; Labar, J.L.; Rodriguez-Viejo, J.J. Ultra-Low Thermal Conductivity in Nanoscale LayeredOxides. Heat Transf. -Trans. ASME 2010, 132, 032402. [CrossRef]Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individualauthor(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury topeople or property resulting from any ideas, methods, instructions or products referred to in the content.http://doi.org/10.1016/j.surfcoat.2020.125763http://doi.org/10.1115/1.4000052 Introduction  Experimental Procedure  Thin Film Deposition  Heat Conduction Equation for the ITR and Thermal Conductivity  Result  Data-Driven Material Selection  Structure  Transmittance  ITR and Thermal Conductivity at Room Temperature  Discussion  ITR  Structural Effect on the ITR  Comparison with Other Transparent Layered Materials  Summary  References