betaFeedback

WorksPhysics and Astronomy

Analytical framework for interpreting thermoreflectance measurements of nanoline gratings under finite spot illumination

Abstract

A frequency-domain thermoreflectance (FDTR)-based technique is presented to measure the effective thermal properties of a material stack whose surface is patterned with metallic nanoline gratings that are heated and probed by lasers. An analytical heat diffusion model is derived for predicting the FDTR phase response that incorporates a finite laser spot radius with a Gaussian intensity profile for the pump and probe lasers, as well as the dimensions of the nanoline grating. The model is validated by comparing with measurements of Au nanoline gratings, with widths ranging from 100 to 1000 nm, fabricated on a-SiO2_2 and cc-sapphire substrates. Due to the short phonon mean free paths (MFPs) in these substrates, thermal transport remains in the diffusive regime for the whole range of nanoline widths. The model and the measurement show excellent agreement across different laser spot radii and nanoline widths, with a bulk thermal conductivity value of cc-sapphire and the thermal boundary conductances at the Au-Cr/a-SiO2_2 and Au-Cr/cc-sapphire interfaces comparable to our prior measurements and the literature. The validated model will serve as a robust baseline for extending this method to measure effective thermal conductivity as a function of the heater size. This approach is particularly relevant for materials where non-diffusive thermal transport regimes enable probing phonon MFP dependent contributions to the thermal conductivity.

Introduction

When the characteristic length scale LcL_c (e.g., film thickness, grain size, heat source size) of thermal transport is comparable to or smaller than the phonon mean free paths (MFPs) in a material, the thermal conductivity, kk, is suppressed relative to its bulk value. A complete description of how thermal conductivity accumulates with phonon MFPs is given by the material-dependent thermal conductivity accumulation function, kaccumk_{\mathrm{accum}}. Time-domain and frequency-domain thermoreflectance (TDTR and FDTR) experiments have been used on thin films to extract kaccumk_{\mathrm{accum}} using the film thickness as LcL_c.[1, 2] Phonon transmission at interfaces and sample-to-sample variations for samples in a thickness progression, however, can complicate the mapping from the measured thermal conductivity kmeas(Lc)k_{\mathrm{meas}}(L_c) to kaccum(Lc)k_{\mathrm{accum}}(L_c). To mitigate these challenges, others used the LcL_c associated with a heat source size to suppress kmeask_{\mathrm{meas}} without altering the physical condition of a bulk sample. Koh et al.³ and Regner et al.⁴ used the thermal penetration depth, which is inversely proportional to the modulation frequency in TDTR and FDTR, as LcL_c to measure the size-dependent kmeask_{\mathrm{meas}} of bulk substrates. The interpretation of such measurements is sensitive to the treatment of the interface between the metal transducer and the substrate.[5, 6, 7, 8] Minnich et al.⁹ used the laser spot diameter in TDTR as LcL_c and demonstrated a suppression of thermal conductivity in silicon as the spot diameter decreased from 60 μ\mum to 15 μ\mum. Johnson et al.¹⁰ applied transient grating (TG) spectroscopy, which uses two intersecting laser beams to form a spatially sinusoidal heating profile in a semiconductor membrane, to measure non-diffusive thermal transport in silicon. The grating period acts as LcL_c, which can be controlled by changing the angle between the intersecting lasers.[11] Cuffe et al. used TG spectroscopy to reconstruct the phonon MFP distribution in crystalline silicon[12] and Kim et al. used a similar approach for amorphous silicon.[13]

The TDTR laser spot diameter and the TG grating period are diffraction-limited. It is therefore difficult to reduce these dimensions below 1 μ\mum using laser light in the visible spectrum,[11] thus limiting the range of phonon MFPs that can be suppressed. To overcome this limitation, nanoline patterns of metallic heaters can be deposited on a substrate to produce heater sizes and grating periods smaller than the diffraction limit. Using electron-beam lithography, the width of the individual nanolines can be reduced to tens of nanometers, lowering the lower bound of LcL_c. Using this approach within the TDTR framework, Zeng et al.¹⁴ measured the kaccumk_{\mathrm{accum}} of silicon, employing heaters as small as 30 nm to probe a wide range of phonon MFPs. Hu et al.¹⁵ used a similar range of nanoline heater sizes to measure kaccumk_{\mathrm{accum}} and reconstruct the phonon MFP distribution for sapphire, GaN, and GaAs. Metal nanolines have also been used as diffraction gratings to study quasi-ballistic thermal transport in silicon.[16, 17, 18] However, in order to achieve similar heater sizes as in previous TDTR studies, this approach often requires the use of ultra-violet or x-ray lasers because the periodicity of the nanolines that can be probed is diffraction-limited.

In this work, we adapt the nanoline grating approach for FDTR measurements. We derive an analytical solution for the heat diffusion equation for a metal nanoline grating that incorporates the effect of the finite spot radii of the laser beams, which was neglected in previous TDTR studies where the spot radius was large compared to the grating period. We validate our model through measurements on nanoline patterns fabricated on amorphous fused silica (a-SiO2_2) and sapphire with the cc-axis perpendicular to the surface (cc-sapphire). These substrates are chosen because their energy carriers exhibit small MFPs in comparison to the nanoline widths and grating periods, to ensure that thermal transport remains in the diffusive regime.[19, 20]

Analytical Model Development

FDTR is a non-contact optical technique in which a continuous-wave pump laser is sinusoidally modulated and focused on the surface of the sample being measured. The FDTR setup is shown schematically in Figure 1(a). The top surface of a single or multi-layer sample is coated with a Au film that absorbs the pump laser such that a sinusoidal heat flux is induced at the top surface. The resulting temperature response on the same surface oscillates sinusoidally with the same frequency (ω0\omega_0) but with a phase lag ϕ(ω0)\phi(\omega_0). Based on the thermoreflectance of Au, the reflected probe intensity gets modulated with the same ω0\omega_0 and ϕ(ω0)\phi(\omega_0), which can be monitored by a photodiode that is read by a lock-in amplifier. A thermal transport model for the sample is needed to derive the transfer function H(ω0)H(\omega_0), which relates the thermal and geometric properties of the sample to ϕ(ω0)\phi(\omega_0) as[21]

Nanoline array description

Figure 1. Nanoline array description: (a) Schematic diagram of the frequency-domain thermoreflectance (FDTR) measurement setup with 405 nm pump and 532 nm probe lasers. (b) Schematic representation of the intensity variation due to the finite spot radius of a Gaussian laser beam over nanolines with a comparable width and separation.

ϕ(ω0)=tan⁡−1{Im⁡[H(ω0)]Re⁡[H(ω0)]}.(1)\phi(\omega_0) = \tan^{-1}\left\{\frac{\operatorname{Im}[H(\omega_0)]}{\operatorname{Re}[H(\omega_0)]}\right\}. \tag*{(1)}

This thermal model is then used to fit the desired material and/or geometric properties based on the measured phase lag. The conventional analytical model used in FDTR[4] assumes a uniform, continuous metallic film as the transducer. In this section, we derive an analytical solution of the heat diffusion equation for metal nanoline grating structures. Our approach builds upon the framework developed by Minnich and co-workers[22, 23] and subsequently adopted by others[15, 24, 25] to investigate thermal transport from metal nanoline gratings in TDTR measurements. The previous measurements approximated the large TDTR laser spot-size with an assumption of uniform spatial heating, as if the spot-size was infinite. We extend their methodology by incorporating Gaussian intensity variation due to the finite spot radius of both the pump and probe lasers into our analytical model for FDTR. For a large laser spot radius (r>100 μmr > 100\,\mu\mathrm{m}), the phase predicted by our analytical model is identical to that predicted by the infinite spot-size analytical model used in the previous TDTR measurements (details in Appendix § 4.2).[15]

We start from the heat diffusion equation of an orthotropic material

kx∂2T(x,y,z,t)∂x2+ky∂2T(x,y,z,t)∂y2+kz∂2T(x,y,z,t)∂z2=ρc∂T(x,y,z,t)∂t,(2)k_x \frac{\partial^2 T(x,y,z,t)}{\partial x^2} + k_y \frac{\partial^2 T(x,y,z,t)}{\partial y^2} + k_z \frac{\partial^2 T(x,y,z,t)}{\partial z^2} = \rho c \frac{\partial T(x,y,z,t)}{\partial t}, \tag*{(2)}

where xx and yy are the two in-plane directions and zz is the cross-plane direction [Figure 1(b)], with kxk_x, kyk_y, and kzk_z being the directional thermal conductivities. TT, ρ\rho, cc, and tt denote temperature, density, specific heat capacity, and time, respectively. We Fourier transform the xx, yy, and tt coordinates of Eq. (2) to obtain an ordinary differential equation in zz as

i2ξx2kxT~(ξx,ξy,z,ω)+i2ξy2kyT~(ξx,ξy,z,ω)+kzd2T~(ξx,ξy,z,ω)dz2=iωρcT~(ξx,ξy,z,ω).(3)i^{2}\xi_{x}^{2}k_{x}\widetilde{T}(\xi_{x},\xi_{y},z,\omega) + i^{2}\xi_{y}^{2}k_{y}\widetilde{T}(\xi_{x},\xi_{y},z,\omega) + k_{z}\frac{d^{2}\widetilde{T}(\xi_{x},\xi_{y},z,\omega)}{dz^{2}} = i\omega\rho c\widetilde{T}(\xi_{x},\xi_{y},z,\omega). \tag*{(3)}

Here, i2=−1i^{2}=-1; [ ]~\widetilde{[\,]} denotes the Fourier transform; and ξx\xi_{x}, ξy\xi_{y}, and ω\omega are the reciprocal space variables corresponding to xx, yy, and tt. Eq. (3) can be rearranged as

d2T~(ξx,ξy,z,ω)dz2=q2T~(ξx,ξy,z,ω),(4)\frac{d^{2}\widetilde{T}(\xi_{x},\xi_{y},z,\omega)}{dz^{2}}=q^{2}\widetilde{T}(\xi_{x},\xi_{y},z,\omega), \tag*{(4)}

where q=iωρc+ξx2kx+ξy2kykzq=\sqrt{\frac{i\omega\rho c+\xi_{x}^{2}k_{x}+\xi_{y}^{2}k_{y}}{k_{z}}}. Eq. (4) admits solutions of the form

T~(ξx,ξy,z,ω)=α1cosh⁡qz+α2sinh⁡qz,(5)\widetilde{T}(\xi_{x},\xi_{y},z,\omega)=\alpha_{1}\cosh qz+\alpha_{2}\sinh qz, \tag*{(5)}

where α1\alpha_{1} and α2\alpha_{2} are constants that are determined based on the boundary conditions. The bottom surface of a thick sample can be treated as adiabatic. In a stack of materials, the Fourier transformed top-surface [z=0[z=0 in Figure 1(b)]] temperature T~t\widetilde{T}_{t} and heat flux Q~t\widetilde{Q}_{t} can be related to the bottom-surface values (T~b\widetilde{T}_{b} and Q~b\widetilde{Q}_{b}) using a transfer matrix approach.[26] For a material layer jj in the stack with thickness djd_{j}, the Fourier transformed temperature and heat fluxes at the top and bottom of that layer are related using the quadrupole matrix,[26] MM as

\begin{bmatrix} \widetilde{T}_{b_j}\\ \widetilde{Q}_{b_j} \end{bmatrix} =M_j\cdot \begin{bmatrix} \widetilde{T}_{t_j}\\ \widetilde{Q}_{t_j} \end{bmatrix}, \tag*{(6)}\end{bmatrix}

where

Mj=[cosh⁡(qjdj)−1kzjqjsinh⁡(qjdj)−kzjqjsinh⁡(qjdj)cosh⁡(qjdj)].(7)M_j= \begin{bmatrix} \cosh(q_jd_j) & -\frac{1}{k_{z_j}q_j}\sinh(q_jd_j)\\ -k_{z_j}q_j\sinh(q_jd_j) & \cosh(q_jd_j) \end{bmatrix}. \tag*{(7)}

For convenience, the explicit dependence of qq, T~\widetilde{T}, Q~\widetilde{Q}, and MM on the variables ξx\xi_{x}, ξy\xi_{y}, ω\omega, and zz has been dropped from the notation. For a stack with NN layers, the relation between temperature and heat flux at the top and bottom of the stack is then

\begin{bmatrix} \widetilde{T}_{b}\\ \widetilde{Q}_{b} \end{bmatrix} = \begin{bmatrix} \widetilde{T}_{b_N}\\ \widetilde{Q}_{b_N} \end{bmatrix} =M_NM_{N-1}\cdots M_1\cdot \begin{bmatrix} \widetilde{T}_{t_1}\\ \widetilde{Q}_{t_1} \end{bmatrix} = \begin{bmatrix} A & B\\ C & D \end{bmatrix}\cdot \begin{bmatrix} \widetilde{T}_{t}\\ \widetilde{Q}_{t} \end{bmatrix}. \tag*{(8)}\end{bmatrix}

Applying an adiabatic boundary condition at the bottom of the stack, which is reasonable for a thick stack, leads to

T~t=−DCQ~t.(9)\widetilde{T}_{t}=-\frac{D}{C}\widetilde{Q}_{t}. \tag*{(9)}

Q~t\widetilde{Q}_t is obtained by a Fourier transform of the heat flux on top of the sample. The heat flux on an array of nanolines due to a unit-power Gaussian laser beam, modulated with a temporal frequency ω0\omega_0, is[22, 23, 27]

Qt(x,y,t)≡Q(x,y,z=0,t)=2πr02exp⁡(−2(x2+y2)r02)[∑n=0∞ancos⁡(2nπxL)]cos⁡(ω0t),(10)Q_t(x,y,t) \equiv Q(x,y,z=0,t) = \frac{2}{\pi r_0^2}\exp\left(-\frac{2(x^2+y^2)}{r_0^2}\right)\left[\sum_{n=0}^{\infty}a_n\cos\left(\frac{2n\pi x}{L}\right)\right]\cos(\omega_0t), \tag*{(10)}

where

an={wLif n=02nπsin⁡(nπwL)otherwise.a_n = \begin{cases} \frac{w}{L} & \text{if } n=0 \\ \frac{2}{n\pi}\sin\left(\frac{n\pi w}{L}\right) & \text{otherwise.} \end{cases}

Here, the summation captures the spatial profile of the nanolines, which is infinite in the yy-direction and periodic in the xx-direction with periodicity LL and nanoline width ww [Fig. 1(b)]. The location x=0x=0 is at the center of a nanoline. The prefactor describes the Gaussian intensity variation of the pump laser spot with radius r0r_0, which is centered at (x=0,y=0)(x=0,y=0). After Fourier transforming both spatial coordinates and time, the heat flux is expressed as

Q~t(ξx,ξy,ω=ω0)≡Q~(ξx,ξy,z=0,ω=ω0)=π∑n=0∞{an2exp⁡(−ξy2r028)[exp⁡(−(ξx−2nπL)2r028)+exp⁡(−(ξx+2nπL)2r028)]}.(11)\begin{aligned} \widetilde{Q}_t(\xi_x,\xi_y,\omega=\omega_0) &\equiv\widetilde{Q}(\xi_x,\xi_y,z=0,\omega=\omega_0) \\ &= \pi\sum_{n=0}^{\infty}\left\{\frac{a_n}{2}\exp\left(-\frac{\xi_y^2r_0^2}{8}\right) \left[\exp\left(-\frac{\left(\xi_x-\frac{2n\pi}{L}\right)^2r_0^2}{8}\right)+\exp\left(-\frac{\left(\xi_x+\frac{2n\pi}{L}\right)^2r_0^2}{8}\right)\right]\right\}. \tag*{(11)} \end{aligned}

Inserting Eq. (11) into Eq. (9) gives

T~t(ξx,ξy,ω)≡T~(ξx,ξy,z=0,ω)=πδ(ω−ω0)∑n=0∞DC{an2exp⁡(−ξy2r028)[exp⁡(−(ξx−2nπL)2r028)+exp⁡(−(ξx+2nπL)2r028)]}.(12)\begin{aligned} \widetilde{T}_t(\xi_x,\xi_y,\omega) &\equiv\widetilde{T}(\xi_x,\xi_y,z=0,\omega) \\ &= \pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\frac{D}{C}\left\{\frac{a_n}{2}\exp\left(-\frac{\xi_y^2r_0^2}{8}\right) \left[\exp\left(-\frac{\left(\xi_x-\frac{2n\pi}{L}\right)^2r_0^2}{8}\right)+\exp\left(-\frac{\left(\xi_x+\frac{2n\pi}{L}\right)^2r_0^2}{8}\right)\right]\right\}. \tag*{(12)} \end{aligned}

Here δ(ω−ω0)\delta(\omega-\omega_0) signifies that for a sinusoidal heat flux with a single frequency component ω0\omega_0, the temperature response recorded at the photodetector will only have the same frequency component. Taking the inverse Fourier transform of Eq. (12) in xx and yy leads to the real-space temperature field oscillating at the modulation frequency ω\omega,

Tt(x,y,ω)=∫−∞∞∫−∞∞T~(ξx,ξy,ω)exp⁡(iξxx)exp⁡(iξyy) dξxdξy.(13)T_t(x,y,\omega)=\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\widetilde{T}(\xi_x,\xi_y,\omega)\exp(i\xi_xx)\exp(i\xi_yy)\,d\xi_xd\xi_y. \tag*{(13)}

The probe laser also has a Gaussian intensity profile with a spot radius r1r_1, centered at (x=0,y=0)(x=0,y=0) at the surface of the sample. The probe Gaussian intensity profile multiplied by the spatial profile of the nanolines acts as a weighting factor for the thermoreflectance response from the top surface of the sample. The transfer function H(ω)H(\omega) quantifies how this spatial weighting maps the measured thermoreflectance to the original pump modulation signal. The corresponding transfer function is

H(ω)=∫−∞∞∫−∞∞Tt(x,y,ω)×2πr12exp⁡(−2(x2+y2)r12)∑m=0∞amcos⁡(2mπxL)dxdy.(14)H(\omega)=\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}T_t(x,y,\omega)\times\frac{2}{\pi r_1^2}\exp\left(\frac{-2(x^2+y^2)}{r_1^2}\right)\sum_{m=0}^{\infty}a_m\cos\left(\frac{2m\pi x}{L}\right)dxdy. \tag*{(14)}

The Fourier coefficients ama_m have the same form as the ana_n in Eq. (10) as they describe the same nanoline pattern. Substituting Eq. (13) into Eq. (14) leads to the a four-dimensional integral expression for H(ω)H(\omega),

H(ω)=πδ(ω−ω0)∑n=0∞∑m=0∞∫−∞∞∫−∞∞∫−∞∞∫−∞∞amanπr12(−DC)×[exp⁡(−(ξx−2nπL)2r028)+exp⁡(−(ξx+2nπL)2r028)]×exp⁡(iξxx+iξyy−ξy2r028−2(x2+y2)r12)×cos⁡(2mπxL)dξxdξydxdy.(15)\begin{aligned} H(\omega)={}&\pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\sum_{m=0}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\frac{a_m a_n}{\pi r_1^2}\left(-\frac{D}{C}\right)\\ &\times\left[\exp\left(\frac{-\left(\xi_x-\frac{2n\pi}{L}\right)^2r_0^2}{8}\right)+\exp\left(\frac{-\left(\xi_x+\frac{2n\pi}{L}\right)^2r_0^2}{8}\right)\right]\\ &\times\exp\left(i\xi_xx+i\xi_yy-\frac{\xi_y^2r_0^2}{8}-\frac{2(x^2+y^2)}{r_1^2}\right)\times\cos\left(\frac{2m\pi x}{L}\right)d\xi_xd\xi_ydxdy. \tag*{(15)} \end{aligned}

As shown in Appendix A, Eq. (15) can be converted from a four-dimensional integral to a two-dimensional integral using Hubbard-Stratonovich transformation[28] to obtain

H(ω)=πδ(ω−ω0)∑n=0∞∑m=0∞∫−∞∞∫−∞∞anam4(−DC)exp⁡[−ξy28(r02+r12)]×{exp⁡[−r028(ξx−2nπL)2−r128(ξx−2mπL)2]+exp⁡[−r028(ξx−2nπL)2−r128(ξx+2mπL)2]+exp⁡[−r028(ξx+2nπL)2−r128(ξx−2mπL)2]+exp⁡[−r028(ξx+2nπL)2−r128(ξx+2mπL)2]}dξxdξy.(16)\begin{aligned} H(\omega)={}&\pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\sum_{m=0}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\frac{a_na_m}{4}\left(-\frac{D}{C}\right)\exp\left[-\frac{\xi_y^2}{8}\left(r_0^2+r_1^2\right)\right]\\ &\times\Bigg\{\exp\left[-\frac{r_0^2}{8}\left(\xi_x-\frac{2n\pi}{L}\right)^2-\frac{r_1^2}{8}\left(\xi_x-\frac{2m\pi}{L}\right)^2\right]\\ &\quad+\exp\left[-\frac{r_0^2}{8}\left(\xi_x-\frac{2n\pi}{L}\right)^2-\frac{r_1^2}{8}\left(\xi_x+\frac{2m\pi}{L}\right)^2\right]\\ &\quad+\exp\left[-\frac{r_0^2}{8}\left(\xi_x+\frac{2n\pi}{L}\right)^2-\frac{r_1^2}{8}\left(\xi_x-\frac{2m\pi}{L}\right)^2\right]\\ &\quad+\exp\left[-\frac{r_0^2}{8}\left(\xi_x+\frac{2n\pi}{L}\right)^2-\frac{r_1^2}{8}\left(\xi_x+\frac{2m\pi}{L}\right)^2\right]\Bigg\}d\xi_xd\xi_y. \tag*{(16)} \end{aligned}

The pump and probe laser beam radii were measured independently using the knife-edge method and were found to agree within 10%. Therefore, we assume that the pump and probe lasers have the same spot radius on the surface of the sample, r0=r1=rr_0=r_1=r, which simplifies Eq. (16) into

H(ω)=πδ(ω−ω0)∑n=0∞∑m=0∞∫−∞∞∫−∞∞anam4(−DC)exp⁡(−ξy24r2)×{exp⁡[−r28(ξx−2nπL)2−r28(ξx−2mπL)2]+exp⁡[−r28(ξx−2nπL)2−r28(ξx+2mπL)2]+exp⁡[−r28(ξx+2nπL)2−r28(ξx−2mπL)2]+exp⁡[−r28(ξx+2nπL)2−r28(ξx+2mπL)2]} dξxdξy.(17)H(\omega)=\pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\sum_{m=0}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\frac{a_na_m}{4}\left(-\frac{D}{C}\right)\exp\left(-\frac{\xi_y^2}{4}r^2\right) \begin{aligned} &\times\Biggl\{\exp\left[-\frac{r^2}{8}\left(\xi_x-\frac{2n\pi}{L}\right)^2-\frac{r^2}{8}\left(\xi_x-\frac{2m\pi}{L}\right)^2\right]\\ &+\exp\left[-\frac{r^2}{8}\left(\xi_x-\frac{2n\pi}{L}\right)^2-\frac{r^2}{8}\left(\xi_x+\frac{2m\pi}{L}\right)^2\right]\\ &+\exp\left[-\frac{r^2}{8}\left(\xi_x+\frac{2n\pi}{L}\right)^2-\frac{r^2}{8}\left(\xi_x-\frac{2m\pi}{L}\right)^2\right]\\ &+\exp\left[-\frac{r^2}{8}\left(\xi_x+\frac{2n\pi}{L}\right)^2-\frac{r^2}{8}\left(\xi_x+\frac{2m\pi}{L}\right)^2\right]\Biggr\}\,d\xi_xd\xi_y. \end{aligned} \tag*{(17)}

FDTR MEASUREMENT

A schematic diagram of the FDTR setup is shown in Figure 1(a). The pump laser (Coherent OBIS 405 nm LX) emits an intensity-modulated beam, with a frequency range of 100 kHz to 5 MHz chosen for these experiments. The pump laser serves as a temporally periodic, spatially Gaussian heat source. The pump laser is absorbed by the Au nanolines, inducing a temperature response that exhibits a phase lag relative to heating, which depends on the thermal properties of the substrate. Based on the thermoreflectance of Au, the phase lag is monitored by a co-aligned probe laser (Coherent Verdi-V6 532 nm) for each modulation frequency.

The process by which the nanolines were deposited on the substrates is described in Sec. S1. The nanoline width and spacing were characterized using scanning electron microscopy [Fig. 2(a)], while thickness was measured using a KLA Tencor P-15 Profilometer, as shown in § 14 of the Supplementary Information. The dimensions of all nanoline patterns measured in this study are listed in Table I. In addition to the nanolines, a 5×55\times5 matrix of square Au patches, each 50×5050\times50 μm2\mu\mathrm{m}^2, was fabricated on the substrate as a control, as shown in Figure 2(b). The lateral dimensions of these patches are much larger than the laser spot radius (r<5r<5 μm\mu\mathrm{m}), allowing the application of the standard FDTR analytical solution for interpreting the measurements on them.

SEM images of a nanoline pattern and a 5×5 matrix of square patches

Figure 2. (a) SEM image of a nanoline pattern. D1 represents gap = L−wL-w, and D2 represents ww. (b) SEM image of 5×5 matrix of square patches.

On the nanoline patterns, the pump and probe beams passing through the gap between the nanolines do not directly affect the thermoreflectance response since both a-SiO2_2 and cc-sapphire are transparent to the pump and probe wavelengths. As a confirmation, no thermoreflectance response was obtained on a cc-sapphire substrate in the absence of a Au transducer, as shown in

Substrateww (nm)gap = L−wL-w (nm)
a-SiO2_2172180
a-SiO2_2524170
a-SiO2_21024164
cc-sapphire234169
cc-sapphire548153
cc-sapphire1052154

Table 1. Measured nanoline patterns.

Figure S4. We did observe that the phase lag was affected by the portion of the pump laser beam that passed through the nanoline gaps and was reflected off the rear surface of both substrates, causing backside heating. The effect was observed as a significant departure of the measured phase lag from the prediction of the thermal model. To mitigate this issue, we applied a thin layer of glycerin, which has a refractive index (≈1.5\approx1.5) similar to that of a-SiO2_2 and cc-sapphire, on the rear surface of the substrates. This thin glycerin layer effectively eliminated internal reflection by allowing the light that passed through the grating to exit through the back surface of the substrate and scatter diffusively in the glycerin. Upon application of the glycerin layer, the measured phase lag became consistent with the thermal model.

Thermoreflectance signal recorded on a c-sapphire sample on the area covered with an Au transducer and on the bare c-sapphire substrate

Figure S4. Thermoreflectance signal recorded on cc-sapphire sample on the area covered with an Au transducer and on the bare cc-sapphire substrate.

The measured phase lag as a function of the modulation frequency was fitted using the analytical model presented in Sec. II to extract the unknown thermal properties. The in-plane thermal conductivity of the Au nanolines in the direction perpendicular to the nanoline orientation (kx,Auk_{x,\mathrm{Au}}) was set to 0 W/m·K to simulate the physical gap between the nanolines. In § 15, we present results from a 3D finite element study to show that the predicted phase lag is insensitive to this simplification. On the a-SiO2_2 substrate, we extracted the in-plane thermal conductivity of the Au nanolines in the direction parallel to the nanoline orientation (ky,Auk_{y,\mathrm{Au}}) and the thermal boundary conductance at the Au-Cr/a-SiO2_2 interface (GAu−Cr/a−SiO2G_{\mathrm{Au-Cr}/\mathrm{a-SiO}_2}). For the Au nanolines on the c-sapphire substrate, we extracted the thermal conductivity of c-sapphire (kc−sapphirek_{\mathrm{c-sapphire}}) and the thermal boundary conductance at the Au-Cr/c-sapphire interface (GAu−Cr/c−sapphireG_{\mathrm{Au-Cr}/\mathrm{c-sapphire}}). A single kc−sapphirek_{\mathrm{c-sapphire}} value was extracted because its in-plane (parallel to c-plane) and cross-plane (normal to c-plane) thermal conductivities are comparable at room temperature.[29, 30]

Two Mitutoyo objective lenses with different magnifications, 10× and 20×, were used to explore the effect of the laser spot radius, which is a critical parameter in the analytical model. We used the square patch arrays on both substrates as reference samples to measure the laser spot radius. The phase lag was fitted to the standard multilayer solution of FDTR[4] to extract the laser spot radius and the thermal boundary conductance at the Au-Cr/substrate interface (GAu−Cr/subG_{\mathrm{Au-Cr}/\mathrm{sub}}). Since the analytical FDTR model does not explicitly include the thermal boundary conductance, the interface was modeled as a thin layer with thickness d=1d = 1 Å and thermal conductivity kk, where G=k/dG = k/d. Since the nanolines were deposited on the same substrate under the same conditions and within the same fabrication process, GAu−Cr/subG_{\mathrm{Au-Cr}/\mathrm{sub}} of the square patches should be the same as GAu−Cr/subG_{\mathrm{Au-Cr}/\mathrm{sub}} for the nanolines. The laser spot radius was then treated as a fixed parameter for the particular objective used during the nanoline measurements. For the nanoline measurements, the sample stage was positioned with the micro-manipulator so that the probe laser was entirely within the nanoline pattern, with the laser center aligned to the center of one of the nanolines.

Results

a-SiO2_2

We first studied aa-SiO2_2 to experimentally validate the analytical solution derived in Sec. II. Phonon MFPs in aa-SiO2_2 are much shorter (≈1 nm)(\approx1\ \mathrm{nm})[19] than both the spot radius and the nanoline widths. As such, thermal transport remains in the diffusive regime, allowing us to assess if the model can accurately capture variations in phase lag arising from changes in spot radius and nanoline width. An analysis was performed to determine the sensitivity of each parameter used in the analytical solution. The sensitivity of a parameter, SϕS_{\phi}, to the phase lag measured in FDTR is quantified by

Sϕ=dln⁡(ϕ)dln⁡(s)=dϕds×sϕ,(18)S_{\phi} = \frac{\mathrm{d}\ln(\phi)}{\mathrm{d}\ln(s)} = \frac{\mathrm{d}\phi}{\mathrm{d}s} \times\frac{s}{\phi}, \tag*{(18)}

which shows the normalized phase change per normalized unit change in the parameter.[31] A high sensitivity of a fitting parameter indicates that FDTR can be used to accurately measure that parameter. If the phase lag is sensitive to an input parameter, then that parameter needs to be externally quantified with low uncertainty to minimize its unintentional influence on the fitted parameters.

The nominal parameters used in the sensitivity analyses for aa-SiO2_2 and cc-sapphire are presented in Table II. The thermal conductivity of the Au-Cr nanoline was initially determined from a witness sample using the four-point probe method. The witness sample was prepared by depositing the Au-Cr layer on a separate aa-SiO2_2 substrate simultaneously with the Au nanoline deposition to ensure identical thickness and deposition conditions. The Wiedemann-Franz law was applied to convert the measured electrical conductivity to the Au-Cr thermal conductivity. The volumetric heat capacity and the thermal conductivity of the aa-SiO2_2 substrate were obtained from the literature.[32, 33]

PropertyAua-SiO2_2c-sapphire
CvC_v (MJ/m2^2-K)2.43 ±\pm 0.03[32]1.58 ±\pm 0.02[33]2.99 ±\pm 0.03[32]
kxk_x (W/m-K)01.4 ±\pm 0.1[33]Fitted
kyk_y (W/m-K)Fitted1.4 ±\pm 0.1[33]Fitted
kzk_z (W/m-K)160 ±\pm 811.4 ±\pm 0.1[33]Fitted
dd75.0 ±\pm 1.3 nm2300 ±\pm 3 μ\mum400 ±\pm 4 μ\mum
rr (μ\mum)-4.1 ±\pm 0.2 (10×\times)<br>2.3 ±\pm 0.1 (20×\times)5.0 ±\pm 0.2 (10×\times)<br>2.5 ±\pm 0.1 (20×\times)

Table II. Parameters used in the sensitivity analysis and the Monte Carlo uncertainty quantification. kx=ky=kzk_x = k_y = k_z for both substrates.

The aa-SiO2_2 sensitivity analysis is shown in Figure 3(a) and indicates sensitivity to ky,Auk_{y,\mathrm{Au}} and GAu−Cr/a-SiO2G_{\mathrm{Au-Cr}/a\text{-}\mathrm{SiO}_2}. The initial fit using ky,Auk_{y,\mathrm{Au}} from the witness sample was unsuccessful, suggesting a possible variation in ky,Auk_{y,\mathrm{Au}} with different nanoline widths. We therefore simultaneously extracted ky,Auk_{y,\mathrm{Au}} and GAu−Cr/a-SiO2G_{\mathrm{Au-Cr}/a\text{-}\mathrm{SiO}_2} for three nanoline patterns and the square patches for each laser spot radius. The analysis is insensitive to the cross-plane thermal conductivity of the Au nanolines (kz,Auk_{z,\mathrm{Au}}), and therefore the value measured from the witness sample was used and treated as a fixed parameter. The analysis is also highly sensitive to the laser spot radius [Figure 3(a)]. The uncertainty of the

Sensitivity analysis for (a) nanolines deposited on the a-SiO₂ substrate and (b) nanolines deposited on the c-sapphire substrate

Figure 3. Sensitivity analysis for (a) nanolines deposited on the a-SiO₂ substrate and (b) nanolines deposited on the c-sapphire substrate.

laser spot size was characterized with previous knife-edge measurements and the reference sample (Au on a-SiO2_2) measurement. As verified by 3D Finite Element Method (FEM) simulations (Figure S7), the laser spot position relative to the nanolines has a negligible effect on the measured phase. The simulations considered a lateral offset of 100 nm from the nanoline center, corresponding to twice the resolution of the micro-manipulator. The analytical fits to the measured phase lag are plotted for laser spot radii of 4.1 ±\pm 0.2 μ\mum [Figure 4(a)] and 2.3 ±\pm 0.1 μ\mum [Figure 4(b)]. The values of ky,Auk_{y,\mathrm{Au}} and GAu−Cr/a−SiO2G_{\mathrm{Au-Cr/a-SiO_2}} extracted from these fits are plotted in Figure 5(a) and Figure 5(b). The uncertainty of the fitting parameters was quantified using a Monte Carlo method.[34] For the input parameters listed in Table II, we assumed Gaussian distributions using their estimated uncertainties as standard deviations. In each Monte Carlo iteration, the input parameters were randomly sampled from each distribution, and the unknown parameters were extracted by fitting the newly constructed analytical solution to the experimental data. This procedure was repeated for 1000 iterations to generate distributions of the fitted parameters. Uncertainties are reported as the central 95% confidence interval to account for possible skewness of the distribution.

Phase lag and analytical solution fits for nanolines deposited on a-SiO₂

Figure 4. Phase lag and analytical solution fits for nanolines deposited on a-SiO₂ with (a) 10× objective (r=4.1±0.2 μmr = 4.1 \pm0.2\,\mu\mathrm{m}) and (b) 20× objective (r=2.3±0.1 μmr = 2.3 \pm0.1\,\mu\mathrm{m}). The legend uses the notation ww/gap (nm); e.g., 172/180 corresponds to a width ww of 172 nm and a gap (L−wL-w) of 180 nm.

Bar charts of extracted thermal properties for nanolines

Figure 5. Extracted (a) ky,Auk_{y,\mathrm{Au}} and (b) GAu−Cr/a−SiO2G_{\mathrm{Au-Cr}/a-\mathrm{SiO}_{2}} for nanolines deposited on a-SiO2\mathrm{SiO}_{2} measured with 10× objective (r=4.1±0.2 μmr = 4.1 \pm0.2\ \mu\mathrm{m}) and 20× objective (r=2.3±0.1 μmr = 2.3 \pm0.1\ \mu\mathrm{m}). The legend uses the notation ww/gap (nm); e.g., 172/180 corresponds to a width ww of 172 nm and a gap (L−wL-w) of 180 nm.

Schematics of the nanoline centerline offset and FEM-calculated phase responses

Figure S7. (a) Schematics of the offset between the nanoline centerline and the center of the laser profile. (b) FEM-calculated phase responses of the 234 nm and 1052 nm nanoline patterns on sapphire with and without a 100 nm lateral offset from the nanoline centerline.

The results in Figure 4(a) and Figure 4(b) show that the measured phase lag in FDTR changes upon changes in the nanoline pattern geometry [variations within (a) or (b)], as well as upon changes in the laser spot radius [variations across (a) and (b) for the same nanoline pattern]. The change in phase lag due to a change in nanoline dimensions was previously measured using TDTR and modeled using an infinite spot radius approximation.[15, 22, 23, 25] Comparing across Figure 4(a) and Figure 4(b), the need to incorporate the effect of the finite laser spot radius is evident. There is an increase in phase lag (i.e., a larger negative value) recorded for the larger spot radius for the same nanoline pattern, which is consistent with the effect of using a larger spot radius in the conventional FDTR heat diffusion model.[35] [36] The laser spot radius dependence is further explored in Appendix § 4.2, where we emphasize differences relative to the infinite spot radius assumption.

As shown in Figure 5, we find that ky,Auk_{y,\mathrm{Au}} decreases by roughly half as the nanoline width is reduced from 1024 nm to 172 nm for both spot radii, based on our set of three measured linewidths. We hypothesize that this reduction is due to a modified grain structure in the Au when it is deposited between the narrow e-beam resist (AR-P 6200.11, CSAR) walls, as noted in Sec. S1. In the most extreme case, the Au is deposited in a 172 nm wide trench between 300 nm tall CSAR walls. ky,Auk_{y,\mathrm{Au}} for the nanoline width of 1024 nm is comparable to that for the square patch, suggesting that the grain structure is less affected at this larger width. GAu−Cr/a−SiO2G_{\mathrm{Au-Cr}/a-\mathrm{SiO}_{2}} for all patterns are comparable to the GAu−Cr/a−SiO2G_{\mathrm{Au-Cr}/a-\mathrm{SiO}_{2}} extracted from the square patch within the experimental uncertainty. These values also fall within the range we routinely obtain (20–60 MW m−2 K−1\mathrm{MW\,m^{-2}\,K^{-1}}) for this interface from reference measurements used to calibrate the laser spot radius prior to experiments.

Sapphire

We next studied cc-sapphire to further experimentally validate the analytical solution within a diffusive thermal transport regime. FDTR phase lag data and fits for the Au nanolines on cc-sapphire are shown in Figure 6(a) (10× objective) and 6(b) (20× objective). We co-fit the thermal boundary conductance at the Au-Cr/cc-sapphire interface (GAu−Cr/c-sapphireG_{\mathrm{Au-Cr}/c\text{-}\mathrm{sapphire}}) and the thermal conductivity of cc-sapphire (kc-sapphirek_{c\text{-}\mathrm{sapphire}}) for the different nanoline patterns. The extracted ky,Auk_{y,\mathrm{Au}} from the Au nanolines on a-SiO₂ was used as a fixed parameter. Because ky,Auk_{y,\mathrm{Au}} varies with the nanoline width on a-SiO2\mathrm{SiO}_{2}, we selected the ky,Auk_{y,\mathrm{Au}} value obtained from the nanolines deposited on a-SiO2\mathrm{SiO}_{2} using the 20× objective with the closest linewidth. We note that FDTR has a much lower sensitivity to ky,Auk_{y,\mathrm{Au}} for the cc-sapphire substrate compared to the a-SiO2\mathrm{SiO}_{2} substrate, as shown in Figure 3(b).

Phase lag and analytical solution fits for nanolines deposited on the c-sapphire

Figure 6. Phase lag and analytical solution fits for nanolines deposited on the cc-sapphire measured with (a) 10× objective (r=5.0±0.2 μmr = 5.0 \pm0.2\ \mu\mathrm{m}) and (b) 20× objective (r=2.5±0.1 μmr = 2.5 \pm0.1\ \mu\mathrm{m}). The legend uses the notation ww/gap (nm); e.g., 234/169 corresponds to a width ww of 234 nm and a gap (L−wL-w) of 169 nm.

The fitted analytical solution plotted in Figure 6(a) and Figure 6(b) shows excellent agreement with the measured phase lag data for both laser spot radii. The extracted kc−sapphirek_{c-\mathrm{sapphire}} for all patterns plotted in Figure 7(a) are within the error margins of kc−sapphirek_{c-\mathrm{sapphire}} from the square patch as well as the reported bulk value of 38 ±\pm 2 W/m-K.37 These results indicate that the thermal conductivity suppression due to ballistic phonon transport is insignificant within the range of MFPs we explored with nanolines, as approximately 90 percent of the bulk kc−sapphirek_{c-\mathrm{sapphire}} is accumulated from MFPs less than 100 nm.[20] As shown in Figure 7(b), GAu−Cr/c−sapphireG_{\mathrm{Au-Cr}/c-\mathrm{sapphire}} from all patterns is comparable to the values extracted from the square pattern as well as to a literature-reported value (390 ±\pm 70 MW/m2^{2}-K).38 Variation in the extracted GAu−Cr/c−sapphireG_{\mathrm{Au-Cr}/c-\mathrm{sapphire}} is likely due to a relatively low sensitivity to GAu−Cr/c−sapphireG_{\mathrm{Au-Cr}/c-\mathrm{sapphire}}, as shown in Figure 3(b).

Bar charts comparing extracted thermal conductivity and interfacial thermal conductance

Figure 7. Extracted (a) kc−sapphirek_{c-sapphire} and (b) GAu−Cr/c−sapphireG_{Au-Cr/c-sapphire} for nanolines deposited on the cc-sapphire measured with 10×10\times objective (r=5.0±0.2r = 5.0 \pm0.2 μ\mum) and 20×20\times objective (r=2.5±0.1r = 2.5 \pm0.1 μ\mum). The legend uses the notation ww/gap (nm); e.g., 234/169 corresponds to a width ww of 234 nm and a gap (L−wL-w) of 169 nm. The shaded lines represent the literature-reported kc−sapphirek_{c-sapphire} and GAu−Cr/c−sapphireG_{Au-Cr/c-sapphire}, which are 38 ±\pm 2 W/m-K37 and 390 ±\pm 70 MW/m2^{2}-K.38

Conclusion

We developed and validated a nanoline-based FDTR technique with a finite spot radius that targets studies of kaccumk_{\mathrm{accum}} without the need for samples of varied thickness. A new analytical solution to the heat diffusion equation was developed to account for finite Gaussian laser spots and nanoline periodicity. To validate the solution, we presented FDTR measurements of Au nanolines with widths ranging from 100 nm to 1000 nm that were deposited on a-SiO2_2 and cc-sapphire substrates, which exhibit diffusive heat transport due to their short energy carrier MFPs. Measurements on both a-SiO2_2 and sapphire substrates showed good agreement between the experimental phase lag data and the analytical model, validating the robustness of the approach. The high sensitivity of ky,Auk_{y,\mathrm{Au}} on a-SiO2_2 highlights the importance of performing sensitivity analysis for each new set of experimental conditions. The present framework can be extended beyond one-dimensional nanoline gratings to two-dimensional periodic arrays, such as square or rectangular metallic features, by incorporating an additional Fourier series expansion in the transverse direction. Moving forward, the established FDTR framework with its commonly used range of spot radii can now be extended to measure phonon MFP contributions to kaccumk_{accum} across a wide range spanning sub-micron to micrometer length scales.

AUTHOR’S CONTRIBUTIONS

Abhishek Pathak: Conceptualization (lead); Data curation (equal); Formal analysis (equal); Methodology (lead); Software (lead); Visualization (equal); Validation (equal); Writing - original nal draft (equal); Writing - review & editing (equal). Seungwon Park: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Methodology (equal); Software (equal); Visualization (equal); Validation (equal); Writing - original draft (equal); Writing - review & editing (equal). Shubham Deshmukh: Formal analysis (support); Methodology (support); Software (support); Validation (support); Writing - original draft (support); Writing - review & editing (support). Alan J. H. McGaughey: Funding acquisition (lead); Project administration (lead); Supervision (lead); Writing - review & editing (lead). Jonathan A. Malen: Funding acquisition (lead); Project administration (lead); Supervision (lead); Writing - review & editing (lead).

ACKNOWLEDGMENTS

This work was supported by the Army Research Office Award No. W911NF2220191. The authors acknowledge the use of the Bertucci Nanotechnology Laboratory at Carnegie Mellon University supported by grant BNL-78657879 and use of the Materials Characterization Facility at Carnegie Mellon University supported by grant MCF-677785.

CONFLICT OF INTEREST

The authors have no conflicts to disclose.

DATA AVAILABILITY

The data that support the findings of this study are available from the corresponding author upon reasonable request.

REFERENCES

Simplifying 4-D integral in transfer function to 2-D integral

Eq. (15) is a four-dimensional integral, which can be simplified into a two-dimensional integral by applying the Hubbard–Stratonovich transformation,[28] which states that for a real, positive scalar γ\gamma,

∫−∞∞exp⁡(−12γx2+iξxx) dx=2πγexp⁡{−ξx22γ}.(19)\int_{-\infty}^{\infty} \exp\left(-\frac{1}{2}\gamma x^2+i\xi_x x\right)\,dx=\sqrt{\frac{2\pi}{\gamma}}\exp\left\{-\frac{\xi_x^2}{2\gamma}\right\}. \tag*{(19)}

To convert Eq. (15) into the appropriate format, we use the identity cos⁡(θ)=exp⁡(iθ)+exp⁡(−iθ)2\cos(\theta)=\frac{\exp(i\theta)+\exp(-i\theta)}{2}, and rewrite the cosine term in Eq. (15) to obtain

H(ω)=πδ(ω−ω0)∑n=0∞∑m=0∞∫−∞∞∫−∞∞∫−∞∞∫−∞∞amanπr12(DC)×{exp⁡(−(ξx−2nπLx)2r028)+exp⁡(−(ξx+2nπLx)2r028)}×exp⁡(iξxx+iξyy−ξy2r028−2(x2+y2)r12)×12{exp⁡(i2mπxLx)+exp⁡(−i2mπxLx)} dξxdξydxdy.(20)\begin{aligned} H(\omega)={}&\pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\sum_{m=0}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\frac{a_m a_n}{\pi r_1^2}\left(\frac{D}{C}\right)\\ &\times\left\{\exp\left(-\frac{\left(\xi_x-\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)+\exp\left(-\frac{\left(\xi_x+\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)\right\}\\ &\times\exp\left(i\xi_xx+i\xi_yy-\frac{\xi_y^2r_0^2}{8}-\frac{2(x^2+y^2)}{r_1^2}\right)\\ &\times\frac{1}{2}\left\{\exp\left(i\frac{2m\pi x}{L_x}\right)+\exp\left(-i\frac{2m\pi x}{L_x}\right)\right\}\,d\xi_xd\xi_y dxdy. \tag*{(20)} \end{aligned}

Noting that ξy\xi_y is independent of yy, the integration of the variable yy can be rearranged in the form of Eq. (19) as

H(ω)=πδ(ω−ω0)∑n=0∞∑m=0∞∫−∞∞∫−∞∞∫−∞∞aman2πr12(−DC)×{exp⁡(−(ξx−2nπLx)2r028)+exp⁡(−(ξx+2nπLx)2r028)}×exp⁡(iξxx−2x2r12)×{exp⁡(i2mπxLx)+exp⁡(−i2mπxLx)}[exp⁡(−ξy2r028)∫−∞∞exp⁡(iξyy−2y2r12)dy]dξxdξydx.(21)\begin{aligned} H(\omega)={}&\pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\sum_{m=0}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\frac{a_m a_n}{2\pi r_1^2}\left(-\frac{D}{C}\right)\\ &\times\left\{\exp\left(-\frac{\left(\xi_x-\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)+\exp\left(-\frac{\left(\xi_x+\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)\right\}\times\exp\left(i\xi_xx-\frac{2x^2}{r_1^2}\right)\\ &\times\left\{\exp\left(i\frac{2m\pi x}{L_x}\right)+\exp\left(-i\frac{2m\pi x}{L_x}\right)\right\}\left[\exp\left(-\frac{\xi_y^2r_0^2}{8}\right)\int_{-\infty}^{\infty}\exp\left(i\xi_yy-\frac{2y^2}{r_1^2}\right)dy\right]d\xi_xd\xi_y dx. \tag*{(21)} \end{aligned}

With γ=4r12\gamma=\frac{4}{r_1^2}, the expression simplifies to

H(ω)=πδ(ω−ω0)∑n=0∞∑m=0∞∫−∞∞∫−∞∞∫−∞∞aman2πr12(−DC)×{exp⁡(−(ξx−2nπLx)2r028)+exp⁡(−(ξx+2nπLx)2r028)}×exp⁡(iξxx−2x2r12)×{exp⁡(i2mπxLx)+exp⁡(−i2mπxLx)}[exp⁡(−ξy2r028)πr122exp⁡(−ξy2r128)]dξxdξydx.(22)\begin{aligned} H(\omega)={}&\pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\sum_{m=0}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\frac{a_m a_n}{2\pi r_1^2}\left(-\frac{D}{C}\right)\\ &\times\left\{\exp\left(-\frac{\left(\xi_x-\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)+\exp\left(-\frac{\left(\xi_x+\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)\right\}\times\exp\left(i\xi_xx-\frac{2x^2}{r_1^2}\right)\\ &\times\left\{\exp\left(i\frac{2m\pi x}{L_x}\right)+\exp\left(-i\frac{2m\pi x}{L_x}\right)\right\}\left[\exp\left(-\frac{\xi_y^2r_0^2}{8}\right)\sqrt{\frac{\pi r_1^2}{2}}\exp\left(-\frac{\xi_y^2r_1^2}{8}\right)\right]d\xi_xd\xi_y dx. \tag*{(22)} \end{aligned}

Further simplification and rearranging the terms leads to

H(ω)=πδ(ω−ω0)∑n=0∞∑m=0∞∫−∞∞∫−∞∞∫−∞∞aman22πr12exp⁡(−ξy28(r02+r12))(−DC)×{exp⁡(−(ξx−2nπLx)2r028)+exp⁡(−(ξx+2nπLx)2r028)}×exp⁡(iξxx−2x2r12)×{exp⁡(i2mπxLx)+exp⁡(−i2mπxLx)}dξxdξydx.(23)\begin{aligned} H(\omega)={}&\pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\sum_{m=0}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\frac{a_m a_n}{2\sqrt{2\pi r_1^2}}\exp\left(-\frac{\xi_y^2}{8}(r_0^2+r_1^2)\right)\left(-\frac{D}{C}\right)\\ &\times\left\{\exp\left(-\frac{\left(\xi_x-\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)+\exp\left(-\frac{\left(\xi_x+\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)\right\}\times\exp\left(i\xi_xx-\frac{2x^2}{r_1^2}\right)\\ &\times\left\{\exp\left(i\frac{2m\pi x}{L_x}\right)+\exp\left(-i\frac{2m\pi x}{L_x}\right)\right\}d\xi_xd\xi_y dx. \tag*{(23)} \end{aligned}

Similarly, integration over xx can be performed by collecting terms as

H(ω)=πδ(ω−ω0)∑n=0∞∑m=0∞∫−∞∞∫−∞∞aman22πr12exp⁡(−ξy28(r02+r12))(−DC)×{exp⁡(−(ξx−2nπLx)2r028)+exp⁡(−(ξx+2nπLx)2r028)}×[∫−∞∞exp⁡{i(ξx+2mπLx)x−2x2r12}dx+∫−∞∞exp⁡{i(ξx−2mπLx)x−2x2r12}dx]dξxdξy.(24)\begin{aligned} H(\omega)={}&\pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\sum_{m=0}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\frac{a_m a_n}{2\sqrt{2\pi r_1^2}}\exp\left(-\frac{\xi_y^2}{8}(r_0^2+r_1^2)\right)\left(-\frac{D}{C}\right)\\ &\times\left\{\exp\left(-\frac{\left(\xi_x-\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)+\exp\left(-\frac{\left(\xi_x+\frac{2n\pi}{L_x}\right)^2r_0^2}{8}\right)\right\}\\ &\times\left[\int_{-\infty}^{\infty}\exp\left\{i\left(\xi_x+\frac{2m\pi}{L_x}\right)x-\frac{2x^2}{r_1^2}\right\}dx+\int_{-\infty}^{\infty}\exp\left\{i\left(\xi_x-\frac{2m\pi}{L_x}\right)x-\frac{2x^2}{r_1^2}\right\}dx\right]d\xi_xd\xi_y. \tag*{(24)} \end{aligned}

Again, using the identity in Eq. (19) to perform integration on the xx coordinates leads to the expression of H(ω)H(\omega) in a two-dimensional integral form as

H(ω)=πδ(ω−ω0)∑n=0∞∑m=0∞∫−∞∞∫−∞∞anam4(−DC)exp⁡{−ξy28(r02+r12)}×[exp⁡{−r028(ξx−2nπLx)2−r128(ξx−2mπLx)2}+exp⁡{−r028(ξx−2nπLx)2−r128(ξx+2mπLx)2}+exp⁡{−r028(ξx+2nπLx)2−r128(ξx−2mπLx)2}+exp⁡{−r028(ξx+2nπLx)2−r128(ξx+2mπLx)2}] dξxdξy.(25)\begin{aligned} H(\omega) ={}& \pi\delta(\omega-\omega_0)\sum_{n=0}^{\infty}\sum_{m=0}^{\infty}\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}\frac{a_n a_m}{4}\left(-\frac{D}{C}\right)\exp\left\{-\frac{\xi_y^2}{8}\left(r_0^2+r_1^2\right)\right\}\\ &\times\left[\exp\left\{-\frac{r_0^2}{8}\left(\xi_x-\frac{2n\pi}{L_x}\right)^2-\frac{r_1^2}{8}\left(\xi_x-\frac{2m\pi}{L_x}\right)^2\right\}\right.\\ &\quad+\exp\left\{-\frac{r_0^2}{8}\left(\xi_x-\frac{2n\pi}{L_x}\right)^2-\frac{r_1^2}{8}\left(\xi_x+\frac{2m\pi}{L_x}\right)^2\right\}\\ &\quad+\exp\left\{-\frac{r_0^2}{8}\left(\xi_x+\frac{2n\pi}{L_x}\right)^2-\frac{r_1^2}{8}\left(\xi_x-\frac{2m\pi}{L_x}\right)^2\right\}\\ &\left.\quad+\exp\left\{-\frac{r_0^2}{8}\left(\xi_x+\frac{2n\pi}{L_x}\right)^2-\frac{r_1^2}{8}\left(\xi_x+\frac{2m\pi}{L_x}\right)^2\right\}\right]\,\mathrm{d}\xi_x\mathrm{d}\xi_y. \tag*{(25)} \end{aligned}

Variation of predicted phase with laser spot radius

In Figure 8, we show how the predicted phase changes upon varying the spot radius for a nanoline pattern on the a-SiO2_2 substrate with w=1024w = 1024 nm and L=1188L = 1188 nm. As we increase the spot radius in our model while keeping other parameters constant, the predicted phase lag increases for all frequencies. Eventually, for a large radius (r>100 μmr > 100\,\mu\mathrm{m}), the predicted phase converges to the values predicted by the previously used infinite radius model that was adopted for TDTR measurements.[15] Our analytical model, used with the bulk thermal conductivity of a-SiO2_2, shows excellent agreement with the measurements, further reinforcing the impact of the finite spot radius.

Spot radius dependence of predicted phase on nanolines

Figure 8. Spot radius dependence of predicted phase on nanolines (w=1024w = 1024 nm and gap (L−wL-w) = 164 nm) deposited on a-SiO2_2 sample from analytical model derived in this work compared against previously used model (r→∞r \to\infty) in TDTR study.[15]

Supplementary Material

Fabrication of nanoline array

To test the validity of the model developed in Sec. II, we deposited Au nanolines on cc-sapphire and a-SiO2_2 substrates and measured their responses using FDTR. The electron-beam lithography process followed by a lift-off procedure that was used to fabricate the nanoline patterns is illustrated in Figure S1. The targeted nanoline widths range from 100 to 1000 nm, with a 200 nm gap between adjacent lines. In the direction parallel to the lines [y in Figure 1(b)], the length of the nanolines was set to 50 μm\mu\mathrm{m} with a periodicity of 100 μm\mu\mathrm{m}, as shown in Figure 2(a). For every value of nanoline width, a 500 μm×450 μm\mu\mathrm{m} \times450\ \mu\mathrm{m} area was patterned.

Nano-fabrication process for metallic heaters

Figure S1. (a) Nano-fabrication process for metallic heaters.

A double-side polished 400 μm\mu\mathrm{m} thick 2-inch wafer of cc-sapphire, purchased from MSE Supplies LLC, was cleaned by sonication in acetone for 10 minutes, followed by rinsing with isopropyl alcohol (IPA) and drying with nitrogen gas. To promote the adhesion of e-beam resist, the substrate was baked at 180°C for 30 minutes, followed by an O2_2 plasma cleaning step in an IPC barrel etcher operated at 750 mTorr and 300 W for 10 minutes. The baking process removes surface-adsorbed moisture, whereas O2_2 plasma eliminates residual organic contaminants. An approximately 300 nm layer of e-beam resist (AR-P 6200.11, CSAR) was spin-coated onto the substrate, followed by a pre-bake at 150°C for 1 minute. Subsequently, an approximately 20 nm conductive film of DisCharge-H2_2O-2X (Dischem Inc.) was spin-coated to dissipate surface charge during electron-beam exposure and to prevent pattern distortion. The lithography was performed using a G100 Electron Beam Lithography system operating at an accelerating voltage of 100 kV with a spot radius of 1.8 nm. Based on the manufacturer-recommended process, the dose-to-clear for CSAR was determined to be 120 μC/cm2\mu\mathrm{C}/\mathrm{cm}^{2} for a 300 nm thick layer on a cc-sapphire substrate. We used a higher dose of 200 μC/cm2\mu\mathrm{C}/\mathrm{cm}^{2} to account for proximity effects and to achieve a slightly negative resist-sidewall profile after development, which facilitates an efficient lift-off process. The same parameters were found to be suitable for the 300 μm\mu\mathrm{m} thick 2-inch a-SiO2_2 substrate.

Following exposure, the DisCharge-H2_2O-2X coating was removed by rinsing with deionized water and the substrate was dried under a nitrogen stream. Pattern development was performed at room temperature using AR600-546 (Allresist) developer for 2 minutes, followed by immersion in IPA for 1 minute to stop the reaction. The patterned substrate was then treated with a mild dose of O2_2 plasma (20 W, 20 mTorr) in a Plasma-therm 790 reactive ion etching (RIE) system for 40 seconds to remove residual resist and enhance metal adhesion in the subsequent step. A 5 nm thick Cr adhesion layer38 followed by a 70 nm Au layer were deposited on the patterns in a Kurt J. Lesker PRO Line PVD75 E-beam Evaporator under high vacuum (≈3×10−7\approx3 \times10^{-7} Pa). The lift-off process was performed by upside-down immersion of the samples overnight in AR600-71 Remover. Although a spatially uniform increase in dose preserved the shape-integrity of the patterns, an artifact of this over-exposure is that the nanolines are about 20 nm too wide on a-SiO2_2 and 35-50 nm too wide on cc-sapphire.

Determination of Au-Cr thickness

Figure S2 shows a sample profilometer scan measured with KLA Tencor P-15 Profilometer on a cc-sapphire square patch to determine the combined thickness of Au and Cr layers. The combined thickness was determined by subtracting the base average height from the step average height, which was measured to be ≈75 nm\approx75\ \mathrm{nm}.

Sample profilometer scan

Figure S2. Sample profilometer scan.

Finite element validation of heat flux boundary condition on continuous film

The analytical model derived in Sec. II approximates the nanoline geometry heated with a Gaussian laser spot by applying a modified boundary heat flux profile on a continuous Au film. The presence of a physical gap is simulated by setting kx=0k_x = 0 for the Au film. To evaluate the effect of this simplifying assumption on the predicted phase lag, we prepared a 3-D finite element model of nanoline on the substrate, as shown in Figure S3. Particularly, we evaluate whether the backflow of heat from the substrate to the continuous Au film and its spread in the yy-direction in the Au film affect the phase lag.

Comparison between the phase lag recorded in the FEM model using continuous film simplification vs physically modeling metal nanolines

Figure S3. Comparison between the phase lag recorded in the FEM model using continuous film simplification vs physically modeling metal nanolines (a) Schematic diagram of the FEM model geometry with w=234w = 234 nm, L=403L = 403 nm, t=75t = 75 nm. (b) The phase lag shows excellent match between the two models.

We used COMSOL to perform a steady-state solution under constant, spatially Gaussian heat flux followed by a harmonic perturbation analysis at several modulation frequencies encompassing the whole range used in our measurements. A 3-D finite element model is computationally expensive, and a few simplifications are made to keep the simulation computationally tractable.

The size of the computational domain is limited to Lx=8 μmL_x = 8\,\mu\mathrm{m} in the xx-direction and Ly=5 μmL_y = 5\,\mu\mathrm{m} in the yy-direction, and the substrate is only modeled as a d=5 μmd = 5\,\mu\mathrm{m} thick layer in the zz-direction. Symmetry boundary conditions are applied for the YY-ZZ plane at x=0x = 0 and the XX-ZZ plane at y=0y = 0, such that only a quarter model is solved. To achieve a steady-state solution quickly, isothermal boundary conditions are applied at x=Lxx = L_x and z=dz = d. However, y=Lyy = L_y boundary is kept insulated to avoid forced heat flow in the yy-direction. These simplifications are chosen to speed up the finite element simulation without affecting the core conclusions regarding the backflow of the heat from the substrate and the yy-direction heat flow in the Au film.

We performed the same analysis by replacing the physical nanolines with a continuous film and applied the same modified heat flux as used in deriving the analytical model (Sec. II) and setting kx=0k_x = 0 in the film. Both the simulations used identical material properties except kxk_x of the Au film. Figure S3 shows the comparison between the phase lag computed for the two models. The maximum absolute deviation in the phase lag between the two simulations is less than 1.1∘1.1^\circ over the entire frequency range. The phase shift corresponds to a 0.3%0.3\% and 2.9%2.9\% change in the extracted thermal conductivity of c-sapphire and the thermal boundary conductance at the Au-Cr/c-sapphire interface, respectively. The analysis clearly shows that the assumption of a continuous Au film with kx=0k_x = 0 excellently approximates the nanolines with a physical gap for our measurements.

Absence of thermoreflectance from bare substrate

Figure S4 shows the thermoreflectance signal obtained on the Au transducer deposited on the cc-sapphire substrate and on the bare cc-sapphire substrate. Since cc-sapphire is transparent to the pump (405 nm) and probe (532 nm) lasers, it does not absorb heat, and the signal obtained for the bare substrate is basically coming from the instrumentation noise of the setup. This also ensures that the laser passing through the gap between the nanolines and incident directly onto the substrate does not affect the thermoreflectance measurement.

Phase lag and analytical solution fitting of square patches

Figure S5 and (Figure S5) show the fitting of the phase lag obtained in the FDTR measurement on one of the square patches deposited on cc-sapphire and a-SiO2_2, respectively. As the size of the square patches (50 μm×50 μm50\,\mu\mathrm{m} \times50\,\mu\mathrm{m}) is large compared to the spot radius used in the measurements, the conventional FDTR model with an infinite uniform transducer film is used for fitting the measurement. For comparison, we also fit the measurements using our analytical solution for the periodic nanoline patterns with w=50 μmw = 50\,\mu\mathrm{m} and L=100 μmL = 100\,\mu\mathrm{m}. The excellent agreement between the two curves shows that our analytical model converges to the standard FDTR solution in the limit of large ww relative to the spot radius.

Phase lag and analytical solution fitting of square patch on (a) c-sapphire substrate and (b) a-SiO2 substrate

Figure S5. Phase lag and analytical solution fitting of square patch on (a) cc-sapphire substrate and (b) a-SiO2_2 substrate.

modelky,Auk_{y,\mathrm{Au}} [W/m-K]ky,Auk_{y,\mathrm{Au}} [W/m-K]TBCTBC [MW/m2^2K]TBCTBC [MW/m2^2K]
model10x20x10x20x
uniform160±4.8a160 \pm4.8^{\mathrm{a}}160±4.8a160 \pm4.8^{\mathrm{a}}42−14+6.842^{+6.8}_{-14}32−5.3+1232^{+12}_{-5.3}
nanoline160±4.8a160 \pm4.8^{\mathrm{a}}160±4.8a160 \pm4.8^{\mathrm{a}}39−4.1+6.339^{+6.3}_{-4.1}37−3.2+7.137^{+7.1}_{-3.2}

Table S1. a-SiO2_2 square patch fitting with nanoline model and uniform film model.

Transmitted energy simulation

Figure S6 shows the schematic of the simulation domain used to calculate the transmitted laser energy through the gap between the adjacent metal nanolines using the COMSOL wave optics module.3 One unit cell of the periodic grating structure was simulated with Floquet boundary conditions applied on the lateral boundaries. The laser is incident perpendicular to the grating surface from air, k⃗\vec{k} shows the wavevector of the incident electromagnetic wave. The polarization electric field E⃗\vec{E} is aligned with the longitudinal direction (into the page in Figure S6). The top boundary is used for applying wave excitation due to a unit power incident electromagnetic beam. Top and bottom boundaries are used to measure the reflected and transmitted power, respectively.

Schematics of the simulation domain and calculated percentages of reflected, transmitted, and absorbed laser energy

Figure S6. (a) Schematics of the simulation domain. The periodicity LL is chosen such that the gap between adjacent nanolines is 200 nm. (b) Calculated percentage of reflected, transmitted, and absorbed laser energy by Au gratings on a-SiO2_2 substrate and (c) cc-sapphire substrate.

modelkc-sapphirek_{c\text{-}sapphire} [W/m-K] 10xkc-sapphirek_{c\text{-}sapphire} [W/m-K] 20xTBC [MW/m2^{2}K] 10xTBC [MW/m2^{2}K] 20x
uniform37−2.6+2.837^{+2.8}_{-2.6}37−2.7+2.737^{+2.7}_{-2.7}338−19+20338^{+20}_{-19}429−17+20429^{+20}_{-17}
nanoline38−2.3+2.538^{+2.5}_{-2.3}38−2.6+2.638^{+2.6}_{-2.6}286−17+22286^{+22}_{-17}392−33+28392^{+28}_{-33}

Table S2. cc-sapphire square patch fitting with nanoline model and uniform film model.

The optical properties of Au nanolines, air and substrates were obtained from ‘Refractiveindex.info database of optical constants’4 for the pump laser wavelength (405 nm), and are listed in Table S3. To be consistent with our nanoline pattern geometry, the gap between nanolines was kept at 200 nm, and the Au thickness was kept at 75 nm. Figure S6 and (c) show how the percentage of incident light reflected, transmitted, and absorbed by the Au nanoline grating changes with nanoline width. The results obtained for a-SiO2_2 and cc-sapphire are qualitatively similar. For a smaller nanoline width, the length of the gap forms a large fraction of the unit cell, and significant light is transmitted to the substrate. As the nanoline width relative to the periodicity increases, the absorption in the Au nanoline rapidly increases and saturates to ≈ 60%. For larger nanolines, the decreasing trend of transmission power and the increasing trend of reflected power balance each other.

MaterialReal part of refractive indexImaginary part of refractive index
Au1.46521.9549
a-SiO2_21.46960
cc-sapphire1.78570

Table S3. Material properties for COMSOL simulation for 405 nm pump laser.40

Although there is ≈ 20% transmission for the smallest nanoline width (200 nm) used in our study, both a-SiO2_2 and cc-sapphire are transparent to the pump wavelength, and no direct substrate heating is expected. Indirect heating of Au nanolines from the light reflected back from the bottom surface is possible. The application of a refractive index-matching liquid can be used to minimize reflection from the bottom surface. We used a thin layer of glycerine (refractive index ≈ 1.5) to achieve this effect.

Uncertainty Quantification of Laser Spot Positioning

To position the laser spot on the nanolines, we used a micro-manipulator to translate the sample relative to the fixed laser spot while monitoring the measured signal. The laser spot was aligned to the nanoline centerline by maximizing the signal amplitude. Because the micro-manipulator has a positioning resolution of 50 nm, a small residual lateral misalignment between the laser spot and the nanoline centerline may remain. We therefore evaluated the influence of this uncertainty on the measured phase using Finite Element Method (FEM) simulations.

In the simulations, the laser spot was displaced from the nanoline centerline along the xdirection, and the resulting phase at the measurement frequencies was calculated. We considered a conservative worst-case lateral offset of 100 nm, corresponding to twice the micro-manipulator resolution. For both the 234 nm and 1052 nm sapphire nanoline patterns, the calculated phase variation was less than 0.36∘0.36^\circ over the entire frequency range. The phase shift corresponds to a 2.1% and 0.7% change in the extracted thermal conductivity of cc-sapphire and the thermal boundary conductance at the Au-Cr/cc-sapphire interface, respectively.

References

  1. [1]A. Negi, A. Rodriguez, X. Zhang, A. H. Comstock, C. Yang, D. Sun, X. Jiang, D. Kumah, M. Hu, and J. Liu, “Thickness-dependent thermal conductivity and phonon mean free path distribution in single-crystalline barium titanate,” Advanced Science 10, 2301273 (2023).DOI
  2. [2]C. Zhang, F. Liu, D. Kim, X. Xu, S. Guo, Y. Pei, K. A. Mkhoyan, T. Feng, B. Jalan, and X. Wang, “Probing phonon mean-free-path distribution via thickness-dependent thermal conductivity in epitaxial srsno3,” Materials Today Physics , 101991 (2025).DOI
  3. [4]K. T. Regner, D. P. Sellan, Z. Su, C. H. Amon, A. J. McGaughey, and J. A. Malen, “Broadband phonon mean free path contributions to thermal conductivity measured using frequency domain thermoreflectance,” Nature communications 4, 1640 (2013).DOI
  4. [5]J. L. Braun and P. E. Hopkins, “Upper limit to the thermal penetration depth during modulated heating of multilayer thin films with pulsed and continuous wave lasers: A numerical study,” Journal of Applied Physics 121 (2017).DOI
  5. [6]K. T. Regner, J. P. Freedman, and J. A. Malen, “Advances in Studying Phonon Mean Free Path Dependent Contributions to Thermal Conductivity,” Nanoscale and Microscale Thermophysical Engineering 19, 183–205 (2015).DOI
  6. [7]K. T. Regner, A. J. McGaughey, and J. A. Malen, “Analytical interpretation of nondiffusive phonon transport in thermoreflectance thermal conductivity measurements,” Phys. Rev. B 90, 064302 (2014).DOI
  7. [8]K. T. Regner, L. C. Wei, and J. A. Malen, “Interpretation of thermoreflectance measurements with a two-temperature model including non-surface heat deposition,” Journal of Applied Physics 118 (2015).DOI
  8. [11]A. A. Maznev, J. A. Johnson, and K. A. Nelson, “Onset of nondiffusive phonon transport in transient thermal grating decay,” Physical Review B—Condensed Matter and Materials Physics 84, 195206 (2011).DOI
  9. [12]J. Cuffe, J. K. Eliason, A. A. Maznev, K. C. Collins, J. A. Johnson, A. Shchepetov, M. Prunnila, J. Ahopelto, C. M. Sotomayor Torres, G. Chen, et al., “Reconstructing phonon mean-free-path contributions to thermal conductivity using nanoscale membranes,” Physical Review B 91, 245423 (2015).arxiv.org/abs/1408.6747
  10. [13]T. Kim, J. Moon, and A. J. Minnich, “Origin of micrometer-scale propagation lengths of heat-carrying acoustic excitations in amorphous silicon,” Physical Review Materials 5, 065602 (2021).DOI
  11. [15]Y. Hu, L. Zeng, A. J. Minnich, M. S. Dresselhaus, and G. Chen, “Spectral mapping of thermal conductivity through nanoscale ballistic transport,” Nature nanotechnology 10, 701–706 (2015).DOI
  12. [16]M. E. Siemens, Q. Li, R. Yang, K. A. Nelson, E. H. Anderson, M. M. Murnane, and H. C. Kapteyn, “Quasi-ballistic thermal transport from nanoscale interfaces observed using ultrafast coherent soft x-ray beams,” Nature materials 9, 26–30 (2010).
  13. [17]K. M. Hoogeboom-Pot, J. N. Hernandez-Charpak, X. Gu, T. D. Frazer, E. H. Anderson, W. Chao, R. W. Falcone, R. Yang, M. M. Murnane, H. C. Kapteyn, et al., “A new regime of nanoscale thermal transport: Collective diffusion increases dissipation efficiency,” Proceedings of the National Academy of Sciences 112, 4846–4851 (2015).
  14. [18]T. D. Frazer, J. L. Knobloch, K. M. Hoogeboom-Pot, D. Nardi, W. Chao, R. W. Falcone, M. M. Murnane, H. C. Kapteyn, and J. N. Hernandez-Charpak, “Engineering nanoscale thermal transport: Size-and spacing-dependent cooling of nanostructures,” Physical Review Applied 11, 024042 (2019).DOI
  15. [19]S. T. Yang, M. J. Matthews, S. Elhadj, V. G. Draggoo, and S. E. Bisson, “Thermal transport in co2 laser irradiated fused silica: In situ measurements and analysis,” Journal of Applied Physics 106 (2009).DOI
  16. [20]H. Zhou, J. Tiwari, Q. Xu, M. F. R. J. Islam, T. Abedien, and T. Feng, “Effective Phonon Mean Free Paths for Thermal Modeling of Modern Semiconductors,” IEEE Transactions on Electron Devices 73, 2430–2439 (2026).DOI
  17. [21]A. J. Schmidt, R. Cheaito, and M. Chiesa, “A frequency-domain thermoreflectance method for the characterization of thermal properties,” Review of scientific instruments 80 (2009).DOI
  18. [22]A. J. Minnich, Exploring electron and phonon transport at the nanoscale for thermoelectric energy conversion, Thesis, Massachusetts Institute of Technology (2011), accepted: 2011-12-09T21:29:23Z.
  19. [23]A. J. Minnich, “Measuring phonon mean free paths using thermal conductivity spectroscopy,” Annual Review of Heat Transfer 16 (2013), 10.1615/AnnualRevHeatTransfer.v16.70.
  20. [24]L. Zeng, K. C. Collins, Y. Hu, M. N. Luckyanova, A. A. Maznev, S. Huberman, V. Chiloyan, J. Zhou, X. Huang, K. A. Nelson, and G. Chen, “Measuring Phonon Mean Free Path Distributions by Probing Quasiballistic Phonon Transport in Grating Nanostructures,” Scientific Reports 5, 17131 (2015).DOI
  21. [25]X. Chen, C. Hua, H. Zhang, N. K. Ravichandran, and A. J. Minnich, “Quasiballistic thermal transport from nanoscale heaters and the role of the spatial frequency,” Physical Review Applied 10, 054068 (2018).DOI
  22. [26]H. Carslaw and J. Jaeger, “Conduction of heat in solids, clarendon,” (1959).
  23. [27]D. G. Cahill et al., “Analysis of heat flow in layered structures for time-domain thermoreflectance,” Review of scientific instruments 75, 5119 (2004).DOI
  24. [28]L. Schäfer and F. Wegner, “Disordered system with n orbitals per site: Lagrange formulation, hyperbolic symmetry, and goldstone modes,” Zeitschrift für Physik B Condensed Matter 38, 113–126 (1980).DOI
  25. [29]L. Tang and C. Dames, “Anisotropic thermal conductivity tensor measurements using beam-offset frequency domain thermoreflectance (BO-FDTR) for materials lacking in-plane symmetry,” International Journal of Heat and Mass Transfer 164, 120600 (2021).arxiv.org/abs/2007.12984
  26. [30]M. Rahman, M. Shahzadeh, and S. Pisana, “Simultaneous measurement of anisotropic thermal conductivity and thermal boundary conductance of 2-dimensional materials,” Journal of Applied Physics 126, 205103 (2019).DOI
  27. [31]B. C. Gundrum, D. G. Cahill, and R. S. Averback, “Thermal conductance of metal-metal interfaces,” Physical Review B 72, 245426 (2005).
  28. [32]T. L. Bergman and A. S. Lavine, Fundamentals of heat and mass transfer, eighth edition ed. (John Wiley & Sons, Hoboken, NJ, 2017).
  29. [33]VDI e. V., ed., VDI Heat Atlas (Springer Berlin Heidelberg, Berlin, Heidelberg, 2010).
  30. [34]J. Yang, E. Ziade, and A. J. Schmidt, “Uncertainty analysis of thermoreflectance measurements,” Review of Scientific Instruments 87, 014901 (2016).DOI
  31. [35]X. Wang, M. Jeong, A. J. H. McGaughey, and J. A. Malen, “Reducing the uncertainty caused by the laser spot radius in frequency-domain thermoreflectance measurements of thermal properties,” Review of Scientific Instruments 93, 023001 (2022).DOI

Paper details

vlnc:W_yktbychq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents