Whether the presence of adsorbates increases or decreases thermal conductivity in metalorganic frameworks (MOFs) has been an open question. Here we report observations of thermal transport in the metal-organic framework HKUST-1 in the presence of various liquid adsorbates: water, methanol, and ethanol. Experimental thermoreflectance measurements were performed on single crystals and thin films, and theoretical predictions were made using molecular dynamics simulations. We find that the thermal conductivity of HKUST-1 decreases by 40-80% depending on the adsorbate, a result that cannot be explained by effective medium approximations. Our findings demonstrate that adsorbates introduce additional phonon scattering in HKUST-1, which particularly shortens the lifetimes of lowfrequency phonon modes. As a result, the system thermal conductivity is lowered to a greater extent than the increase expected by the creation of additional heat transfer channels. Finally, we show that thermal diffusivity is even more greatly reduced than thermal conductivity by adsorption.

No abstract

We show that the use of sub-nm adhesion layers significantly enhances the thermal interface conductance at metal-dielectric interfaces. A metal-dielectric interface between Au and sapphire (Al 2 O 3 ) was considered using Cu (low optical loss) and Cr (high optical loss) as adhesion layers. To enable high throughput measurements each adhesion layer was deposited as a wedge such that a continuous range of thickness could be sampled. Our measurements of thermal interface conductance at the metal-Al 2 O 3 interface made using frequency domain thermoreflectance show that a 1 nm thick adhesion layer of Cu or Cr is sufficient to enhance the thermal interface conductance by more than a factor of 2 or 4, respectively, relative to the pure Au-Al 2 O 3 interface. The enhancement agrees with the Diffuse Mismatch Model-based predictions of accumulated thermal conductance versus adhesion layer thickness assuming that it contributes phonons with wavelengths less than its adhesion layer thickness, while those with longer wavelengths transmit directly from the Au.

No abstract

GPT-3 shows remarkable in-context learning ability of large-scale language models (LMs) trained on hundreds of billion scale data. Here we address some remaining issues less reported by the GPT-3 paper, such as a non-English LM, the performances of different sized models, and the effect of recently introduced prompt optimization on in-context learning. To achieve this, we introduce Hy-perCLOVA, a Korean variant of 82B GPT-3 trained on a Korean-centric corpus of 560B tokens. Enhanced by our Korean-specific tokenization, HyperCLOVA with our training configuration shows state-of-the-art in-context zero-shot and few-shot learning performances on various downstream tasks in Korean. Also, we show the performance benefits of promptbased learning and demonstrate how it can be integrated into the prompt engineering pipeline. Then we discuss the possibility of materializing the No Code AI paradigm by providing AI prototyping capabilities to nonexperts of ML by introducing HyperCLOVA studio, an interactive prompt engineering interface. Lastly, we demonstrate the potential of our methods with three successful in-house applications.

Abstract

The first systematic measurements on the impact of interdiffusion between a metal overlayer and adhesion layer on the thermal interface conductance (G) at the metal bilayerdielectric interface are reported. Composition depth profiles quantify the interdiffusion of a Au-Cu bilayer as a function of Cu adhesion layer thickness (0-10 nm), annealing time, and annealing temperature. Optical pump/probe measurements of G quantify the effect of Au-Cu interdiffusion on thermal transport across the (Au-Cu)-Al 2 O 3 interface. The enhancement of G between Au and Al 2 O 3 through the addition of a Cu adhesion layer decreases as Au-Cu interdiffusion occurs. For example, annealing a 41 nm Au film with a 4.7 nm Cu adhesion layer on Al 2 O 3 at 520 K for 30 minutes, results in a 52 ±16% drop in G. An analytical model of the composition profile is derived with inputs of annealing time, temperature dependent permeabilities of the Au-Cu interface to each species, and the initial thicknesses of the Au and Cu layers. Integrating this model with a Diffuse Mismatch Model defines a new methodology for the prediction of G that accounts for interdiffusion in metal bilayers on dielectric substrates, and can be used to evaluate the degradation of G over a device's lifetime.

No abstract