Simulation of thermal Interface Resistances: ENAble first principle molecular dynamics via heat transients – SIRENA

To achieve quantitative predictions of thermal interface resistances and conductivities, the idea is to use ab initio molecular dynamics that allow to describe the structure and dynamics of materials in excellent agreement with experimental measurements. The predictive power of this approach is mainly due to the explicit account of the electronic structure, computed within the density functional theory (DFT) framework and used to calculate interatomic forces instead of using empirical potentials requiring parameters that are particularly unsuitable for two separate materials in contact, as is in the case of our interfaces. However, this predictivity has as a drawback a computational cost that makes it impracticable for the calculation of thermal properties. The approach used in this project is to simulate the transient regime of the heat equation, in a way similar to an experiment on a computer. This allows to determine the relevant quantities, conductivities, and thermal interface resistances. The starting point of SIRENA was a first test aimed at calculating the thermal conductivity of a model of amorphous GeTe4 atoms of approximately 2 nm. This has been successful and used as a template for further developments.

The results of the SIRENA project show in particular that the thermal conductivity is reduced in disordered materials similarly to what occurs in crystals, which have been studied much more from this point of view. The origin of these effects has yet to be elucidated to determine if they are due to propagative modes existing despite the disorder. But their consequence must be considered in order to optimize the heat management of the devices they compose, such as PCMems.

The studies on the thermal conductivity of amorphous materials will be continued during a project of the ITI QMAT starting the 1st of october 2022. More specifically, in the case of the amorphous SiN, one of the objectives is to be able to bridge the gap between the sizes accessible by our calculations and the experimental dimensions. This will be done by developing a so called «Machine Learning» interatomic potential from the FPMD simulations carried out during SIRENA. Another objective is to be able to validate the nanometric effects obtained by calculation with experimental measurements. Initial tests are underway for both thermal conductivities and interface thermal resistances.

The activities of the SIRENA project have resulted in several peer-reviewed journal articles, and three more articles have yet to be written with the results obtained during the project. The activities resulted in numerous invitations of the members of the consortium in international congresses. Finally, many proposals and projects have come out of SIRENA, including a project that will start in October 2022 and that opens up prospects for PCMems and qubits.

The issue addressed in the present SIRENA proposal is the quantitative modelling of heat transport at heterogeneous interfaces, i.e. interfaces between different materials. This situation is commonly encountered in present and future nanotechnologies. For instance, in integrated circuits, the heat generated in the transistor channel (silicon, SiGe, III-V) flows through several interfaces with other insulating/semiconductors materials towards the back side, and with metals of the interconnections towards the front side, respectively. Atomistic computer simulations are an invaluable tool to elucidate the thermal transport in nanostructures by focusing on the fundamental mechanisms of thermal exchange in targeted model systems. We propose atomistic simulations to achieve a breakthrough in the quantitative understanding of thermal interface resistance. To this end, we will combine first-principles molecular dynamics (FPMD) and a simulation framework developed recently to address heat conduction by molecular dynamics (MD) in large-scale systems, the approach to equilibrium molecular dynamics (AEMD). AEMD greatly reduces computational costs via the study of temperature transients. The proof of concept and benchmarking will be performed on two systems selected because i) their size and expected time-dependent thermal response are compatible with a first-principles treatment and ii) they have been studied recently by the most advanced experimental techniques of nanocharacterization, thereby providing an ideally suited experimental counterpart. The first system is an alkane self-assembled monolayer (SAM) sandwiched between a gold and a silicon surface, currently considered experimentally at the IBM Zurich Research Laboratory by B. Gotsmann. This system is particularly well suited for a proof of concept, as it appears that classical MD cannot provide a quantitative estimate of the thermal resistance in this case, due to the many different types of chemical bonds involved.
The second system is the interface of a chalcogenide glass, Ge2Sb2Te5 (GST) with a crystal or an amorphous material, following the study we made on another glass, GeTe4. In that case, published very recently (2017) the FPMD/AEMD methodology was put to good with a frank success to obtain the thermal conductivity. We shall focus on the thermal interface resistance at the interface between GST and silica/silicon thin films, by obtaining a fundamental parameter impacting the heat transfer in GST-PCM memories. The choice of glassy GST stems from the availability of FPMD structures for GST featuring different degrees of chemical order (amount of defects). By focusing on these two classes of systems (molecular interface layer and direct interface between two dense materials), we shall exploit at its best the capabilities of the FPMD/ AEMD approach in order to obtain a quantitative determination of prototypical interface thermal resistances in systems technologically relevant.