Si/diamond thermal boundary conductance enhanced by amorphous carbon interlayers

Research by Khalid Zobaid Adnan, Tanvirul Abedien, and Tianli Feng

Department of Mechanical Engineering

Modern AI hardware faces a critical challenge in heat dissipation. To address this, Khalid Zobaid Adnan, Tanvirul Abedien, and Tianli Feng propose interlayer engineering to enhance thermal properties. They utilized simulations inspired by molecular dynamics (MD), followed by state-of-the-art finite element method (FEM) modeling. Although FEM lacks certain intrinsic atomic-level properties, the combined results clearly demonstrate that inserting an interlayer material significantly lowers peak operating temperatures.

Interfacial atomic structures of silicon and diamond heterostructures — Interfacial atomic structures for (a) Si/diamond, (b) Si/SiN_x/diamond, and (c) Si/a-C/diamond heterostructures obtained from *ab initio* molecular dynamics.

Comparison of Tersoff and MLIP interfacial atomic structures and bonded carbon distribution functions — (a) Interfacial atomic structure of Si/diamond obtained by using (i) Tersoff potential and (ii) machine-learning interatomic potential (MLIP). (b) Distribution function that indicates the average number of carbon atoms that are bonded to Si atoms at the interface, calculated using density-functional theory (DFT), MLIP, and Tersoff potential.

Hot-spot temperature rise as a function of thermal boundary conductance — Hot-spot temperature rise in the active channel layer as a function of thermal boundary conductance (TBC) for (a),(b) silicon-on-diamond (SOD) architectures and (c),(d) diamond-on-silicon metal-oxide-semiconductor field-effect transistors.

The researchers utilized general GPU compute nodes for both their molecular dynamics simulations and MLIP training. The VASP-based ab initio molecular dynamics (AIMD) calculations were run separately on CPU nodes.

Attribution: This content was provided by the researchers and edited by staff at the CHPC.