Data-Driven Rheological Characterization of Thermal Interface Materials

Thermal management plays a crucial role in maintaining optimal temperatures and improving energy utilization of high-power dense electronics and data centers. Electronics thermal management techniques employ thermal interface materials (TIMs) that help reduce junction temperatures and improve heat removal across the solid-solid interface. TIMs are complex paste or gel-like mixtures comprised of a base polymer and conductivity-enhancer particles (ceramic or metallic). However, these complex fluids have complex rheological behavior. Their rheology is non-Newtonian and cannot be represented simply by conventional shear-thinning models. We present a data-driven approach to rheological characterization of TIMs via rheology-informed neural networks (or, RhiNNs, proposed by Mahmoudabadbozchelou and Jalali, Sci. Rep., 2021. The simplest RhiNN considers the TIM as an elasto-visco-plastic (EVP) complex fluid with shear relaxation (via an elastic shear modulus), a yield stress, and a shear viscosity upon flow. Rheological experiments on TIMs under steady and unsteady forcing provide data to feed the RhiNN and solve the inverse problem of rheological parameter identification. Start-up flow experiments are performed at constant shear rates and are repeated for a range of shear rates. Validation of the RhiNNs is carried out by comparing their prediction to experimental data that is not a part of the training set.