Multi-Element Thermal Modeling of Interconnects Derived from a Projection-based Leaning Algorithm

An approach combining domain decomposition and model order reduction enabled by a data-driving learning algorithm is developed for thermal simulation of interconnects. The approach accounts for variations of heat sources, boundary conditions (BCs) and material properties. This approach is applied to construct the thermal model of a generic element for a group of interconnects that are used to wire FinFET standard cells. The interconnect structure in a FinFET IC is then partitioned into several elements, each modeled by the generic element model. The developed multi-element thermal simulation of the interconnects is demonstrated and its accuracy is examined in terms of the metal/via routings and BCs.