Efficient full-wave broadband modelling of interconnects with graphic processors

The paper deals with the problem of the efficient full-wave modeling of full 3D interconnects, in a broadband range (from DC to tens of GHz). The full-wave electromagnetic equations are recast in terms of an EFIE-like surface integral formulation. The adopted numerical model is the SURFCODE, derived from the above formulation by using a suitable null-pinv decomposition of the basis functions describing the currents. In order to lower the computational of this scheme, the assembly of the most cumbersome matrix (the inductance matrix) is performed by means of a parallel computation approach based on GPU (Graphics Processor Units) architecture. A case-study is analyzed, referring to a differential pair of nets in a package structure. The achieved speedup to conventional serial approach is 54x.