Leakage current modeling for GPGPU systems

In the state-of-the-art VLSI designs, the importance of leakage current estimation is significantly growing. This paper presents a new approach to leakage current modeling, which is to be used for GPGPU (General-Purpose Computing on Graphics Processing Units) systems. In the traditional leakage estimation, statistical approaches are used to compromise with the huge computational complexity that the Monte-Carlo (MC) analysis requires. However, these approaches are known to be inaccurate due to the inherent modeling difficulty. In contrast, the recent development of computing technology is making the use of GPGPU widely popular where parallel processing is possible, and MC analysis is certainly one of the very promising application areas. In this paper, we investigate the modeling accuracy of two piecewise polynomial interpolation methods, i.e., piecewise linear and cubic spline, for leakage estimation in comparison with traditional statistical approaches. The experimental results illustrates that the proposed approach provides much higher accuracy than the statistical approaches, exhibiting less than 5% and 2% errors in case of piecewise-linear and cubic spline interpolation methods, respectively, when compared to HSPICE estimation results.