Statistical leakage analysis by parallel Monte-Carlo programming on a CUDA platform

The Monte-Carlo simulation based leakage current analysis provides very accurate results, but it has high computational complexity. In this paper, we present the Monte-Carlo based leakage estimation method, which is implemented on a GPU and a CUDA platform. Thereby using the Monte-Carlo method on a GPU, we can expect not only high simulation accuracy but also high estimation efficiency. Because a GPU contains hundreds of computational cores, it can perform the Monte-Carlo simulation for leakage estimation efficiently if we parallelize the algorithm. When the GPU executes the program in parallel, a large number of threads are generated. Each sample of the Monte-Carlo simulation is assigned to each thread and then all threads are executed at the same time. Because the proposed method is based on the Monte-Carlo simulation, any leakage model and any distribution of process parameter variation can be used. In the experiment, for the simplicity of simulation, the 1^st order polynomial model is used. The simulation results imply that our approach is faster than the conventional Monte-Carlo simulation by about 130 times.