Research on Mapping method for Electrical Parameters to Model Parameters in IC Aging Simulation Flow

With the emerging of semiconductor technology, the typical analytical method does not have the ability to accurately map parameters in the typical aging simulation process, like OMI/TMI, because many non-ideal effects occurred. In this paper, we proposed two multi-task machine learning architectures to map parameters with promising accuracy and speed. One of the architectures is the parallel network, which can handle the interactions between multiple tasks. In order to further digger task correlation and improve accuracy, the cascade network utilizes ensemble learning is proposed. Evaluated on the data generated by the aging simulator, both of the architectures can achieve errors of less than 0.5%.