A behavioral modeling concept and practice of CNN-UM VLSI implementations

Abstract

In this paper we introduce a novel simulation time bounded behavioral modeling technique that optimally selects the incorporated block models. The method has been specially developed for fast performance evaluation of large mixed signal image processing arrays. The time domain accuracy is optimized under the simulation time constraint by automatic selection of various user supplied block models. A dedicated environment also has been developed for efficient numerical simulations. Utilizing the proposed methodology, a bridge has been built for the CNN-UM VLSI implementations between the device level and the high-level functionality.