Deriving Performance Bounds for Conditional Asynchronous Circuits Using Linear Programing

This paper presents accurate performance bounds for conditional asynchronous circuits which demonstrate mode-based behavior. Analyzing the performance of these circuits is challenging as the critical paths cannot be identified without knowing the exact sequence of modes of operations which is generally unknown during design time. We used Markov chain processes to model mode switching and unique choice Petri nets to model performance. We adopt a performance analysis scheme based on decomposing the behavior of the Petri net into marked graph components to reason about performance. The bounds are derived using linear programming based on the probability matrix of the Markov chain and can be incorporated in performance-aware optimization of conditional asynchronous circuits. To evaluate the accuracy of the bounds, the theory is applied to a new set of benchmark circuits that includes randomly generated conditional circuits (ISCAS89a) and an industrially-inspired example.