On the Performance Differences of SystemC and SystemC AMS: A Guideline for Real Number Modeling and Beyond

Abstract:
In modern complex systems virtual prototyping with SystemC has proven to be an effective method to tackle heterogeneity. Different methodologies have arisen to introduce analog/mixed-signal models into the virtual prototypes. In this paper, we aim to analyze two common methodologies in the context of virtual sensor prototypes regarding their simulation performance: classical real number modeling in SystemC and value-continuous time-discrete simulation using timed data flow (TDF) models in SystemC AMS. The major difference in the approaches is that SystemC works fully event-based whereas SystemC AMS uses a static schedule with equidistant timesteps. We conducted experiments to separate the different sources for performance differences to allow designers to derive good estimates based on their actual system properties. Thereby, we additionally consider the dynamic features introduced in SystemC AMS 2.0 and its option to include linear ordinary differential equations.