FLECSim-SoC: A Flexible End-to-End Co-Design Simulation Framework for System on Chips
Tim Hotfilter, Julian Hoefer, Fabian Kres, Fabian Kempf, Juergen Becker · 2021
Hardware accelerators for deep neural networks (DNNs) have established themselves over the past decade. Most developments have worked towards higher efficiency with an individual application in mind. This highlights the strong relationship between co-designing the accelerator together with the requirements of the application. Currently for a structured design flow, however, it lacks a tool to evaluate a DNN accelerator embedded in a System on Chip (SoC) platform.To address this gap in the state of the art, we introduce FLECSim, a tool framework that enables an end-to-end simulation of an SoC with dedicated accelerators, CPUs and memories. FLECSim offers flexible configuration of the system and straightforward integration of new accelerator models in both SystemC and RTL, which allows for early design verification. During the simulation, FLECSim provides metrics of the SoC, which can be used to explore the design space. Finally, we present the capabilities of FLECSim, perform an exemplary evaluation with a systolic array-based accelerator and explore the design parameters in terms of accelerator size, power and performance.