Designing Parameterizable Hardware IPs in a Model-Based Design Environment for High-Level Synthesis

Abstract

Model-based hardware design allows one to map a single model to multiple hardware and/or software architectures, essentially eliminating one of the major limitations of manual coding in C or RTL. Model-based design for hardware implementation has traditionally offered a limited set of microarchitectures, which are typically suitable only for some application scenarios. In this article we illustrate how digital signal processing (DSP) algorithms can be modeled as flexible intellectual property blocks to be used within the popular Simulink model-based design environment. These blocks are written in C and are designed for both functional simulation and hardware implementation, including architectural design space exploration and hardware implementation through high-level synthesis. A key advantage of our modeling approach is that the very same bit-accurate model is used for simulation and high-level synthesis. To prove the feasibility of our proposed approach, we modeled a fast Fourier transform (FFT) algorithm and synthesized it for different DSP applications with very different performance and cost requirements. We also implemented a high-level-synthesis (HLS) intellectual property (IP) generator that can generate flexible FFT HLS-IP blocks that can be mapped to multiple micro-/macroarchitectures, to enable design space exploration as well as being used for functional simulation in the Simulink environment.