Abstract

Multicore platforms are becoming increasingly pervasive in modern embedded systems. System level modeling techniques have enabled creation of fast software models of multicore platforms, commonly known as Virtual Prototypes, for early functional validation of embedded software, before the hardware is available. On the other hand, for accurate performance validation, the complete multicore platform can be implemented as a physical prototype on FPGA. Both virtual platforms and FPGA prototypes have their respective pros and cons. Virtual platforms have the advantage of high speed functional simulation and, typically, scale well with the number of cores. However, the accuracy of performance estimation is sacrificed. FPGA prototypes provide cycle-accurate performance estimation, because the software executes directly on an FPGA implementation of the target cores. However, it takes a significant amount of time to design, implement and test the inter-core communication architecture on the FPGA.
In this thesis we propose to design a novel system-level modeling framework, called Hybrid Prototyping. Our goal is to provide the benefits of both virtual platforms and FPGA prototypes. It aims to provide early, fast, and scalable models, similar to virtual platforms, along with the cycle-accuracy of FPGA prototypes. Using hybrid prototyping, embedded software designers will be able to create concurrent applications and accurately analyze the performance implication of their optimizations before the chip is delivered. At the same time, multicore architects will be able to modify the platform model without having to do full system prototyping. Therefore, hybrid prototyping will enable early and reliable multicore embedded system design, resulting in huge productivity gains for both embedded software designers and multicore chip architects.