Using the application modeling and mapping methodology for system-level performance analysis

René van den Berg, Walter Tibboel, Rob Wieringa, & Martin Klompstra - NXP Semiconductors

September 26, 2010

René van den Berg, Walter Tibboel, Rob Wieringa, & Martin Klompstra - NXP Semiconductors

This article describes our experiences using the Application Modeling and Mapping methodology (AMM) based on commercial tooling from Synopsys. This methodology is valuable at the technical and organizational level for investigating the feasibility of new electronic products.

Technically, the methodology reduces the risk by giving architects a clear understanding of the application and features in an early stage of the project. This is related to system performance, hardware and software allocation on available resources, software scheduling scenarios and architecture dimensions and decisions (what-if scenarios).  

On the organizational level, the methodology facilitates the early collaboration of system architects, software developers and hardware designers based on an executable specification of the product. Within this article the AMM methodology is discussed and applied to a dual DAB reception application. For the different aspects as described above, the benefits and disadvantage are shown and discussed.  

The continuous integration of more and more applications into a single device imposes huge challenges on the design of electronic products. Given the enormous cost of an SoC design project, the system architect has a very difficult task to define the right system architecture that supports all the desired application use-cases in an optimal way:

• Over-designing the system leads to excessive cost and non-competitive products;
• Under-designing causes expensive chip re-spins until performance requirements are met. Therefore, there is a high risk associated with the dimensioning of the SoC architecture. 

Unfortunately the performance analysis of such complex systems is not trivial because, at execution time, multiple application use-cases utilize different resources of the SoC architecture. In addition there are multiple levels of resource sharing in the different components of the architecture:

• Multiple tasks compete for programmable processing elements;
• Multiple components compete for access to the interconnect;
• Multiple ports of a memory controller compete for access to the shared memory.

According to our experience, the usage of spread-sheets is no longer appropriate to analyze the performance of a dynamic application workload on a resource-shared architecture. To cope with the increasing complexity, we have developed a simulation based performance analysis methodology that allows the systematic exploration of application use-cases and SoC architectures. Our goal is to more accurately estimate the performance of the final product and significantly reduce the risk of over- or under-design.  

In this article we present the results of a product analysis, where the methodology has been applied to a feasibility analysis in the area of Digital Audio Broadcasting (DAB). The goal of this analysis was to evaluate if an existing SoC architecture can support two concurrent DAB radio streams. In the following sections we will first introduce the performance analysis methodology as well as the application domain and the reference architecture. After that we discuss the results and conclusions of the feasibility analysis.