A statistical application model for fast MPSoC simulation

ERDSE: efficient reinforcement learning based design space exploration method for CNN accelerator on resource limited platform

A large body of work can be found in literature on Design Space Exploration (DSE) methods for distributed embedded system architecting (DESA). However, almost none of these methods is successfully adopted in automotive industry. To clarify the reasons, this paper 1) analyzes the current state of the art (SOTA) on DSE methods for DESA through a systematic literature study, focusing on the assumed architecting process and concerns; 2) investigates the state of practice (SOP) on DESA in the automotive industry through a literature study and interviews with experienced system architects from five different automotive manufacturers; and 3) analyzes the gap between SOTA and SOP, and thereby discusses potential improvements of DSE methods.

Daedalus is a system-level design flow for the design of multiprocessor system-on-chip (MP-SoC) based embedded multimedia systems. It offers a fully integrated tool-flow in which design space exploration (DSE), system-level synthesis, application mapping, and system prototyping of MP-SoCs are highly automated. In this paper, we describe our first industrial deployment experiences with the Daedalus framework. Daedalus is currently being deployed in the early stages of the design of an image compression system for very high resolution cameras targeting medical appliances. In this context, we performed a DSE study with a JPEG encoder application, which exploits both task and data parallelism. This application was mapped onto a range of different MP-SoC architectures. We achieved a performance speed-up of up to 20x compared to a single processor system. In addition, the results show that the Daedalus high-level MP-SoC models accurately predict the overall system performance, i.e., the performance error is around 5%.

Multi-processor system-on-chip (MPSoC) simulators are many orders of magnitude slower than the hardware they simulate due to increasing architectural complexity. In this paper, we propose a new application sampling technique to accelerate the simulation of MPSoC design space exploration (DSE). The proposed technique dynamically combines simultaneously executed phases, thus generating a sampling unit. This technique accelerates the simulation by allowing the repeated combinations of parallel phases to be skipped. A complementary technique, called cluster synthesis, is also proposed to improve the simulation acceleration when the number of possible phase combinations increases. Our experimental results show that this technique can accelerate the simulation up to a factor of 800 with a relatively small estimation error.

This paper presents an approach advocating abstract clocks to represent data-intensive applications executed on multiprocessor systems-on-chip (MPSoCs) for facilitating the exploration of large design spaces. By using abstract clocks, the advocated method characterizes applications defined by multiple loop nests, as well as, useful loop transformations that contribute to an efficient application execution. It combines the advantages of optimizations provided by loop transformations and the precision of information on scheduling captured by the abstract clocks. As a result, it favors a rapid, and yet accurate design space exploration (DSE) of data-intensive systems.

The complexity of modern embedded systems, which are increasingly based on heterogeneous multiprocessor system-on-chip (MPSoC) architectures, has led to the emergence of system-level design. To cope with this design complexity, system-level design aims at raising the abstraction level of the design process from the register-transfer level (RTL) to the so-called electronic system level (ESL). However, this opens a large gap between deployed ESL models and RTL implementations of the MPSoC under design, known as the implementation gap. Therefore, in this chapter, we present the Daedalus methodology which the main objective is to bridge this implementation gap for the design of streaming embedded MPSoCs. Daedalus does so by providing an integrated and highly automated environment for application parallelization, system-level design space exploration, and system-level hardware/software synthesis and code generation.

Multiprocessor Systems-on-Chip (MPSoCs) are a promising solution to fulfill the requirements of high performance computing applications, such as image processing or bioinformatics. Many different MPSoC architectures are introduced by industry and academics, such as Intel's Single Cloud Computer (SCC) or IBM's Cell. To achieve a high efficiency on such MPSoCs a high computing performance in the processing elements (PE) is not the only parameter. The bandwidth of the memory and especially the on-chip communication infrastructure are extremely important to achieve the requirements of the applications. Furthermore, for embedded System-on-Chip solutions, flexibility is extremely important for providing a good tradeoff between performance and power consumption. Especially, for general purpose MPSoCs, designed for a wide range of applications with different communication demands, this flexibility is essential for a good energy efficiency. A simple bus or Network-on-Chip (NoC) with fixed bandwidth and communication paradigm is not sufficient in this varying application scenario and its design space. In this paper novel features for the heterogeneous and runtime adaptive Star-Wheels Network-on-Chip are presented. This Network-on-Chip is used within a runtime adaptive MPSoC, which enables to tailor processors and accelerators to the requirements of the applications. For optimizing the Star-Wheels Network-on-Chip and its communication protocol at runtime to the application requirement the following novel features have been added: the combination of the circuit- and the packet-switched communication paradigm and an algorithm-based placement of the processing elements within the NoC. The integration of the novel algorithm into the design methodology and the exploitation of the novel features within the special purpose runtime operating system of a runtime adaptive MPSoC are also introduced.

The design of a Multiprocessor System-on-Chip (MPSoC) is a complex task, including steps as application development, platform configuration, code generation, task mapping onto the platform and debugging. An integrated environment covering most of these steps is a gap in the literature. The present work first details an MPSoC architecture, which supports the execution of distributed applications, including an operating system enabling multitask execution at each processing element. The MPSoC is heterogeneous, due to the support to different processor architectures. Then, a framework able to cover the design steps previously mentioned is presented. The framework enables the design space exploration for applications to be executed in the MPSoC, varying for example the number and type of processors, the memory size, the task mapping. Results demonstrate the correct operation for different MPSoC configurations, generated from the proposed framework. Such open-source framework enables the research community to investigate new subjects related to MPSoC and Network on Chip (NoC) design, as well as evaluate distributed applications in a multiprocessor environment.

Industrial Cyber-Physical Systems (CPS) are com-plex heterogeneous and distributed computing systems, typically integrating and interconnecting a large number of subsystems and containing a substantial number of hardware and software components. Producers of these distributed Cyber-Physical Systems (dCPS) face serious challenges with respect to designing the next generations of these machines and require proper support in making (early) design decisions to avoid expensive and time consuming oversights. This calls for efficient and scalable system-level Design Space Exploration (DSE) methods for dCPS. In this position paper, we review the current state of the art in DSE, and argue that efficient and scalable DSE technology for dCPS is more or less non-existing and constitutes a largely unchartered research area. Moreover, we identify several re-search challenges that need to be addressed and discuss possible directions for targeting such DSE technolozy for dCPS.

The design of efficient multiprocessor systems on chip (MPSoC) is a daunting challenge. This challenge is driven by the increasing demand in achieving high performance, low power, and reconfigurable applications onto the same platform. As a consequence, there is an emerging need for fast, and efficient design space exploration, while providing abstract models for a wide range of applications, and types of processing units (PUs). Therefore, this paper proposes a framework for a fast MPSoC generation with joint task and core mapping with the objective of minimizing the average power consumption. The proposed framework considers the static, dynamic, reconfiguration and communication power components. A tool for high level MPSoC task and core mapping (MPMAP) is built based on the proposed framework. MPMAP provides a flexible XML interface that provides a high level description of the different PU architectures, and different applications scenarios. Additionally, the paper presents a case study of real-life applications that can be adopted in future heterogeneous wireless systems.

This article proposes an original design flow for the fast prototyping of image processing on a MP-SoC (MultiProcessors System on Chip) architecture. Developing processors network systems tailored to a particular application domain is critical and design-time consuming in order to achieve highperformance customized solutions. The effectiveness of such approaches largely depends on the availability of an ad hoc design methodology. This paper illustrates a new design flow that enables to instantiate a generic Homogeneous Network of Communicating Processors (called HNCP) tailored for a targeted application. The HNCP is generated with a tool that avoids fastidious manual editing operations for the designer. Specific lightweight communication functions have been developed to fasten the programming of the MP-SoC network. A case study (image texture analyzes) is presented to illustrate the proposed MP-SoC design methodology and enables to focus on architecture exploration, instantiated scheme of parallelization and timing performance.

This article proposes an original methodology for the fast prototyping of image processing on a generic MP-SoC (Multi-Processors System on Chip) architecture. To define a processors network adapted to a particular application is critical and design-time consuming in order to achieve high-performance customized solutions. The effectiveness of such approaches largely depends on the availability of an ad hoc design methodology. This paper illustrates a new methodology that enables to instantiate a generic Homogeneous Network of Communicating Processors (called HNCP) tailored for a targeted application. The HNCP is generated with a tool that avoids fastidious manual editing operations for the designer. Specific lightweight communication functions have been developed to fasten the programming of the MP-SoC network. A case study (image texture analyzes) is presented to illustrate the proposed MP-SoC design methodology and enables to focus on architecture exploration, instantiated scheme of parallelization and timing performance.

High-level design space exploration for adaptive applications on multiprocessor systems-on-chip

This paper presents principles and tools to facilitate multi-processor system on chips (MPSoCs) design and modeling, and to speed up cycle accurate SystemC simulation. We describe an effective way to build an hardware architecture virtual prototype, using a library of SystemC simulation models based on communicating synchronous finite state machines. This modeling approach supports a fully static scheduling strategy, based on the analysis of the combinational dependency graph. Our static scheduling algorithm has been implemented in the SystemCASS simulator, and provides speed-up of one order of magnitude versus the standard event-driven SystemC simulation engine. The modeling approach proposed in this paper has been adopted by the SoCLIB French National Project, that is an open modeling and simulation platform for multi-processors system on chips.

Convolutional Neural Network (CNN) models for modern applications are becoming increasingly deep and complex. Thus, the number of different CNN mapping possibilities when deploying a CNN model on multiple edge devices is vast. Design Space Exploration (DSE) methods are therefore essential to find a set of optimal CNN mappings subject to one or more design requirements. In this paper, we present an efficient DSE methodology to find (near-)optimal CNN mappings for distributed inference at the edge. To deal with the vast design space of different CNN mappings, we accelerate the searching process by proposing and utilizing a multi-stage hierarchical DSE approach together with a tailored Genetic Algorithm as the underlying search engine.