Evaluating, benchmarking and classifying embedded systems are challenging tasks. By definition, an embedded system is application specific. Furthermore, an embedded system is designed to fit within certain constraints dictated by their target application areas. Aspects such as power consumption, real time awareness, reliability, accuracy of the computation, cost, device and physical size are all metrics under which an embedded system can be classified and that have a direct influence on the overall system performance. All these contrasting metrics define how well an embedded system can address its target application.
An additional challenge derives from the fact that modern embedded systems are composed of a multitude of heterogeneous processing elements and components. They include general-purpose processors, digital signal processors, graphic processors, application specific accelerators, reconfigurable logic and more. These components can be either off-the-shelf or custom designed. Embedded systems often present disparate types of memory hierarchies. Also the software toolchain, which includes compilers, synthesizers and optimization tools, may have significant impacts on the metrics.
This workshop is related to a new DARPA initiative in this arena called SEAK (Suite of Embedded Applications and Kernels), whose goal is to define a new, open suite of benchmarks, together with a novel methodology to evaluate in terms of performance and power end-to-end embedded systems for DOD’s application areas.
The workshop aims to involve, from the very early stages of SEAK, the benchmarking, simulation and modeling community to devise novel systematic approaches to evaluate and classify embedded systems, with a particular focus on the power/performance tradeoffs. Its objective is to provide a forum for discussing methodologies and approaches to evaluate embedded systems, including appropriate metrics, benchmarks applications, microbenchmarks, power/performance instrumentation.
Areas of particular interest for this assembly are research topics that identify and analyze novel ideas rather than providing incremental advances on the following themes:
- Benchmark and micro-benchmark creation, analysis, and evaluation issues
- Characterization of relevant workloads for embedded systems - Modeling of application and system behavior
- Tools for analyzing power and energy with different granularities and scope from hardware (e.g. component, core, system) or software views (e.g. threads, tasks, processes, etc) or both (end-to-end systems)
- Definition, identification of metrics that determine the suitability of an embedded system for its target applications, and their collection methodologies
- Characterization of current state-of-the art embedded systems in terms of power and performance
This full day workshop will involve invited talks, a session with selected posters and a closing panel. Invited talks include a keynote from DARPA, stakeholders of SEAK and teams responsible of other benchmarking initiatives. A session will introduce of a proposed methodology and benchmark suite from the SEAK team at PNNL.
The workshop will host a session of rapid-fire talks from embedded system vendors and design automation companies. Posters will be chosen through a peer-reviewed selection of short position papers from academia and industry on the topics of evaluation, characterization, benchmarking and modeling of embedded systems or their components for power and performance.
The panel will involve key people from the previous sections, ideally closing the day-long discussions.
DAC is the premier conference devoted to the design and automation of electronic systems (EDA), embedded systems and software (ESS), and intellectual property (IP).
DAC 2017 will be held in Austin, Texas, at the Austin Convention Center. Get details about travel, hotels, and area attractions in one convenient spot.