This project addresses curriculum development in the field of embedded computing which requires expertise from a number of disciplines: digital circuit design, VLSI, real time systems, applications, algorithms and programming. Of the two major classes of embedded systems, one consists of extremely large volumes of simple embedded systems (toasters, doorknobs, toys, controllers, etc.), and typically does not require high performance. The devices are inexpensive, mature, and stable, cross-compilation environments exist and so products can be designed and marketed fairly reliably. This project, HiPHiPECS, focuses on a related, more challenging and growing segment of the market (telephones, digital cameras, automobile sensor data processing, satellite control, information appliances, network routers, photocopiers, printers, scanners, fax machines, etc.), that require extremely high performance. Developing these high performance embedded applications necessitates the exploitation of parallelism, imposes increasingly severe constraints on bandwidth, power/energy consumption, computation time, and throughput. For such systems, there is an increasing trend towards Systems-on-a-Chip (SoC) where multiple implementation fabrics (re-configurable logic, processor cores, DSP 's and ASIC 's) are combined on a single heterogeneous chip. According to the International Technology Roadmap for Semiconductors (ITRS), there is an exponential "design gap" (the roadmap identifies this as the gap between the ability of circuit designers to effectively use silicon resources and the number of transistors available on the chip) in this area. This design gap necessitates raising the level of abstraction involved in digital circuit/system design. Although this has been identified as a need, few efforts are directed toward addressing the associated educational and training needs in a concerted manner. There is a critical need to train circuit/system designers in these skills that encompass such areas as: Expertise in the application domain (signal/image processing, networking, control systems, numerical analysis, algorithms), System engineering and design (software engineering, modularity, stepwise refinement, formal methods for correctness), Programming skills, data structures, algorithm analysis, Parallel processing (parallel algorithm theory, parallelism detection, parallel programming languages, multi-threading, data-parallelism), Operating systems, especially real-time OS, Computer architecture, instruction set architectures, DSP 's, instruction level parallelism, and SIMD architectures, Circuit level design, performance analysis, precise modeling of power/energy consumption, throughput, area and time. The main contributions of HiPHiPECS will be the development of a curriculum for high performance embedded systems. It will target FPGA based re-configurable computing platforms, since their re-configurability allows for the exploration of other implementation fabrics in a pedagogic setting. The proposed curriculum will focus on the computation-and data-intensive parts of embedded applications, primarily because this is where high performance solutions are critical, with the goal of enhancing instruction in the area of High Level Programming for High Performance Embedded Computing Systems. The project, which involves industrial collaboration, addresses a critical national need, it is founded on current research in High Performance Computing, and it represents a major shift (as opposed to an incremental change) in education in this area.
