Grant: Strategic Projects - Group

Modern society is increasingly driven and improved by advancements in digital systems. Field-programmable gate arrays (FPGAs) provide access to state-of-the-art IC fabrication technology to smaller companies (like many of those in Canada) who could not otherwise afford the risk, cost, or time to develop their own custom-fabricated designs. FPGAs are programmable, and can be programmed to implement any digital circuit within days or weeks, compared with months or years for custom-fabricated hardware. However, FPGAs present two challenges to designers. First, they are difficult to program compared to programming a general-purpose processor with software: FPGAs are normally programmed using hardware description languages (HDLs, such as Verilog or VHDL), requiring the expertise of a hardware designer to design and debug the implementation. Second, as FPGA transistor density has grown, so has the time required to compile and synthesize HDL which can now take a whole day for a large design (even if only a small part of the design changed), compared to a few minutes for compiling software. To overcome this drawback, Profs. Lemieux and Steffan have recently proposed soft vector processors (SVPs). Vector processing is sufficiently general-purpose that it can be used for a variety of embedded and multimedia-oriented tasks. The applicants have demonstrated significant speedups over scalar soft processors, closing the performance gap between soft processors and hand-crafted HDL from 1000x down to 20x. Through the research described in this proposal, soft vector processors for FPGAs will enable new tools, products, and potentially lead to new jobs and revenue for Canadian industry. By the end of the project, we will have developed a complete environment - a manycore SVP generator, compiler system, and simulation environment - that works on real hardware (both Altera Stratix and Xilinx Virtex platforms). Due to the intrinsic commercial value, we will investigate technology transfer to industry and provide it free to researchers for non-commercial purposes.