The future of computing lies in achieving increasingly higher performance with greater energy efficiency and a compact form. This will power the revolution in intelligent edge-computing applications like self-driving cars and autonomous drones. To do this, software must integrate specialised hardware that dramatically speeds up computations and enhances reliability while reducing memory requirements and power consumption. Until now, it has been quite a challenge. All this is about to change, thanks to the pioneering full-system simulation tool SLX, which considers hardware and software simultaneously. EU funding of the FuSy project supported significant enhancements, optimisation and validation.
Streamlined implementation of accelerated computing
Accelerated computing relies on massive parallel processing and ‘accelerators’ that are highly efficient for certain types of operations and algorithms. The latter include graphics processing units (GPUs) and artificial intelligence accelerators. Offloading common tasks onto these rather than using the central processing unit (CPU) increases computational speed and enhances energy efficiency. In addition, adaptable hardware like field-programmable gate arrays (FPGAs) and adaptive compute acceleration platforms (ACAPs) open the door to programmable hardware accelerators and customised processing. Until now, orchestrating such a symphony of heterogeneous devices has relied on laborious manual coding and off-the-shelf ‘hard-wired’ simulators. This challenging and time-consuming task required high-level hardware expertise. According to Raphael Bruns of Xilinx (formerly of Silexica) and FuSy project coordinator, the FuSy project set out to support the ‘democratisation’ of accelerated computing, enabling software developers to efficiently programme and effectively use these heterogeneous devices.
Automation accelerates application development
The FuSy project supported extension of Silexica’s performance simulation tool to include design space exploration on target devices, user guidance and automation of the software analysis, optimisation and integration processes. What would take months using current state-of-the-art methods can now be accomplished in just a few hours. Bruns provides concrete proof: “The Silexica solution outperformed the hand-optimised algorithm designed for post-quantum cryptography at the National Institute of Standards and Technology in the United States by 43 % and reduced 3 months of manual efforts to a couple of days.” Further, the design space exploration revealed solutions with low latency but high hardware consumption on the targeted FPGA. The solution allows users to define the degree to which latency is prioritised versus area consumption.
Blazing a trail as part of a new team
The company and its full-system simulation tool were a perfect fit for Xilinx, the pioneering silicon vendor company in the US that invented the FPGA, the programmable System-on-Chip, and the ACAP. Silexica is now a subsidiary whose software solutions support the programming of Xilinx’s hardware devices. Still based in Cologne, Germany, the team is expanding rapidly, acquiring talent that will drive its evolution to a centre of excellence for compilers, machine learning, and artificial intelligence. “We are accelerating our democratisation efforts through ever higher levels of abstraction and automation. The world is about to experience a massive revolution of intelligent technologies that will transform our society, our economy, and our industry. Silexica and Xilinx intend to power this revolution with solutions for next-generation high-performance and adaptive computing platforms,” Bruns concludes.
FuSy, computing, simulation, accelerator, accelerated computing, FPGA, Silexica, software, hardware, ACAP, Xilinx