Compute-Oriented FPGA device architectures and tools

现场可编程门阵列（FPGA）器件是用于构建硬件系统的新型半导体器件。然而，缓慢的编译和有限的容量极大地限制了FPGA的生产力。 对于大型设计，FPGA编译可能需要数小时甚至数天。这限制了硬件工程师的设计生产力，并使使用FPGA进行计算任务的软件程序员感到沮丧。自然有限的设备容量也使得将应用、特定计算任务映射到FPGA变得繁琐和困难。这些限制极大地阻碍了FPGA在通用计算等新市场中的应用。 这项拨款提案的目标是设计新的可编程计算平台，通过将其纳入类似FPGA的框架，超越传统CPU或GPU的传统组织。特别是，这项工作将创建新的FPGA器件架构，这些架构比传统FPGA更友好，并提供更高的器件容量和更快的编译时间。这项研究将产生一种比传统FPGA更易于使用、功耗更低、速度更快的新型FPGA，并为将软件和硬件电路编译成所开发的新型FPGA提供新的工具和方法。可重构计算已经使用FPGA来加速许多重要的研究应用，如诊断医学成像、药物发现、通信、分子模拟、天气和海洋预报、气候变化、金融建模、娱乐、安全和人工智能。这项工作的进展将加速研究之外的更广泛受众的采用，并有助于使其成为主流。

Field-programmable gate array (FPGA) devices are novel semiconductor devices used to build hardware systems. However, slow compilation and limited capacity greatly restricts productivity with FPGAs. FPGA compilation can take hours or even days for large designs. This limits the design productivity of hardware engineers, and frustrates software programmers using FPGAs for computing tasks. The naturally limited device capacity also makes it tedious and difficult to map applications, particular computing tasks, to FPGAs. These limitations greatly hinder the adoption of FPGAs into new markets such as general-purpose computing. The goals of this grant proposal are to design new programmable platforms for computing which go beyond the traditional organization of traditional CPUs or GPUs by adopting them into an FPGA-like framework. In particular, the work will create new FPGA device architectures that are more compute-friendly and offer higher device capacity and faster compile-times than traditional FPGAs. This research will result in a new type of FPGA that is easier to use, lower power and faster than traditional FPGAs, as well as offering new tools and approaches for compiling both software and hardware circuits into the novel FPGAs developed. Reconfigurable computing already uses FPGAs to accelerate many important research applications such as diagnostic medical imaging, drug discovery, communications, molecular simulations, weather and ocean forecasting, climate change, financial modeling, entertainment, security, and artificial intelligence. Advances in this work will accelerate the adoption beyond research to a more general audience and help make it mainstream.