SHF:Small:Design Validation Using Multiple Concurrent Abstract Models and GPGPUs

SHF:Small：使用多个并发抽象模型和 GPGPU 进行设计验证

• 批准号: 1422054
• 负责人: Michael Hsiao
• 金额: $ 41.83万
• 依托单位: Virginia Polytechnic Institute and State University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2014
• 资助国家: 美国
• 起止时间: 2014-07-01 至 2018-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1422054&HistoricalAwards=false
• 关键词: SHF; Small; Design; Validation; Using

A critical bottleneck in the design of large, complex hardware systems is verification and validation, whose goal is to check if the implementation meets the specifications. An ill-verified design compromises not only reliability, but also security of the hardware. Therefore, having a sound and effective verification framework plays a critical role in designing today's large scale circuits. However, the exponential growth in circuit complexity over the past few decades has made verification and validation an extremely daunting task. Hence, in many projects the level of effort needed to verify the correctness of systems often far exceeds efforts spent on design. Thus, much-needed verification and validation breakthroughs hold the key to ease this mounting challenge. The objectives of this project is to address this need via four coherent tasks: multiple abstractions for extracting various core functional behaviors and diversifying search perspectives; swarm-aggregate learning of branching behavior and necessary loop repetitions; combined particle swarm optimization with ACO in generating long sequences; and GPGPUs for enhancing performance and scalability. Together, these tasks elicit the collective power of diverse perspectives, thus aiming to advance the knowledge of verification. The proposed approach is flexible and is not restricted by the inherent depth limitation imposed by deterministic methods. Furthermore, with swarm-aggregate learning applied to GPGPUs, the computational cost can be significantly reduced. It is expected that the synergy from simulation, swarm intelligence, multiple concurrent abstractions, and GPGPUs will bring out the best from each domain to achieve a common goal.Making significant strides in the field of verification and validation will not only reduce time to market products, but also will increase national competitiveness both from the technical and economic standpoint using design optimizations previously deemed unattainable. Through the Multicultural Academic Opportunities Program (MAOP) at Virginia Tech the PI will advise a diverse team of graduate students, including both women and minority students under-represented in engineering. The project also plans continued expansion of downloadable resources (tools, benchmarks, etc.), benefiting both industry and academia.

大型复杂硬件系统设计中的一个关键瓶颈是验证和确认，其目标是检查实现是否符合规范。未经验证的设计不仅会损害硬件的可靠性，还会损害硬件的安全性。因此，一个完善有效的验证框架在当今大规模电路的设计中起着至关重要的作用。然而，在过去的几十年里，电路复杂性的指数级增长使得验证和确认成为一项极其艰巨的任务。因此，在许多项目中，验证系统正确性所需的工作量往往远远超过设计所花费的工作量。因此，亟需的核查和验证突破是缓解这一日益严峻的挑战的关键。这个项目的目标是通过四个连贯的任务来满足这一需求：用于提取各种核心功能行为和多样化搜索视角的多个抽象；分支行为和必要循环重复的群聚合学习；在生成长序列时将粒子群优化与蚁群算法相结合；以及用于增强性能和可扩展性的GPGPU。这些任务共同激发了不同观点的集体力量，从而旨在促进核查知识的发展。该方法是灵活的，不受确定性方法固有的深度限制的限制。此外，将群体聚集学习应用于GPGPU，可以显著降低计算代价。预计模拟、群体智能、多个并发抽象和GPGPU的协同作用将使每个领域发挥最大作用，实现共同的目标。在验证和确认领域取得重大进展不仅将缩短产品上市时间，而且将利用以前被认为无法实现的设计优化，从技术和经济角度提高国家竞争力。通过弗吉尼亚理工大学的多元文化学术机会计划(MAOP)，PI将为不同的研究生团队提供建议，包括女性和在工程学中代表性不足的少数族裔学生。该项目还计划继续扩大可下载资源(工具、基准等)，使工业界和学术界都受益。