Statistical Analysis of Interconnect-Limited Systems

互连受限系统的统计分析

• 批准号: 9414187
• 负责人: Phillip Christie
• 金额: $ 19.37万
• 依托单位: University of Delaware
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1995
• 资助国家: 美国
• 起止时间: 1995-03-01 至 1999-02-28
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9414187&HistoricalAwards=false
• 关键词: Statistical; Analysis; Interconnect; Limited; Systems

The growth in size and complexity of modern computers is matched by the need for more accurate models of their internal wiring structure. The task of modeling this complexity is ideally suited to a recent development in statistical physics called the re- normalization group. It has the power to attack the wiring optimization problem because it takes this one very difficult problem and breaks it up into a very large number of much simpler problems. Rather than attempting to attack the entire range of partitioning and placement requirements from the transistor level to the back-plane, an optimization procedure need only be found for a highly restricted wire length range. Once this has been derived, the procedure is renormalized for all other length scales in a manner which generates a system-wide solution. This project will investigate how such an approach may be used to develop much better estimates of interconnect-limited system performance and how this theory leads to new optimization algorithms. In contrast to other theories the predictions made by this technique will grow ever more accurate as the systems grow larger and more complex.

现代计算机的大小和复杂性的增长与更准确的内部布线结构的模型相匹配。 对这种复杂性进行建模的任务非常适合统计物理学的最新发展，称为归一化组。 它具有攻击布线优化问题的能力，因为它将这个非常困难的问题带入了很多简单的问题。 与其尝试攻击从晶体管级别到背面的整个分区和放置要求的范围，不需要在高度限制的电线长度范围内找到优化过程。 一旦得出了这一点，该过程将以生成全系统范围的方式对所有其他长度尺度重新归一致。 该项目将如何使用这种方法来开发更好的互连系统性能估计值，以及该理论如何导致新的优化算法。 与其他理论相反，随着系统的增长更大，更复杂，该技术的预测将变得更加准确。