Power is the major challenge threatening the progress of very large scale integration （VLSI） technology development. In ultra-deep submicron VLSI designs, clock network size must be minimized to reduce power consump...Power is the major challenge threatening the progress of very large scale integration （VLSI） technology development. In ultra-deep submicron VLSI designs, clock network size must be minimized to reduce power consumption, power supply noise, and the number of clock buffers which are vulnerable to process variations. Traditional design methodologies usually let the clock router independently undertake the clock network minimization. Since clock routing is based on register locations, register placement actually strongly influences the clock network size. This paper describes a clock network design methodology that optimizes register placement. For a given cell placement result, incremental modifications are performed based on the clock skew specifications by moving registers toward preferred locations that may reduce the clock network size. At the same time, the side-effects to logic cell placement, such as signal net wirelength and critical path delay, are controlled. Test results on benchmark circuits show that the methodology can considerably reduce clock network size with limited impact on signal net wirelength and critical path delay.展开更多
对多核片上系统(MPSoC)而言,随着集成度和性能的提升,时钟网络的结构愈发重要。研究了基于结构建模的多路全局/局域时钟网络的结构建模与分析。通过建立多级级联,分别从主干、支干和接入三层对时钟网络的结构进行建模。针对运算单元接...对多核片上系统(MPSoC)而言,随着集成度和性能的提升,时钟网络的结构愈发重要。研究了基于结构建模的多路全局/局域时钟网络的结构建模与分析。通过建立多级级联,分别从主干、支干和接入三层对时钟网络的结构进行建模。针对运算单元接入数、单行中肋排数目、运算单元中输入时钟数目以及时钟区域数等几方面,评估了时钟网络性能。以Stratix V E FPGA为例对时钟网络综合分析,分析结果表明,四象限的对称结构权衡了多项性能指标,是最优的时钟网络结构,可以作为一种通用结构应用在目前主流MPSoC上。展开更多
基金the National Natural Science Foundation of China (No. 60776026)
文摘Power is the major challenge threatening the progress of very large scale integration （VLSI） technology development. In ultra-deep submicron VLSI designs, clock network size must be minimized to reduce power consumption, power supply noise, and the number of clock buffers which are vulnerable to process variations. Traditional design methodologies usually let the clock router independently undertake the clock network minimization. Since clock routing is based on register locations, register placement actually strongly influences the clock network size. This paper describes a clock network design methodology that optimizes register placement. For a given cell placement result, incremental modifications are performed based on the clock skew specifications by moving registers toward preferred locations that may reduce the clock network size. At the same time, the side-effects to logic cell placement, such as signal net wirelength and critical path delay, are controlled. Test results on benchmark circuits show that the methodology can considerably reduce clock network size with limited impact on signal net wirelength and critical path delay.
文摘对多核片上系统(MPSoC)而言,随着集成度和性能的提升,时钟网络的结构愈发重要。研究了基于结构建模的多路全局/局域时钟网络的结构建模与分析。通过建立多级级联,分别从主干、支干和接入三层对时钟网络的结构进行建模。针对运算单元接入数、单行中肋排数目、运算单元中输入时钟数目以及时钟区域数等几方面,评估了时钟网络性能。以Stratix V E FPGA为例对时钟网络综合分析,分析结果表明,四象限的对称结构权衡了多项性能指标,是最优的时钟网络结构,可以作为一种通用结构应用在目前主流MPSoC上。
