芯粒集成逐渐成为不同场景下敏捷定制深度学习芯片的高可扩展性的解决方案,芯片设计者可以通过集成设计、验证完成的第三方芯粒来降低芯片开发周期和成本,提高芯片设计的灵活性和芯片良率.在传统的芯片设计和商业模式中,编译器等专用软...芯粒集成逐渐成为不同场景下敏捷定制深度学习芯片的高可扩展性的解决方案,芯片设计者可以通过集成设计、验证完成的第三方芯粒来降低芯片开发周期和成本,提高芯片设计的灵活性和芯片良率.在传统的芯片设计和商业模式中,编译器等专用软件工具链是芯片解决方案的组成部分,并在芯片性能和开发中发挥重要作用.然而,当使用第三方芯粒进行芯片敏捷定制时,第三方芯粒所提供的专用工具链无法预知整个芯片的资源,因此无法解决敏捷定制的深度学习芯片的任务部署问题,而为敏捷定制的芯片设计全新的工具链需要大量的时间成本,失去了芯片敏捷定制的优势.因此,提出一种面向深度学习集成芯片的可扩展框架(scalable framework for integrated deep learning chips)--Puzzle,它包含从处理任务输入到运行时管理芯片资源的完整流程,并自适应地生成高效的任务调度和资源分配方案,降低冗余访存和芯粒间通信开销.实验结果表明,该可扩展框架为深度学习集成芯片生成的任务部署方案可自适应于不同的工作负载和硬件资源配置,与现有方法相比平均降低27.5%的工作负载运行延迟.展开更多
The drastic difference in healing capacity between the anterior cruciate ligament and medial collateral ligament injuries is still largely unexplained.So,studying the molecular expression after anterior cruciate ligam...The drastic difference in healing capacity between the anterior cruciate ligament and medial collateral ligament injuries is still largely unexplained.So,studying the molecular expression after anterior cruciate ligament and medial collateral ligament injury could be a new direction to explain such differences.In this study,eighteen male 3-month-old New Zealand white rabbits were randomly divided into six groups by injury modeling to study the molecular expression after anterior cruciate ligament and medial collateral ligament injury.The expression of molecular genes and proteins was examined using reverse transcriptase chain reaction and Western blotting.Marked differences were found in the postinjury changes in gene levels and proteins in the anterior cruciate ligament compared to the medial collateral ligament.The results show that there were drastic differences in angiogenesis,collagen and matrix metalloproteinases after anterior cruciate ligament and medial collateral ligament injury.It can be found that the repair ability of the anterior cruciate ligament and medial collateral ligament after injury may be related to the differences in molecular expression.展开更多
This paper addresses the problem of test response compaction. In order to maximize compaction ratio, a single-output compactor based on a （n, n-1, m, 3） convolutional code is presented. When the proposed theorems ar...This paper addresses the problem of test response compaction. In order to maximize compaction ratio, a single-output compactor based on a （n, n-1, m, 3） convolutional code is presented. When the proposed theorems are satisfied, the compactor can avoid two and any odd erroneous bits cancellations, and handle one unknown bit （X bit）. When the X bits in response are clustered, multiple-weight check matrix design algorithm can be used to reduce the effect of massive X bits. Some extended experimental results show that the proposed encoder has an acceptable-level X tolerant capacity and low error cancellations probability.展开更多
文摘芯粒集成逐渐成为不同场景下敏捷定制深度学习芯片的高可扩展性的解决方案,芯片设计者可以通过集成设计、验证完成的第三方芯粒来降低芯片开发周期和成本,提高芯片设计的灵活性和芯片良率.在传统的芯片设计和商业模式中,编译器等专用软件工具链是芯片解决方案的组成部分,并在芯片性能和开发中发挥重要作用.然而,当使用第三方芯粒进行芯片敏捷定制时,第三方芯粒所提供的专用工具链无法预知整个芯片的资源,因此无法解决敏捷定制的深度学习芯片的任务部署问题,而为敏捷定制的芯片设计全新的工具链需要大量的时间成本,失去了芯片敏捷定制的优势.因此,提出一种面向深度学习集成芯片的可扩展框架(scalable framework for integrated deep learning chips)--Puzzle,它包含从处理任务输入到运行时管理芯片资源的完整流程,并自适应地生成高效的任务调度和资源分配方案,降低冗余访存和芯粒间通信开销.实验结果表明,该可扩展框架为深度学习集成芯片生成的任务部署方案可自适应于不同的工作负载和硬件资源配置,与现有方法相比平均降低27.5%的工作负载运行延迟.
文摘The drastic difference in healing capacity between the anterior cruciate ligament and medial collateral ligament injuries is still largely unexplained.So,studying the molecular expression after anterior cruciate ligament and medial collateral ligament injury could be a new direction to explain such differences.In this study,eighteen male 3-month-old New Zealand white rabbits were randomly divided into six groups by injury modeling to study the molecular expression after anterior cruciate ligament and medial collateral ligament injury.The expression of molecular genes and proteins was examined using reverse transcriptase chain reaction and Western blotting.Marked differences were found in the postinjury changes in gene levels and proteins in the anterior cruciate ligament compared to the medial collateral ligament.The results show that there were drastic differences in angiogenesis,collagen and matrix metalloproteinases after anterior cruciate ligament and medial collateral ligament injury.It can be found that the repair ability of the anterior cruciate ligament and medial collateral ligament after injury may be related to the differences in molecular expression.
基金supported in part by the National Basic Research Program of China(973)(Grant Nos.2005CB321604 and 2005CB321605)in part by the National Natural Science Foundation of China(Grant Nos.90207002 and 60576031).
文摘This paper addresses the problem of test response compaction. In order to maximize compaction ratio, a single-output compactor based on a （n, n-1, m, 3） convolutional code is presented. When the proposed theorems are satisfied, the compactor can avoid two and any odd erroneous bits cancellations, and handle one unknown bit （X bit）. When the X bits in response are clustered, multiple-weight check matrix design algorithm can be used to reduce the effect of massive X bits. Some extended experimental results show that the proposed encoder has an acceptable-level X tolerant capacity and low error cancellations probability.
