一种SRAM辅助新型非易失性缓存的磨损均衡方法

发布时间：2018-03-03 03:13

本文选题：非易失性存储器　切入点：缓存　出处：《计算机学报》2017年03期 　论文类型：期刊论文

【摘要】：随着半导体工艺的发展,处理器集成的片上缓存越来越大,传统存储器面临着存储密度低和漏电功耗高等问题日益严峻.近年来,新型非易失性存储技术展现出漏电功耗低、存储密度高和可扩展性强等优点,是最有潜力构建大容量缓存的新技术.然而,非易失性存储器的写操作次数有限,作为缓存将限制其寿命.同时缓存上的写操作是不均匀的,存在缓存组间和组内的写波动,这一访问特点将导致缓存的每部分磨损不均衡,现有的缓存管理策略却不能感知缓存的写波动.为解决这一问题,提出了SRAM辅助新型非易失性缓存的磨损均衡(SRAM-assistEd weAr Leveling,SEAL)方法,该方法包含写波动感知的缓存块迁移算法(Write Variation-aware blOck Migration,WVOM)和阈值指导的缓存块迁移算法(Threshold gUided Block mIgration,TUBI).SEAL方法重点关注写波动大的缓存组和写强度高的缓存单元,着力减少这部分缓存单元的写压力.WVOM感知缓存组间写波动并迁移写强度大的缓存组,用于解决组间磨损均衡问题.TUBI迁移缓存组内写局部性高的缓存块来达到组内磨损均衡.实验结果表明,SEAL方法与基准配置相比,缓存的磨损程度平均减少了34.2%,缓存的平均寿命提升了175%,性能平均提高了0.735%,系统的动态功耗平均降低了5.65%.同时,SEAL方法与最新的研究成果相比,缓存的磨损程度平均减少了13.1%,缓存的平均寿命提升了20%.
[Abstract]:With the development of semiconductor technology, the on-chip buffer of processor integration becomes larger and larger. Traditional memory is faced with the problems of low storage density and high leakage power consumption. In recent years, new non-volatile memory technology has shown low leakage power consumption. The advantages of high storage density and high scalability are the most potential new technologies for building large capacity cache. However, the number of write operations of non-volatile memory is limited, and the lifetime of non-volatile memory is limited as a cache. At the same time, the write operation on the cache is not uniform. There are write fluctuations between and within cache groups, which will cause uneven wear and tear in every part of the cache. However, the existing cache management strategies are not aware of write fluctuations in the cache. A new wear equalization method, SRAM-assistEd weAr LevelingSEAL, is proposed for SRAM assisted non-volatile buffer. The method consists of write Variation-aware blOck migration algorithm (write Variation-aware blOck migration) and threshold guided buffer block migration algorithm (threshold). The method focuses on writing volatile cache groups and high-strength buffer units. Focus on reducing the write pressure of this part of the cache unit. WVOM perceives write fluctuations between cache groups and migrates the cache groups with high write strength, It is used to solve the problem of inter-group wear equalization. TUBI migrates the cache blocks with high write locality in order to achieve intra-group wear equalization. The experimental results show that the proposed method is compared with the reference configuration. The average wear degree of the cache has been reduced by 34.2 percent, the average lifetime of the cache has increased by 175, the average performance has increased by 0.7355.The dynamic power consumption of the system has been reduced by an average of 5.65. at the same time, the SEAL method is compared with the latest research results. The average wear of the cache has been reduced by 13. 1 and the average lifetime of the cache has increased by 20.
【作者单位】：武汉大学计算机学院;武汉大学软件工程国家重点实验室;合肥工业大学计算机与信息学院;
【基金】：国家自然科学基金(91118003,61170022,61373039,61402145,61502346,61640220,61662002) 湖北省自然科学青年基金(2015CFB338) 安徽省自然科学青年基金(1508085QF138) 江西省教育厅科技项目(GJJ150605)资助
【分类号】：TP333

【相似文献】