Research On Efficient Cache Management Methods For Non-volatile Processors

In recent years,with the empowerment of wireless communication technology,new embedded devices such as smart wearables and implantables have emerged,which have attracted widespread attention in academia and industry.New embedded devices can obtain environmental energy for energy supply through energy harvesting technology.To address intermittent execution issues caused by unstable ambient energy,new embedded devices employ non-volatile processors(NVP)to architect non-volatile memory(NVM)to back up volatile data in the event of a power outage.However,NVP and surrounding devices need to exchange information,and unreasonable data backup will increase a lot of I/O operations and rollback overhead,affect the overall performance of the system,and increase energy consumption.Therefore,this thesis designs an efficient cache data management algorithm based on a performance-oriented self-powered NVP system.The main research contents and contributions of this thesis are as follows:(1)This thesis adopts a hybrid cache based on volatile memory SRAM and non-volatile memory STT-RAM.In order to make full use of the storage density of STT-RAM,a multi-level cell(MLC)mode STT-RAM is constructed in the hybrid cache.STT-RAM is logically partitioned to reduce its write cost.The cache model is designed by analyzing the access cost of soft and hard blocks formed after MLC STT-RAM logical partitioning.(2)This thesis proposes a cache data state mechanism based on retention time.This mechanism is used to identify the data state of SRAM and soft blocks in the cache.Guides the design of cache data migration replacement and backup policies by analyzing the behavior of cached data as reflected by data state mechanisms and taking advantage of the principle of locality of cache accesses.(3)This thesis proposes efficient cache management methods,including a performanceoriented cache management method when the energy is sufficient,and a pre-backup and backup method when the energy is insufficient.Two energy thresholds are designed and analyzed to distinguish between cache management schemes performed at different stages.Secondly,in the performance-oriented cache management method,the system performance is guaranteed by designing a reasonable data migration and replacement strategy.Finally,in the pre-backup and backup method,the retention state method is adopted to reduce the energy consumption caused by frequent backup and restore operations.The experimental results show that,compared with the existing schemes,the scheme proposed in this thesis can effectively improve the overall performance of the system and reduce energy consumption.