Research Of Write Energy Consumption Optimization Strategy Based On MLC NVM

Stastic power consumption has become a major bottleneck in the development of CMOS memory technology with the increasing integration of integrated circuits and the shrink of process dimension.This problem is very prominent in both dynamic random access memory(DRAM)and static random access memory(SRAM).Since DRAM needs keep the data by refresh operations,the static energy consumption accounted for more than 40%of the DRAM energy consumption.For SRAM,the storage cell needs to be power continuously to save the data.Non-volatile memory(NVM),is one of the new storage technologies to solve the problem of the technical bottleneck of traditional storage technologies.Its main feature is that the data can still be stored in storage cells after power off.The mainstream nonvolatile memory technology includes Phase-Change Memory(PCM)and Spin-Transfer Torque Magnetic RAM(STT-MRAM)and etc.Most NVM technologies have the advantages of high-density,high-reliability,and nonvolatile.Since PCM has higher access delay with DRAM,it may replace DRAM as the main memory in the future.STT-MRAM has slightly lower storage delay than SRAM.In recent research,STT-MRAM has been studied as the last level on-chip cache.Multi-level cell(MLC)is that more than one bit,usually two bit,can be stored in a memory cell.Compared with Single-level cell(SLC),MLC has higher storage density.MLC NVM has almost no static energy consumption like SLC NVM,but it has a higher dynamic energy consumption.In this paper,we optimize the write energy consumption of MLC NVM.For MLC PCM and MLC STT-MRAM,the average energy consumption of flip the left bit of a 2-bit data is much higher than that of not flip the left bit,and the average energy consumption of writing states '01' and '10' is higher than that of writing states '00' and ' 1'In this paper,we modify the dynamic state remapping scheme and take the characteristic of the write energy consumption of MLC PCM and MLC STT-MRAM into consideration.We analyze the trace of memory access and improve the algorithm of getting the mapping decisions.This scheme first counts the number of each transition and gives each transition the weight of energy consumption,then calculates the difference of the energy consumption of each remapped data and the original data,and final chooses the optimal one of remapped data.Experimental results showed that our scheme reduces 6.17%write energy consumption on avergy compared with the basic write scheme for MLC PCM and reduces 12.17%write energy consumption on avergy compared with the basic write scheme for MLC STT-MRAM.