Research On Energy Consumption Optimization Algorithms For Phase Change Memory

Phase Change Memory(PCM),a type of non-volatile random access memory on the rise,is a promising candidate of universal memory since that it has many advantages such as high access speed,non-volatility,high density,simple technology and bit-alterability.However,high write energy consumption is one of the main challenges for its adoption in memory hierarchy.Different from flash memory,there is no erasure step in the write operation of PCM due to its bit-alterability,so that when an input bit is the same as the corresponding target bit,this write operation will be skipped.Therefore,one of the effective methods to address this problem is to reduce write activities on PCM.In this paper,we target at PCM-based secondary storage and present a partition-based write mechanism to increase the flexibility of write operation.In this write mechanism,both of input data block to write and target data block in PCM are partitioned into data sub-blocks in the same manner,so that these input data sub-blocks can be flexibly written to any one of target data sub-blocks.Targeting at secondary storage write mechanism,we propose a basic block write algorithm to reduce write activities on PCM.In this basic write algorithm,if the match between input data block and the corresponding target data block is less than 50%,then this input data block will be flipped before written.Moreover,based on partition-based write mechanism,we propose two kinds of write algorithms called BMW-Greedy and BMW-KM,respectively.Based on greedy algorithm,BMW-Greedy obtains an approximating optimum result by allocating each input sub-block to an optimal available target sub-block.Based on kuhn-munkras algorithm,BMW-KM obtains an optimum result by finding a perfect matching between input sub-blocks and target sub-blocks.Moreover,both of the above two algorithms adopt the basic block write algorithm to write each input data sub-block into the corresponding target data sub-block.Experimental results show that,the proposed BMW-Greedy algorithm achieves 11.29%average energy saving(up to 18.63%)compared to the DCW algorithm and 13.30%average energy saving(up to 20.71%)compared to the Flip-N-Write algorithm;the proposed BMW-KM algorithm achieves 12.80%average energy saving(up to 20.86%)compared to the DCW algorithm and 14.81%average energy saving(up to 22.93%)compared to the Flip-N-Write algorithm.