Data Allocation Optimization For Resistive Morphable Memory In Embedded Systems

In recent years,with the development of communication network technology and intelligent technology,real-time embedded equipment has become an indispensable element in economic society.Dynamic random access memory(DRAM),often used as the main memory of traditional real-time embedded system,has become a key problem for the further development of real-time embedded system due to its disadvantages such as high static leakage power consumption,lack of reliability and easy data loss due to power failure.New non-volatile memory(NVM),represented by variable resistor random access memory(RERAM),has become the key storage technology to replace DRAM because of its advantages such as non-volatile,low leakage power consumption,high reliability and close to DRAM read-write performance.However,it is worth noting that Reram also has its disadvantages.On the one hand,ReRAM has the disadvantages of asymmetric read and write performance and limited write durability.On the other hand,RERAM storage unit can be dynamically converted between SLC/MLC.SLC mode has high performance and high durability,while MLC mode has large storage capacity and low durability.Therefore,the performance of different storage modes varies greatly.In order to promote the wide application of RERAM in embedded system,this thesis focuses on the data allocation optimization problem based on the storage medium,so as to make full use of the performance advantage of the memory and avoid its defects,so as to optimize the embedded system delay and energy consumption.The main research contents are as follows:1.In the real-time embedded system,based on the variable ReRAM main memory architecture,a data allocation strategy aiming at optimizing the time delay is proposed.Based on the idea of dynamic data allocation,considering the difference between the read-write performance and the storage density of the variable ReRAM SLC/MLC,and considering the fast dynamic conversion characteristics of SLC/MLC,the data allocation problem was constructed into an integer linear programming model by mathematical modeling method,and an approximate algorithm was proposed based on this,so as to optimize the system performance.2.In the real-time embedded system,based on the variable ReRAM main memory architecture,a data allocation strategy with wear balance is proposed on the basis of the above research.Because the above study did not take wear equilibrium into account,ReRAM is easy to be broken.Based on this,the unit wear degree and average wear degree are introduced into the data distribution method,and wear equalization optimization is carried out for the distribution algorithm studied above,so as to prolong the write life of non-volatile ReRAM.3.The corresponding simulation experimental platform is built to carry out relevant experiments.The experimental results confirm that the above technology can effectively improve the system performance,reduce system energy consumption and extend the life of RERAM memory.To sum up,this thesis makes full use of the application characteristics of real-time embedded system and carries out research on data allocation optimization based on variable Reram memory,in order to overcome the application problems of variable NVM in real-time embedded system and further promote the application of embedded devices.