Studies On CRS Crossbar Based Single-Instruction Multiple-Data Stream Computing Architectures

Modern computer are facing the great challenge of speed limitation on processing named "storage wall" due to the synchronization of development between memory technology and processor technology.One of most representatively traditional way to solve the storage wall is Processing In Memory(PIM).The main idea of the PIM is to integrate the memory with multiple microprocessors in the same chip,which not only would improve the reading/writing speed of memory,but also improve processing efficiency by multiple microprocessor parallel processing.However the application of PIM is subject to the chip area,power consumption and other factors.Today,the emergence of the memristor brings new development direction for PIM technology.Memristor has advantages such as small size,low power consumption,and it can be used for storage and computing.As more memristor based storage unit can be placed on chip in unit area,this new type of storage technology bring PIM technology probability of greater scale of data processing and parallelism of data processing.Based on this,this thesis puts forward a new SIMD architecture which uses complementary resistive switch(CRS)crossbar as a foundational part.Compared to conventional design whose microprocessor and memory are separated,the innovation point of this architecture is to put forward a kind of crossbar structure which is used as memory and processor at the same time.It can not only save the area of the microprocessor,power consumption,but also can save the time consumption by eliminating information interaction between microprocessor and memory,getting over the wall storage.In order to verify the effectiveness of the proposed structure,this thesis also designs the corresponding algorithm of solving Boolean linear equations.In the field of engineering,solving Boolean linear equations has been widely used.Based on the storage and logic operation characteristics of CRS crossbar unit,this thesis designed solving method of the Boolean linear equations of Gauss-Jordan.elimination method.This thesis verified the arithmetic logical correctness in the process of the algorithm by PSpice software simulation and calculated the communication time of system by gpnocsim software simulation.Combined with statistical algorithm steps,we counted the overall execution time of processing of this algorithm.Compared with the time consumption of solving linear equations by traditional computer,the proposed structure has great advantage in computing speed.The time consumption of our architecture is about three orders of magnitude less than traditional architecture,Showing a great advantage in computing speed.More deeply understanding about memristor characteristic will greatly change the PIM structure design method,providing a new direction and train of thought for designing and manufacturing PIM architecture which has higher integration,lower power consumption and the stronger computing function,which would provide effective support for promoting the development of process of high performance computing.