Design Of General Purpose Processorfor Processing-in-Memory

Processor and storage are main fields in computer design,representing compute and data,respectively.The main part of traditional computer architecture is processor,which leads a significant development of processors optimization in the last few decades.However,with the change of application scenarios,computers have to process larger data size.Data access is occupied a higher and higher proportion compared with compute.Thus,the problem of access delay has been more serious caused by existing storage structure.More power consumption is hard to neglect considering frequent data access.Processing-in-memory integrates computing unit into memory,achieving the function that data could be calculated directly.It shortens the distance of data access and reduces power consumption of access delay,compared with traditional architecture.In the processing-in-memory architecture,computing center is integrated into the memory with a method of 3D package.Given 3D package demands a high thermal property,power consumption of computing center needs to be strictly restricted.Moreover,the target of processing-in-memory design is further boosting computer performance.Thus,the module of bulk data processing demands higher computing center performance.This paper designs a processor that is suitable for computing center with processingin-memory architecture on the basis of improving computing center performance.The processor,with high-performance as well as generality,is based on ARM instruction set.It achieves relatively high instruction-level parallelism relying on relatively simple pipeline architecture.This paper designs one architecture that use the processor as the computing unit.Then we implement the two parts of the architecture on two FPGA development boards respectively,while the two parts connect each other through interconnection between two development boards.FPGA simulation testing on the processor testifies that the processor is able to improve performance effectively and the architecture is effectiveness.We use programs with different data correlation,such as MapReduce and Mpeg2 decode,to test the architecture on the simulation platform and found that performance increases about 60% and power consumption decreases about 95%.Comparting with traditional computing structure,which could demonstrate sufficiently that the processor is well-suited to the processing-in-memory architecture.