Design And Research Of Multi-Granularity Reconfigurable Array Structure For Multi-Domain Applications

With the continuous development of information society,more and more applications take a large number of data processing as the main demand,and require higher and higher computing power for the hardware platform.Because of the natural limitations in the field of parallel computing based on the Von Neumann structure,the traditional processor cannot provide strong computing power,so the reconfigurable array has become a research hotspot again.Most of the existing reconfigurable arrays are only used in specific fields,and lack of sufficient support for multi field applications.In addition,most of the existing structures are based on dynamic configuration mode and a single interconnection granularity,which has limitations in complex and variable application fields.To solve the above problems,this paper studies a multi granularity reconfigurable array for multi domain applications.By summarizing the common properties of multi domain applications,it supports and optimizes the execution mode of reconfigurable array in three aspects: loop,branch,and access to memory.Based on the execution mode,the whole structure of reconfigurable array and the PE(process Element)micro architecture are designed.In order to solve the problem that the current reconfigurable array has weak support for complex control logic,this paper designs a multi granularity PE,which realizes logic control through differentiated finegrained units,reduces control overhead,and enables the array to independently complete the entire computing task.In order to meet the demand of computing power for multi field applications,this paper designs based on the static configuration mode.Through lightweight configuration storage and low configuration switching overhead,it improves the proportion of computing resources and processing performance,and optimizes the disadvantage of static configuration through cyclic multi transmit and dynamic data flow technology.In order to evaluate the multi granularity reconfigurable array proposed in this paper,based on high-level language modeling,the algorithm in multi domain applications is simulated and tested.The experimental results show that the array architecture proposed in this paper has an average performance improvement of 3.73 times compared with cortex-a7 processor,and can provide better support for applications with complex control logic.In addition,this paper also compares with HRe A,which is also oriented to multi domain applications,the experimental results show that the reconfigurable array proposed in this paper has higher computing power and has advantages in more computationally intensive applications.In conclusion,the work of this paper provides a reference for the design of reconfigurable array structure in the future.