The Optimization Research Based On ScratchPad Memory For Embedded System

With the rapid development of embedded systems, their application has become increasingly widespread, which brings on more strict requirements on performance, energy consumption and real-time response on such a kind of systems. The hardware and software need to be efficiently coordinated in order to meet the applications' requirements. As researches on embedded systems based on MPSoC and SoC is continually progressing and the utilization of on-chip area steadily increasing, there are more and more on-chip integrated devices available that boast better performance and larger capacity. Therefore, how to use high-performance on-chip devices to improve the overall performance of the embedded system emerges as one of the hottest focuses of current research.With the increase of available on-chip area, more on-chip memory can be embedded into the chip and accessed by high-speed on-chip bus, which contributes to narrow the access delays between the processor core and memory thus improves the overall performance. Scratchpad Memory is the name of SRAM integrated on chip but not as cache. Through software control methods, code and data can be stored in Scratchpad Memory and transferred to the processor. The Scratchpad Memory can be controlled by software, response quickly, takes up small on-chip area and consumes low power, so has been increasingly employed in embedded systems.This thesis concerns the effective usage of Scratchpad Memory, and investigates on how to make use of Scratchpad Memory to optimize embedded systems, accelerate programs, reduce delays in the memory access time, lower the energy consumption, thereby enhancing the overall performance of embedded systems.This thesis makes the following contributions:(1) It presents the optimization methods of embedded operating systems based on Scratchpad Memory. The scheduling algorithm in the embedded operating system is improved and the scheduling module of the embedded operating system optimized. The scheduling module is allocated to run on Scratchpad Memory. The embedded operating system is cut out as a microkernel with different components to enhance the efficiency.(2) It depicts the optimization of multi-program shared Scratchpad Memory. The embedded applications are analyzed and the memory objects are generated. When the memory objects start to run, they can be allocated to Scratchpad Memory. Thus the execution time and the energy consumption can be reduced by utilizing the features of Scratchpad Memory.(3) It proposes the Scratchpad Memory optimization for MPSoC architecture. The memory objects are extracted from the traditional embedded applications by compiler. Then multi-programs can share the Scratchpad Memory coordinated by the special Scratchpad Memory Management. The applications are threaded for parallel execution and at the same time the threads are pipelined to utilize the Scratchpad Memory and achieve the optimization.This thesis evaluates the proposed Scratchpad Memory optimization. These optimization methods in this thesis are tested and verified through the embedded hardware platform and simulation platform. Experimental results show that these optimization methods can effectively improve the performance of embedded systems, reduce energy consumption and optimize system performance by more than 20% in performance and more than 25% in energy consumption.This research fully utilizes the characteristics of Scratchpad Memory, and designs an optimization framework with embedded operating systems and multi-program share; it also explores on how to optimize ScratchPad Memory for multi-core based embedded systems. In this way, this work can effectively improve the whole performance of embedded operating systems, and can promote furthermore research of Scratchpad Memory applications on embedded operating systems.