Program Sections Allocation to Scratchpad Memory based on Frequency Analysis

Scratchpad memory is popular on-chip storage in embedded systems. The process in scratchpad is mainly attributable to performance and energy consuming. The researches on scratchpad memory focus on the allocation for the whole program, so the program runs in a new type of memory simply as a whole. However it ignores the limitation of the scratchpad memory´s capacity. So it is not practical using scratchpad memory in that way. In this paper, we propose frequency analysis techniques to find out the valuable sections of program and allocate them to scratchpad memory to improve the performance of the whole embedded system. The results show it can make a reduction of the execution time by 20%.