WCET-Aware Energy-Efficient Data Allocation on Scratchpad Memory for Real-Time Embedded Systems

Scratchpad memory (SPM) is a software-managed, small, on-chip form of memory. For real-time embedded systems, worst case execution time (WCET) is more important than average-case performance. We address the problem of allocating program data variables between main memory and SPM to minimize the energy consumption due to data variable accesses, while respecting a given upper bound on a program's WCET. We present an optimal branch-and-bound algorithm; and an efficient heuristic algorithm. Our approach provides a flexible framework for the designer to perform tradeoff analysis between the program WCET and the energy consumption based on application requirements.