Resumen
Embedded system designs have changed greatly owing to rapid developments in both hardware and software technology. Typical design should consider hardware limitations, such as size, weight, or battery capacity. In other words, the designs are heavily dependent on the hardware component. Since hardware can deteriorate and degenerate, hardware-aware software design is needed to achieve power-efficient embedded systems. Studies usually focus on the microprocessor in terms of optimizing power consumption. Besides computation, however, the system also consumes power when executing programs. A lot of memory accesses result in the entire execution, it should be considered to minimize for more efficient designs. Modern embedded systems often use heterogeneous memory to benefit from different characteristics of memory devices. This study aims to optimize the power efficiency of heterogeneous memory in embedded systems. We have proposed a detailed function complexity concept to identify specific function units in a program that consume less power in migrated memory. Using the function complexity, function selection algorithm is proposed to select a unique function which improves most after the migration. Experiments and quantitative analyses with various benchmarks have been performed to prove the validity of the proposed algorithm. Power consumption is successfully minimized by migrating certain function of a program in low-power memory.