Region compression: a new scheme for memory energy minimization in embedded systems

Power consumption in microprocessor-based embedded systems can be reduced by decreasing the number of memory accesses needed to fetch instructions from memory. We propose a code compression approach that reduces instruction memory bandwidth by assigning short code words to frequently-executed instructions. Our technique is called “region compression” because we use a dedicated symbol, called “mark”, to flag transitions between regions of consecutive compressed instructions and consecutive uncompressed instructions. In practice, the hardware complexity of instruction decompression is directly related to the number of different instructions that are compressed in a program. Our technique performs remarkably well even when the hardware complexity of the decompression unit is tightly constrained, and consequently only a small number of different instructions can be compressed