Power Efficient Embedded Memory Design for Mobile Video Applications

Abstract

This thesis mainly addresses the issue of low-power technology for streaming media applications. In order to ensure high output video quality under low-voltage supply, the proposed 8-bit pixel memory is sized by different bit positions. A novel MSEpixel estimation method is then developed according to bit failure rates to directly evaluate the video quality for every 8-bit sizing combination. Based on this estimation, one area-priory and one quality-priority mobile video applications are proposed by SPIDER algorithms. The results show that both luma and chroma data should be considered. More than 70% power is saved in memory units by using sizing-priority SPIDER algorithms. And the proposed SPIDER design methodology for low-voltage application is a feasible and efficient trade-off between the memory reliability and area overhead. Besides, a sample SRAM chip is designed for tape-out for further verification of the proposed SPIDER methodology.