Low power design of block-based video codecs

Low Power Design of Block-Based Video Codecs The improving display resolution of new video appliances continuously increases the throughput requirements of video codecs and further complicates the challenges encountered during their cost-efficient design. In contrast, the energy and heat dissipation limitations of mobile appliances create the demand for low-power implementations. This PhD proposes a memory and communication centric design methodology to reach an energy efficient dedicated implementation. The high level steps of this design flow combine memory optimizations and algorithmic tuning on a sequential executable description. Then, a partitioning exploration introduces parallelism using a cyclo-static dataflow model. To maintain the effect of the high-level optimizations, also implementation specific aspects of communication channels, like using a kind of shared buffers, are expressed without extending the model of computation. Consequently, all analysis potential at design time is preserved. Towards dedicated hardware, these channels are implemented as a restricted, but sufficient set of communication primitives. They allow exploiting the principle of separation of communication and computation and lead to an automated RTL test and development strategy enabling rigorous functional testing. In this way, the design time is reduced. The introduced methodology is applied to the design of a high-performance MPEG-4 video encoder. The fully dedicated video pipeline exploits the inherent functional parallelism of the compression algorithm. It has a tailored memory hierarchy; uses burst accesses to external memory and supports realtime processing of 30 4CIF frames per second. The effect of the high-level optimizations on the power-efficiency is demonstrated through power simulations. The core consumes only 71 mW in a 180 nm, 1.62V UMC technology. This energy efficiency, achieved without using a low-power CMOS library, is equivalent to the state of the art for high resolution video encoders.