An H.264/MPEG-4 audio/visual codec LSI with module-wise dynamic voltage/frequency scaling

Abstract

A single-chip H.264/MPEG-4 audiovisual LSI for mobile applications including terrestrial digital broadcasting systems such as ISDB-T and DVB-H with a module-wise dynamic voltage/frequency scaling architecture is fabricated in a 90nm 6M CMOS technology. This LSI operates even during the voltage/frequency transition, so there is no performance overhead. Voltage/frequency scaling is realized by a dynamic deskewing system and an on-chip voltage regulator with slew rate control. Figure 7.3.1 shows a micrograph of the chip equipped with four optimally configured RISC processors, dedicated hardware accelerators for specific signal processing, 32Mb embedded DRAM and interfaces for camera, display, audio and network streaming for mobile multimedia applications. The power consumed when decoding QVGA (320x240) H.264 baseline profile level 1.2 video streams at 15frames/s and MPEG-4 AAC LC is only 90mW. The chip features are summarized Fig. 7.3.2.H.264 and MPEG-4 standards play essential roles in the field of mobile multimedia. H.264 is a video compression standard adopted for terrestrial digital broadcasting. Demands for larger image size, higher frame rate and higher image quality are ever increasing. These demands require larger memory capacity and higher operating frequency, both resulting in higher power consumption, which is unacceptable for battery-powered mobile equipments. The LSI described in this paper decodes CIF (352x288) H.264 baseline profile at level 2, or encodes VGA (640x480) MPEG-4 SP @L4a video stream at 30 frames/s while encoding/decoding audio/speech streams and multiplexing/demultiplexing them at 180MHz. Figure 7.3.3 shows the block diagram of the chip.There are four major modules: video frontend, video backend, audio/speech and multiplexer/demultiplexer. Each of the modules consists of an optimally configured 32b media-embedded processor (MeP) core [1] and dedicated hardware accelerators for its specific operation. These modules, peripheral interface units for camera, display, mic/speaker, network, etc. and embedded DRAM are connected via a 64b main bus. As for voltage and frequency, this chip slows down the audio module independently from the rest of the chip. The audio module is decoupled by a voltage/frequency socket from the main bus and an on-chip voltage regulator and a dynamic deskewing system (DDS) for the dynamic voltage/frequency scaling.H.264, compared to previous video standards, requires very high programmability so it is very difficult to implement in dedicated hardware accelerators. By studying data processing procedures, elementary processes, are allocated to the processor core or the hardware accelerators, including a context adaptive variable length decoder. Cooperation between the processor core and dedicated hardware accelerators have successfully reduced a large amount of operation time and power consumption without losing the programmability necessary for H.264. Traditional power reducing techniques such as embedded DRAM, clock gating and gated I/O are implemented as in previous w...

Low Power SRAM with Boost Driver Generating Pulsed Word Line Voltage for Sub-1V Operation

“…A V DD -hopping scheme [2] where the supply voltage is dynamically controlled according to workload of a processor is known as one of the representative low power design methodologies employing hardwaresoftware cooperatively oriented algorithm. In addition, a DVS (Dynamic Voltage Scaling) [3] and a DVFS (Dynamic Voltage Frequency Scaling) [4], lowering a frequency and supply voltage as long as lower performance is allowed, are becoming a standard approach for reducing active power consumption when performance requirements vary. The design style of voltage scaling which is moving from coarse-grained approaches to fine-grained ones even aims at scaling the threshold voltage (V th ) in small steps [5].…”

Process variation aware transcoding for low power H.264 decoding

“…Many of such works are based on scaling voltage and frequency on the logic gates, which make use of the performance margins available from dynamic variation in the performance requirements of the system [2], [3]. While the primary assumption of such techniques was to keep 100% correctness of the circuit functionality, later proposals [5], [6] exploited the error resiliency within the logic paths of the video codec circuits to push the voltage levels below the reliable operating points imposed by variabilities and imperfections in the fabrication process of sub 100 nanometer technologies.…”

RTOS and Codesign Toolkit for Multiprocessor Systems-on-Chip

“…This is mainly because increasing the number of processors is generally more power/performanceefficient than increasing the clock frequency. Specifically, multiprocessor systems-on-chip (MPSoCs) are considered to be a promising solution to achieve both high-performance and lowpower consumption and to be used in a wide range of embedded systems in future [1,2]. On the software side, real-time operating systems (RTOSs) have become commodity tools in order to manage the growing complexity of embedded software.…”