SESSION TP 14
SALON 8, Tues, 8
1:30 PM

SIGNAL PROCESSING FOR MULTIMEDIA

Chair: S. Molloy, Luxxon Corp., San Jose, CA
Associate Chair: T. Yamazaki, Sony Electronics, San Jose, CA

14.1 A 60MHz 240mW MPEG-4 Video-Phone LSI with
16Mb Embedded DRAM 1:30 PM
T. Nishikawa, M. Takahashi, M. Hamada, T. Takayanagi, H. Arakida, N. Machida, H. Yamamoto, T. Fujiyoshi, Y. Matsumoto, O. Yamagishi, T. Samata, A. Asano, T. Terazawa 1, K. Ohmori 1, J. Shirakura 1,
Y. Watanabe 1, H. Nakamura 2, S. Minami, T. Kuroda, T. Furuyama
Toshiba Corp., Kawasaki, Japan
1Toshiba Microelectronics Corp., Kawasaki, Japan
2Toshiba Information Systems Corp., Kawasaki, Japan
A single-chip MPEG-4 video-phone system LSI in a 0.25 um CMOS, quad-metal technology with 16Mb embedded DRAM integrates three 16b RISC processors and consumes 240mW at 60MHz. Optimization of the embedded DRAM configurations and architectural efforts reduce power dissipation by 78% compared to that of a conventional design.


14.2 A 30Frames/s Megapixel Real-Time
CMOS Image Processor 2:00 PM
D. Doswald, B. Schreier, S. Oetiker, J. Häfliger, P. Blessing,
N. Felber, W. Fichtner
ETH, Zurich, Switzerland
A 30frames/s 1024x1024 pixel real-time 0.35 um CMOS image processor ASIC corrects black current and white imbalance pixelwise, interpolates images from a 1-CCD camera head over nine lines and performs a color space transformation for true-color motion images.


14.3 A Parallel Vector Quantization Processor Eliminating Redundant Calculations for Real-Time Motion Picture Compression 2:30 PM
T. Nozawa, M. Konda, M. Fujibayashi, M. Imai, T. Ohmi 1
EE Dept., Tohoku Univ., Aobaku, Sendai, Japan
1New Industry Creation Hatchery Ctr., Tohoku Univ.
A parallel vector quantization processor eliminating redundant calculation handles 2048 template vectors (16-elements/12b) by a single chip and is applicable to real-time compression of 30frames/s full-color VGA images. A real-time motion picture compression system employs this chip.

BREAK 3:00 PM

14.4 A 200MHz 0.25W Packet Audio Terminal Processor for Voice-over-Internet Protocol Applications 3:15 PM
B. Martin, I. Buckley, P. Bednarz, D. Chrissan, K. Amiri, C. Burnett,
C. Eddington, W. Lin, G. Maturi, H. Subagio, G. Teng, J. Waite,
Y.-T. Yong
8x8, Inc., Santa Clara, CA
Voice-over-IP (VoIP) terminals used in phone terminal applications require a low-power implementation suitable for the limited chassis area of these devices. This 200MHz VoIP terminal processor in 0.18 um CMOS implements a complete IP phone solution from audio samples to compressed TCP/IP packetized network signals with 2Mb of on-chip RAM.


14.5 A 720 uW 50MOPS 1V DSP for a Hearing Aid Chip Set
3:45 PM
P. Mosch, G. van Oerle 1, S. Menzl 1, N. Rougnon-Glasson,
K. Van Nieuwenhove 2, M. Wezelenburg 2
Xemics SA, Neuchâtel, Switzerland
1Phonak AG, Stäffa, Switzerland
2Frontier Design, Leuven, Belgium
The flow and techniques are used to design a digital signal processor chip containing 1.3M transistors and operating at 2.5MHz for hearing-aid applications. Its DSP consumes 720mW at 1.2V supply in 0.25 um CMOS, and performs 50M 22b OPS. The DSP achieves 0.015mW/MOPS performance, a factor 6 better than prior results.


14.6 A 4-Way VLIW Embedded Multimedia Processor
4:15 PM
A. Suga, T. Sukemura, H. Takahashi, K. Wada, H. Miyake,
Y. Nakamura, Y. Takebe, K. Azegami, Y. Hirose, M. Kimura,
H. Okano, T. Shiota, M. Saito, S. Wakayama, T. Ozawa, T. Satoh 1,
A. Sakurai 1, T. Katayama 1, K. Abe, K. Kuwano 2
Fujitsu Labs Ltd., Nakahara-ku, Kawasaki, Japan / 1Fujitsu Ltd.
2Fujitsu LSI Technology Ltd.
A 350MHz 5.6GOPS 1.4GFLOPS 4-way VLIW embedded multimedia microprocessor occupies 7.5x7.5mm 2 die in 0.18 um 1.8V five-layer-metal CMOS. VLIW, SIMD, 3-valued predicate mechanism, dual-load operation and dual-fetch contribute to performance.


14.7 A 7.1GB/s Low-Power 3D Rendering Engine in 2D Array Embedded Memory Logic CMOS 4:45 PM
Y.-H. Park, S.-H. Han, J.-S. Kim, S.-J. Lee, J.-H. Kook, J.-W. Lim,
R. Woo, H.-J. Yoo, Je.-H. Lee 1, Ja.-H. Lee 1
Korea Advanced Inst. of Science and Technology, Taejon, Korea
1Hyundai Electronics Ind. Co., Ltd., Kyoungki-do, Korea
A 52mm 23D rendering engine with 7.1GB/s bandwidth and 11.1M polygon/s drawing speed uses 0.35 um embedded memory logic technology and 2D array architecture. Main features of embedded frame buffer are partial word line activation and sequential block activation to reduce power consumption of nodal memory to <20mW.

14.8 Heterogeneous Multi-Processor for Composition and Enhancement of Video 5:00 PM
M. Strik, A. Timmer, J. van Meerbergen, E. Waterlander, F. Harmsze, A. Vaasen, L. Sevat, M. Oosterhuis, G. J. van Rootselaar,
H. van Herten, E. Jaspers, J. Janssen, G. Essink
Philips Research Labs, Eindhoven, The Netherlands

A system implements a set of video applications on a single die for TV display processing requiring up to 10 GOPS. Multiple data-driven processors work in parallel and communicate on-chip. Real time video processing is performed on a maximum of 10 video streams in parallel.

CONCLUSION 5:15 PM