To: J Fieb who wrote (37376 ) 11/25/1998 3:11:00 PM From: BillyG Read Replies (2) | Respond to of 50808
MPEG-4 project in Europe achieves wavelet silicon By Peter Clarke EE Times (11/25/98, 11:08 a.m. EDT) LEUVEN, Belgium — An MPEG-4 multimedia research program at the Interuniversities Microelectronics Center (IMEC) has achieved working silicon on a wavelet-compression-based visual texture-coding chip. Called Ozone, the chip could become a key intellectual-property (IP) core for third-generation mobile terminals. The work was coordinated at IMEC with support from corporate partners Ericsson (Stockholm, Sweden), National Semiconductor Corp. (Santa Clara, Calif.) and Alcatel Microelectronics (Brussels, Belgium). The Ozone hardware is designed to perform embedded zero-tree coding and adaptive binary arithmetic coding, both of which are useful techniques for encoding and decoding such visual objects as meshes, textures and shapes. The developing MPEG-4 standard covers a set of technologies for the delivery and combination of images, sounds and positional information. It is intended to be a scalable, object-based multimedia system that can also take into account the capability of the receiving terminal, which could vary from a device as simple as a cellular telephone to a powerful computer client or a multimedia TV or PC. MPEG-4 is thus a superset of the established MPEG-2 standards for the encoding and decoding of two-dimensional audiovisual data using discrete cosine transforms. For other graphical information, scalable wavelet compression is recommended; that's where the Ozone chip fits in. The program, set to run for three years, has seen some setbacks in its first full year of operation. One fundamental delay was the achievement of a working device: Jan Bormans, section leader of the multimedia image-compression group at IMEC, said the project had expected to tape out a wavelet-compression chip in March but was forced to re-spin the silicon. Another setback has been the project's failure to attract Japanese semiconductor partners. Japanese participation had been a founding goal of the project, which Bormans had thought would appeal to the country's consumer-electronics giants. But the Asian economic crisis foiled those plans, Bormans said. That's not to say that Bormans is dissatisfied with the partners IMEC has secured. "Ericsson is clearly interested in mobile MPEG-4. They want to do videoconferencing and send still images to mobile terminals; therefore they are interested in a low-power solution. Ericsson is looking for applications for UMTS [Universal Mobile Telecommunications Systems]." Eyeing games on PC National Semiconductor's interest is "for games on the PC," Bormans said. Games could be played over the Internet using avatars to represent a large community of interacting players. Alcatel Microelectronics, meanwhile, was the foundry for the experimental chip. Low power is one motivation for developing a dedicated silicon approach, since MPEG-4 data streams could, in theory, be decoded on powerful desktop PCs. "If you run MPEG-4 on a 333-MHz Pentium, it runs slow. You will need the capability of a future Pentium, but that, including the memory subsystem, would consume something like 60 W of power," said Bormans. The Ozone chip has been fabricated by Alcatel in a 0.5-micron CMOS three-layer metal process and consumes 30 mW at 32 MHz. The 70k-gate, 250k-transistor chip is also somewhat slow, typically processing 12 to 19 CIF (352 x 288) 8-bit frames/second. But the block is within striking distance of full-frame-rate color processing, which could come with a modest increase in clock frequency and a shrink to 0.35-micron process technology without substantially increasing power consumption. Ozone is configured as a decoder, but Bormans said that the algorithms and the logic blocks used to implement them are highly reusable. That should allow an encoder to be built relatively easily, he said. Beyond the project's exploration of dedicated MPEG-4 silicon, Ozone is exploring memory-centric design and working on experimental system-level design tools developed at IMEC. Part of Bormans' thesis is that an analysis of most systems will reveal that most of the transistors are memory elements and that, therefore, system-on-a-chip design is best served by a memory process that can support logic, rather than a logic process that can support memory. eet.com