
| All Todays Press Releases Click Here |
|
High
Performance Multimedia Technology For 3G Phones |
|
4th
November 2002 |
|
The Performance Challenge According to a recent report by Weili Su, an analyst at market research firm IDC (International Data Corp., Framingham, MA), "Shannon's Law states that the number of mathematical calculations required to process current data on wireless handheld devices is set to grow ten-fold, every four years. This is at odds with Moore's Law, which states that processor performance will increase ten-fold, every six years. Hence, the burden clearly rests on processor vendors to ensure that the performance gap between algorithmic complexity and processor performance is bridged." "The industry is seeking new architectural approaches to solving the issues of performance and power. The concept of combining storage and processing into a uniform structure is promising in this context," Su concluded. Parallel Processing … “When we look at nature’s response to sensory information we see an interesting parallel,” explained Dr. Avidan Akerib, general manager of NeoMagic Israel and a principal inventor of the APA architecture. “The ear is a sequential processor. It processes sound, which comes to the eardrum as a sequence of pressure waves. The eye is a parallel processor. Its retina is an array of rods and cones, which receive and process thousands of concurrent light waves in parallel. We can take some architectural cues from this natural phenomenon – visual information is better processed in a highly parallel structure,” he noted. … in an Intelligent
Cache Increasing the bandwidth to the processor from the memory and back again, has been a challenge to design engineers for decades,” said Dr. Akerib. “The most common response has been to create a cache, a smaller memory structure implemented with comparatively faster memory technology, which resides close to the processor. This approach requires more frequent movement of the data between the main memory and the cache, and between the cache memory and the processor. Any sequential processing pipeline has the basic characteristic that data computations require the data to be moved multiple times. Dr. Akerib noted, “Within the physics of semiconductors, power is used every time data is moved. It may not be much power, but when a processor runs hundreds of millions of cycles per second, small amounts of power add up quickly. Additionally, there are relations between the elements of imaging, video or graphical data, which means that the sequential processing engine may need to load and reload the same data element many times to complete its task. The result is a rapid escalation in the power consumed by multimedia tasks.” The APA resolves this data movement problem at the architectural level by providing a single structure which both stores and processes the data. Serving much like a data cache, the APA loads a significant amount of data from the main memory. But unlike traditional approaches, the APA processes that data within the caching memory structure, performing multiple processing steps and saving interim values as needed without unloading or reloading the data. When the processing of the data is complete, it can then be returned to the main memory, or held for the processing needs of subsequent data. APA’s Roots The theory, structure, and practical implementation of the Associative Processing Array are protected by multiple U.S. and international patents. The MiMagic Product Roadmap The APA architecture will
be a key element of future MiMagic Applications Processors. The initial
APA technology applications will be MPEG-4 video compression and decompression.
The company plans to also develop other forms of digital imaging and
digital video, 3D graphics, and continuous speech recognition capabilities.
MiMagic products incorporating the APA technology are expected to become
available in the first half of 2003. |
| All Todays Press Releases Click Here |
| All Material Subject to Copyright. All logos, graphics and trademarks are the property of their respective owners. |
| m |
|
|