ARM's v6 balances power, efficiency

ARM's v6 balances power, efficiency
By Crista Souza, EBN
October 19, 2001 (11:36 a.m. EST)
URL: http://www.eetimes.com/story/OEG20011019S0070

With power/performance considerations taking center stage at last week's Microprocessor Forum here, ARM Ltd. shed more light on its upcoming v6 architecture, which is designed to give embedded system OEMs a big boost in megahertz without compromising power efficiency.

While specific speed and power consumption levels won't be disclosed until the first v6-based implementation-a soft core-debuts in mid-2002, ARM outlined a number of features added to enhance future embedded products.

“Power/performance has always been the ARM story,” said John Rayfield, director of research and development at the Cambridge, England, company. “The interesting thing is the high-performance computing people are now hitting the same problems we've had in the embedded space for years.”

As the complexity of embedded systems grows, it's becoming more important to achieve high performance and low power simultaneously, he said.

“We're trying to pragmatically do the right thing to get the best efficiency possible,” Rayfield said.

Among the features to increase efficiency, the company has added ARM and Thumb instructions to support mixed-endian states, both for little-endian interfaces such as PCI and USB, and big-endian data payloads like TCP/IP and MPEG. “In the new connected world, you have to be a little of both,” he said.

By implementing the instructions in hardware instead of software, ARM claims to achieve better code density and power efficiency.

For multiprocessor designs, a shared- memory attribute was added to improve communication between processors while using fewer megahertz. The shared-memory feature supports multiple ARM designs, as well as designs that use an ARM in combination with DSPs or other embedded processors.

Intel Corp. and Texas Instruments Inc. in July licensed the v6 architecture to take advantage of this feature in designing their next-generation processors, Rayfield said.

Additionally, a new cache architecture, called V irtual Memory System Architecture, improves performance 20% to 30% by eliminating forced cache flushing by the operating system, according to Rayfield. OR