San Jose, Calif. — As a memory buyer for Intel Corp. in the days of the DRAM drought, Jeff Ravencraft knew how to get difficult things done in a big company. Those scrappy skills could be handy as he tries to drive the Universal Serial Bus interface into the wireless world for Intel.
Ravencraft, a technology strategist for Intel and president of the USB Implementers Forum, announced last week that the 1.0 wireless USB specification had been completed. The work is but one of several tricky steps ahead before Intel hits its goal of pulling the wires from a host of computer and consumer devices.
The wireless USB spec promises untethered point-to-point links using the multiband orthogonal frequency-division multiplexing version of ultrawideband technology at 480 Mbits/second at 3 meters or 110 Mbits/s at 10 meters. The UWB physical-layer chip is expected to consume about 130 to 160 milliwatts, with first interoperable chips and devices arriving by year's end.
Wireless USB will adopt as much of the existing wired USB technology as possible, including a topology supporting up to 127 device links to a single host. The wireless spec adds security, requiring that all traffic use AES-128 hardware encryption.
Thanks to a "convergence layer" of software being developed by the WiMedia Alliance, the underlying ultrawideband radio will support software stacks for 1394, Bluetooth, Internet Protocol or USB protocols. Most observers agree USB is the first logical target since as many as 2 billion USB ports have already shipped, including 500 million on PCs alone.
Dramatically demonstrating his goal, Ravencraft ripped a rat's nest of wires from the back of a rack of consumer electronics gear in his keynote before some 300 engineers at the Wireless USBDevelopers Conference here last week. User frustration with wired links could lead OEMs to ship up to 140 million WUSB links in 2009, market watcher In-Stat Inc. estimates.
But several hurdles stand between today's WUSB spec and mainstream deployment. Standard compliant chip sets must pass interoperability tests. Some parts of the spec and several layers of software have yet to be written. Costs need to come down dramatically from expected initial levels of about $10 per node. And regulations in Europe and Asia need to open the door for designs roughly similar to those approved by the FCC.
All this has to happen while Freescale Semiconductor touts its own, direct-sequence spin of wireless USB.
The next milestone comes in September at the Intel Laboratories in Hillsboro, Ore., where testing starts for the first 1.0-compliant WiMedia MAC and PHY chips. As many as seven PHY and perhaps 40 MAC designs are in the works, said James Wright, a director of marketing for LeCroy Corp., which announced an ultrawideband protocol analyzer at the conference.
Companies that want to deploy WUSB products in time for the 2006 holiday season are designing their own MACs now, but "this time next year no one will want [discrete MAC/PHY chips]," said Wright. That could spell problems for ultrawideband startups such as Alereon, Staccato and Wisair if they cannot deliver intellectual property or roll discrete chips quickly.
Philips Semiconductors said it will roll a 90-nm WUSB end-device MAC based on an ARM7 that includes a third-party PHY and sells for less than $10. NEC showed its MAC with a third-party PHY and said it has its own PHY in development. And Wipro Technologies showed a generic UWB MAC supporting AES-128 and plans a WUSB end-device MAC early next year.
All the chips on display last week were based on a 0.9 version of the WiMedia MAC and PHY specs and were implemented in FPGAs. "The biggest issue is getting to interoperability," said Eric Broockman, chief executive officer of Alereon Inc.
A representative of Wipro said WUSB's AES capability can require up to 30 percent of the gate area of a MAC. End-device chips also require a microscheduler feature.
Soft hurdles
Several pieces of the spec and its associated software are still in progress. The WiMedia MAC spec will not be finished for about another month. It will be the end of September before the WiMedia convergence software layer, on which the WUSB and other protocol stacks are built, is complete.
The IP-over-ultrawideband protocol stack for peer-to-peer traffic should be complete about the same time. A wireless 1394 spec for WiMedia chips will be finished by year's end.
Another issue is how WUSB devices will identify each other. A special task group is working on it and hopes to complete its spec by the end of September. It's considering as many as five association models including matching numbers of a host and end device or requiring a PIN.
Intel conducted roughly five focus groups, each with as many as eight consumers, to study association models and found the winner to be near-field communications (NFC), another wireless technology still in its infancy. "It was unanimous," said Ravencraft. "Ultimately, we'd like to get down to a couple of [OEM-selectable optional] approaches with a path to NFC."
The WUSB group is also debating how to users or systems select protocols for particular jobs. That could be handy since Bluetooth is the most power-efficient protocol for jobs such as streaming low bit-rate traffic like music, while UWB saves power when transferring large blocks of data.
Microsoft's next version of Windows, dubbed Longhorn and due late in 2006, is not expected to support UWB or WUSB natively, so Microsoft is asking for help from OEMs in developing beta drivers. Microsoft said it has successfully tested 13 of its 15 USB class drivers for wireless use.
"In this age of online Windows Update capabilities, we are in much better shape for getting software support," said Ed Beeman, a USB expert at Hewlett-Packard Co. He called good driver support one of his top concerns.
WUSB backers expect two early markets to bootstrap the industry. First products in 2006 are likely to be optional dongles and adapters for PCs or peripherals. They could be followed by internal PCI- or Express-based adapter cards for high-end systems once the chips hit the $5 price point expected in late 2006.
But if they are to hit mainstream use, the WUSB features need to become blocks integrated into PC chip sets and controllers for digital cameras and MP3 players. It's there that "a Synopsys can come into play," Beeman said.
The migration path "is not going to be a chicken-and-egg thing," he said, but it won't be as smooth as the USB 2.0 transition, which used the previous generation's software and went from spec to chip set integration in about a year.
Global regulations
Getting Europe and Asia to open up spectrum for ultrawideband could be one of the trickiest tasks. The European Commission told Intel that it's unlikely the EC rules — expected before June 2006 — will be the same as the FCC's, said Stephan Wood, an Intel technology strategist and president of the WiMedia Alliance. Tightening or blocking out spectrum bands are easy changes to handle, "but if they bring the power levels way down it could scrap some planned applications," he said.
South Korea has already issued test licenses for UWB products and could issue full licenses by year's end. Wood hopes other Asian governments will move forward by the midyear 2006.
Ravencraft rebuffed suggestions that competitors will crimp his WUSB plans. The Freescale products, he said, are a straight port of USB 2.0 to a wireless medium, rather than a "ground up" redesign for wireless, and lack the support of the USB Implementers Forum.
Eighteen years ago, during a DRAM shortage, Ravencraft joined Intel as a purchasing agent. After memory prices tanked, Intel was about to write off inventory of a warehouse of PCs built with expensive DRAMs. Ravencraft offered to sell the memories they contained to recoup a $1 million investment. Such resourcefulness may serve him well as he negotiates the path ahead.
