Something in the air

Intel makes its money from computer chips. So why is it now interested in radios? Pat Gelsinger talks to Charles Arthur about the birth of 'smart' networks

Pat Gelsinger knows what the future's going to be like. Or at least the part that involves computer chips. As chief technology officer for Intel, the Godzilla of microprocessors, he's really quite excited about where we're heading. Give it a couple of years and he can see us having ultra-fast networks around the house, speech recognition that really works, and phones, laptops and other digital devices that will be able automatically to recognise the best signal for whatever wireless signal happens to be around, and will latch on automatically.

Let's start with the ultra-fast networks, which will use a technology called Ultra Wide Band, or UWB. You think that Bluetooth, the wireless system for linking phones, PCs and printers, is fast? Pah. "That delivers about half a megabyte a second. We expect that UWB will run at about 500 megabytes per second, perhaps up to a gigabyte per second, over distances of about 10 metres, at a lower cost and power than Bluetooth."

Using that, he reckons we'll be able to link up our DVD players to our PCs to our hi-fis to our other PCs to our digital cameras and effectively be able to stream and share our data all around our homes. "Consumer electronics and PC companies are both very excited about this," he says. "We expect the standard to be finished in the second half of 2004, which means products should be available using it in 2005 or 2006, on the basis that stuff usually comes out the year after the standard is finalised."

What about the speech part - because that has proved one of the thorniest computing tasks for decades? Gelsinger responds: "Try closing your eyes and covering one ear. Now listen to me talk. See how hard it is when I move my head?" - his voice became less distinct, although we are sitting across a table - "and move around? That's what it's like at the moment for a computer trying to recognise speech."

Gelsinger notes that speech recognition today is "pretty good - if you have a perfect signal". But he thinks that adding sight to the sound will lead to the next major breakthrough in this technology, so that we will truly be able to address our computers wherever we are, and they will hear, see and - more importantly - understand us. "You know, there's five billion people in this world who've never touched a keyboard, and most are using phonetic languages. This is a pretty big area. But it just takes a lot of processing power to do it."

But why is Intel, the world's biggest manufacturer of microprocessors, so interested in building what can only be called radios? Look at its Centrino chip, launched in March: it marks the first time in the company's history that a new generation of CPU has a slower clock speed than its forebears. That's because Centrino concentrates on wireless ability - the circuitry needed to connect to a Wi-Fi (802.11b) network is etched into the chip.

The reason is that radio is now a major focus at Intel, one that Gelsinger himself has been key in promoting under the tag "Radio Free Intel". Why?

Gelsinger joined Intel in 1979 aged 19 and became its CTO in September 2001 at the age of 40. But even before then he had been pushing the wireless future hard. "About four years ago we made a decision to go big on wireless. It was prompted really by what was happening with 802.11 [the "Wi-Fi" revolution] and the services in Japan offered by DoCoMo, such as i-Mode. And the wireless LAN standard emerged. When we saw how Wi-Fi was taking off, well..."

Radio Free Intel, unveiled in May, aims to put wireless capability on to all sorts of mobile chips. "Centrino is the first step towards building radio into everything," Gelsinger says. "You'll have dynamically reconfigurable radios - that means they'll attach to any network seamlessly, whether for voice, data, for your laptop, PDA, mobile phone..." The intelligence to do that will be in the devices, not the networks, he says.

This is part of an ongoing debate in the computing industry - whether the best way to build vast networks is to have a "smart" network that recognises the devices attaching to it, or a "dumb" network that just transmits data to smart devices at its endpoints. Gelsinger is certain that the answer is dumb networks, smart devices. "The internet is a dumb network. We believe in that model of a simple network with intelligent endpoints."

But building radios into digital chips isn't as easy as saying it. Radios are inherently analogue devices: frequencies and voltages vary smoothly over a range. The digital world is 1 or 0, and it takes a lot of ingenuity with silicon to produce something that appears analogue with a digital signal.

But doing that begins to make everything possible, and more importantly, affordable. UWB, which is presently only a nascent technology, could be built on to a chip alongside Wi-Fi and other wireless capabilities.

So is Intel going to give up the pursuit of speed? Not at all, says Gelsinger: "It's not either-or," The week before Comdex, in January, Intel release the 3Ghz Pentium 4, and Gelsinger said: "Today we add 25MHz to the product every week. That used to take three years. Soon we will add 25MHz every day." How? Because as he sees it, they have to. "Our job as technologists at Intel is to 'obey Moore's Law [that processing power doubles every two years]'."

Moore's Law is one thing, but there's another law that still reigns supreme - Murphy's. Being an exalted executive in a technology company that promotes wireless must mean that his seven-PC home (of six people) uses Wi-Fi, right? He answers vaguely uncomfortably. "Er, actually I installed a lot of [Ethernet] cable. The Wi-Fi signal isn't very good in the house," he confesses. "Someone can walk into the room and it affects the signal so badly that I lose contact... I have to sort it out one day."