Adam L. Penenberg
Fast Company, 10/2008
DESIGNING A MICROPROCESSOR LIKE THE ATOM, Intel’s smallest chip ever, is like planning a city so tiny it could fit into a single bacteria. First, architects map out which routes go where so that millions of switches — the transistors — can direct traffic in the form of ones and zeroes that shoot from transmitters to transreceivers across silicon expressways (called “buses”). Once the schematics are plotted, designers create microscopic mock-ups of each layer, or “mask,” and then test them on powerful workstations, mimicking all the functions of the chip. Much depends on getting the map right: Once the chip is built, its 47 million transistors, which are so minute that 2 million of them could sit on the period at the end of this sentence, switch on and off up to 300 billion times per second. If just one of them malfunctions, the entire processor spits up a hair ball.
The chip’s fabrication phase is another logistical nightmare, some 300 steps involving chemicals, gas, and light. It begins with a land grab in the form of purified beach sand — also known as silicon. The sand is melted and grown into cylinders, then forged into thin wafers the size of LP records, which are shined until their surfaces function as “perfect mirrors.” With photolithographic “printing,” the transistors and electrical passages are layered onto the wafers. Things get really complicated from here. To ready the chip for mass production, the wafer is blasted with heat and coated with silicon dioxide and light-sensitive photographic film. The masks are then overlaid, with more layers ladled on top, and the whole thing etched, bombarded with chemicals, and covered with layers of metal. Each and every gate on each and every transistor is fed a positive- or negative-charged ion that will determine whether its job is to open or close. The wafers are then cut into chip-size bits using a precision saw.
Building transistors that are only slightly larger than the silicon atom itself is a dazzling display of design and engineering, made more so by the sheer pace of technological innovation over the past six decades. An early transistor, created by Bell Labs, was about an eighth of an inch in diameter; today, 2,000 transistors placed side by side equal the width of a human hair — and the cost has fallen to about one-millionth of what it was in 1968. Mooly Eden, general manager of Intel’s mobile platforms group, contrasts his business with the automobile industry: If Ford or Toyota had innovated at a similar pace, he says, “a car would probably go half the speed of light,” and if you were going to San Francisco for a week, it would be “cheaper to throw out the car and buy a new one” than to feed quarters into the parking meter.
At least as impressive as the Atom’s size, however, is how little power it uses. Over the years, while Intel and its rabid rival AMD engaged in a game of one-upmanship over who could produce faster, more powerful chips, they didn’t pay much attention to battery drain or heat — both natural side effects of all that increased processing power. And that is what makes the Atom a tectonic shift in strategy. It’s not faster than previous processors, nor does it do more. In fact, it does less. But it uses a fraction of the battery power — 10 times less, according to Intel — and therein lie the seeds of a revolution in mobile technology.
Most people with handheld devices such as a BlackBerry don’t need a processor powerful enough to edit video, create Pixar-like effects, or deconstruct the human genome. They just want a smartphone that lets them send and receive email, open attachments, surf the Web, take and store photos, and perform other basics without burning through the battery. (Steve Jobs, take note.) Of course, that is easier said than done. Today’s handhelds gorge on battery power, but since they can’t digest Flash, the basis for most of the media online, they still don’t allow users to surf the entire Internet. With an iPhone, for instance, you can visit YouTube, but you can watch only those videos that have been specially recoded. The Atom, by contrast, was designed from the bottom up to provide the entire Internet experience with minimal battery drain. And even with that quantum leap in capability, the first set of devices running on Atom chips get between four and six hours of battery life, according to Pankaj Kedia, Intel’s ecosystems manager. He predicts those numbers will only improve in future generations.
Martin Reynolds, a vice president at Gartner Inc., says that while the Atom won’t rival desktop chips for speed or power, it’s certainly quick enough to do what it was designed to do — and, he adds, it’s “really cheap.” Its teeny transistors allow Intel to pack memory and basic system controllers into one package. “That means all the complicated work you do on a circuit board can now occur in a chip. This will make the devices smaller.”
It seems obvious today that any chip would be a slam dunk if it improves battery performance while offering the full mobile Internet. But at Intel, the decision to develop the Atom was controversial. It’s much easier to improve an existing product line, after all, than to create a new chip from scratch, and in 2004, when the project began, the handheld market was just emerging. The Atom would siphon billions of dollars in development costs from other programs and require hundreds of engineers working full time for four years.
Once the commitment was made, however, a series of new engineering and design features were built into the chip. Foremost among them was a special type of metal transistor gate to replace the tried-and-true silicon. The new gates drastically cut down on the inefficiencies (read, missed signals) that cause information leakage and therefore extra battery drain. But they also set up a nasty obstacle for competitors such as AMD that have not yet found a way to achieve the Atom’s efficiency. “Without metal gates, they can’t match the power of Atom,” Reynolds says. “Intel has a couple of years before AMD can enter the space.”
While the Atom was developed specifically for the handheld market, Intel discovered along the way that the chip could also power a whole new category it dubs “netbooks” — mini PCs that are lightweight, cheap (as low as $250), and capable of running basic PC functions such as word processing, email, and Web surfing. Intel’s Eden views the netbook as a disruptive technology that could create whole new markets: Kids as young as 5 will play on netbooks, he predicts, and why should students haul 12 pounds of books to school when they could simply carry a 2-pound computer? Emerging economies in China and India alone could mean a huge opportunity as the Internet becomes a staple of life in those countries — especially since the Atom will power not only Intel’s Classmate PC (the controversial free-market foil to the One Laptop Per Child machines developed by MIT’s Nicholas Negroponte) but also dozens of other cut-rate contenders. Here in the United States, Mike Feibus, an analyst with TechKnowledge Strategies, sees the price for some Atom-powered netbooks dropping below $199 by Christmas 2009, a price low enough to close the gap between PC haves and PC have-nots. Intel estimates there will be about 50 million netbooks in circulation by 2011. And because the Atom packs oomph without getting superheated, there’s no need for bulky fans to keep the insides cool, which means the chip could soon be driving everything from handheld gaming machines to GPS gizmos, e-book readers, Internet tablets, and pocket video and music devices — all juiced with mobile Internet capability.
Perversely, Intel’s temporary technological advantage may mean trouble down the road. While Atom sales are jamming, according to the company, Gartner’s Reynolds calls it a “price-point enabler — it could burn Intel by driving the margins out of the [chip] market. If everyone switches to Atom, Intel’s revenue would go down, especially if it starts leaking into the notebook-and-desktop market. That could hurt.”
Eden doesn’t think that will happen. “Let’s not fool ourselves,” he says. “If you want to manipulate pictures, if you want to look at high-definition video, don’t do it on the Atom. It is not fit for this.”
Because the Atom isn’t Intel’s fastest or most powerful chip. And that’s precisely the point.
Copyright 2010 Adam L. Penenberg (penenberg.com)