The ‘micro’ enterprise that is chip repair
Rodrigo Alvarez sits silent in a darkened room. He stares through his computer screen at a gray landscape, faintly rippled like windblown sand.
Mr. Alvarez, a PhD student here at Stanford University, is looking at the polished surface of a silicon chip magnified 15,000 times under an electron microscope. That chip sits a few feet away, sealed inside a Frankensteinesque machine with wires protruding from all sides.
Alvarez is at Stanford’s Focused Ion Beam Laboratory, about to do a microscopic repair. It’s the equivalent of cutting and reconnecting wires in a toaster oven 1/100th the size of a grain of sand. Earning his PhD depends on it.
Silicon chips sit at the heart of all things electronic, from computers to iPods to cellphones. But this chip – Alvarez’s PhD project – represents an entirely new vision. He’s designed it to mimic a part of the human brain called the cerebellum. This chip, he hopes, will someday allow intelligent robots to move gracefully to explore Mars, rescue kittens from burning buildings, or cook omelets.
But that’s all in the future. Today, he has a repair job to do. Alvarez has found a fatal flaw in this chip that he spent two years designing.
Two million transistors crowd onto this silicon wafer the size of the nail on your pinkie. They are connected by 250 feet of microscopic wires, which crisscross over one another in six layers, like the stacks of an endless freeway interchange. Among those millions of wires, Alvarez has discovered eight misrouted wires that cause the entire chip to lock up like a crashed computer.
He can’t afford to order a new set of chips from the factory – not at $35,000 a set. So he’ll fix them by hand.
He’ll burn his way down to those offending wires, through the upper layers of the chip, using a high-energy ion beam as a microscopic blowtorch. He’ll cut and weld the tiny wires – without, he hopes, nicking any others.
It’s 10 o’clock on a Thursday morning. Alvarez glances at the corner of the chip displayed on his computer screen. Starting from this reference point, he turns a knob and the image on the screen shifts sideways. He maneuvers to the place where he’ll cut, as though counting paces to a buried treasure, each step measured in millionths of a foot: 400 steps to the right, 150 steps down. Dig here.
He clicks the computer mouse to highlight a rectangle much smaller than the smallest speck of dust visible to a human eye – the spot where he’ll cut.
He clicks a button. A low whir comes from the Gothic horror machine. Inside, a beam of positively charged gallium atoms, traveling at 650,000 miles per hour, sears into the chip.
•••
You might call Alvarez a geek cog in the machine of high-tech innovation – or better yet, a Renaissance man in training.
By designing an entire chip by himself and then repairing it to boot, he’s learning the jobs that 10 different specialists would do at a big company like Intel. That training is meant to produce people who will come up with tomorrow’s big ideas.
“In a startup company you’re going to have the same guy doing everything,” says Stanford professor Kwabena Boahen – Alvarez’s PhD adviser. “That’s what it takes to innovate. Inventions don’t come from teams.”
The version of the chip that Alvarez is repairing today is strictly a research tool for deciphering how the cerebellum works. Some later version of it could find its way into battlefield robots or unmanned submarines, with patents that could net millions of dollars. But all of that is far away. The hard work comes now”
•••
By 2 p.m. Alvarez has made 15 cuts with the ion beam, burning gradually deeper into the chip. He speaks hardly at all.
The computer screen shows a rectangular pit excavated like a limestone quarry. Stripes run across the pit’s walls, like layers in a slice of birthday cake – alternating layers of metal and insulator buried in the chip, each 1/200th the width of a human hair.
Only by counting those layers can Alvarez tell how deep he is. If he burrows too deep, he’ll ruin the chip. He squints and pauses.
“I’m really at a loss,” he says. “I have no idea what these layers are.” He glances at a diagram of the chip, combs his fingers through his hair. “I guess let’s just make another hole and see what’s underneath.”
In this work, the manual dexterity that Alvarez developed from years of handling circuit boards counts for nothing. Instead, the laws of physics must serve as surrogate hands. Alvarez focuses the ion beam to a point using a magnetic field. He adjusts it by turning knobs, slow turns between thumb and forefinger. He is infused with Buddha-like calm and the concentration of a bomb-squad officer.
Now and then, problems appear.
Metal vaporized by the ion beam billows up in tiny clouds and condenses on the walls of the pit, like steam on a bathroom mirror. The metal condensation forms ornate curves. These “decorations” obscure the layers that Alvarez is straining to read. “It’s very annoying,” he whispers, slowly, as though not very annoyed at all.
•••
As a child in Mexico City in the 1970s, Alvarez used to dawdle about his father’s plastics factory. Machines whined as they extruded noodles of molten plastic, flattened them on rollers, and stretched them into shopping bags. Little Rodrigo inhaled the burning-candle aroma as he wandered the factory floor. By college, he was helping his father redesign those machines to improve their efficiency. The lessons he learned were as much philosophic as scientific.
When his girlfriend’s father challenged him to solve a Chinese solitaire game that had stumped him for months, Alvarez didn’t bother touching the game board. He went home, wrote a computer program to solve the game, and within 30 minutes had a set of winning moves. He demonstrated them to the older man the next morning. He was pretty annoyed, says Alvarez. “He never gave me credit for knowing how to use a tool to solve the problem.”
After graduating from Ibero-American University in Mexico City, Alvarez worked at a social networking website, Chilangolandia.com, founded by some friends. With his earnings, he set up a laboratory in his parents’ basement and spent two years working to design a chip that would mimic the brain and – he hoped – launch him into robotics. But Mexico City offered little opportunity. In 2003 he arrived as a PhD student in Professor Boahen’s laboratory.
Not every PhD student has to learn the obscure skill of chip repair. But not every one gets to learn it either. Alvarez has been forced to learn a lot in a hurry. He has already worked on 10 of his chips. The procedure takes about seven hours, and only one of those 10 chips has survived the procedure so far. He needs six more to finish his PhD.
•••
By 4:30 p.m., Alvarez has uncovered his eight wires. He slices them with a single pass of his ion beam.
Now he welds the eight wires together – a bizarre process in this little world. Pressing a button, he puffs a cloud of gas containing platinum over the chip. He zaps the gas with his ion beam, triggering a microscopic rainstorm. Platinum atoms shower down on the chip at the spot of his choice. A glob of metal accumulates over the wires, like fresh cement. “You have this incredible piece of machinery,” says Alvarez – high-end, but still “very hacky, very primitive.”
As though to illustrate his point, several round divots have appeared around at the edges of the pit – marks left by sparks of static electricity. A human would barely feel those sparks, but here in this tiny Whoville they leave massive craters, as though from errant smart bombs.
Those lightning bolts probably haven’t harmed the chip – or then again they could have singed hundreds of transistors, ruining the chip that Alvarez has worked for hours to save.
He hopes for the best. But not until he solders the chip to a circuit board and plugs it into a computer to test it will he learn that he’s succeeded, and that moment is weeks away.
For now it’s one chip down, six to go.
