Wafer Madness
Twenty-four hours a day at Motorola’s MOS 11 factory outside Austin, workers race to build the complex computer chips that power our brave new world. For two weeks, I was one of them.
(Page 4 of 4)
Raining Metal
The third shift starts at 11:30 P.M. and lasts till 8 a.m. During the night, the area outside the fab is deserted except for robots that move along the gleaming linoleum floors carrying finished wafers over to the shipping area. Not surprisingly, given the hour, the strangest things that happen at MOS 11 usually happen on the third shift. Most recently, a romantic fever seems to have swept through the fab: There is a rumor, neither confirmed nor denied by Motorola, that two employees were caught having sex in a conference room and two others were caught having sex in a car. Such exertion aside, everybody who works the third shift typically functions in a state of perpetual jet lag. And yet the third shift shovels wafers along. “On first shift, you’ve got managers running around, maintenance people working on equipment—lots of issues to deal with,” said Andrew Venson, the 27-year-old training coordinator who showed me around for several nights. “On third shift, you don’t have as much of that going on, and you can just do your work.” Venson moved here from California and tends to stay calm despite the frenzied pace inside the fab, which often makes other employees seem like they’ve been drinking too much coffee. On the weekends, he reverses his sleeping schedule to normal so he can be active in a Jehovah’s Witnesses church. He used to work on the first shift, but when his wife, Jessica, started working at MOS 11, she was assigned to the third shift, so Andrew asked to work then too.
Venson used to be an operator in etch, where wafers are sent from photo. In etch, acids or plasma are used to carve away part of the last layer that was grown on the wafer in diffusion’s furnaces—leaving behind minuscule parts of transistors, capacitors, or diodes. “Think of a chisel chipping away at wood,” said Venson. The acid or plasma removes material except where photoresist remains on the wafer. Afterward, the last of the photoresist is removed; by then, the material it protected has taken on the shape of the pattern that was photographed onto the wafer. In this fashion, the rudimentary parts of a circuit are built. Because etches are irreversible, many of the problems that cause chips to fail occur in this area. Incomplete etching of metal layers, for example, causes electrical shorts in the minute circuitry.
There are two basic types of etches. Wet etches are performed by dunking wafers into hoods that hold corrosive liquids such as sulfuric or nitric acid. Dry etches, which take place in a machine called an etcher, are more complicated. Venson and I went into one of the maintenance areas to look at the etchers. The machines resembled large chrome boxes, and the area was full of skinny pipes carrying chemicals up from below; we were standing in an aisle of etchers that were working on metal layers, and those machines were using chlorine—one of the more toxic gases in the factory. Electrodes inside the etchers excited the chlorine into plasma form, which is a powerful, electrically conducting state (lightning and neon are both examples of plasma). An electrical coil inside the machines then aimed the plasma at the wafer, burning away the material that was bare of photoresist. There was a window on the side of one of the etchers we were looking at, and it was emitting a purple light—the glow of plasma eating metal.
From etch, many wafers go back to diffusion to begin another round through the factory’s divisions—they go back to diffusion’s furnaces, back to the lithocells, and back to etch’s hoods. Sometimes, however, instead of going back to diffusion, wafers acquire additional layers in a part of the factory known as films. Amsale Nunu, the training coordinator who formerly worked as an operator there, is a petite 37-year-old woman with a quiet, engaging manner. Nunu grew up in Ethiopia and immigrated to this country at age 22. She got a high school degree by taking an equivalency exam and began working at Motorola in 1987. Nunu explained that late in the assembly process, layers of metal such as aluminum and titanium (which are used to wire together all the transistors, capacitors, diodes, and other devices that have been built) are sputtered onto the wafers by shooting beams of ionized gas at big tablets of metal, causing metal to rain down onto the wafers in a thin sheet.
The most complicated and dangerous machines in films are the ion implanters, where phosphorus, boron, and arsenic are fired into the wafers. Maintenance crews that work on them are screened for arsenic poisoning once a year, and the operators who run them wear badges that measure radiation levels. Nunu and I went into a maintenance area to look at an ion implanter. Inside the machine was a large metal wheel that holds the wafers as they are implanted. When the machine is operating, the wheel spins at 1,200 revolutions per minute to make sure the ions are distributed evenly, and it goes so fast that the individual wafers become a shiny blur. To create ions, gaseous forms of elements such as phosphorus, boron, or arsenic are piped into the machine, where electrodes excite them until some of their electrons are knocked off. The remaining molecules—or ions—have a charge, and the direction and speed of their movement can be governed by an electrical field. The ion implanter uses such a field to accelerate the ions to high speeds and slam them into the wafers. Once implanted, the ions alter the electrical properties of the wafer, transforming electrically neutral material into a good conductor of electricity.
The Detectives
In the forty to eighty days that it takes for a wafer to make it all the way through the assembly process, as many as fifty layers of material will be grown and patterned and etched. The finished wafers are sealed in a protective glass coating and carried down to the testing area on the factory’s first floor in a dumbwaiter. There the chips are fired up with electrical current to see if they work. All the components of the chip are wired to metal pads that rim its border, and the testing machines apply current to these pads through minute needles while a computer evaluates the chip’s performance. When I was there, one screen showed a problematic wafer on which only 20 percent of the finished chips were working. “Must be PowerPC,” quipped Russell Reed, a wise-cracking engineering student from New York who was the training coordinator who showed me around the test floor. The PowerPC chip goes through twice as many steps as some of the other devices that MOS 11 makes, and not surprisingly, the factory’s yields on the PowerPC are lower than those for less complicated chips. The finished wafers are sent to assembly sites around the world, where they are cut into individual chips with a diamond saw.
Whenever operators on the test floor discover wafers with chips that fail in a regular pattern—sometimes they have “dead edges,” when all the chips at the border fail, or “dead centers,” when the chips in the middle don’t work—a team of device engineers will try to figure out what went wrong. (One of the greatest sources of contamination is the factory’s equipment, which needs oil and grease to function and can have parts that flake.) Every morning during the 8:30 staff meetings, device engineers report to the factory’s managers about the latest yields of every chip that MOS 11 makes. One morning, device engineer George Campbell told the assembled managers about problems with a certain kind of memory chip. In essence, too many of the chips had poorly defined circuits, and current wasn’t flowing where it should. “We had high leakage fallout,” Campbell told the group. In the days to come, Campbell had to figure out why this was happening.
Campbell and the other device engineers are like detectives. They spend a lot of time poring over data from the testing machines, and they spend a lot of time staring at chips under microscopes. Sometimes they will painstakingly remove one layer at a time from a wafer, which is called deprocessing, in search of the defect that has caused chips to malfunction. Because the causes of failure are almost always as small as the components themselves, it can be maddeningly difficult to discern what went wrong. Once they think they have isolated the cause of a systematic failure, the device engineers will repeatedly run batches of wafers through the step where they suspect it occurred, to determine whether their hunch is correct. If it is, then the assembly process is changed so that the source of failure is removed—hopefully without causing any new problems. The trick is to do all this in a matter of days. “The whole time, either part of the factory has ground to a halt and is waiting for you to finish the work, or else they are running with unacceptably low yields,” said Chris Magnella, the process engineering manager. “We can’t solve a problem in months—it isn’t cost effective.”
As soon as MOS 11’s employees get really good at building chips of a certain scale, Motorola asks them to make devices with even smaller components, to keep up with the semiconductor industry’s perpetual revolution. That allows the factory to jam more transistors and capacitors onto the chips, making them faster and giving them more power. When MOS 11 was first built, the factory warmed up by making chips with features that were 0.8 microns in size. Now MOS 11’s managers are concentrating on producing higher yields of chips whose component parts are 0.65 microns, while they also build chips with features that are only 0.5 microns and push the factory to start building components with parts that are even smaller. “You just keep working on the product,” said Bill Walker, MOS 11’s creator. “You tweak the process and identify what we call the manufacturing window, and then you get the process established to the point where it’s well controlled, so that you know you can repeat it, time after time, for ever and ever.”
My Library
