Rudy Rucker penetrates the temple of the computing machine - the clean room at the chip fabrication plant - and discovers the place where robots reproduce themselves.
I fell in love with the Silicon Valley word "fab" the first time I heard it. This short, moderne word means "chip fabrication plant." A manager might, for instance, say, "What kind of outs are we getting from the fab?" In the '50s and '60s, of course, fab was short for "fabulous," as in the detergent Fab, or as in the lines in "Bob Dylan's 115th Dream": "I ran right outside and I hopped inside a cab. I went out the other door; this Englishman said 'fab.'" Gear! Kicky!
After exceedingly many phone calls, I managed to get inside two fabs in Silicon Valley, one belonging to the chip giant Intel, and the other owned by Intel's small challenger, Advanced Micro Devices. AMD recently won a court battle with Intel over its right to produce the popular 486 processor chips for DOS- and Windows-based personal computers. Very much a "we try harder" company, AMD was the first to let me into a fab - a quarter-billion-dollar building in Sunnyvale, California, called the Submicron Development Center.
A micron is a unit of measurement equal to one-millionth of a meter. A typical human hair might be a hundred microns wide. The scale of chips is discussed in terms of the size of the smallest features of the patterns on the chip. Today's chips use features about half a micron in size; hence, they are said to be using "submicron" designs.
AMD's Submicron Development Center was originally intended to be purely a research facility, but the demand for AMD 486 chips is such that the facility is now also being used for commercial production. It turns out to be crowded and a bit hellish in the AMD fab, which feels about the size of a wide office-building corridor plus maybe six offices on either side.
Something I hadn't initially realized is that being a fab worker is like being any other kind of assembly-line worker. It's a rigorous blue-collar job. Most of the workers are Asian or Latino. The AMD fab is open 24 hours a day, every day of the year except Christmas - and in the Intel fab they work on Christmas, too. The workers pull 12-hour shifts, with three shifts one week and four shifts the next, for an average of 41 hours a week. Although some of the fab workers are highly paid engineers, starting pay for a simple technician is around US$24,000 a year, which comes to about $12 an hour.
What actually goes on in a fab? A fab buys blank silicon wafers and draws complicated patterns on them. This boosts a wafer's value from $200 to $30,000 or more. It's almost like printing money. The catch is that each of the many machines used in a fab costs more than a million dollars. And buying machines for a fab is complicated.
When a fab finishes a wafer, it is shipped to another plant, where the wafer is sawed up into chips and the chips are put into those familiar plastic cases with wires coming out. Such secondary plants are mostly in Southeast Asia - Silicon Valley fabs are solely concerned with printing the chips onto the wafers. To avoid dust, the wafers are shipped in vacuum-sealed bags.
The fab is a place for chips, not for people. People are dirty. Their bodies flake and crumble, generating showers of dust. One dust particle can ruin a chip, for instance, by shorting out the separation between two nearby submicron circuit lines.
With the current prehistoric state of robotics, there is no hope of fully automating a fab, especially given the fact that the process technology is subject to being changed over and over. So to deal with the necessary dirty people, the fab must be maintained as a clean room.
The room's cleanliness is specified in terms of the number of particles larger than one micron that can be found in a cubic foot of air. An average no-smoking restaurant might have a few hundred thousand particles per cubic foot. In a surgical operating room, the level is brought down to about 20,000. In the outer hallways of a fab building, the level is 10,000; in the wafer-handling areas of the fab itself, the level is brought down to one.
How? Here's the AMD procedure:
Dan Holiga, a member of the AMD corporate training division, is responsible for instructing new workers on clean-room procedures and for arranging science courses for them at local colleges. Dan leads me into the pre-gowning room. The floor inside the door is covered with sticky adhesive. I sit down on a bench and put some blue booties over my shoes so as not to track dirt into the locker room. The woman behind the counter can't find Dan's special fab badge, so she gives him a visitor badge like mine. We select building suits in our sizes: these are two-piece suits that look like tight-cuffed blue pajamas. The woman gives us each some white plastic shoes that resemble bowling shoes.
In the pre-gowning room, we stash our street clothes in the lockers and put on the blue building suits and the white plastic shoes. We wash our hands and put on hair nets and safety glasses. Dan has brought a camera with him. We walk through a corridor into the outer hallway of the fab building. This is the 10,000-particles-per-cubic-foot zone, and the air feels cleaner than any I've breathed in a long time.
We pass a break room where some of the fab workers are having nondusty snacks like apple juice and yogurt. Then we go into a second locker room. I thought we were already dressed for the fab, but that was just the start. The second locker room is the gowning room proper.
Here we put on latex gloves. Then we wipe off our safety glasses and our visitor badges and Dan's camera - wipe everything three times with lint-free, alcohol-soaked cloths. We put on white hoods and "bunny suit" overalls made of Fibrotek, which is a sandwich of nylon and Teflon. We pull "fab booties" over our shoes, and we put on face masks. We pull vinyl gloves over our latex gloves. This is starting to feel a teensy bit ... obsessive. I'm reminded of the "environmentally ill" fanatics you see in Berkeley natural food stores, shopping while wearing gas masks and elbow-length gloves. They'd love it here in the gowning room. But, I remind myself, this isn't about fanaticism here, this is about objective scientific fact: getting down to one micron-sized particle of dirt per cubic foot of air!
Now Dan leads me through the air shower, a corridor lined with air nozzles blasting away. We hold up our hands and turn around, letting the air wash us all over. The invisible particles still adorning our bodies fall to the floor, where they are sucked away. In the air shower and in the fab, the floors are coarse grates, and the ceilings are filled with fans. There is a constant flow of air from above to below, with any showers of filthy human particles being sucked out through the floor grates. The air in the fab is completely replaced 10 times a minute.
I step out of the air shower and, fully purified, enter the fab. As the Bible says, "I was glad when they said unto me, let us go into the house of the Lord." I am in the temple of the God-machine of Silicon Valley. The lights are yellow to avoid clouding the photo-resist emulsions. This gives the fab a strange, underworld feeling. Air streams past me from ceiling to floor. Other white-garbed figures move about down the corridor; all of us are dressed exactly the same.
On the sides of the corridor are metal racks holding boxes or "boats" of wafers waiting for the next stage of processing. The racks have wires instead of shelves - there are in fact no flat horizontal surfaces at all in a fab; such surfaces collect dust and interfere with the air flow.
The only hint of human contamination is the meaty smell of my breath, bounced back to me by the white fabric face mask I'm wearing. I wish I could tear off the mask and breathe the clean, pure air of the chips. But then I would exhale, and the wafers wouldn't like that - detectors would notice the increased number of particles per cubic foot, and lights would flash.
The layout of a fab is a single main corridor with bays on either side. To keep the bays clean and uncluttered, most of the machines are set so that their faces are flush to the bay walls, with their bodies sticking out into sealed-off corridors called "chases." Like people, machines have bodies whose exigencies are not fully tidy. The chases are clean only to a 10-particles-per-cubic-foot level.
As we move down the main corridor to start our tour, people recognize Dan and come over to pat him on the back or on the arm. Dan's theory is that in the clean room, people can't see each other's faces, so they touch each other to fill in nonverbal communication. Another factor could be that, since everyone is clean, there is no fear of getting yourself dirty through human contact. Or maybe it's just that you have fewer inhibitions with someone who is dressed exactly like you. In any case, the fab workers seem to have strong team spirit and a sense of camaraderie. They're like happy termites in a colony.
The craft of getting a hundred 486 or Pentium chips onto a silicon wafer involves laying down about 20 layers of information. It's a little like printing a silk-screen reproduction with 20 different colors of ink. At each step, a fresh layer of silicon dioxide is baked on, parts of the new layer are etched away, and metals or trace elements are added to the exposed areas.
As well as having to be positioned to an accuracy of one-tenth of a micron or better, the successive layers need to have a very specific thickness. Rather than being measured in microns, the thickness of the layers is best measured in nanometers, or billionths of a meter. Each layer is about 10 nanometers thick.
The process takes as long as 12 weeks for a completed wafer's worth of chips. It's not so much a linear assembly line as it is a loop. Over and over, the wafers are baked, printed, etched, and doped. At AMD, workers carry the boats of wafers up and down the corridor; at Intel's plant there is a miniature overhead monorail on which the boats move about automatically, like gondolas in a scale model of an amusement-park ride.
At AMD, I visit the etching bay first. There are a series of sinks filled with different kinds of acid piped up from tanks located on the story below the fab. In the bad old days, you could recognize fab workers by the scars on their necks from splashes of acid, but now there's a small industrial robot arm to dip the chips. I'm happy to see the arm; this confirms my sci-fi notion that fabs will ultimately be places where robots reproduce themselves: robot obstetric wards.
The acid baths are for removing the photo-resist masks after the etching itself is done. The etching is typically done "dry" - that is, a fine dust of ions is whipped into a frenzy with powerful radio frequency signals to make a submicron sandblaster. The idea is to dig out parts of the chip so that metal conductors and metal-doped semiconductors can be patterned in to make up the wires and transistors of the integrated circuit that the chip is to become.
The real center of a fab is the photo-lithography bay. Here the gel called photo-resist is sprayed onto the wafers, and then the wafers go into a "stepper," which is the machine that projects the circuit diagrams onto the wafer's chips. The projector is called a stepper because it projects the same image 100 times or so onto each wafer, moving the wafer in steps to receive each successive image. Steppers are the most expensive devices in a fab. The images projected by the steppers are found on transparencies called "reticles." Reticles are based on circuit diagrams created by engineers using computer-drafting techniques.
Once a wafer gets out of the stepper, a developer chemical removes the photo-resist that was exposed to light, leaving masks shaped like the dark regions of the reticle. This is a very efficient process, because although a reticle may have thousands of features on it, projecting its image onto the wafer puts all those features there at once.
The better the stepper, the smaller the images it can make. Smaller chips run faster, use less power, and can be produced in larger batches - more chips per wafer. In order to handle very small feature sizes, steppers need to use light with very short wavelengths - the current ones use deep ultraviolet, and, to get much smaller, the steppers will have to start using X-rays.
The light is mellow yellow in the bay with the steppers, where I find the most people clustered. This is the heart of the temple. Some of the workers are debugging a problem with one of the machines that sprays on the photo-resist; one of them is lying on the grated floor with a laptop computer. It strikes me that in this world, the floors are not dirty.
There are a couple of men with an electron microscope looking at wafers. One of them is holding a handful of wafers, some of them cracked. "I guess those ones are no good?" I ask. The man looks at me oddly and finally grunts, "Yeah." Seeing only my visitor badge and not my expression, he thinks I'm an executive being sarcastic, but Don explains that I'm a journalist. The guy warms up then and has his co-worker show me some wafers under the electron microscope. There's a nice, clear image on a TV screen next to the microscope. It shows something like your usual image of a chip but with lots of parts missing. This is just one or two layers' worth.
"These things," the man with the microscope says, pointing to some fat short rectangles, "we call these the hot dog buns. And these other things," he points to some longer, thinner rectangles overlaid onto the fat short ones, "we call these the hot dogs. We check that the hot dogs are on the buns."
We peek into a few more bays. One especially cute little industrial robot catches my eye. Jerky and articulated like a shore-feeding bird, it folds its tail and pecks wafers out of their cartridges to slide them into some machine's maw. It reminds me of the animated Disney film of Alice In Wonderland - the part where Alice is lost in the woods near the Cheshire cat, and a little bird that looks like a pencil with two legs comes running up to her.
Dan takes some pictures of me, and then we go out into the gowning room to take off our face masks, gloves, and Fibrotek suits. It feels good to get out of that suit; I was getting hot. Also, it's great to stop inhaling my own breath. It sure would be tough to have to spend 12 hours at a time in a fab. And for $24,000 a year! As a communist friend used to tell me in grad school: the secret of capitalism is that the less you get paid, the harder you have to work.
Now we're in our building suits again, and Dan wants to show me the sub fab, which fills the entire floor below the fab. As we go out into the building hall, a security guard in a clean-room suit runs up to us and asks our names. He writes our names on his glove; he's too excited to get the spelling right. He doesn't recognize Dan - we're both wearing visitor tags - and Dan is carrying a camera. Uh-oh. While the guard hurries off to make a report, Dan hustles me down the stairs to the sub fab.
The sub fab is a techno-dream. It holds all the machinery that supports the machines of the fab. The electrical generators are here, the plumbing, the tanks of acids, the filtering systems, the vacuum lines, the particle monitoring equipment - miles of wires and pipes and cables in an immaculate 10,000-particle-per-cubic-foot concrete room. This is the ultimate mad scientist's lab. I'm enthralled.
Now here comes the clean-room security guard again. "You have to come with me." Dan wants to take some pictures first. "You have to come right away." The clean-room guard leads us out into a hall off the sub fab. Three unsmiling uniformed guards are there. Dan explains about his lost fab badge; they phone the pre-gowning room to go into Dan's locker and check out his ID; finally they decide it's OK, and we're back on our way.
"They thought maybe we were from Intel," Dan says. "Someone who doesn't know me saw us taking pictures in the clean room."
When I'm finally out in the dirty real world again, I'm grateful and glad. It feels as if I've been in the underworld, in some sort of gray limbo that seems too mechanical, a world in which people are totally out of place. I don't feel like turning on a computer again for several days. But I'm happy to have seen the central mystery.
Two weeks later, Intel finally comes through with a fab tour for me as well. My guide here is Howard High, of Intel Corporate Communications. The fab layout is quite similar to AMD's, although Intel's fab is much bigger - perhaps the size of a football field, and with high 15-foot ceilings to accommodate the wafer-boat-carrying monorails overhead.
The vibes in the Intel fab seem more relaxed than at AMD. Intel is ahead, and AMD is trying to catch up. At Intel, for instance, I don't have to exchange my clothes for a building suit: I'm allowed to just put the clean-room bunny suit on over my clothes. Because of dust, I wasn't allowed to use any paper on my AMD tour, but Intel issues me a spiral notebook of lint-free paper.
Rants aside, the more I learned about the fabs, the more I was amazed that they work. The intricacy of the system is reminiscent of the complexity of a biological process like photosynthesis. Nobody could have designed one of today's fabs from scratch - these are giant industrial processes that have evolved, a step at a time, from earlier, simpler versions. There is a very real sense that these processes are the synthetic biology from which planet Earth's next great species may arise.
Even journalists have a tough time getting a fab tour - at Intel they told me the last person let in had been the vice premier of China. But Intel does have an informative museum near its fab, in the main building at 2200 Mission College Boulevard, off Montague Expressway, which is in turn directly off US 101 in Santa Clara (near San Jose). The museum is open to the public from 8 a.m. to 5 p.m., Monday through Friday. For info or tours, call +1 (408) 765 0503.
