SOMEWHERE BETWEEN OPENING presents and the post-Christmas eggnog hangover, we had the opportunity to tour the Micron DRAM and Flash memory fabrication facility in Manassas, Virginia. While a fab is a fab and the net result, like anywhere else, is a groovy 300mm platter of chips that you cannot eat, we’ll walk you through Micron’s production process, profile the company, and give you a bit of insight into the challenges it faces and its opportunities in the near future and years to come.
At the risk of dating myself, my first encounter with Micron occurred circa 1996 when my family purchased a Pentium 100, Micron Millenia computer system to replace our decade old 286 IBM PC-AT. Micron has since exited the personal computer business, but its lineage goes back even farther to 1978. The company was founded by four partners in Boise, Idaho who rented out the basement of a dentist’s office to design DRAM chips. While originally funded by some small-time local businessmen, the partners got their big break when potato tycoon J.R. Simplot threw down enough dough to fund the building of their first DRAM fabrication facility in 1981. To this day, merely pointing out the obvious progression from potato chips to microchips is enough to get you beaten with a silicon ingot and tossed out the nearest airlock.
Micron is unique in that it is the last remaining DRAM manufacturer in the United States. Throughout the 1980s and early 90s, an onslaught of cheap Japanese DRAM was enough to force the other major players (including Intel) to sell off or board up their DRAM operations in the states. As the company was focused on its core mission of providing cost effective DRAM, it was able to withstand the assault and pick up the pieces of some of its competitors. While Micron has fabs around the world, its U.S. roots and “in-house” manufacturing capacity make it an obvious choice for sensitive US government contract work. To date, most government fabrication orders go to competitors such as BAE Systems, but Micron is actively flaunting its home-grown pedigree to Washington in hopes of landing some lucrative contracts to spur growth in the future.
While Micron is primarily a fabrication company, they do own several subsidiaries that offer up retail kit you are likely familiar with. Crucial, Lexar and SpecTek are all brands under the Micron tent that can be purchased just about anywhere. In the last couple years Micron has been on an acquisition spree purchasing stakes in firms like Inotera, Displaytech and Numonyx.
The Fab Tour
Upon pulling up to front door of Fab 6 in Manassas, you are greeted by an interesting trio of buildings. To your right is a large, white, modern looking facility (the fab) with a column of high-tech duct work jutting out of the far-right side. In the center, linked to the fab by a walkway, is a brown office and shipping complex with several industrial looking stacks protruding from the roof, and to your left is what appears to be a large mobile home. According to our guide, the mobile home was actually a cafeteria, as the need to expand production at the fab has consumed what used to be the lunch spot for the company’s bunny-people. Fab real estate at the Manassas facility is a precious asset as nearly all available space is now occupied with equipment, and future expansion is questionable to impossible without building additional facilities.
Regrettably photography was not allowed inside the fab, but for a general (if dated) idea of the fab, there are some images posted here. We began our tour in a small office room as our guide tossed two 300mm wafers onto a center table (one DRAM, one flash) and explained the process of coating and etching the wafers with the assistance of a university recruitment diagram.
The process begins in a far off land where mystical beings grow silicon ingots. While much could be written on this process, the quick and dirty version of how it works is as follows. Crystalline chunks of Silicon are melted down in a crucible at around 1400 degrees Centigrade and mixed together with a small amount of “doping” agent, often Boron, Phosphorous, Antimony, or Arsenic. A silicon “seed” is then lowered into the mixture, spun, raised, and dipped again. The process is not unlike that used in candle making where you lower a wick into a vat of melted wax until you achieve a desired diameter. As the molten silicon cools, its molecular structure aligns in such a way as to make it an ideal semiconductor for creating modern ICs. Once the ingot reaches a desired length and diameter it is cooled, then ground down to the appropriate diameter (300mm in our case).
After the growth and grinding are completed the ingot gets sliced up, lapped (to remove saw marks and defects), polished, and inspected to be shipped for production. This is where our journey picks up at, as Micron does not grow and cut its own wafers on-site. During our office-room briefing the question of a wafer’s value was asked. The response that it was confidential information was unsurprising, but we were given a few interesting off-the-cuff estimates of a wafer’s value. As the blank wafers arrive to the fab they are worth a few hundred dollars according to our guide, but after the fabrication process is completed we were told that we could go down the street to the Toyota dealership and purchase a couple cars with that wafer.
Now whether we’re talking about two brand new, fully equipped Avalons, or a couple used 1985 Tercels is unclear, but we imagine the truth lies in the middle of that range. Assuming you can fit 1000 working dies on a wafer (a generous estimate) at current prices a wafer of 8Gb chips would be worth somewhere around $5,000-$8,000 USD and a thousand 64Gb chips would likely fetch between $8,000 and $20,000 depending on trading conditions, but in a back alley such a wafer might fetch even more as competitors would love to have such a specimen to dissect. We were told an amusing anecdote about an unnamed employee who brought home such a wafer to adorn his walls, where during a house party his boss noticed the décor and fired him on the spot. As far as we can tell, the moral of the story was to never invite your boss to a house party.
Resuming our tour, we were escorted down a long corridor to the fabrication area. The walls of the corridor were peppered with photos of famous visitors to the facility, with former president George W. Bush being specifically pointed out. At the end of the corridor, seemingly guarding the entrance to the fab, was a manikin dressed in a white bunny suit that the employees affectionately refer to as “Chip.” Once past the aptly named guardian we were brought down yet another corridor, this time with many windowed doors to peer through. On the other side of the doors were the massive amounts of equipment responsible for scrubbing the air, processing waste materials from the various processes, and maintaining proper temperatures for all the equipment. We were told that this was all in fact the underbelly of the fabrication plant and after a few more hallways full of windowed doors we were led up a flight of stairs to where the magic actually happens.
Once the cut and polished wafers arrive at the fab, they are placed into transparent orange cartridges internally referred to as “FOUPs” (Front Opening Unified Pod.) Each of these FOUPs hold 25 wafers which are then grabbed by a robotic system of tracks and carriers suspended from the ceiling throughout the fab and delivered to various machines for fabrication. The fab itself consists of rooms that essentially perform the same few tasks over and over again: coat the wafer with a conductive or insulating material, scan it, polish it to make sure the wafer surface remains perfectly flat, blast it with UV radiation through a mask to etch out circuits, polish it again, rinse, and repeat until all the layers are in place. We were told the entire process can take between 45 and 90 days depending on what is being fabricated.
The various fabrication machines vary greatly in size from washing machine sized air scrubbers and polishers to a city bus sized 20-something nanometer immersion lithography scanner. No exact process size was given for fear of revealing some secret, but we were told the bus sized machine was currently the most advanced of its kind in the world and that Samsung could suck it (ok maybe we paraphrased that last part.) Unfortunately for consumers most NAND manufacturers are touting “20nm-class” products, which tells us virtually nothing. About the only information we can glean from this lack of forthrightness is that at 20nm, etching options become so limited that divulging such information would give away valuable clues about the wavelengths of light and materials the process uses. It is however known that Intel and Micron have partnered to make 25nm flash products together, so that is likely the ballpark we are talking about for this city bus machine. The currently held belief in the industry is that Samsung, Hynix, Micron and Toshiba are producing NAND flash at 27nm, 26nm, 25nm, and 24nm respectively.
Once the wafers have finished their long journey of etching and polishing, they are deposited at the testing facility where specialized machines place probes on contact pads to test the whole wafer at once, and then test each die individually. Each die is manufactured with redundancy so that a single defect in a die need not ruin the entire thing. For instance a flash memory cell may be marketed as having 32 columns, but in reality it is manufactured with 35, and if the testing machine discovers a defect in one column, it blows a fuse in the chip to bypass the defective column and one of the built-in extras automatically picks up the slack. This process is used industry wide and greatly increases the all-important yield percentage.
After testing is complete, the wafers have completed their mission at Micron’s Manassas fab. They are packed up, complete with the required information to identify bad dies on each wafer and shipped out to customers. The dies will later be cut up, packaged, and turned in to various products but these are all processes that happen elsewhere.
Micron, Today and Tomorrow
We got the distinct impression that Micron has a bad case of product envy directed at Samsung. Sammy is viewed as their number one competitor and perhaps even a role model for a couple main reasons. First, and most obviously is that Samsung is currently the number 1 manufacturer of NAND flash memory in the world, a title that Micron no doubt would love to steal from them. The other is that Samsung’s diverse product portfolio allows them to use the chips they produce in actual consumer products from phones to televisions. This is where Micron wants to be in the future, and their current foot-in-the-door consumer products are in the SSD, DRAM, and pico-projector markets. In addition, they believe that diversifying their product portfolio will allow for added security in the event that one specific market such as flash or DRAM crashes.
Given their history of survival attributed to one product focus, and failed consumer business units such as their PC venture it might seem counter-intuitive that Micron wants to diversify their own portfolio now. The reality is that they do not have much of a choice. Short of winning some massive government contracts, their revenues are closely tied to wild fluxuations of the DRAM and flash memory markets, which is not a spot any number-crunching corporate type would want to be in. The emergence of SSDs in popular computing has given them an opportunity in the form of their (Crucial’s) RealSSD line of drives, but Samsung is also playing this game with its excellent 470 series consumer SSDs. Because Micron and Samsung produce their own flash memory, they are somewhat insulated from the effects of the open memory markets, but Micron has its eyes set on other non-storage related markets as well. The other harsh reality is that in the coming years we will reach the limits of silicon photolithography which may stagnate the semiconductor industry and force a paradigm shift in how ICs are designed and built. To quote one of my favorite guilty pleasure movies, Antitrust, “Those who do not innovate are doomed to die.”
In 2009 Micron acquired DisplayTech to snag their FLCOS (Ferroelectric Liquid Crystal on Silicon) technology which allows them to essentially fabricate a projector onto a piece of silicon. This enables the production of cheap pico-projectors which can be embedded into portable electronics such as laptops or cellphones. The hope is that such devices gain popularity with the everyday consumer and that Micron (or a subsidiary) can create branded products directly instead of solely relying on third party sales of the chips for revenue. On the tour we were told that Micron is looking for its “iPhone,” a breakout product to put them on the map in the consumer space and perhaps pico-projectors will be their first attempt at such a device.
Going forward, Micron faces some formidable challenges on the world stage. Samsung is the second largest of Korea’s “Chaebol” companies, (companies run by private families, but aggressively financed by the government and given virtual monopoly status to ensure growth.) All else being equal this gives them a huge advantage if things get rough, whereas Micron has an uphill battle on their hands to prove to the US government that it is “too big to fail.”
Even inside the United States it is worried internally that the government might look beyond Micron for high security contracts due to the extreme diversity of those employed by the company. Because of the growing educational gap between the United States and other countries, and the specific technological focus of countries such as India, Micron finds many of its highly skilled employees abroad which may sadly make them look less desirable for projects involving national security despite being the sole U.S. DRAM manufacturer.
Finally, Micron is coming off of three straight years of losses to post a net income of 1.94 Billion in FY2010 and if it can continue that momentum into 2011, things might be on the up and up in the near term. Even if pico-projectors don’t pan out to be their “iPhone” product, increased demand for flash memory and DRAM in the coming years should buy them some time to continue the search. S|A
EDIT: Spelling of “FOOP” corrected to “FOUP”