The semiconductor chip industry has been in deep recession for the last few years, but the last year has been especially bad. Recent reports have revenue down 30 per cent from last year. In an industry with huge capital investments, and excruciatingly thin profit margins, this constitutes a disaster.
A semiconductor wafer is a round disk made from silicon dioxide. This is the form in which batches of semiconductor chips are manufactured. Depending on the size of the individual chip and the size of the wafer, hundreds of individual semiconductor chips may be made from a single wafer. More complex chip designs can require more than 500 process steps. After the wafer has been processed, it will be cut into individual die, and these die assembled into the chip package. These assemblies are used to make build computers, cell phones, iPods, and other technology products.
Transitions to larger wafer sizes have always been a normal evolution of the semiconductor industry. In 1980, a modern fab used wafers that were only 100 mm in diameter (1 inch = 25.4 mm). The transitions in the 1980s were in increments of 25 mm. Motorola MOS 11 in Austin (1990) was the first 200 mm fab, and this was the first time that an increment had been skipped (175 mm).
WAFER SIZE
Metric English (approximate)
100 mm 4 inch
125 mm 5 inch
150 mm 6 inch
200 mm 8 inch
300 mm 12 inch
450 mm 18 inch *
* proposed
It has always been a challenge to be an early adopter of a new wafer size. The larger surface area makes it more difficult to maintain process consistency across the wafer. Often the process tool vendors will be late to transition, and lose market share. Lam Research (LRC) grew tremendously at the transition from 125 mm to 150 mm, since their largest competitors at the time, Applied Materials and Tegal, did not offer tools at the new wafer size. Intel and AMD were the first two chip companies with 150 mm fabs, and both companies had little choice but to choose Lam. LRC quickly grew and permanently acquired the market.
Another factor in the transition to larger wafers is process technology. When the semiconductor industry moves to a new wafer size, the newest process technologies developed by the tool companies will sometimes be offered only on the largest wafer size tools. If a chip company wants to remain on the leading technology edge, it can be more difficult if it does not manufacture with the newest wafer size.
The last wafer size increase occurred in 2000 with the first 300 mm volume chip production facility. This was built by Infineon in Dresden, Germany. At the time, 200 mm wafers were the standard. It may not sound like a large change, but a 300 mm wafer has 250 percent more surface area than a 200 mm wafer, and surface area directly relates to production volume.
By the end of 2008, worldwide, there were 84 operating 300 mm fabs, with 14 more fabs expected online by the end of 2009. Fab is short for "fabrication", and is what the semiconductor industry calls their factories. In the second quarter of 2008, 300 mm wafers fabs passed 200 mm wafers fabs in production volume.
A 300 mm fab is substantially less expensive than a 200 mm fab for the same capacity of chip production. Intel estimates that they spent $1 billion less on 300 mm capacity in 2004 than the same capacity would have cost instead by building 200 mm wafer fabs.
The problem is many small and medium size companies do not need the volume of production that a 300 mm fab generates, and they may not be able to afford the expense for a 300 mm fab ($3-4 billion). It is not reasonable to invest this amount of money and not fully utilize the fab. Since the 300 mm fab is inherently more efficient than the smaller diameter wafer fabs, there is pressure for a solution.
For the small and medium size companies, the solution has often been to close their manufacturing facilities, and hire a third party with a 300 mm fab to manufacture their product. This is what is known as going "fabless", or "fab-light". The companies that perform the third party manufacturing are called foundries. Most foundries are in Asia, especially Taiwan.
Ironically, 300 mm was developed by Motorola and Infineon at a project called Semiconductor3000 in Dresden, Germany. This was a small pilot line that was not capable of volume production. These two companies have suffered with their peers from their lack of fore-sight. In 2000, Motorola operated 18 fabs and was the 5th largest semiconductor company in the world. Today, Motorola has divested their manufacturing into a company called Freescale that now operates just 6 fabs. Infineon divested their manufacturing into a company call Qimonda. Qimonda has filed for bankruptcy.
Companies like AT&T (Lucent), LSI Logic, Hewlett-Packard and Xilinx have already eliminated chip manufacturing. Companies like Texas Instruments and Cypress Semiconductor have set paths for the eventual elimination of most of their fabs. AMD (GlobalFoundries) and Motorola (Freescale Semiconductor) have separated their manufacturing divisions into independent companies, and profess a plan to be free of fabs. Even Intel outsources its newest hot product, the Atom (used for "Netbooks"), to a foundry.
More than half of the fabs in operation at the beginning of the decade are now closed. With 20-40 fabs closing every year, there is a glut of used production tools on the market, most selling at bargain basement rates.
Recently three of the largest semiconductor companies, Intel (microprocessors), Samsung (memory), and TSMC (foundry) have been planning a transition to 450 mm wafers. A 450 mm fab should have approximately the same advantage over a 300 mm fab, that a 300 mm fab has over a 200 mm fab. It is undoubtedly a strategic decision to create a situation where other-than-huge companies will be at a competitive disadvantage. Intel had $12 billion in the bank at the end of 2008. Can AMD (GlobalFoundries), or comparably sized companies, afford a 450 mm fab ($6-10 billion)? No.
If the industry continues to progress along the current path, competition will disappear. The largest memory manufacturer will control memory, the largest microprocessor manufacturer will control microprocessors, and the foundry business will be controlled by one company. These companies already have advantages of scale over their competitors, but their existing manufacturing advantage will grow significantly.
No comments:
Post a Comment