My theory on why there will not be another semi cycle
Why there will not be another "chip" cycle
Presently, some analysts think there will be a chip glut in 2002-3. Indeed, Salomon Smith Barney claims that the semiconductor industry will peak in 6 to 9 months. The slow down is a result of over production of chips. In the past, cycles were occurred when numerous IDCs simultaneously built next generation fabs. When these fabs came online, there was a sudden increase in production capacity, resulting in a glut of chips. Clearly the chip glut was preceded by large semi equipment spending. A dropping book to bill ratio typically signaled the beginning of the glut. Chip companies are in a price war and they fear losing market share to a competitor. Thus if a competitor is building a new fab to produce a similar chip at 30% lower cost, others must follow. Interestingly, it was also said that the glut was never caused by a drop in demand, but rather by over capacity. I don't agree. The next chip glut will occur only when the global economy sinks, not by over capacity. Here are the arguments:
Fab throughput (supply not as high as feared)
It is true that the next 300mm wafers with 0.13µm gates can contain approximately 5x the chips that the present 200mm 0.18µm wafers carry. Yet it is also true that because of the single wafer processing, and more process steps, that same 300mm wafer will take longer to process. I have not read an estimate of the total increase in throughput of a 300mm fab, ie chips/minute. Clearly it should be higher than 200mm, but probably not 5x.
Sudden supply increases vs. gradual increases (supply to grow smoothly)
In the past, Sematech was indirectly a major cause of the cycles. Sematech kept all of the semi industry on the same technological page. When the next generation of chip technologies came to fruition, all the companies had access to it simultaneously, and the race to put the technology to use in new fabs was started. The result was a simultaneous build up of chip capacity. When these fabs came online, there was almost a step function in chip production capacity. Sematech does not have everybody in step anymore. There is more dissention and variation in technology adoption and realization. Not all the chip companies have embarked on 300mm, and those that have are on different time schedules, and can halt 300mm build out should over capacity raise its head. Additionally, there is still much life in 200mm Fabs. Fab manufacturing strategies and technologies (more on this later) developed for 300mm can be/is applicable to 200mm, boosting capacity and lowering prices. It is estimated that tools in 200mm fabs have an 35 -50% operational efficiency. Since tools in 300mm fabs are so expensive, the operational efficiency of each tool must be increased with good scheduling software. These same systems that help 300mm fabs can and are used in retrofitting existing 200mm fabs as well as in new 200mm fabs coming on line in the near future. 300mm fabs will also take much longer to bring on line. While this all points to increased capacity down the line, the various approaches, strategies and time schedules for increased capacity, indicate that general production capacity will increase in a more smooth and gradual manner. This is beneficial since it gives fabs more time to react to changes in demand and prevent overproduction. If supply appears too high in 2001, some fab construction and equipment purchase could be delayed. Additionally, it has been pointed out by Sue Billat at Robertson Stephens, some semi equipment is in tight supply and will naturally slow down capacity increases.
An increasing diversity of market segments and chips (different supply and demand)
When analysts speak of past cycles, they are primarily referring to DRAMs & CPU's. This is because, until recently, DRAMs and CPUs dominated the semiconductor market. If Intel produced too many CPUs or the demand for computers dropped, the result was a chip glut. Today, there are many more chip types and markets. To name a few, there are new system on a chip (SOC), flash memories, cell phone baseband chips, embedded systems, reprogrammable logic chips, RF chips, Bluetooth chips, power conditioning chips, DSP, MP3 chips, CCD camera chips, GPS, automotive chips, etc. These new markets and chip types now constitute a large portion of the chips produced. The current estimates of market size are PCs 25%; communications 22%, Consumer 16%, Automotive 6%; and other is 31%. CPUs and DRAMs are no longer the dominant chip types. There may be slumps in DRAM for example, but demand for Bluetooth, flash, and high speed CDMA data chips could be in high demand then. When DSP chips slump, embedded processors, CPUs, and DRAMs might be hot. The auto industry is adopting many new chips for cars (Drive by wire brakes and throttle, anti-collision systems, and intelligent transportation initiatives). Industrial manufacturing, energy production, and other "old" industries are going through a renaissance as information technologies and chips give them new efficiencies. New MEMS sensors and devices are taking off. Technology is mushrooming. New ideas are flourishing and creating new capabilities. New capabilities means new markets. Each new innovation spurs advances in numerous other areas.
Cheaper processing power means demand for more chips (trend for increasing demand)
Along with system on a chip, there will soon be system on programmable chip (SoPC). Why are these developments important? Because as processing power becomes cheaper, computation and intelligence can be pushed down to the functional hierarchy. In other words, at a relatively low cost, simple devices can be empowered. This is not confined to the consumer space, but it is easiest to illustrate - consider the lowly vacuum cleaner or microwave. It may seem excessive to current home owners to have a smart microwave that, responds to voice commands, can learn how hot one likes tea, or knows exactly how to weigh and defrost a frozen chicken. But microwave manufacturers will try to differentiate their products by making them smarter as long as the costs are not too high, which can be accomplished by using system on a chip, reusable cores, and mixed signal chips. How about an autonomous vacuum cleaner that roams around the house when no one is home? These are mundane, old economy consumer items, but by pushing computation and intelligence down to lower functions, new markets and products are created. This principle exists in other areas as well, such as manufacturing, distribution and warehousing. It would be interesting to observe how much computing power is required to produce a new Pentium in a 300mm fab. In all, the demand for new chips will clearly increase over time.
New Markets (diversity in demand)
It is very difficult to know what new chips and technologies will be available in 2003, making assessments of chip demand difficult. It is quite possible, for example, that by 2003, silicon based integrated optical switches will be in large demand and foundry capacity will be in tight supply to produce these switches. Consider, how strange would an MP3 player be, less than 3 years ago? Perhaps MP3 chips are currently still a small market, but they are climbing as MP3s become more ubiquitous. "The shipment of systems with Bluetooth wireless connections will grow to 1.4 billion units by 2005 after the first products begin appearing on the market this year, predicted a new research report from Cahners In-Stat Group." In the business world, companies are developing internal nervous systems to coordinate internal activities (called knowledge management, ie need more chips), they developing external senses to better judge demand (still more chips, see B2B below), and outsourcing more none core operations, requiring additional coordination and remote monitoring (yet even more chips). This is in addition to the booming communications and the internet markets. The point is that there is a cornucopia of new chips and chip markets and applications - not just consumer applications either. It is very difficult to predict chip demand three years from now.
Uncorrelated semiconductor market segments (hedging risk)
Like never before, these new and different chips belong to different market segments. One might correlate CPUs and large DRAMs demand (and I would say the two are becoming less correlated), but large DRAM demand does not necessarily correlate with embedded processors, cell phone, or DSP chips. We already saw this recently with flash and cell phone chips. Demand for these chips blossomed while CPUs and DRAMs were laggards. Now cell phone chip industry is "supposedly" slowing down, but DRAM market is soaring despite what appears to be average CPU sales. Around the corner is Bluetooth which will spur another industry boom for small consumer chips with embedded ARM processors and embedded memories. Around the same time, Transmeta web pads will totally change the way consumers use computing power and spur another wave of semiconductor demand particularly in flash memories. In 2002-3, when DRAM is thought to be in oversupply (shortly after Infineon and AMD 300mm flash memory fabs come online), the high speed wireless data rates (CDMA) will be in full force and everything from additional servers, wireless appliance chips, communications chips, SOC with embedded ARM and flash, will be in demand. There could very well be chip gluts in particular segments, but even these will be minimized by B2B.
B2B (better demand and supply information)
Information technologies are being rapidly adopted in order to increase efficiencies. B2B is the aspirin for cumbersome procurement costs and inefficiencies all through the supply chain. Additionally, companies can now shop for lower prices more easily at internet marketplaces and exchanges. This has been the main driving force for adopting B2B, but in the bigger picture, I believe, this is not the main benefit of B2B. Its like saying that eyes are good for sensing if it is day or night out, so that a person can dress accordingly. There is so much more information to be gleaned in the online markets, auctions, and exchanges. Information about demand curves and price elasticity - previously thought to be mere academic notions. But online markets, auctions, and exchanges, and new tools such as futures (imaging buying futures contracts on pork bellies and on 256MB DRAMs) provide information and signals that will help industries scale production accordingly. If in 3 years, DRAM is becoming overproduced, DRAM manufactures could obtain the signals much earlier than in the last chip slump. Perhaps some might argue, the time to build capacity is so long, that getting an overproduction signal early will not help. But here too, the 300mm fabs are special. These fabs will be able to shift production to other chips. Future 300mm fabs will be nimble because 300mm fabs will be completely automated.
Blurring of foundries and dedicated fabs (supply flexibility)
Today, foundries already have to make different chips within a fab. They already contend with the inefficiencies of being a general production fab. Being general in any realm is more difficult than being specialized since specialists are usually better at any given task. Consider a general vs a dedicated fab, a PLD or FPGA vs. an ASIC, a normal human being vs. Deep Blue chess program, MS Excel vs. Intuit Quicken. So when is it an advantage to be a generalist? There are two conditions which push the competitive advantage toward generalists, 1) when the market/task is variable or changing rapidly, and 2) when the cost of being a generalist is not significantly higher than being a specialist. Both conditions are becoming extremely significant. Setting up a specialized fab to make DRAMs, for example, seems to make sense since the DRAM market is cut-throat and the manufacture of DRAMs needs to be as streamlined as possible. What happens, though, when there is a glut of DRAM (consider AMD/Fujitsu and Infineon fabs coming online simultaneously). The whole fab investment becomes a rope to hang by.
However, the 300mm fabs by design need to be highly automated. Doing so incurs higher cost and more complexity initially, as is true with any general operation. The 300mm FOUPs (carriers which store 25 wafers) are too heavy for humans to carry from one tool to the next. Thus by design, the wafer transport within the fab will be automated using robots, monorails and other transport schemes. Such robotic transport is expensive, and was not implemented in 200mm fabs. Since it is required in 300mm, the door is now open for complete fab automation, including batch and wafer tracking, fully integrated transport, routing, scheduling and tool usage optimization software, tool drift, recipe management, and maintenance scheduling software, reticle management systems, etc. Such a system could handle multiple chip products with multiple recipes sharing similar and dissimilar tools. While having high initial costs and complexity (probably also extending fab ramp up times), such a Fab will survive well in a rapidly changing environment. Such systems can make the introduction of a new process for new chips almost painless. Such a system which handles and tracks nearly every function of the fab including tool drift and recipe modifications, can handle low priority engineering runs of new chip processes with minimal interference on the current high priority chip output. 300mm fabs are already being designed using a ballroom rather than the traditional bay-chase layouts. Ballrooms are much more open and amenable to various processes. Thus in the case of the DRAM fabs, the fab could also make other chips which do not correlate with DRAM demand. Should DRAM prices fall, a few new tool purchases and a few new processes in the fab automation software, and the result is a reduced DRAM output (stabilizing DRAM supply), and increased production of another uncorrelated chip, ie DSP, cell phone, or Bluetooth chips. Foundries are already doing this to a large extent in response to customer demand, and this already helps prevent gluts in particular chip area. In the future fabs, both foundry and IDMs, can adjust production between various chip types based on market signals acquired through B2B exchanges.
Conclusions
While it is technically possible to have another chip glut, it is improbable that the semi cycle will continue. There is now a diversity of markets, new production capacity will come online more slowly than in previous cycles, there will be better market information, new chip markets will drive new demand, and new flexible 300mm fabs will be able to adapt to demand and supply in different chip markets - preventing overproduction. |
