Economic Forecast: Slowly Turning Upward

The semiconductor manufacturing equipment market should turn upward in 1999 and return to health in 2000.

Overcapacity has driven DRAM prices downward, even though demand for memory has continued to grow at ~75% per year. Demand is expected to absorb this overcapacity by mid-1999, though forecasts vary. DRAM prices have started to go up again, giving the initial signal of an upturn.

The cycles historically have been very predictable (Fig. 1). When possible, it makes the most sense to build a new fab, so it is ready for full-flow production right when the market is starting a boom period. Unfortunately, this would require spending hundreds of millions of dollars at a time when the market is down, like the present.

The practice of waiting for the market to be up for a number of months before committing to an expensive new fab project would normally be considered prudent. However, it limits the time that the new fab can be used to take advantage of that up market. The advantage to starting a new fab project when the market is down, however, is only available to companies that can afford to do so.

Another variable that comes into play is the specter of 300 mm technology. A common conception is that building one of the first 300 mm fabs would be too risky, and building one of the last 200 mm fabs would be committing to a facility that will become too obsolete to pay for itself. It is very possible that many companies will put off adding to their own capacity. This would lead to an undercapacity situation, raise IC prices and set up the next boom period. Unfortunately, these companies would then build enough new fabs just in time to set up the next overcapacity condition.

Fig. 1. The semiconductor industry follows predictable economic cycles. (Source: IC Insights)

Bill McClean, IC Insights 'There is nothing wrong with demand; it is all supply and pricing issues. You can kill a good market. If your market is doubling every year, that is only part of the story. The other part of it is how much supply is growing, and if supply is growing faster than demand, whatever demand is, there will be tight margins and tough pricing. Currently, bit volume usage is growing faster than in the boom periods of 1993-1995. The only difference is capacity and pricing, and that is really the whole story of the market.'

George Burns, Strategic Marketing Associates 'We are hitting the bottom of the cycle of new fab activity. I expect that new fab construction activity this year will increase by as much as 20% over 1998. We will get some increased spending from that. I do see signs of a DRAM recovery. As a result, I think that we are going to see increased capital spending, even though we are seeing people drop out of the DRAM business. Plus, there will be increased foundry spending. Put all that together, and I think capital spending in 1999 will increase by about 10%, which will be back loaded rather than front loaded.'

Risto Puhakka,VLSI Research 'It is going to be a pretty mild recovery. We think on the equipment side that the industry will be pretty much flat compared to 1998. On the semiconductor side we see a slight growth rate compared to 1998. I think it is going to grow about 5.4%, after a decline of about 12% in 1998. To really get reasonable growth rates, we need to go all the way to the year 2000. Beyond that, we do expect the markets to really pick up again, looking at the long term forecast. We do not think that it is going to be a really strong recovery until we get to the year 2000.'

John Schuler, SEMI 'I think what has happened is that all of the companies have lost sight of basic Economics 101, and they have all overproduced. The technologists have done an outstanding job of decreasing linewidths. That is what has pushed out 300 mm, and to a certain extent the over production has also pushed it out. In addition, I think equipment companies are trying to recoup their 300 mm costs by plowing the technology, automation and production efficiency they developed for 300 mm back into 200 mm machines.'

Sue Billat, BancBoston Robertson Stephens 'We expect the recovery will be selective and technology driven. Chip makers will look beyond being able to execute shrinks and start looking at other ways to enhance their productivity and products. I think we are going to see a lot of activity in copper, and CMP will continue to outperform the overall sector as memory makers convert over to CMP. I think that companies in the test area are going to lead more than they usually do coming out of the downturn. We expect to see a technology driven bookings recovery in 1999 followed by capacity purchases in 2000.'

Gunnar Miller, Goldman Sachs 'We are looking for 10% growth in the semiconductor market in 1999. However, our view is that equipment and capital spending is likely to remain flat in 1999, though there is a lot of variability in that number. I think there are two megatrends in the general outlook. One is increased capacity outsourcing to the foundries. Number two is the loss of what we call sovereign drivers. Sovereign entities are not giving direct or indirect subsidies to semiconductor manufacturers like they used to; while this is likely to increase the efficiency of the industry in aggregate, it could also result in a drop in global capital intensity of the semiconductor industry.'

Frank Dickson, Cahners In-Stat Group 'You are going to start seeing some recovery next year. We are actually saying that 2000 is going to be the peak, but there will still be double digit growth in 2001 and 2002. Whatever happens to the semiconductor device industry happens exponentially to equipment suppliers. So, I do not think you are going to see any firm improvement in the machine book-to-bills or in equipment sales until the third or fourth quarter of 1999. The only way to lessen the impact of the boom-bust cycles is through deliberate planning, and as the players get bigger, you will see more deliberate planning. You are still going to have boom-bust cycles though; it just cannot be helped.'

Clark Fuhs, Dataquest 'We see overcapacity continuing in the DRAM area until late 1999, early 2000 and in the foundry area until the middle of 2000. We are entering into a 0.18 µm technology spending pattern that will provide an uptick to the equipment market over the next couple of quarters. This 0.18 µm migration is less broad and a little bit smaller than the 0.25 µm migration was in early 1997. So, the uptick and potential recovery here is going to be slower than in the past. We are not going to hit an elbow to the curve until we start getting capacity buying near the middle of 2000.'

Carl Johnson, Infrastructure 'The equipment business is heading for a U-shaped recovery, and for the first couple of quarters in 1999 we will probably see more downsizing and reduction in earnings. There will be technology purchases for low- and high-k dielectrics and the migration to copper interconnects, and we will see a lot more use of planarization. The sector positioned to benefit most from movements to smaller feature sizes will be photomask manufacturing, given that 248 nm stepper technology will be extended using either OPC or phase shift technology. The lack of greenfield construction will really hurt the big automation companies; they will have to focus on retrofit business.'

One of the safety valves to protect against undercapacity for companies that wait are the foundries. Foundry fabs have primarily served fabless semiconductor companies, but now they are also becoming second-source fabs for traditional semiconductor companies. In some cases, foundry companies are even building fabs dedicated to particular customers. Foundries should be taken into account when evaluating overall capacity, and there is some speculation about how much of the current overcapacity is due to improper accounting of the foundries.

The tendency in the market is to overreact in both phases of the cycle. Companies tend to add too much capacity in the boom periods and cut back too much capacity in the bust periods. This deepens the boom-bust cycle.

Many analysts are forecasting that the semiconductor market will start turning up sometime in 1999. The semiconductor manufacturing equipment market may turn up by the end of 1999.

300 mm tools

Approximately $4 billion has been spent by tool manufacturers to develop 300 mm technologies, and return on that investment is not imminent. Tools for 300 mm processing will likely not be sold until late 1999 or early 2000. Some speculate that it will be even longer. Much of this is due to economic conditions for chip manufacturers.

There are productivity driven reasons to push out 300 mm processing as well. One reason has to do with planning, scheduling and execution on the fab floor. If a company was to use the same practices for 300 mm production that are used in 200 mm production, that company would probably go broke. Wafers will be too expensive to dedicate 10% of them for test and control purposes, and automation for enabling continuous processing has yet to be fully developed.

Another reason has to do with how the number of devices per wafer affects cost of ownership. Part of the planning process for executing wafer size increases is to keep the number of die per wafer above a minimum value. As IC devices become more complex, they take up more area. Not only does the number of devices per wafer go down, but the amount of wasted space on the wafer also goes up. In most cases, as long as each wafer is yielding 100 devices, a wafer size increase is not considered necessary, though it may be welcome.

The push to 180 and 150 nm processing will shrink die sizes and keep the die per wafer count up. This has likely been accounted for in 300 mm technology roadmaps. The increased use of CMP allows for more metal layers, providing a third dimension for adding complexity to a device. This may slow the horizontal growth in die size from the historical norm. Systems-on-a-chip (SOC) may offset these effects and make die sizes grow even faster.

When 300 mm fabs are built in larger numbers, one conservative strategy for 300 mm fab construction is to build it half the size of a typical 200 mm fab. It would provide the same capacity as a full size 200 mm fab at ~60% the cost, if the 1.3X price goal is met for 300 mm tools. This strategy might only be valid for a 300 mm facility under construction right now. By the time 300 mm fabs start to appear, a half-sized fab may not provide enough capacity to be justified.

Things to look for

As a general rule, back-end capacity is absorbed faster than front processing capacity. This is mostly because back-end capacity is measured in units of devices, and front-end capacity is measured in wafers. Also, lead times for back-end equipment are generally shorter than front-end equipment. Increased sales of back-end equipment, like ATEs, wire bonders and die bonders, are viewed as the sign that the equipment market will start to pick up.

Aside from the major forces that shape the semiconductor market, there are other things to look out for. One is flat panel displays coming down to the 2X price point compared to CRTs. If flat panels start replacing CRTs on desktop computers and other appliances, equipment companies with flat panel manufacturing technology may see more of an upturn.

If the most popular way of solving the Y2K problem turns out to be buying new computers, then demand for semiconductor devices in computers may absorb the existing overcapacity sooner than expected. There may be room for equipment companies to benefit from the Y2K problem, if IC manufacturers follow the example of General Motors in its dealings with EDS. GM recently offered EDS a monetary reward for making all of its systems Y2K compliant before Jan. 1, 2000, rather than threatening to sue if they were not.

Conclusion

The semiconductor markets operate on predictable cycles. Even with the current regional economic troubles, most of the reason for the downturn in the semiconductor industry is still overcapacity. The coming upturn may be slower than past upturns, but the coming boom period promises to be strong.