A crisis of confidence has hit the chipmakers, but are their fears justified?
At the IET / GSA International Semiconductor Forum held in London last month, executives complained of the years of profitless prosperity they have endured since the collapse of the dot-com bubble at the end of 2000. Industry statistics show robust growth in units since 2002 and there seem to be few signs of a halt in the encroachment of silicon into everyday life.
Maria Marced, president of TSMC Europe, said: "Silicon is absolutely driving the growth."
Eric Mayer, vice president of production partner management at Infineon technologies, described how changes in automotive technology have increased demand for semiconductors. "In a full-hybrid car, we are talking about a 300mm wafer of semiconductors going into such a car," Mayer claimed.
But, as the amount of silicon shipped continues to climb, the chipmakers are facing immense problems trying to sell it profitably. Countries such as China and India expect to see car use rise. "What we hear from our partners in China is that they don't want to sacrifice on amenities. Their cars will have lots of gadgets onboard," said Mayer. At the same time, vehicles such as the Tata Nano put a severe upper limit on how much the car manufacturers will spend on semiconductors.
"The Nano has the promise of high-volume growth. But how can you bring it to the market for $2,000? That is a challenge for our customers," Mayer added.
What suffers is the money that chipmakers can collect for their products. The continuing decline of the average selling point - spiking upwards only occasionally in the last five years - has left many of the chipmakers feeling practically destitute.
Ajit Manocha, who runs operations for memory maker Spansion, remarked: "In this industry I very rarely hear about price going up. It is always coming down."
The chipmakers used to be sanguine about the crunching recessions that they suffered in the past, because good times were always around the corner, Manocha noted. "Ten years ago, if you met someone at a conference they might say there is a down-cycle now but it will recover next year," he said. The difference now is that people don't talk about the recovery part: just whether they might ever make some profit.
At the same time, it is simply getting more and more expensive to compete in the biggest markets. The decision by three of the largest fab operators to work on a move to 450mm wafers only served to underline just how expensive future process technologies will be to develop and run. For those able to sell in high volume, 450mm should offer lower costs overall. But the cost of entry is enormous. Even the fourth largest fab operator, Toshiba, said it would work to improve efficiency on 300mm lines rather than attempt to move to 450mm.
Design and marketing costs are soaring. Estimates vary on just how expensive chip design is for a 65nm or 45nm, but it is clear that it easily runs into tens of millions. Even though mask production costs are now measured in millions of dollars, those costs are dwarfed by the money it takes to get high density chips designed in.
Several years ago, Handel Jones, president of International Business Strategies, performed a survey that showed that the cost of developing software and reference designs for complex system-on-chip parts eats up the lion's share of the chip-development budget. These are costs that product-design customers took on ten years ago. In order to guarantee access to their target markets, the chipmakers have taken on much of the software development burden but have shown few signs of being able to charge for the additional value they provide. Instead, software and design-support has become just another part of the cost of staying in the market.
A further problem for the makers of high-integration parts is that they are finding fewer sockets into which they can sell their products.
Some contend that the pricing environment has to improve. "We expect to see some ASP rationalisation as profits are suffering," said Marced.
Malcolm Penn, president of analyst firm Future horizons, believes that the price recovery is underway and that the situation will improve thanks to cutbacks in capital expenditure by the chipmakers.
In the meantime, many executives have adopted a bunker mentality, believing that promises of a profit recovery are no more than wishful thinking. At events such as the International electronics Forum organised by Future horizons, executives worried about the state of the industry and if it is entering a shakeout phase.
In a maturing industry, consolidation seems inevitable. "In the top-50 companies, scale matters," claims Manocha.
Theo Claasen, vice president for business development at NXP, believes that the number of chipmakers - whether they are integrated device manufacturers or fabless operators - should shrink from the almost 500 in the market today to no more than 50. Manocha predicts much greater focus on joint ventures as well as consolidation as the chip industry wrestles with its maturation.
Yet, such consolidation has not taken place. If anything, the trends point to a different evolution. Mayer and Marced point to 'horizontalisation' as a more important trend in semiconductors in which vertically integrated IDMs shuffle off operations they cannot support full-time and simply buy in those services as they are needed.
The most obvious beneficiaries of the move away from vertical integration have been the foundries, principally TSMC, which is currently accelerating away from the rest of the pack in terms of acres of silicon it can ship each day. However, there are many other operations that the IDMs have been outsourcing, such as chip and semiconductor intellectual property (IP) design. Some have even proposed a model of the 'designless' company that only specifies what a given chip should do. Everything else is parcelled off to subcontractors. It has yet to happen in semiconductors but most of the pieces needed to make it happen are in place.
It is possible to dominate a market but have very little in the way of capital. The first wave of companies to embrace this model have done well, growing faster than the IDMs with which many of them compete. In some sectors, such as programmable logic, the fabless companies drove the IDMs out of the market at quite an early stage.
"The fabless model is accounting for the lion's share of the growth in the semiconductor industry. They have woven themselves in the top 25. They will displace the IDMs as the historical leaders," claimed Jack Harding, president and CEO of design-services company eSilicon.
One consequence of the rise of the fabless company has been not consolidation but differentiation. Less value is concentrated in the top 25 today than was the case at the beginning of the decade, according to league tables by analysts such as iSuppli and IC Insights. In 2002, the top 25 chipmakers accounted for more than 80 per cent of the total IC market: the other 400 or so took in less than 20 per cent of the almost $200bn of silicon sold that year. At the end of 2007, the top 25 accounted for just under two-thirds of the total market. The top ten slipped by a couple of percentage points: the companies at the top suffered from price wars in memories and PC processors. The average revenue for companies in the second 15 slipped from $1.9bn in 2001 to just over $1.5bn in 2007.
The big are not getting bigger, with only a few notable exceptions, such as Samsung. Instead, the industry is becoming populated with highly streamlined specialists.
"There is a change in the distribution of small companies in the market," said David Baillie, CEO of Cambridge Semiconductor.
"Just look here in Europe with the number of IDMs getting of out of the fab business," said Harding. "There is no doubt that the fabless model will outpace the IDM model."
Marced said companies are evolving away from the 'old Parthenon model' in which large IDMs had pillars each representing a vertical market. "They looked for synergies among the business units and the roof of the Parthenon was the sales and marketing teams." the temple's foundations were the manufacturing and design operation. "The new form we are seeing is with the industry completely horizontalised."
The foundations and roof are stripped back to the bare minimum, with many of the functions going to external suppliers. In some cases, which may form the trend for the next few years, the pillars are being sold off or augmented with small purchases in order to build a leadership position in a chosen vertical market.
In some of those segments, scale is likely to be important. For example, the huge market for cellular handsets is one where iSuppli analyst Francis Sideco believes a supplier has to have a billion-dollar business to stay in the game. It was that kind of consideration that led STMicroelectronics to take over NXP Semiconductors' wirelesschip operation.
Texas Instruments is focusing more on analogue-oriented parts and the OMAP applications processors rather than trying to extend its position in baseband silicon where a number of suppliers are trying to compete.
Ron Slaymaker, vice president and manager of investor relations at TI, told analysts in April: "As you see customers shift toward standard products in the form of their modem function or the digital baseband, our strategy more and more in wireless will be led by the OMAP application processor."
In other markets, scale will be less important than responsiveness. The result is likely to be that consolidation will take place in sectors where ultra-high integration is favoured but that specialisation will take over in those areas where there remains a market for discrete silicon. There are few markets where one chip can do it all. Peter Frith, CTO of Wolfson Microlectronics, points out that even so-called single-chip mobile phones contain five or six ICs.
Will there be a big crunch in which the giants of the industry seek shelter with each other? There will certainly be semiconductor supergiants and they are likely to be the same three pushing for 450mm wafer fabs.
In ten years we may not see much of a reduction in the number of suppliers. however, within each of those 300 or so segments, it is unlikely to be worth the investment for more than three or four suppliers to compete head-on rather than the ten or 20 in some situations, like wireless communications.
So, the chipmakers will spend the next few years trying to work out what they want to keep and what they can sell to other players as they deal with the new reality. Many could accelerate their demise by staying too long in markets in which they cannot succeed, hobbling the business units they use to subsidise their unrealistic ambitions.