Less is more
Consumer demand for electronic gadgets is placing huge demands on the global microchip industry.
"We cannot afford it," admits Frans van Houten, president and chief executive of NXP Semiconductors. He's referring specifically to the most advanced manufacturing process in the semiconductor industry today - the use of 300mm-wide silicon 'wafers' to produce the microchips that have become ubiquitous in modern life.
When you consider that NXP is Europe's second largest and one of the world's top ten microchip makers, you can only imagine how much of a say its hundreds of smaller competitors are having in the development of tomorrow's cutting-edge chip manufacturing technologies.
As microchips, or integrated circuits, continue to increase their density, performance and speed, the high-precision tools required to build them are becoming expensive. The price tag of a modern 'stepper' - a key tool in the photolithography process that is used to stamp the millions of microscopic features on the silicon surface - can be as high as $30m. And that's just one of the hundreds of pieces of machinery that a 300mm-wafer chip fabrication plant, or fab, would need to operate.
Intel opened one such fab last October in the US near Phoenix, Arizona. The second of its factories to produce 45nm microprocessors, it cost a whopping $3bn to build. That might be fine when you're the world's largest semiconductor vendor; and, indeed, Intel will be opening two other similar facilities later this year in the US and Israel. But for pretty much everyone else, these numbers are simply getting out of reach.
"In order to justify building a $3bn fab, you've got to be shipping $5bn worth of products every year," estimates Will Strauss, principal analyst with electronics market research firm Forward Concepts. "Well, there are only about five companies in the whole world who are shipping that kind of volume."
In fact, there are more than five companies with revenues of $5bn or more, although Strauss is probably right about how many actually have that amount of money generated exclusively by the sale of products and not other services and software. Globally, the top five chip companies are considered to be Intel of the US, Samsung Electronics (Korea), Toshiba Semiconductors (Japan), Texas Instruments (US) and STMicroelectronics (Italy/France).
Even when you do manage $5bn in revenue, it might not be enough to justify building a state-of-the-art fab. After all, NXP, formerly Philips Semiconductors before it was sold off by Philips in 2006, reported annual sales in excess of $7bn for 2007. And yet the Dutch-based company's chief executive insists they can't afford such a luxury.
"I'll tell you why," says van Houten, who is also president of the European Semiconductor Industry Association. "Typically, a semiconductor company has a portfolio of manufacturing process nodes. So some of your sales will be for older nodes and some will be for the most advanced nodes. The proportion for the most advanced nodes for us is far too small to justify having our own 300mm, advanced CMOS [complementary metal-oxide semiconductor] wafer fab.
"That means that, if we were to build one, it would be under-loaded. And the most criminal thing to do is to have an under-loaded fab: you would lose a lot of money. We have calculated that we would not fill a single advanced CMOS fab ourselves and, as a consequence, we will never own one."
Strauss says that big players are turning to the sub-contractor 'foundries', the independent chip makers who apply the designs of their customers. "To have a world-class fab you've got to have some big, expensive chips on them. Or else, you've got to be prepared to act as a foundry and have people essentially come to you [with their designs]. As a foundry you don't really care about what it is your customers are making - you just fabricate the chips and make sure you have a reasonable yield."
NXP has done what an increasing number of major integrated device manufacturers (IDMs) are doing to produce particularly those chips demanding capital-intensive manufacturing processes: outsource. IDMs are those giants that, historically, have designed, made and sold their own chips.
Texas Instruments, STMicroelectronics, Infineon Technologies and Freescale Semiconductors have all recently announced what is known in semiconductor industry jargon as 'fab-lite' strategies. These are a sort of medium ground between remaining a full-blooded manufacturer and becoming a 'fabless' firm. Going fab-lite almost invariably means outsourcing to a foundry in the Far East.
In NXP's case, the company chose to partner with some of the key players in this sub-contracted fab-making sector. As part of a joint venture with TSMC for 200mm wafer fabrication, NXP has a collaboration agreement with the Taiwanese foundry for 300mm wafers.
"That's not to say we don't invest in process technologies," van Houten stresses. "Even though we are not investing in a 300mm fab, we have a research partnership with TSMC to work already today on 32nm geometries. Even as a fab-lite company, it is important to be first-to-market with the latest CMOS node."
Like any fab operator, the foundries also have to ensure that their expensive production facilities are kept fully utilitised. Foundaries such as TSMC, UMC and Chartered have developed so-called 'shuttle services', whereby multiple fab-lite clients can have their chips fabbed on a single wafer.
"They'll take the digital files for each product line and put them on a single mask [the transparent plate containing the photographic image of wafer patterns], so a single wafer might be making products for two, four or more companies," Strauss explains.
Although shuttle services have been around for three or four years, Strauss says that they have now become the only way for small companies to have their chips produced.
As fewer IDMs are able to build their own high-end chips, a new production picture is emerging. "I see several companies clustering around either TSMC or the IBM Alliance, and then maybe Intel is big enough to do it itself," van Houten predicts.
"Eventually, we will end up with a couple of manufacturing clusters and then a few independent agents like the other foundries. The semiconductor industry is going to be more mature and will require very large-scale manufacturing set-ups. There will be only three, four or five of these clusters around - and that's it."
Strauss agrees. "We'll either have to find a cheaper way to make wafers or we'll all end up with one huge semiconductor factory in the middle of China," he warns.
Consolidation is not only happening at the production level. Entire semiconductor companies are being forced to consider being acquired by a larger firm or risk going out of business. For an industry that was already expected to grow well below its long-term average of 7 per cent during 2008, the now-evident worsening of the global economy won't do much to help.
"Some companies are going to find themselves in more financial difficulties than others, especially some of the smaller companies that venture capitalists have backed. Investors may see it's a bad year to do IPOs [initial public offerings] and might well opt to have these companies acquired," Strauss says.
"I'm seeing some very innovative smaller companies at the moment that are introducing, I wouldn't say revolutionary but, certainly, breakthrough products for very small areas of the market. However, I don't think they will make it through the year without being acquired, mainly because the venture capitalists who are backing them will want to get their money back as soon as they can."
The smaller companies are facing other pressures. Some semiconductor vendors - usually the larger ones - are increasingly integrating into a single chip some features that, by themselves, provided niche markets for standalone components manufacturers.
"A good example of that is happening right now in digital television," van Houten says. In the past, you would typically have a channel decoder, a graphics engine and picture quality improvement all in separate components. At that stage you saw companies working separately on a part of the solution. Now, you see completely integrated analogue/digital, one-chip TV solutions, and many companies cannot follow that.
"We think that we are one of the integrators in this market - one of maybe only two or three - and that there will be a big shake-out in the digital TV market in the next 12 months or so."
While the Far East continues to accumulate an ever-growing proportion of the manufacturing side of the industry, the crucial design of the chips remains dominant in the US, Japan, western Europe and Korea. The only question is: how much longer? "Slowly - or maybe not so slowly - China is getting its own capabilities," says Strauss. "Largely, this is an area where Western technology dominates. But China is acquiring it, both legally and illegally.
"There are now hundreds of chip-design houses in China. Most are fairly small, but there are some that are already shipping in the hundreds of millions of dollars worth of chips a year with their own designs. All of these companies have global ambitions, and some of them have significantly cut into the market share of US companies."
He gives the example of Actions Semiconductor, a Chinese fabless firm which has grown to become the world's number one designer of MP3 player chips. Which shows that China is not going to be just about wafer making for a great deal longer.