Sometime in the 1990s, the CEO of Cypress Semiconductor, TJ Rodgers, claimed: "Real men have fabs." His argument was that, although some companies had seemed to be doing OK by renting space at other people's chipmaking plants, to get real control over what you are doing you need to have a fab you can call your own.
Until recently, most companies with fabs agreed with Rodgers. They saw themselves as having the keys to unlock Moore's Law. To put yourself at arms length with the technology that drives electronics would be commercial suicide. But, in the years that followed Rodgers' claim, a growing number of companies have cheerfully bought processed wafers either from old-style integrated device manufacturers (IDMs) - companies like Cypress - or from specialist 'pure play' foundries. And they have not done too badly. This is what Scott McGregor, president and CEO of fabless chipmaker Broadcom sought to tell executives from other companies in the industry at the recent International Semiconductor Executive Forum in Munich, organised by the Institution of Engineering and Technology (IET) and the Fabless Semiconductor Association (FSA).
McGregor used to run Philips Semiconductors, one of the biggest IDMs in the world. But Broadcom has convinced him that everyone should go fabless. Maybe not tomorrow, but sometime. He decided to engage in a bit of mythbusting, listing ten statements that people have used, apparently, to argue against the fabless chipmaking model. Some of them are familiar; some seemed to be strawmen designed to be dismissed easily and quickly.
McGregor's first myth was: "fabless companies can't achieve significant revenue growth". He plotted the sales figures of fabless vendors and IDMs since 1995 and showed that, on average, the fabless companies grew more. However, as fabless companies started off small at the start of that period, anything less than stellar growth would be a victory for the IDM model.
The second myth declared: "fabless companies can't achieve acceptable profit and gross-margin levels". This one is less of a strawman argument. The argument on the side of the IDM is that you get to keep the money that fabless companies lose in the profit margin they have to pay to their foundry suppliers. McGregor put up figures claiming that gross margins on a basket of IDMs and fabless suppliers he selected were 8 per cent better on the fabless side: 50 per cent for those without their own plants; 44 per cent for those with. McGregor did not go into detail on it at this point, but a lot of this has to do with fab utilisation - a theme I will return to often. IDMs have to pay for their fabs whether they are busy or empty. When fabs are empty, they cost a lot of money. That can put a serious dent in your gross margins.
The third myth was "fabless companies can't multisource". It seemed a strange myth to me. Was there something in contracts with foundries that said fabless companies can't go to the competition? Xilinx has been using at least two foundries for many years. Similarly, McGregor said Broadcom has made ample use of different foundries for the same products to help guarantee supplies. "We can tape out the same product to any one of these guys," he said, flashing up a list. "If this foundry is full, we can shift the load. If there is an earthquake at one, we can tell customers what we can do to restore supplies."
Fourth on McGregor's list was: "fabless companies can't develop integrated mixed-signal circuits". Again, a rather easy one to knock down for anyone. Plenty of companies selling products with large proportions of analogue circuitry on them do not have fabs. McGregor said this myth came about because the IDMs have specialised, exotic process such as BiCMOS and silicon germanium bipolar that they don't offer to other people. Curiously, the same day, Infineon Technologies said it would offer access to its more exotic processes.
"They say they can do really cool stuff, unlike the fabless guys. People say you need those fancy processes or you can't do high-speed chips," said McGregor. He countered by claiming that Broadcom had done advanced mixed-signal chips in ordinary CMOS processes available at any of the top foundries. Examples were chips that could support 10Mbit/s interfaces that were implemented using 130nm CMOS from foundries. "In working with the foundries, if you have a good engineering team, there is no limit on what you can do," he claimed.
The fifth myth cited by McGregor was: "fabless companies can't develop innovative products on time". The argument from the IDMs here was that owning a fab got you engineering samples faster from new processes. Again, the Broadcom boss declared reality is different - the foundries are, in some cases, providing access to advanced processes before many of the IDMs. In reality, you have to have good connections with the fab in either case to get samples through quickly using what the industry calls 'super hot lots'.
That fabless suppliers can't get traction from investors was McGregor's sixth myth. "They can deal with VCs but when it comes to Wall Street it's different," he said. This was relatively easy to deal with. Even I was thinking of the way that shareholders tend to dislike capital-intensive businesses with highly cyclical returns. "The IDMs suffer on Wall Street," said McGregor. However, some of the best performers on the graph that he showed did own fabs. Above a line drawn as the sector average were companies such as Maxim Integrated and Microchip Technologies. The link between those companies lay in small, specialist analogue or mixed-signal products. And very, very old fabs. That is something that has not escaped the other IDMs.
Myth number seven was similar: "fabless companies can't gain the trust of the major OEMs". There was a hint of straw about this myth too. Major OEMs dislike changing to companies with little or no track record, unless they get a good deal that makes sense for them. After a while, they get used to dealing with new kids and start treating them as long-term suppliers. This myth was quickly pushed aside. McGregor said this is no more difficult for a fabless company than for an IDM.
The eighth myth was more interesting, that "fabless suppliers can't get enough product during allocation". This myth had a particular resonance at the conference. Supplies were getting scarce and the managers there were thinking of previous booms, such as in 1995 and 2000, when customers had to be told that there were no more chips to be bought. In 1995, the allocation problem was particularly severe for fabless companies. There were stories of fab owners selling access to wafers twice over. Companies that thought they had guaranteed access to production found their wafers had been sold to some other company. It was at that time that the first batch of big fabless companies found themselves getting tied into deals that netted them 10 or 20 per cent of a fab in new consortia.
McGregor himself did not allude to 1995 but to the more recent bubble of 2000, when he was at Philips. "I don't have a detailed statistic on this but in 1999 and 2000, the toughest years of the last capacity crunch, we had no allocation to some customers during 2000. But the fabless companies did get their wafers," he said.
Myth number nine stated: "fabless companies will outgrow their foundries". The idea behind this one is that some fabless companies will get so big, they might as well build a fab because they will be able to fill it just with their own products. "I often get asked that by investors," claimed McGregor. He cited an 11 per cent annual growth rate in foundry capacity as one reason why there would always be enough rented fab space to go round. And declared that Broadcom would not build a fab even with sales of more than $10bn - a level more than healthy enough to justify sole ownership of an advanced fab by an IDM. "We can remain healthy and growing as long as the foundry industry is healthy and growing," he said.
And the tenth myth claimed: "fabless suppliers can't gain access to the latest and greatest technology". This was more or less like the fifth myth. However, McGregor cited Taiwanese foundry TSMC as the prime example of why IDMs did not have all their own way. "TSMC leads the ITRS Roadmap," he said, alluding to the document prepared by research consortium Sematech that is regularly updated to show when new processes are expected to come through. If you look at the schedules from the major fab owners, TSMC is up there with the leaders, but is behind some and ahead of others.
Having dealt with the myths, McGregor looked at the core problem for IDMs: "If you have a fab and it's only 80 per cent loaded, you think a lot about how you are going to fill it." That means selling chips for less than they would like, simply to get the utilisation levels in the fab up. Eighty per cent is not a good number - it is the kind utilisation level people in the industry associate with downturns. "The IDMs imagine they have a margin contribution but they don't. They should be realistic about that," said McGregor.