The world's largest chipmakers are gambling heavily on manufacturing in an attempt to take advantage of their market muscle.
"We are going to look back and see 2012 as a breakpoint," declares Bill McClean, head of analyst firm IC Insights. "The IC industry is changing structurally."
A bold assertion, but the evidence for this change, according to McClean, lies in the amount of money the largest chipmakers are spending on expanding production capacity versus their smaller competitors. For 15 years, Intel has sat atop the IC manufacturing leaderboard due to its dominance of the PC industry.
In the early 1980s, when IBM decided to use Intel's x86 architecture, the company's fortunes turned around dramatically. Up to that point, Intel had specialised in making memories and was the prime supplier during the 1970s, but was facing stiff competition from rising Japanese companies such as Hitachi and NEC.
As Intel struggled against companies that were undercutting it by at least 10 per cent on price in what had been up to then its most lucrative market, IBM decided in 1983 to secure its supply of processors by investing $250m in the company. By 1985, NEC was the top supplier of ICs worldwide; the three other Japanese memory makers - Hitachi, Toshiba, Fujitsu - were fourth, fifth and sixth.
That year, in the face of a punishing recession in the business after the collapse of the first home computer boom, Intel quit making DRAMs and began its dramatic turnaround - and then it was the turn of the Japanese companies to falter. Earlier this year, Elpida - the DRAM specialist formed by the merger of the memory operations of Hitachi and once dominant NEC in 1999 - filed for bankruptcy in Japan.
The company that now dominates the DRAM business, Samsung, barely had a microelectronics operation in the mid-1980s, but grew to be the largest supplier of memory chips by the end of 1992. The company is now the world's second largest chipmaker, double the size of nearest competitor Texas Instruments (TI). It outpaces Elpida in terms of total chip sales almost tenfold. The Korean giant has become the other agent of change in the semiconductor world, according to IC Insight's McClean.
With an estimated total of $12.2bn to spend in 2012, Samsung is likely to have spent more on expanding production at its semiconductor fabs in the year to date than Elpida made in chip sales in all of 2011. "Samsung over three years will spend more than $30bn on capital expenditure," says McClean.
Intel is taking a similar approach, with plans to spend $12.5bn. Only TSMC, the world's largest foundry, which supplies chips to companies such as AMD's ATI graphics division, and nVidia, is close to keeping-up, with just over half that amount.
"The spending levels fall away real fast. The number four capital-spending company is Globalfoundries at $3bn," McClean adds. "It's unbelievable. I have never seen anything like this in 30 years in the industry. It shows just how top heavy this is. These massive increases have little to do with the market. These are strategic increases. These companies are breaking away from the pack. Intel and Samsung are really serious about this. This is a breakpoint in the industry. It will not be the same from this point forward."
For much of the past three decades, since the Japanese companies rose and fell, the shape of the semiconductor industry has flattened. In 1985, the top-ten suppliers accounted for 60 per cent of chip sales worldwide. After that, fabless companies began to take a hold.
By subcontracting manufacturing out to foundries such as TSMC - and not having to spend hundreds of millions on their own fabs - it became possible for much smaller players to make money out of designing and selling ICs. Only in the past few years did fabless companies such as Broadcom and Qualcomm make it into the top ten players. An increasing number of companies have been joining them in getting out of the fab business.
Even TI, which is now only slightly smaller than TSMC, has stepped back from leading-edge fabs, choosing to focus its much smaller capital expenditure on increasing production on older processes used for its analogue circuitry. Malcolm Penn, president of Future Horizons, chastises this fab-light strategy chosen by many of the larger players because of the potential loss of control over future technology. "Eventually, it turns into a fabless strategy," Penn says.
By 2005, the proportion of total chip sales by the top ten companies fell to 45 per cent. This has masked the increasing power of the few at the top in an industry that has acquired a structure similar to the English football leagues where the top teams collect an increasing share of the money and prizes.
For 25 years, the top three suppliers accounted for just under a quarter of the worldwide total. At the end of the last decade, that number began to change. Now the top three are responsible for more than 30 per cent of sales, and that number is likely to increase over the next few years as Samsung's competitors run into financial problems and Intel tries to make the most out of what amounts to a lead of one process generation over its competitors.
One of the reasons Intel and Samsung are having to spend so much is the rapidly increasing cost of the equipment needed to define sub-30nm features on a chip. The smaller companies are being forced out of the market because they cannot use their existing kit to support a shrink in feature size. "Those that can't are being killed on margin," Penn says. "The DRAM industry has never actually made money. But Samsung has. They have taken all the money. Which means the others have lost an awful lot." According to McClean at IC Insights: "Nanya and the other DRAM suppliers in Taiwan: they're dead."
A large part of Samsung's business now lies in flash memory. Although Japan-based Toshiba is stronger in this technology than Elpida is in DRAM, the Korean company has gradually been pulling ahead on the back of a fast-growing market, fuelled by successful mass-market products such as the Apple iPhone and iPad.
"NAND is growing fast - and there isn't much capacity left. There is just insatiable demand for this," reports Future Horizon's Penn. "It is unbelievable the amount of memory we are looking at." The third strand of Samsung's business lies in its becoming a competitor to TSMC and, thanks to its huge bulk, a stronger player than the current number-two in its sector, Globalfoundries.
McClean says: "It is interesting that Apple and Samsung are very much competitors at the corporate level. Yet Apple is happy having Samsung producing its devices. Samsung IC guys are trying to do everything they can to keep this business. Apple is talking about moving to TSMC but the problem is that, through much of last year, TSMC was working close to 100 per cent."
He continues: "Apple's volumes are such that they would need a dedicated fab. Even if Apple wants to move they can't. I believe they want to give this business to other foundries but any shift is not going to be dramatic. And Samsung is going to try to make it attractive to stay with them. Samsung is being very aggressive in this field."
Intel is also going after the foundry business, albeit with substantially smaller clients. Last year, the chip giant signed up tiny programmable-logic start-up Achronix.'Two more deals followed this year,'including another maker of programmable logic, Tabula. Even if it signed Apple up itself, foundry sales would remain a small fraction of Intel's sales: the company's main ambition lies in what follows the PC, an area where the company has struggled up to now. "Intel is trying to get into mobile area. They are going after it big time with technology," says McClean at IC Insights. "They have 22nm now. In a couple of years' time they will have 14nm."
Mike Bryant, CTO of Future Horizons, says Intel is positioning itself to take advantage of the increasing range of problems that face chipmakers. "Moore's Law can't continue in its current form most people are struggling. There has to be a slowdown." Although the schedule for Intel's 22nm Ivy Bridge slipped by just over a quarter to be launched at the end of April, the company now has parts in the market. They have arrived long before competitive devices will ship from suppliers such as TSMC, Globalfoundries or, indeed, Samsung. And Intel has a new transistor technology in the shape of the finFET running in volume; that took lots of money, says Bryant.
"Intel ran 20,000 wafers to get finFETs working. So far, TSMC has run no wafers in a production environment for finFETs," Bryant explains. Potentially, Intel could have a major advantage in terms of the core technology inside its processors for four or more years. If Achronix sticks to its schedule, it could have its products out on 22nm well before either Altera or Xilinx, who lead the market and which use Samsung and TSMC to make their parts.
TSMC is not expected to introduce finFETs until the 14nm generation, although it is possible that the foundry will attempt to speed up introduction to try to fend off the threat from Intel. The massive uptick in spending from the two runaway leaders at the top of the semiconductor market could lead to problems for TSMC, which ST-Ericsson, among others, uses heavily. Now running at about half the turnover of Samsung's chip division, even that company may find it problematic keeping up technologically.
The main benefit of the finFET is that it reduces power without sacrificing as much performance as when using conventional CMOS transistors. This presents a problem for ARM, says Future Horizon's Bryant, which has historically maintained an advantage in low-power, mobile processors. "The 22nm trigate technology will never be available to ARM," he claims.
Some respite, however, will come from Intel's own strategy. "When it comes to low-power devices, Intel is not yet throwing its top technology at the problem. They are using a very safe technology," says Bryant, referring to the 32nm Medfield processors.