The not-so-brave new world of semiconductor manufacturing
As Moore's Law begins to break down, chip makers risk making billion-dollar bets on next-generation technology with consequences for the entire electronics value chain.
The semiconductor industry has a problem with two four-letter words: cost and risk. After a couple of decades of boom-and-bust cycles, the industry settled into a period of modest yet relatively stable growth in the early part of this century. Driving the vast majority of the industry's revenue was the personal computer, sales of which grew steadily year after year. More recently, however, the world's ardor for PCs has cooled as consumers and businesses have gravitated to mobile devices, principally smartphones and tablet computers. The growth rates for smartphones and tablets, at least at the high end of those markets, are beginning to slow, as buyers shift to lower-priced products, especially in emerging markets, such as China. The problem for the semiconductor industry is that as the PC market has slumped, mobile devices aren't taking their place as the profitable drivers of increasing chip sales.
In the midst of this product transition, the industry has an expensive and tricky juggling act - continuing to shrink the dimensions of semiconductors while reducing the manufacturing cost of those chips. Scaling, as this shrinking is called within the industry, proceeds apace, but the cost of scaling is becoming a significant concern. (See figure below.)
For nearly five decades, the semiconductor industry has conformed to the widely known and often misunderstood phenomenon known as Moore's Law, which predicted that the number of transistors on a semiconductor device would double every year (later revised to every two years). Gordon Moore postulated the trend in 1965 while he was working at Fairchild Semiconductor; three years later he would become a co-founder of Intel. Moore also speculated that the cost of these multiplying transistors would continue to fall as their volume grew, a prediction that has generally held true for decades. Until now.
As the tiny features of microchips shrink to line widths of 28 nanometers, 20 nanometers, and even 14 nanometers, the cost of making these chips that contain billions of transistors continues to rise. (A strand of human DNA is about 2.5 nanometers in diameter and the width of a piece of paper is about 100,000 nanometers, according to the U.S. National Nanotechnology Initiative.) This breakdown of the cost-density relationship is a new wrinkle and a serious departure from what the industry has come to count on. Indeed, it threatens the financial viability of many chip companies, their material and equipment suppliers, as well as their customers. That's the risk part of the problem.
Some market observers believe the answer to the cost problem is migrating to larger silicon wafers (the round disks that can contain thousands of individual microchips). In the last couple of decades, the diameter of wafers has increased from 200 millimeters to 300 millimeters, which has become the standard size typically produced around the world today. On the horizon is the promise of 450-millimeter wafers, which could contain many more chips on a single wafer. However, that promise is threatened due to the sheer cost of producing 450-millimeter wafers. (See sidebar "Dim prospects for 450-millimeter semiconductor wafers" below.)
For the first time in its history, the semiconductor industry may not be able to reduce prices on its next generation of microchips because of the increasing complexity of manufacturing. For many years, the critical dimensions of semiconductors could be processed with one layer, and photolithography equipment (also known as microlithography systems) could do its work with one exposure for a chip. With the current generation of 193-nanometer immersion lithography - a deep-ultraviolet technology - it's becoming necessary for chipmakers to require multiple lithography exposures in the manufacturing process.
Extreme-ultraviolet lithography (EUV), the next generation in photolithography, has been in development for several years, principally by Netherlands-based ASML Holding and its partners. Chipmakers will probably be able to turn out 14-nanometer chips with the current generation of 193-nanometer lithography systems, but EUV may be necessary for the next generation of microchips with line widths of 10 nanometers and even 7 nanometers.
Indeed, EUV is the semiconductor industry's only reasonable hope to minimize the risk of rising manufacturing costs, since it promises a return to one-step lithography. Still, at 10 or 7 nanometers, manufacturing costs are expected to increase yet again, perhaps by as much as 15-20%. At 10 nanometers, semiconductor manufacturers may have to tell their customers, "Costs are going up." And that will be a hard pill to swallow for an industry that has been able to offer regular cost reduction for decades.
At the same time, it's not clear whether EUV will even fit the bill, since ASML and its Cymer subsidiary have been unable to date to produce an energy source powerful enough for volume throughput of wafers. ASML has been shipping beta EUV tools to customers since about 2005, to help them develop processes. All of the top chipmakers have collaborated on EUV development and have essentially placed all of their eggs in that one basket. If EUV can't produce semiconductors in volume - even at higher prices - it will be a disaster for the continued evolution of the extended semiconductor value chain.
What is essentially at risk for the industry is the ability to enable new products that don't just provide incremental improvement from previous electronics. Producing a chip that can improve the battery life of mobile devices by 10% isn't very impressive. Consumers crave products that are essentially free to purchase and operate, yet that's not practical. Semiconductor manufacturers need to help their customers hit price points that are attractive to consumers and ultimately to corporations and government agencies that make volume purchases of electronic equipment.
Battery technology, which is essentially stuck in the 20th century, has proved to be a major concern for the evolution of mobile devices and wearable electronics, two potentially large-volume, high-growth markets. The challenge in the next few years is to produce advanced semiconductors that will be the basis of cool new products, perhaps in areas that have not even been conceived of, which will be priced low enough for mass-market commodity products. That's the make-or-break value proposition of the near future for chipmakers.
Consolidation the only recourse
Consolidation is a growing trend among semiconductor manufacturers and their suppliers. For instance, Texas Instruments acquired National Semiconductor in 2011, and smaller chip companies have been active in mergers and acquisitions since then. Similarly top-ranked semiconductor equipment maker Applied Materials last year agreed to merge with Tokyo Electron, one of its largest rivals in semiconductor production equipment. In other high-dollar deals in the semiconductor equipment sector Lam Research bought Novellus Systems in 2012 and ASML purchased Cymer last year. In February, Entegris agreed to acquire ATMI.
The simultaneous development and implementation of three process nodes - 28nm, 22nm, and 14nm - is placing phenomenal pressure on leading chipmakers, integrated device manufacturers and silicon foundries alike. The 14-nanometer node involves more than the usual scaling in that it requires novel chip design, such as adding silicon 'fins' to traditional field-effect transistors, called FinFETs, produced with silicon-on-insulator substrates.
Keeping up with the most advanced manufacturing technology has become a bet-the-company, multibillion-dollar decision for the biggest chipmakers. This year, Intel, the world's largest semiconductor manufacturer as measured by annual revenue, decided against equipping Fab 42, its new wafer fabrication facility in Chandler, Arizona, after completing construction of the building. Fab 42 was meant to be the company's premier 14-nanometer FinFET fab, slated to begin production at the end of 2013. Instead, Intel has indefinitely mothballed the facility and will upgrade other fabs in Chandler to handle 14-nanometer chip manufacturing, after processes are qualified at the company's D1X plant in Hillsboro, Oregon. Evidently, Intel CEO Brian Krzanich, who came from the manufacturing side of the company, looked at the cost of equipping Fab 42 and blinked.
Changing of the guard
For years, bragging rights in the semiconductor industry went to the company that could turn out the fastest, most powerful microprocessor. As long as Windows-based PCs were being sold in significant volumes, Intel was generally king of the hill, holding about 80% of the processor market. The majority of the remaining market share went to Intel's long-time rival Advanced Micro Devices, which designed microprocessors based on an architecture licensed from Intel.
As the PC market fades, fast processors have become déclassé. Today's watchword is "low-power", particularly in the applications processors that are at the heart of smartphones. When it comes to low-power chip designs, UK-based ARM Holdings has long held the crown in that category. Apple, Nvidia, Qualcomm, Samsung, and others have incorporated ARM's processor cores, which they license, into their chips. ARM gets revenue coming and going - first from licenses for its technology and then from royalties on each chip sold by its licensees.
The microprocessor is being eclipsed by the system-on-a-chip (SoC) device, which incorporates logic, memory, and other functions on one chip. SoCs can be more versatile in a variety of applications. They could relegate microprocessors to being more like microcontrollers, the workhorse chips typically found in computer printers and microwave ovens, among other products.
The Next Big Thing
What can replace the PC as the driver of semiconductor sales and profits? Smartphones and tablets would have been the predictable answer in 2013, yet those products are maturing, at least at the high end. While Apple and Samsung were able to set records with sales of their mobile devices last year, the growth rates for those products began to flatten, and that trend is expected to continue this year.
The semiconductor industry is anxiously awaiting the Next Big Thing to emerge that will drive demand for cutting edge chips. Some speculate it could be wearable electronics. The 2014 International Consumer Electronics Show in January, for example, featured a host of wearable electronics including smart watches, fitness trackers, health monitoring devices, and the like. The Google Glass wearable computer has inspired a host of copycat products. Augmented-reality headsets for playing video games are also commanding a lot of consumer interest. Pricing for most of these products remains high, however, and it's expected to take a couple of years before price tags settle at a comfort level for the average consumer and demand starts to accelerate.
Qualcomm last year came out with the Toq smartwatch, priced around $300, which uses its Snapdragon processor and Mirasol display technology. The IC design company made it clear that it didn't want to compete in the smartwatch market with the likes of Samsung, Sony, and (it is rumored) Apple, but wanted to demonstrate how such a product could be made.
The cost and risk factors facing chipmakers are, naturally, a concern to their customers as well. Whether chip companies can cheaply and easily produce advanced semiconductors will affect developments in a variety of market sectors including automotive electronics, industrial equipment and medical systems. The semiconductor industry has always answered the call for better and cheaper chips, yet it now seems like "cheaper" will be a hard promise for the industry to keep, with possibly dire consequences for the entire electronics value chain.
Dim prospects for 450-millimeter semiconductor wafers
Which leading semiconductor manufacturers will be building 450-millimeter wafer fabrication facilities? It's always been a short list, but it may be getting shorter still in the near future thanks to the astronomical cost for equipping a next-generation fab.
Intel, Samsung Electronics, and Taiwan Semiconductor Manufacturing (TSMC) have long been the leading proponents of the proposed move to the bigger wafers, which can produce up to 225% more chips on one substrate than the present generation of 300-millimeter wafers.
With PC sales slowly but surely declining along with demand for the microprocessors that power them, Intel isn't likely to lead the charge for 450-millimeter wafer fabs. Similarly, producers of memory chips, such as Micron Technology, Samsung, and SK Hynix, aren't likely to adopt the larger wafers any time soon.
TSMC, the world's largest silicon foundry measured by annual revenue, perhaps has a better cost incentive to make the move, which will give it a cost advantage over its foundry competitors. TSMC's production capacity can provide enough volume to justify the 450mm transition, yet Intel, Samsung, and GlobalFoundries are all seeking more foundry business, limiting TSMC's volume.
Even so, unless wearable electronics or another high-growth semiconductor-based product can be manufactured cheaply and profitably, the 450-millimeter fab is likely to remain a pipe dream for the semiconductor industry. Both Intel and TSMC have put up "shell" buildings that could be equipped for 450-millimeter manufacturing, yet they are letting those facilities go unused at present. Intel is constructing its D1X Module 2 facility in Hillsboro, Oregon, for 450-millimeter development fabrication, which is scheduled for completion in 2015. But even if Intel, Samsung, and TSMC all go full-steam ahead with their 450-millimeter fab plans, it still may not present enough business for the producers of semiconductor manufacturing equipment to justify their research and development expenses for 450-millimeter gear.
Equipment suppliers have long and unhappy memories of the industry transition from 200-millimeter to 300-millimeter wafer fabrication. Their customers urged them to plunge into 300-millimeter R&D and to turn out 300-millimeter-capable equipment, only to be extremely stingy or tardy when it came time to place orders for the equipment.
Indeed, the semiconductor foundry market is where the DRAM market was a decade ago - full of entrants eager to claim market share and ready to spend on the factories to gain that share. TSMC's limit may be the willingness of its shareholders to accept billions of dollars in capital expenditures for a 450-millimeter fab or two. The world's No.2 foundry, GlobalFoundries, is privately held, owned by an Abu Dhabi investment fund, and could spend whatever its investors are willing to bear
There are indications that 450-millimeter volume wafer production will be pushed out from 2018 to 2019 or even 2020. Before the 450-millimeter wafer fab becomes a reality, cost reductions and technical breakthroughs in semiconductor manufacturing are clearly needed.
Len Jelinek Senior Director and Chief Analyst, IHS Technology