The Future of Foundry: Part 1
Future Fab Intl.
A thought leadership project from MazikMedia, Inc.
|The Future of Foundry: Part 1
|The foundry model is one of the great success stories of the semiconductor industry. One-quarter of worldwide semiconductor production is now outsourced to foundries (see Figure 1). However, foundries and their customers are under increasing pressure from rising IC design and process development costs and compressed product life cycles. These pressures are forcing foundries to change the way they engage with customers and invest in their future growth. How will foundries meet the needs of their customers while maintaining profits? The answer lies in increasing specialization and an expansion of service offerings.
In this first of a two-part series, we examine the trends that are driving changes in the foundry market. In the second part, we look at how the foundry model will adapt to these changes.
Forces Reshaping the Foundry Industry
Gartner has identified five “megatrends” that will shape the future development of the semiconductor industry. As foundries seek new opportunities for growth, they must take account of the major trends that are reshaping their industry:
The first three of these global forces have the most direct and immediate impact on the foundries. In addition, foundries are experiencing higher costs of developing manufacturing processes and a shifting customer base resulting from the adoption of “asset-lite” manufacturing strategies by many leading IDMs.
Increasing Device Integration
Each successive process node enables more transistors to be incorporated into a single chip design, which increases the complexity, and the cost, of designing ICs. The interaction between digital IC design and manufacturing process has become critical, requiring sophisticated lithography techniques and advanced design for manufacturing (DFM) methodologies and tools, increasing costs dramatically.
Consider the example of successive generations of DVD players and recorders. Figure 2 shows how the cost of designing system-on-chip (SoC) devices for DVD players/recorders has risen steeply, while the retail price of the end product has continued to fall. Part of the reason for the sharp increase in IC design costs is that the embedded software component is now much larger.
Note that the rate of decline in average selling prices (ASPs) is steeper for each generation of this consumer product. The combination of steeply rising design costs and steeply falling ASPs make it harder for IC suppliers (many of which are fabless companies) to get a return on their investment in product design. There are two inevitable consequences: 1) the number of design starts is falling, and 2) fewer designs can justify the cost of leading-edge processes. There was a dramatic decline in the number of design starts in about 2000, and the rate of increase for designs for leading-edge process technologies is slower than in the past. By 2010, we estimate that only about 10 percent of design starts will be for leading-edge processes.
Increasing Scale of Semiconductor Manufacturing
Along with Moore’s Law, increasing manufacturing scale has been an important driver for foundries. In essence, foundries act as consolidators of semiconductor production into large-scale fabs that operate with improved economies of scale. Foundries assume the risk of underutilized fab capacity, and they minimize that risk through diversification of demand across many customers and products serving a range of applications. With the move to 300 mm diameter wafers, the price tag for an advanced production fab has become out of reach for all but the largest IDMs. The foundry model has given fabless companies and IDMs access to such manufacturing capability at a reasonable cost.
Shift to Consumer Markets
More semiconductor production is going into consumer electronic equipment, such as DVD players/recorders, mobile phones and video game consoles. The DVD player/ recorder example described earlier is indicative of the challenges foundry customers will face as design costs increase while ASPs are under intense pressure in these competitive consumer markets. There will also be time-to-market challenges as the life cycles of consumer electronic products continue to shorten.
Higher Cost of Process Development
Developing a production process for the 65 nm process node requires an investment of about $1.5 billion. The investment needed for the next node (45nm) is about double that. An increasing number of IDMs, such as Texas Instruments, now rely on foundries for digital CMOS process development, shifting the burden for funding the development to the foundries. Generating sufficient return on such a large investment requires a very large-scale foundry operation.
Shifting Customer Base
The rapid growth of the foundry market through 2007 was driven largely by the even-more-rapid growth of fabless semiconductor companies. Between 1997 and 2007, fabless revenue grew at a compound annual growth rate (CAGR) of 17 percent, much faster than the semiconductor industry as a whole. In 2007, total fabless revenue was estimated to be $42 billion, accounting for more than 15 percent of total semiconductor revenue (up from about 6 percent in 1997).
However, the fabless business model is changing, and this will inevitably have an effect on the foundry market. Some fabless companies are now very large. The largest is Qualcomm. Each of the top seven fabless companies now has revenue of more than $1 billion. The success of the fabless model and the rise of the IDM asset-lite model are causing a concentration of demand, particularly for the more-expensive leading-edge processes, into larger foundry customers that can drive the high volumes needed to generate an adequate return on investment (ROI) in design.
Fewer and fewer products will drive the need for foundry services at the extreme leading-edge process nodes. The number of IC designs with sufficient volume to justify the higher nonrecurring engineering expense for leading-edge processes is decreasing as these costs increase. The high cost and complexity of design raise the bar for fabless start-ups looking to produce such leading-edge designs. The cost of building a fab used to be a barrier for semiconductor start-ups. The fabless model removed that problem, but the new barrier is now the cost of leading-edge design. Foundry customers are beginning to use multiple foundry sources. This practice helps mitigate risk by spreading production across two or more foundries. Qualifying IC designs at multiple foundries is more expensive, but IDMs and fabless companies with large production volumes are adopting this strategy because the reduced supply risk outweighs the added cost of qualifying a second fab. As the use of DFM methodologies becomes more widespread beyond the 65 nm process node, the higher qualification costs will enhance the “stickiness” of the foundry/customer relationship, at least for a given IC design, and this trend will make it more expensive to implement a multisource strategy with such advanced designs.
Foundries will have more difficulty achieving acceptable returns on their investments in leading-edge fab capacity. At the same time, increasing costs demand ever-larger fabs to deliver the needed economies of scale. This leads to the conclusion that there will be fewer foundry fabs built for the most-advanced, leading-edge processes in the future. Many foundries will choose to wait until a new process technology is in high-volume production before investing in expanding fab capacity, serving demand for process technology that is a half to a full node behind the leading edge. In Part 2, we will look at some of the strategies foundries will employ to deal with these changes.
About the Author