450mm Wafers: Not the Best Solution
By Iddo Hadar, Applied Materials -- EDN, July 11, 2006
As the semiconductor industry debates the merits and drawbacks of a possible future move to 450mm wafers, a movement has emerged to significantly enhance and extend the 300mm wafer size generation. Called 300mm Prime, this cross-industry effort advocates the use of efficient, high-mix manufacturing operations to keep the industry on the Moore’s Law cost curve. If properly planned and executed, this effort could push out or even eliminate the need for larger wafers, thereby benefiting the entire semiconductor ecosystem.
A conventional assumption is that larger wafers lead to lower production costs—and are required to maintain industry economics. Interestingly, a detailed “apples-to-apples” analysis indicates the opposite is true—the 300mm generation seems to have increased processing costs relative to comparable 200mm technology (Figure 1). At the same time, modeling of average processing cost history and projections shows that the rate of reduction in cost per transistor can be maintained, within the analysis' margin of error, without 450mm wafers (Figure 2).
Similarly, an industry-wide analysis shows that a move to 450mm wafers is likely to be no more productive than 300mm, and very likely even less productive, in terms of dollars per transistor. Furthermore, the industry is unlikely to recoup its 450mm investment. At an assumed 15 percent to 20 percent ROI, it would take approximately 10 to 15 new 450mm fabs every year over at least 15 years to pay back the possible $15 billion to $20 billion in development costs—even with optimistic cost savings. This level of investment is very unlikely to happen and is a poor foundation for the industry to move to 450mm. Given the constraints on development resources, the industry should instead drive fab productivity in higher-impact areas.
The Real Need
In the consumer-driven economy, demand is increasing for short life-cycle, smaller-run products, driving a high-mix production environment. Many chip designs’ total production runs fit in one 300mm FOUP, or even a fraction of a FOUP. Small-lot manufacturing is critical to allow a larger number of chip designs to efficiently use advanced technology—and it can also increase overall factory efficiency for large-lot fabs through faster cycle time. As such, 300mm Prime should focus heavily on addressing three key bottlenecks to this vision as a means to elevate fab productivity above what larger wafers could provide:
1) Small Lot Manufacturing
Chipmakers realize that inefficient cycle time—how long wafers take from beginning to end of processing—is draining profitability. Wafers spend more than half their time in the fab simply waiting. This long wait time is reducing fab productivity, as fabs are required to constrain utilization to meet acceptable cycle time. It also slows learning cycles.
A critical driver of queue time is lot size. The larger the lot size, the harder it is to efficiently manage the flow of wafers to and from process tools. Each generation of larger wafers is more affected by this wasted time, because each wafer carries a higher percentage of production than before.
The key is to reduce lot size. There are proposals to cut this from the current 25 wafers to two wafers, or even single wafer lots; but they entail an order-of-magnitude increase in the volume of transactions. 300mm Prime hinges on a fast and reliable material-handling system to enable small-lot manufacturing.
2) Universal Single-Wafer Processing
Increasing use of single-wafer processing contributes to efficient use of small-lot technology. Single-wafer tools are commonplace, but interspersing them with batch tools results in a worst-case queuing challenge for the fab. Fortunately, single-wafer wet clean, LPCVD, implant and other technologies are beginning to replace their batch counterparts, offering better performance and productivity. Only with a complete single-wafer tool set can the benefits of small lots be fully realized. 300mm Prime should provide a transition plan for single-wafer processing across the fab in a way that best complements small-lot handling solutions.
3) Tight Equipment Characterization and Intelligent Systems
Processing variance is a key contributor to long cycle times. Tight specifications, advanced equipment control, chamber and tool matching, fault detection/classification, quality management, and predictive maintenance are all vital to small-lot, high-mix fabs.
Additionally, to reduce process and tool variance, it is necessary to embed more intelligence in processing systems, primarily through Advanced Process Control and integrated sensors/metrology. Real-time, high-bandwidth information and advanced control frameworks allow reduced variation across chambers/systems and over time, and are critical for high-mix production environments. 300mm Prime should offer a framework, toolkit, and solutions to enable this migration.
In conclusion, 300mm Prime provides the industry with an economical and lower-risk means of continuing to drive Moore’s Law, as opposed to moving to 450mm. By addressing key productivity bottlenecks, it represents a set of high-leverage opportunities to create step-change improvements in 300mm fab productivity consistent with the changing fab environment. If the semiconductor industry focuses its combined resources on these three initiatives—in combination with an unrelenting commitment to pushing process technology forward—fab productivity improvements can be delivered for many years to come.
Iddo Hadar is the CTO and chief marketing officer of the Foundation Engineering Group in Applied Materials.
