Issue

Equipment life after shrinks

05/01/1999

Gordon Moore predicted some 30 years ago that the number of transistors on an IC would double every 18 months. Remarkably, the industry is still on this course. To keep the law in effect, semiconductor companies have made impressive productivity gains in just about every part of chipmaking.

Today, however, it is questionable whether chip manufacturers can continue gaining productivity from the same areas that have enabled Moore's Law in the past, namely yield, wafer size in creases, feature size decreases, and capital utilization. As these traditional sources of productivity gain start to meet diminishing returns, the industry must look to improve capital productivity of semiconductor equipment (i.e., increase throughput or reduce capital cost). Interestingly, capital productivity was emphasized more 10 years ago.

While it might be more intuitive or comfortable for the industry to continue making productivity gains in traditional areas, increasingly there will be physical or economic limits:

Yield improvements.

Die yields have improved so much over the last 10 years that there isn't much room left. Years ago, the industry ramped up 64Kbit DRAM very slowly, taking several years to move from initial 20% die yield to an 80% yield at product maturity. Today, according to industry forecasters, next generation 256Mbit DRAM could achieve up to 90% die yield in less than one year after going into full production. Yield simply cannot go much higher than that or get there much more quickly.

Design shrinks.

Shrinks have been the preferred and most highly publicized method for increasing productivity for several years. In fact, as the other factors listed here have had reduced effectiveness, manufacturers have actually been accelerating design shrinks ahead of the SIA Roadmap in a desperate attempt to maintain the economics of Moore's Law. The design shrink now dominates all other factors in terms of influence. We are now actually printing features below the wavelength of light, literally doing what was forecast 10 years ago to be impossible. How long we can continue to do this is uncertain. Potential technology show-stoppers include the physical limitations of new lens materials for steppers, problems with resist technology, difficulty with masks, and other limitations of optical lithography. At some point we may accelerate through the last "easy shrink" and hit a technological brick wall.

Increased wafer sizes.

In the past 10 years, the industry has found it relatively straightforward to increase the size of wafers from 100mm to 150mm to 200mm. During those transitions, the process equipment remained very similar. However, the move to 300mm requires a complete redesign of all equipment, and the change is taking much longer than expected. In fact, the historical rate of change of wafer sizes is slowing down, and it's not clear that there will be another wafer size increase after 300mm. If there is, it will take a long, long time.

Increased wafer sizes.

Fab utilization.

In the mid to late 1980s, fab utilization hovered at around 50% compared to more than 80% today. The utilization rate will never reach 100% because a wafer fab does not process wafers in a balanced continuous flow and some slack is needed to ensure that equipment needed will be available to process wafers. Most people expect the rate to peak at around 90%. Manufacturers are now running many fabs 24 hours/day, 365 days/year. There is literally almost no room for further improvement.

Fab utilization.

When analyzing semiconductor processing and looking at traditional areas for productivity gain, experts are surprised that there are a few areas where the industry has made only marginal gains or even lost ground. With so much growth, it is hard to believe we have not improved across the board. Consider the industry's vastly wasted potential for productivity gain. The individual drivers for productivity gain made in the last 10 years include: a 300% productivity gain from yield increases, a 400% gain from decreased feature size, a 400% gain from increased wafer size, and a 50% gain from better capital utilization, altogether a 72-fold increase in productivity. However, when factoring capital productivity (throughput/dollar spent) into the equation, as well as rising equipment costs (300%) and diminished equipment throughput (30%), the overall result is that the industry's total productivity gain was just one-fourth what it could have been over the past decade.

Remarkably, not much has been said about equipment productivity. It is as if it has been accepted that it won't improve. Why is this? Equipment makers can improve tool throughput to keep the industry ahead of Moore's Law. The problem is the new business paradigm required: sell fewer systems with higher productivity.

To design high-throughput systems at the lowest possible cost, some equipment companies have looked to alternative process chamber and wafer-handling designs. Process chambers evolved from batch reactors in the early days to single-wafer processing systems and then to cluster tools. The transition to single-wafer tools was driven by the need for improved process control, but the transition to cluster tools was mainly economic. If you put multiple chambers on one system, then you could reduce the cost of robots, the control system, and even reduce system footprint.

What is the next evolution in equipment design? My speculation, believe it or not, is a transition back to batch processing, but with a different kind of batch system, a multistation batch process. Multistation systems, which process multiple wafers simultaneously in one chamber, offer the high productivity of batch-processing systems with the process performance and control of single-wafer process tools. Just as multichamber cluster tools are saved by reducing redundant hardware, multistation process chambers can go one step further by reducing the number of rf power supplies, gas boxes, heater blocks, vacuum pumps, control systems, etc., that are required in a cluster tool. It is impossible for cluster tools to match the economics of multistation batch-processing systems.

Novellus was the first company to successfully commercialize the multistation concept with its Concept One system. Mattson Technology followed by extending this concept from CVD into photoresist strip and RTP systems. Applied Materials, the leader of the opposing archi-tecture, finally developed a high- throughput platform in 1998, responding to competitive pressures. It may seem hard to believe that Applied, the world's largest semiconductor manufacturing equipment supplier, would embrace an idea that means they will sell less equipment (higher throughput means selling fewer tools), but the company adapted to the challenge rather than clinging to old ways.

Today, having good technology alone is not enough to compete successfully. Companies must also emphasize throughput as a way to differentiate equipment. As this happens in each niche of semiconductor equipment, overall productivity should increase, which will allow the industry to keep pace with Moore's Law.

The toughest part of this is for equipment companies to get used to selling less equipment as a strategy. Just as companies in other industries have adapted after reaching maturity, semiconductor equipment companies must prepare to adopt a new business paradigm or prepare for the entire industry growth to be diminished through the failure of Moore's Law.