By Roy White, RAVE LLC, Delray Beach, FL USA

Photomask costs are a painfully visible issue in today's competitive semiconductor market. This puts substantial pressure on the profitability of photomask makers. Since the "mask makers' vacation" ended some 15 years ago, it seems every paper you read has another exorbitant estimate of future mask costs.

Photomask makers point to the litany of issues they must overcome. The usual suspects are lined up and aggressively blamed (they would have been shot, but the shot counts are already too high) – high capital costs, yield issues, slow throughput, increased complexity, and ever-rising materials costs. Perhaps more important than the origins of the photomask cost issue is what to do about it. Firms typically combat eroding profit margins by increasing price or market share, reducing costs, or pursuing new markets.

However, photomask unit volume is steady and forecast to remain so – a sure sign of a mature market. To gain market share in a mature market you have to take it from another supplier, which is a long, difficult, and capital intensive proposition – and often leads to consolidation. The buying power of large semiconductor companies precludes any significant price increase, and new markets hold little promise to drive unit growth. Cost reduction is left as the only reasonable path to improved operating results.

The primary drivers of mask cost are capital equipment cost and yield. A next-generation equipment line costs more than $100M. These rising fixed costs drive the need for increased production output per site to achieve ROI (another factor in consolidation). While photomask equipment suppliers make every effort to control these fixed costs, recent papers (Hector, Lercel, et. al.) have illustrated the challenges for the photomask equipment supplier. These challenges are compounded somewhat by soaring demand (and thus skyrocketing prices) for raw materials like copper, stainless steel, and nickel as developing nations industrialize. Attacking the variable costs is therefore the best approach – that's where photomask repair comes in.

Repair converts scrap to product, directly enhancing the bottom line. Despite this, repair capability is often purchased 1-2 years (or more) after the rest of the next-generation equipment line, with mask shops accepting lower yields as a result. Figure 1 shows production ramps by node, normalized with Year 0 being the year of introduction. Ramp rates are very consistent node to node. For any given ramp, prototyping (Year 0) is typically done on existing equipment. Mask shops typically purchase write, inspect, and process tools in Year 1, and repair tools in Year 2 or 3. But, given the predictability of the ramp and high product costs, does this make sense?

RAVE used financial modeling to investigate that question. To construct the model, we used Dataquest's estimate of worldwide mask shipments, published estimates of the percentage of masks that require repair, and repair yields and scrap Paretos by mask type from Mask Industry Assessments. We conservatively estimated total scrap due to defects by applying a standard "S" curve for repair yield (assuming a minimum repair yield of 30%).

We applied our model to the 45nm node to assess cost impact, estimating cost to build a layer at $42,500 (very conservative) and assuming that the 45nm node ramp follows the average ramp rate of the last four nodes. Figure 2 shows this graphically. The different cost impact lines represent various market shares.

The model shows that by improving repair yield from 30% to just 60% on high end product, a company with a10% market share exceeds a 40% rate of return for Year 1 (5 year straight line depreciation) and payback in less than two years on a $7.0M repair tool. Opportunity costs, increased capacity, and having improved repair processes in place when the ramp accelerates strengthen this argument and make it reasonable for market shares as low as 4%. Additionally, the model is sensitive to mask cost, so if the actual cost of a 45nm layer is higher than our conservative estimate, savings will rise by about the same percentage. Some of our customers, for instance, have reported ROI in less than 12 months.

Many mask shops, however, indicate they do not want to invest in mask repair –they prefer to spend money developing "defect-free" processes. A noble pursuit indeed – but is it reasonable, or a Quixotic quest for a Holy Grail? Over 60% of masks produced today require some form of repair. So, even if mask shops invest large sums of money in defect free manufacturing, it will take several years to make significant inroads. Also, defects have consistently accounted for ~50% of rejects each year since industry Pareto charts were first published, despite existing defect reduction efforts.

Worse, both defect densities and the relative force of adhesion of particles on a surface increase exponentially as particle size decreases. And, environmental concerns and increasingly fragile patterns require that we implement less aggressive process technologies. So unless you have Harry Potter on staff or can otherwise alter the laws of physics, repair will become an increasingly important element in a profitable mask shop. Those with the best repair capability will have a significant operating advantage.

So what factors are most important when selecting a repair tool (or tools) for the next node? First and foremost is predictability. Photomask and semiconductor manufacturers must have predictability to run their businesses effectively, which means predictable deliveries, acceptances, and uptime. The ability to repair the highest margin product is also important, so removing material on AAPSM and high transmission MoSi is another key ingredient. You also want flexibility and capability for rapid process development – including the ability to repair iteratively, handle defects with irregular topography, and provide a large depth of focus for the repair to provide the largest possible process window at the wafer fab.

Additionally, the value of repair for high end masks was not the only finding of the study. The model also illustrated the high overall cost of defects. For example, defect losses cost more than $10M annually for a mask shop having a 4% market share — one third of that cost coming from the 130nm node and larger. This illustrates that scrap is significant at the oldest technologies, even using conservative S curves. This suggests that conducting joint development projects, investing in or acquiring a mask repair company, and/or buying newer repair technologies (e.g. replacing older laser "zappers" with newer femtopulse lasers) are all likely profitable strategies.

Repair has proven to be the fastest, lowest risk method to reduce mask costs. Recent publications (Lercel, et. al.) call out improved repair capability as a main factor keeping mask costs well below the alarmist predictions of the not-too-distant past. Photomask suppliers should prioritize purchase of improved repair tools, and build close partnerships with a supplier fully dedicated to the mask repair business. Sharing confidential data like product mix and defect type would lead to vastly improved repair tools in a joint development project. Wafer fabs should also pressure, and even partner with their photomask suppliers to invest early in repair capability, and in improving repair capability. Improved mask repair is a very low risk, high reward approach to dramatically improved operating efficiency.

Acknowledgment
The author thanks David Brinkley for valuable discussions in creating the model and assistance with edits.

Roy White received his MBA and MS in industrial engineering, both with honors, from the U. of Florida, and his BS in materials science from Carnegie Mellon. He is currently the director of operations for RAVE LLC, Suite 7, 430 S. Congress Ave., Delray Beach, FL 33445 USA; ph 561/330-0411, x287; e-mail roy.white@ravenano.com.