Complex masks: A problem or a solution?
06/01/2002
by Kurt Kimmel
International Sematech
The challenges of maskmaking have broadened in scope and deepened in complexity in the last decade as the industry forges ahead with the mantra "faster, denser, cheaper." There have been some hurdles to clear along the way in device physics, materials reliability, design systems, and, perhaps most significantly, lithography extendibility. The issues for mask users, however, go beyond mask cost and cycle time — the goal is cost-optimized image delivery.
Since the early 1970s, when binary chrome-on-glass masks were introduced, masks have undergone relatively few and infrequent changes.
 Max. cost/"enhanced" mask (assuming four new masks) above the "basic" mask cost, which achieves a break-even point assuming a $1660/wafer profit. |
Since the mid-1990s, however, the pace has quickened as the limits of 365nm technology have tested lithographers, while the burgeoning costs of 248nm lithography have tested fab managers and investors. The industry responded and introduced dramatic changes in materials with phase-shifting attenuators fori-line and deep ultraviolet; and enhanced patterning with optical proximity correction (OPC) features from simple line-end extensions to serifs and sub-resolution assist bars. Pattern transfer processing with plasma-based etching and chemically amplified resists, plus strong-shifter alternating phase-shift masks (PSM) were developed. These huge infrastructure development investments, along with increased mask complexity, have driven overall mask costs to higher levels at a steeper rate than at any other time.
So, while wafer fab managers lament the escalating costs of advanced masks, mask fab managers invest huge portions of their meager profits to buy equipment and develop processes to make needed mask enhancements possible. Collaboration in development, partnerships with customers to access capital, and consortia-led or government-led equipment infrastructure programs all aid in reducing overall development risk and cost. These efforts may not be enough, though, to meet the demands of accelerating product schedules, increasing capability requirements, and rising costs.
Furthermore, the maskmaking process has room for improvement. Many of the sophisticated yield characterization methods, adaptive process control feedback loops, and logistics systems that are integral parts of any competitive wafer fab are mostly just dreams in the mask fabs. Yet the challenge of mask yield improvement is unlike incremental wafer yield learning. A mask's yield can only be zero or 100%, so quantum yield gains are necessary. Mask yield is primarily driven by equipment capability at the patterning and repair sectors, plus defect minimization, requiring capable inspection. One fundamental problem for mask fabs is a lack of suitable equipment to meet technology roadmap demands. Maskmakers are thus forced to stretch their equipment to dwell in the low-yield tail of performance distribution.
The business case for equipment manufacturers to enter or continue investment in mask fabrication equipment is generally weak. The market is small and the technical risk is high. International Sematech (ISMT) has compared ratios of investment-to-product lifetime revenue between the mask and wafer equipment supplier industries, and the difference is between one and two orders of magnitude. The return-on-investment period is relatively long considering that throughput and profits in this business are poor, which at least partly accounts for the lack of motivation to aggressively fund development.
Underfunding can lead to late equipment introduction or higher-priced equipment to compensate suppliers for engineering development costs. ISMT seeks to impact these issues by focusing its funding on critical-path elements to maximize development investment leverage, and has, since its inception in 1987, spent more than $200 million on mask-related technology development.
The demands on mask equipment suppliers have become more severe as masks have become more complex. The comparison of prices for a leading-edge mask of just a few years ago to one today is not a comprehensive, business-oriented way to either justify or condemn the latter's cost.
The industry points to poor yields, ever-tightening specifications, and waning productivity as pattern generators and inspection systems get bogged down by growing design density. These claims may be true, but the test of the value of a technology is whether there is a business case to support its use. If a "complex" mask (i.e., PSM, OPC) costs $20,000 or $40,000 more than the equivalent nominal ground rule simple binary mask, that mask should produce additional end-product profit that compensates for the added cost. A common question many advanced mask users ask is, "Are they worth it?"
A first-order analysis of this yields some insights, but clearly, the variables in fully characterizing the problem are many and varied from business to business. Where microchips are the end product, the overwhelming variables are: 1) profit/wafer, which is nominally proportional to yield; 2) percentage increase in realized yield attributed to the use of complex masks; and 3) mask utilization, i.e., the total number of wafers to be exposed per complex mask over the product lifetime.
Assuming an industry average profit of $1660/200mm equivalent wafer derived from the 2001 SIA Annual Databook, and picking various mask utilization values to span the segments from ASICs and specialty devices running at 100 wafers/mask, to memories running at 10,000 wafers/mask, one finds that above ~5000 wafers/mask, the mask price premium tolerable for a break-even business case is multiple $100,000s/ mask. This is true even for a modest, base-yield increase of 10% attributed to the use of complex masks. So, for example, a memory or high-volume microprocessor fab running an advanced product with poor yield at 50% can choose to address a lithography-limited yield problem by using complex masks at some critical levels to improve the yield 10-55%, and can afford to pay $200,000 and more per mask. This result is not surprising, since the mask cost is amortized over many wafers.
The more stringent and interesting cases are for low-mask-utilization businesses.
Data collected by ISMT from multiple mask industry sources in July 2001 show that, for the 130nm node, the average difference in cost between a binary mask with "critical" specifications and a binary with "very critical" specifications plus OPC is $19,800 (see figure). The average difference from a binary mask to an attenuated PSM with the same "very critical" specifications is $40,000.
Using the profit assumption from the SIA and the case of 100 wafers for the product run, it will be difficult to justify spending the additional money for an enhanced mask, unless very large changes in yield are anticipated — on the order of 50%. However, for the utilization cases of 500 and 1000 wafers/mask, the nominal break-even-cost threshold levels are achieved at reasonable yield improvement values. Thus, for 500 wafers per/mask, the $19,800 and $40,000 thresholds are reached at about 10% and 19% yield increases, respectively. For 1000 wafers/mask, the thresholds are only about 5% and 10% yield increases, respectively.
Keeping in mind that this is not an increase in absolute composite yield points, but a relative increase from whatever the base yield is, these yield increase thresholds are modest. Again, these examples are based upon a profit of $1660/200mm equivalent and do not include other benefits like enhanced reliability and speed-sort binning distribution derived from the use of complex masks.
Mask purchasers must look beyond the price, overcome "sticker shock," and consider whether the value proposition makes sense for the end-product business case. The analysis shows in broad terms that complex masks appear to justify their increased cost for many business profiles. It may be time to include business case parameters along with the data file and the specifications for every mask order that reaches the mask fab. Business and technology-based decisions about mask complexity level, such as compliance to specifications and the fabrication process, could then be optimized for the true bottom line: profit at the silicon level.
A strong partnership between lithographers and maskmakers will achieve the goal of cost-optimized image delivery; perhaps masks will then be uniformly recognized as a solution and not a problem.
Kurt Kimmel is mask strategy program manager at International Sematech, 2706 Montopolis Dr., Austin, TX 78741; ph 512/356-3500.