The
07/01/1999
Tim Bailey, Jeff Clark, Auto-Soft Corp., Salt Lake City, Utah
Automated individual wafer tracking can eliminate misprocessing and lower material costs by detecting and correcting wafer-handling errors. In addition, automated checking of test-wafer contamination levels precludes the possibility of tool or lot contamination. Wafer-level tracking also delivers the visibility needed for thorough and accurate process analysis, and in some cases it eliminates the need for inscribing a second wafer identification number. Such savings at the wafer level go right to a company's "bottom line."
Fab managers are always looking for ways to automate and reduce manufacturing costs. During downturns or times of excess fab capacity, however, automation systems that reduce material costs or enhance yield are far more important than ones that increase throughput. The latter is not as crucial in downturns, since the fab can already produce more than it can sell. In such situations, increasing throughput has a small effect on profits because it does not reduce equipment depreciation charges.
In contrast, reducing direct and indirect material costs, which account for as much as 20-30% of IC costs, has a very favorable impact on profits - as long as throughput is not compromised. This favorable profit impact holds true in both good and bad economic times. In fact, lowering material costs is key to keeping chipmaking costs dropping at the expected rate of 25-30% per year.
The challenges
Conventionally, material and processing costs/IC are reduced by increasing the number of die/wafer. Beyond this straightforward gain, semiconductor manufacturing presents a number of challenges to lowering material costs:
* For maximum fab yield, a cassette carrier must contain all of the product wafers it is supposed to have and no others throughout the fab process. Unfortunately, mixing wafers between lots can occur at various process steps, resulting in misprocessing, scrap, rework, or tool and lot contamination. The only sure way to prevent wafer mixing is to track work in process (WIP) down to the wafer level.
* Maximum test wafer usage, which minimizes test wafer cost, requires contamination level tracking of individual test wafers. To maximize usage, test wafers that have become too contaminated for further use in a particular tool can be "stepped down" in contamination classification and reused on other tools with more tolerance for contamination. The reclassified wafers must be strictly tracked and controlled, however. Accidental use of test wafers with incorrect contamination levels will cause tool and lot contamination, resulting in scrapped wafers and significant downtime to clean contaminated tools.
* Effective process analysis, which is directly related to improving yield, requires accurate logging of individual wafer data through every step in the fab process. Manual entry of analysis data is undesirable because it is error-prone, time-consuming, and far from comprehensive.
Addressing these challenges is complex because they must be resolved without decreasing throughput. Even though throughput improvements are less important in downturns, throughput reductions are totally unacceptable at any time. Also, throughput improvements will again become a key concern when excess capacity no longer exists. So, investments that reduce material costs must at least maintain throughput to produce good returns.
Wafer-level tracking
An automated wafer-level tracking system (see "Elements of automated wafer tracking," above) addresses the above challenges associated with lowering material costs. Such a system combines optical character recognition (OCR) hardware, such as Recif`s IDLW8 noncontact OCR reader, with wafer tracking software, such as Auto-Soft`s WaferTrax. (This combination can align, read and verify a full cassette of 25 wafers in <90 sec, and it supports alphanumeric, bar code, and 2D matrix laser scribes on the front or backside of the wafer.)
By ensuring correct processing and accurate tracking, this system has the potential to lower material costs and facilitate yield improvements. For example, wafer tracking eliminates misprocessing and thus reduces material costs by detecting wafers that have been placed in the wrong lot due to equipment or operator error. Automated tracking also reduces material costs by facilitating additional reuses of test wafers without risk of tool or lot contamination. Safe test wafer reuse is made possible by automatically verifying the correct contamination level of individual test wafers before a lot is mounted on a tool.
Tracking to the wafer level additionally provides a historical database of wafer-to-lot relationships and wafer slot positions for every lot split, lot merge, and tool operation in the fab. This level of tracking ensures accurate process characterization and simplifies process analysis, resulting in higher product yields and lower material usage.
Another benefit of automated wafer tracking is that it eliminates the need for inscribing a second wafer ID number. In many fabs, each wafer is inscribed with its own "second ID number" that is short and simple, because operators cannot reliably identify wafers using the long (12-18 characters) and unintelligible ID numbers of wafer suppliers.
Although the second ID is necessary for reliable wafer identification, it is added at considerable cost. First, this scribed ID takes up valuable wafer real estate, resulting in as much as 1.5% fewer die/wafer. Second, adding the second ID also adds laser scribing and bath wash steps to the process, increasing wafer contamination and decreasing process throughput.
With an automated wafer-level tracking system, only the wafer-vendor ID number is needed to identify wafers accurately and to verify lot integrity. Automated tracking systems, in conjunction with OCR readers, can easily identify the long, esoteric wafer-vendor ID numbers that are not comprehensible to operators. And since wafer vendors can and will inscribe a unique ID number on each wafer, a second wafer ID number is no longer necessary.
Elimination of the second wafer ID obviously eliminates the costs associated with its inscription. Probably the largest cost saving is the ability to map up to 1.5% more die on each wafer, which increases unit output by the same percentage with virtually no increase in production cost. That alone could be worth millions of dollars, but additional savings are realized through eliminating the laser scribe and bath process steps. With fewer process steps, wafer contamination and production cycle times are reduced.
Independent vs. MES tracking
Although some manufacturing execution systems (MES) can track to the wafer level, wafer tracking by the MES is usually not an acceptable alternative because it slows down MES response time. With MES wafer-level tracking enabled, the MES must now track 25 wafer IDs in addition to the lot ID. As a result of the increase in MES data traffic and analysis, MES response times slow down and throughput decreases, often to unacceptable levels.
Off-loading the wafer tracking burden from the MES using an independent wafer-tracking system avoids slowing MES response times. A separate wafer-tracking system only requires the lot ID from the MES to ensure the integrity of every wafer in the lot. The 25 wafer IDs of a lot do not need to be passed over the network to the MES and then analyzed by the MES to verify that all wafers are correct. Thus, wafer tracking systems that are not part of the MES can ensure individual wafer integrity and still maintain MES throughput.
An MES also falls short as a wafer-tracking system because it does not have built-in interfaces and cell control capabilities for OCR equipment. If wafer tracking is to be done by an MES, a separate OCR or sorter cell controller must be developed and interfaced to the MES system.
Addressing misprocessing
As noted above, one of the greatest benefits of wafer-level tracking is the prevention of misprocessing that results from wafer mixing at various process steps. For example, wafer mixing can happen when wafers are processed in batch processing systems, such as diffusion furnaces. After batch processing steps, wafers must be returned to their proper cassettes. Unfortunately, when 200 wafers in a batch tool are being returned to their proper carriers, some wafers may be placed in the wrong cassettes. With automated wafer-tracking software and an OCR reader, any misplaced wafers can be quickly identified and replaced in the correct lot carriers prior to the next operation.
Wafers can likewise be misprocessed when monitor wafers used to calibrate process tools are placed in the first or last slot of the wrong lot carrier. A wafer-tracking system can quickly correct such a mistake by reading and verifying whether the wafers in the first and last slot positions of a cassette are correct.
Wafers processed sequentially, such as on a resist process track system or a stepper, can also be misplaced and result in misprocessing. For instance, if a stepper cannot align the last wafer or two in a lot, the unprocessed wafers may inadvertently get left in the stepper and eventually be placed in the wrong carrier. Once again by verifying lot integrity with wafer-level tracking, any misplaced wafers can be found and returned to their proper lot before any damage is done.
We estimate that prevention of the misprocessing caused by wafer mixing can save 2 to 4 lots/month. In a production operation that produces 100 good chips/wafer, with a chip price of $100 each, two lots/month would equate to a saving of at least $6 million annually.
Test wafer reuse
Although it is not common, some wafer-tracking systems also include the capability to extend the number of times test and fill wafers may be reused. Every day, dozens of test wafers are prematurely scrapped or reclaimed. Because of the risk of tool contamination, many fabs elect to throw out or pay to reclaim test wafers rather than lowering the contamination level classifications of those wafers and reusing them on other process tools. Premature scrapping occurs because manual wafer tracking systems cannot ensure that process equipment will never be accidentally contaminated by reclassified test wafers with the wrong contamination level.
On the other hand, automated wafer contamination level tracking allows some test wafers to be stepped down in classification and safely reused on tools with less demanding contamination requirements. This minimizes the risk of accidental tool contamination because all test wafers are verified for the correct contamination level before each process step. As a result, reclassified test wafers can be reused at other process steps without fear of contaminated lots or unscheduled downtime for cleaning contaminated tools.
Since acceptable contamination limits vary by tool type, a test wafer can be reclassified and reused on as many as two to three different tool types before it must be reclaimed. This can result in a large cost saving since it is estimated that test wafer costs are approximately equal to 10% of raw production wafer costs [1].
For example, reclassifying test wafer contamination levels to reuse them on one additional tool type before reclaiming would reduce test wafer costs by 50%. A fab with 6000 production wafer starts/week would need to purchase 600 new or reclaimed test wafers every week or 31,200 test wafers/year. Assuming an average new or reclaimed wafer price of $80, the 50% savings from reclassifying test wafers would amount to $1.25 million annually.
Process yield analysis visibility
Increasingly, process analysis requires an accurate and comprehensive historical database that tracks individual wafers through the fab process. Unfortunately, tracking to the wafer level is complicated by the multiple lot splits and merges required for thorough process analysis. Twenty or more lot splits and merges may be necessary to characterize a process accurately. With so many splits and merges, it is difficult to track individual wafers manually without errors occurring. Accurate analysis requires an automated system that tracks changes in wafer-to-lot relationships after every lot split or merge (Fig. 2).
Tracking wafer slot positions in carriers is necessary to determine if process yield is affected by the location of a wafer in a process tool or carrier. For accurate analysis, a fab needs automated slot position tracking since many tools reposition wafers in different carrier slots after processing. For example, some process tools "compress" the positioning of wafers in lot carriers (i.e., they fill up empty carrier slots between wafers by replacing wafers in contiguous slot locations, leaving all empty slots together at one end of the carrier).
 FIGURE 3. An example of how wafer position changes can be tracked when processed by tools that compress or reverse the wafer order in a cassette. |
Other tools reverse the order in which wafers are placed in slots (i.e., a wafer removed from slot No. 1 is placed in slot No. 25, No. 2 in No. 24, etc.). In addition, whenever operators get involved in placing wafers in cassettes, automated tracking is the only sure way to know which wafers are in which slots at each process step (Fig. 3).
For accurate process characterization, the exact path each wafer takes through the fab must be accurately documented. Time-stamped wafer tracking data on wafer-to-lot relationships, lot carriers and wafer slot positions in carriers must be recorded for individual wafers at each process step. Only an automated wafer-level tracking system can provide the data integrity required to simplify process analysis and ensure accuracy.
Fast payback, low risk
When capital is tight, as during industry downturns, there are typically some additional constraints on fab investments. First, large returns and fast payback are absolutely mandatory. Second, all investments must be very low risk.
Proven wafer-level tracking systems meet both criteria, offering large, quick returns at very low risk. As the previous examples of cost savings illustrate, a wafer-tracking system, including software and OCR readers, can pay for itself in as little as three months, providing excellent return in the same year it is implemented.
In light of the many advantages of a wafer-tracking system, some may be tempted to develop and maintain their own custom tracking system. However to lower investment risk, consider implementing a commercial tracking system that is already in use at other fabs and has a reliable track record. A commercial tracking system tends to be more robust and much less expensive to develop and maintain than a custom system. A commercial system can typically be implemented in less time, as well.
Conclusion
Clearly, automated wafer tracking provides key visibility into fab operations; after all, you cannot change what you cannot see. Wafer-level tracking lets you see down to the smallest unit of production in the fab. With that kind of visibility, users now have the ability to:
* correct errors before misprocessing occurs,
* reuse test and fill wafers without risk of contamination,
* improve processes through detailed wafer analysis, and
* increase die counts by eliminating a second wafer inscription.
When wafers were smaller than 150mm and accounted for much less revenue/wafer, tracking only to the lot level made sense. Having to scrap one or two of those smaller wafers due to misprocessing did not result in a large dollar loss. Now that wafers are larger, however, and a single processed wafer can be worth tens of thousands of dollars, the penalty for misprocessing even a single wafer is huge. To prevent such losses and reduce material costs, tracking or getting visibility to the wafer level is a necessity. n
Reference
1. K.V. Ravi, "Wafer Requirements - Logic Devices," Proceedings of the First Semi Silicon Wafer Symposium, p. G-5, 1998.
Tim Bailey received his BSEE and MBA from Brigham Young University. He has 22 years of experience marketing hardware, software and electronic instrumentation products. Bailey is market development manager for wafer-tracking products at Auto-Soft Corporation, a subsidiary of Daifuku Ltd., 5245 Yeager Rd., Salt Lake City, UT 84416; ph 801/736-3316, fax 801/322-1846, e-mail [email protected].
Jeff Clark received his BSEE from the University of Louisville. He has more than 20 years of software development experience. Clark is a software engineering manager for Auto-Soft Corp., 1580 N. Fiesta Blvd., Suite 100, Gilbert, AZ 85233; ph 480/632-7870, fax 480/545-7997, e-mail: [email protected].