Lessons Learned from Certifying a SMIF Fab
Michael Gatov and Bruce Whitefield, LSI Logic, Gresham, Ore. -- Semiconductor International, 6/1/2000
 |
At a Glance | | |
|
| ||
| |||
Choosing a method of integration
A properly functioning SMIF system allows products that normally would require better than Class 1 ambient conditions to be built in a Class 1000 room.1 SMIF equipment generally includes a Pod Loading System (PLS), SMIF Pods and a minienvironment that isolates wafers from external room conditions. The minenvironment typically has its own filtered air supply.2 Minienvironments can allow a device manufacturer substantial savings in cleanroom construction and operating costs, though these are largely offset by the cost of SMIF equipment. Other benefits include yield improvements, prevention of cross-contamination between tools, flexibility and control of fab operations, and greater personnel comfort through relaxed gowning protocol. The minienvironments allowed a 50% reduction in the number of filters and lower airborne contamination levels due to reduced airflow.
Interestingly, though SMIF minienvironments were first installed in fabs over a decade ago, not all tool suppliers have developed SMIF solutions for their equipment.3 To meet this need, the device manufacturer can hire a full-service SMIF integration company to address design and integration issues; add SMIF to process tools as an internal project; or buy equipment only from suppliers willing to include SMIF integration.
By hiring a SMIF integrator, the manufacturer gains a consistent minienvironment design and a single source to resolve design and performance issues. Drawbacks to this approach include:
•Little chance that after-market modifications to tool design can provide better SMIF integration than a tool designed for SMIF.
•Difficulties in integrating interlocks with the host tool's software.
•Typical use of "arm" and "box-over" type designs that are less reliable than indexer solutions, occupy more cleanroom space and have poorer ergonomic attributes.
•Solutions limited by the expertise, methods and components of a single supplier.
•Divided responsibility for system performance between tool supplier and integrator, reducing the ownership commitment of both parties.
In the second approach the customer serves as SMIF integrator, allowing control of consistency between designs. In this way, the manufacturer develops internal expertise in maintaining and improving the minienvironments. Drawbacks include those that apply to the first method, along with the need for additional personnel requirements when resources are needed to support factory startup. In addition, fab engineers typically are not as qualified as tool manufacturers to understand particular tool requirements such as vibration, heat loads and attachment points.
The last alternative contractually requires each tool supplier to be responsible for providing a complete SMIF solution. Suppliers can select the SMIF components that best meet their needs; competitive bidding enters the purchasing process, and tool responsibility is clearly designated. As the SMIF approach has gained industry acceptance, this option has become more feasible. Some concerns remain, however, because:
•Some tool suppliers have not previously constructed a SMIF solution, so they lack expertise in Pod automation and minienvironment design.
•Tool vendors frequently underestimate the SMIF task.
•Many suppliers lack the equipment, expertise and/or desire to effectively test their minienvironment designs.
•Some suppliers SMIF components on a price basis only. The user can end up with shoddy equipment and an uphill battle to achieve redress.
•Many different designs and solutions are generated, requiring support for a wide variety of hardware.
Considering these points, as well as other issues of factory automation and supplier proximity, LSI Logic chose this last option. The company provided specifications to aid in the design, testing, shipping and installation of the machines, and established a comprehensive test and certification strategy. Each supplier was given "The Mini-Environment and Pod Loader System Specification," outlining the physical, communication, performance and reliability requirements for SMIF automation and minienvironments, as well as "A Mini-Environment Testing & Certification Procedure," detailing test procedures that must be passed before tool installation documents are signed off for final payment.
These documents are sent to the vendors before an order is placed as part of the equipment acceptance specification. Requirements are covered during design reviews, with LSI Logic engineering expertise made available to suppliers as needed.
Since process tools were going to be installed and tested at the same time the factory staff was being hired, certification services were contracted through Pentagon Technologies (Fremont, Calif.).
SMIF solutions delivered
Under this approach, LSI Gresham encountered many minienvironment solutions including complex multi-chambered add-ons, simple single-chambered add-ons, integrated OEM designs and simple SMIF-ARM attachments. LSI installed 39 designs from 25 suppliers. The majority of suppliers subcontracted SMIF integration, while approximately 20% implemented the minienvironment themselves. Since the tool suppliers tended to use the same subcontractors, only five different PLSs were deployed. Only seven of the minienvironment designs were of least-desired roll-up arm variety, and these were limited to metrology tools.
Minienvironment certification results
We applied 14 tests to assure compliance with the minienvironment specification (Table 1). If done consecutively without encountering a failure, the entire testing process could be completed in four hours for simple tools and 40 hours for complex tools.
| Table 1. Minenvironment Certification Tests | |
| Test name | Purpose |
| Functionality | Verifies that lights, switches, access panels, gauges and fans are functional. |
| Air velocity | Measures air velocity and uniformity through the filters to ensure they meet specifications. |
| Filter leak | Tests filter media and gasket seals to ensure no leaks in the media, media/filter frame, or filter frame/grid structure. |
| Pressurization | Verifies proper pressure relationships are maintained in three equipment status conditions (normal closed condition, minor door open and major door open). |
| Airborne particle counts | Measures particle counts in wafer transport areas and other critical locations. |
| Vibration | Verifies acceptable vibration levels transmitted to the tool. |
| Noise level | Verifies equipment generated noise below specified levels. |
| Airflow visualization | Confirms no dead zones, vortices or other airflow conditions that could cause cross-contamination or entrapment problems. |
| Recovery rate | Measures system's ability to recover from a contamination event or normal servicing following exposure to the ambient environment. |
| Surface particle accumulation | Verifies the minienvironment has been properly cleaned and is able to maintain cleanliness levels over time. |
| Static charge | Determines if inside minienvironment surfaces are holding an electrical charge in locations exposed to wafers. |
| Witness wafer test | Verifies that particle fall-out in the wafer transport zone does not exceed specified levels. |
| PWP (Particles per wafer per pass) | Measures particle gain on transported wafers from wafer handling systems. |
| Trace Chemical | Identifies chemical contaminants that may be detrimental to a particular process step. Specific tests needed are defined for each process. |
Our minienvironment certification process led to a first-pass failure rate of nearly 100%. Even after "first-of-a-kind" tools passed certification, follow-on tools required the majority of tests as many failures were caused during construction or installation. A Pareto chart of the failure modes (Fig. 1) indicates filter leaks and pressurization problems caused the bulk of failures.
|
|
Failures were traced to errors of design and execution. Design errors included poor minienvironment design, incorrect design assumptions and undefined set-points. Execution problems included incorrect installation, damage during installation, inadequate cleaning of the tool and incorrect set-points.
Of the poorly designed systems, some common root errors were evident, including:
•A misconception that if you can draw a picture of the desired airflow on paper, it will follow that path in the machine. In reality, air tends to find the path of least resistance.
•Turbulence or entrapment zones caused by placing a rectangular filter on top of an irregular inlet plane (for example, a polygon). To minimize turbulence, the inlet plane must have 100% filter coverage, or a diffuser membrane can be placed downstream of the filter to even out airflow.
•The presence of positive pressure alone does not prevent the intrusion of particles through gaps. While this is true of gaps near the downstream air exhaust areas, the opposite effect is evident when the gap is near the filter where high air velocity can cause the aspiration of particles through the gaps.
Minienvironments that fail due to design problems are not quickly "tweaked" to meet specification. Overall, correction times varied from a few hours to more than nine months, with redesigns requiring an average of 41/2 months. Ownership questions between suppliers and subcontractors caused some of the longest schedules. The time required to meet full conformance varied greatly (Fig. 2). We determined that most tools performed adequately to support factory qualification early in the process.
|
|
We observed some common themes with execution issues. Typically the installation crew was not trained or supplied with instructions, resulting in tools being released for certification with visible levels of dust, fan speed set-points not checked or pressure sensors not functional. During shipping or installation, some filters and filter frames were damaged.
These observations indicated many equipment suppliers do not have the design, testing or procedural expertise to provide tools with turnkey SMIF and minienvironments. In one design, for example, the filter was mounted to the air-box backwards, and no gasket or seal was used between the filter and frame. Other designs did not account for the fab environment. LSI's Gresham fab uses a raised-floor- ballroom configuration. Minienvironments designed and tested for slab-on-grade facilities require modifications to tune the airflow.
Many filters were damaged due to careless handling during disassembly at the factory, shipping or assembly at the fab. Some companies devised "Band-Aid" solutions for serious design flaws, such as caulking every surface that could leak instead of finding the source of leaks and rectifying the design.
On the bright side, the more responsive suppliers saw the project as an opportunity to gain such skills. The truly proactive suppliers initiated internal engineering teams to address issues and apply the knowledge across their product lines.
Corrective action
Determining the root of many issues starts with the level of SMIF/minienvironment technology and commitment by the equipment suppliers. Many vendors did not read the specifications or take them seriously. Although all suppliers were required to respond to the specifications, a response indicating compliance did not necessarily translate into acceptable performance. In some cases, specifications had not been sent from the sales department to the engineering group and were not addressed until late in the build cycle.
The degree of ownership for the minienvironment varied widely from supplier to supplier. Some companies took their responsibility seriously and mobilized resources to fix the problems. Others complied only when pressed, while a few refused to take responsibility unless forced contractually.
Minienvironment corrective action tended to follow a common process. Although we have been told we are the only company that ever experienced this procedure, we nonetheless summarized the steps (Table 2) as a service to other companies embarking on similar projects.
| Table 2. Minienvironment Corrective Action Process | |
| Stage | Typical comments |
| Denial | Show us that specification. Did we ever get this spec? The Sales Group never gave that spec to Engineering. |
| Cajoling | Are you serious about meeting that spec? |
| Wheedling | Do you really need positive pressure? If we have Class M1 conditions in half of the minienvironment, is that good enough? If the filter tear is away from the wafer area, do we still have to fix it? |
| Peer pressure | None of our other customers has complained. What's your problem? |
| Evasion | We didn't make the minienvironment. You can't expect us to be responsible for it! |
| Tears in the beer | We aren't making any money on this tool. We can't spend a penny more. |
| Indignation | We wash our hands of this entire affair. You are being unreasonable. |
| Horse trading | We've just developed a great solution, and we will sell it to you for only $xx,xxx. |
| Resignation | OK, we will meet our commitments. |
| Relief | Whew, it finally passed the certification. |
| Awards dinner | We are writing a paper on our great new minienvironment. |
Lessons learned
Every minienvironment needed to be tested. The fact that a similar tool passed certification does not necessarily ensure additional tools of similar design will pass. This relates mostly to filter damage and the need to properly set up the minienvironment. Even companies with excellent certification success had problems, such as filter leaks, which showed up on follow-on tools.
SMIF and minienvironment concerns rarely drive process tool selection. A poorly performing tool with a well-designed minienvironment will not be purchased for this reason alone. Likewise, a tool with excellent process performance is not going to be rejected due to lack of a proven minienvironment. Fortunately, with perseverance, even the most poorly designed and executed minienvironment can be brought into compliance.
Vibration and noise tests usually are not needed. Vibration issues typically are process limiting and therefore are covered by process performance specifications. Noise levels were either obvious or irrelevant.
Use of a third-party certification company proves valuable since there can be no suspicion of an unfair agenda. Pentagon Technologies performed a valuable service and was able to back up its data when challenged.
Scheduling time for the tests in the midst of many other fab start-up priorities proved difficult. This was compounded by the need for retesting after corrective action. Considering the time required to correct a defective design, certification tests for first-of-a-kind designs should be given high priority; second-of-a-kind tools can be tested later since installation issues are quickly repaired.
In our case, initial purchase orders did not require minienvironment testing prior to source inspection or supplier-provided set-points. This contributed to suppliers being able to ship without fully testing their product. In cases where the minienvironment design was tested prior to final shipment, deficiencies were corrected quickly. We modified the purchase specifications accordingly.
Conclusions
LSI Logic built a designed-for-SMIF semiconductor facility using supplier-integrated SMIF minienvironments. The approach achieved a toolset with more of the desirable SMIF configurations, resulting in reduced footprint and more reliable equipment.
The solutions currently provided by most equipment manufacturers require comprehensive testing to ensure performance. With a first-pass failure rate approaching 100%, the need for customer testing will remain a critical part of the tool qualification process until the level of expertise and industry standards improve. •
Bruce Whitefield has aBS in chemical engineering from Oregon State University In 20 years of semiconductor manufacturing experience he has had process engineering or engineering management responsibilities for wafer fab processes, equipment maintenance and assembly operations, including several facility start-ups. He is the process and equipment engineering manager for operations at LSI Logic's newest wafer fab in Gresham, Ore.e-mail: brucew@lsil.com
Michael Gatov has a BS degree in mechanical engineering from Brigham Young University. He has served in equipment design and contamination control projects during 19 years in the industry. He has been with LSI since 1996 and is the contamination control staff engineer at LSI's manufacturing facility in Gresham, Ore.
e-mail: mgatov@lsil.com
REFERENCES
- S. Abuzeid, "Comparing Particulate Contamination in Conventional and Minienvironment-Based Operating Cleanrooms," Microcontamination, July 1993, p.33.
- B. Liu, S. Yoo, "Isolation Ratio and Particle Performance Measurement of a SMIF System," Journal of the Institute of Environmental Sciences, Nov./Dec. 1997, p.23.
- B. Klumpp, J. Schliesser, O. Herzog, "Experiences with SMIF-Integration for Semiconductor Fabrication," IEEE/SEMI Advanced Semiconductor Manufacturing Conference, 1995, p.375.
