The last 15 years
Automation grows through adolsecence in the turbulent '90s.
Staff -- Semiconductor International, 6/1/1999
I
n the 1980s, the semiconductor front-end industry envisioned highly automated
fabs with
lots of robots and few people -- approaching the esoteric goal of "lights out
automation." The
industry downturn, starting in 1985, effectively derailed most of this vision
for the rest of that decade. Many of the sophisticated automation
prototypes displayed at SEMICON during that recession never made it to a
production fab environment. Since then, motivations for wafer handling
automation have not changed, though industry economic fluctuations have
altered the priorities for implementation.
The last 15 years
The impact on chip yields caused by handling 150 mm and larger wafers manually with vacuum wands and tweezers was a big lesson to learn. Once realized, nearly all wafer transfers became mechanized. Bulk, mixed mode, and single wafer transfer machines became standard practice for getting wafers from one cassette to another via suppliers like Proconics, Precision Robots, Mactronix, Faith and Microglass. Virtually all process tools became cassette-to-cassette mechanized, either with dedicated mechanisms or robotics. Wet sinks were automated, lithography systems were linked and multi-step metal sandwiches were integrated and automated with cluster tools.
Automation meant that the tools needed to know exactly "what was where" before any motion commenced to insure wafers, carriers, robots, process tools and people remained undamaged. The "wafer sleuth" strategy of randomizing lots, then collating data and analysis, put heavy demand on total wafer traceability. Numerous companies attempted to address the various parts of the fab logistics software universe. The degree of integration achievable with so many parts coming from so many sources has been limited.
 |
| OEM wafer handling modules, such as the Titan, must address the reliability and repeatability requirements of wet wafer processing, e.g., copper and CMP. (Source: Ayst Technologies) |
Over the past 15 years, relatively small companies with critical automation technologies have merged into larger companies, giving complementary technology availability with strong business infrastructure. This has enabled high growth rates, dynamic product development and a high level of customer satisfaction. The stage has been set for more mature fab automation (Fig. 1).
|
| Fig. 1. Transport and stockers comprise 70% of the current fab automated material handling market (AHM). (Source: VLSI Research Inc.) |
The next step: intrabay
With the success of interbay AMHS in the 1990s, the next major step is (by industry consensus) automated material transport among tools within the bay to attain maximum control and productivity in high volume fabs. Increased emphasis on this aspect has led to the development of technologies such as overhead, drop-down FOUP delivery systems for use in 300 mm fabs. I300I and related worldwide 300 mm programs have addressed compatibility issues collaboratively, leading to far more standardization in future generations of equipment. Emerging tool designs now have SEMI compliant interfaces essential to intrabay automation.
In the past, substantial variations were found in load and unload port interfaces from one tool to another. In fact, a 1991 study showed that of all the major tools in use in fabs, not one met the criteria of SEMI Standard E15, "Specification for Tool Load Port." Designing a universal intrabay AMHS system for such a chaotic situation would be a daunting task under any circumstances. As a result, the concept of People Guided Vehicles (PGVs) emerged, combining the flexibility of human handling with some of the advantages of automated tool-to-tool handling.
Recent impact...
So what impact have the recent "bad years" had on wafer handling automation? For starters, widespread 300 mm implementation that was to have been the real driver behind tool-to-tool automation implementation has been pushed back from ~1998 to beyond 2000. When 300 mm fabs do start rolling, "smart" PGVs and intrabay automation should be ready. The delay has given the industry a bit of a chance to assimilate the wafer handling automation progress made to date. In-tool wafer handling robotics have become more sophisticated and reliable. Things that were somewhat speculative two years ago are now more certain. For example, SOP/FOUPs and tool-mounted buffers seem certain for 300 mm.
This "slow period" has given equipment suppliers a chance to test and implement improvements on new 200 mm designs that were originally expected to show up directly on 300 mm systems. Most significant in this context is SEMI (E15) compliance -- a key enabler to tool-to-tool automation. This pause has made possible bridge tools for 200 mm wafers that can also be converted to 300 mm when the time comes. These tools are benefiting from the improved robotics that testing of 300 mm and FOUP handling has dictated. Since multi-wafer loading/ unloading of FOUPs as originally envisioned is proving problematic, suppliers are developing faster means of single wafer handling.
Productivity
The advantage of wafer handling automation has always been clear: an effective means of improving overall wafer fab productivity. It impacts each of the elements historically targeted for improving productivity, yields, circuit-density and wafer-size. Yields, in particular, are affected by reliable automated wafer and wafer carrier-handling systems because of the isolation provided from sources of contamination. The improved yield capability has, in turn, facilitated rapid feature size shrinks and the migration to higher functional densities on these chips. With larger wafers, wafer-handling mechanization is necessary to eliminate yield-limiting handling damage.
|
| Fig. 2. In this example, OEE is 43% of total machine time with IEE comprising 58% of OEE; thus value is being added only 25% of the time. (Source: International SEMATECH) |
However, a typical mistake in automating the handling of wafers through the fab is simply replacing people with robots. This is often very costly and inefficient. With engineering analysis of the process flow, dramatic reductions in handling content, as much as 75%, can be realized, boosting productivity through reductions in cycle-time. This provides a solid foundation for further productivity improvement through automation.
In-tool handling
In-tool, also referred to as intra-tool, wafer handling is already big business at about a billion dollars per year. While in-tool wafer handling apparatus has historically been made in-house (by the process tool suppliers), use of OEM modules for the "front-end" and central wafer handling robot in new process tools is becoming common. These in-tool wafer handling systems have become very sophisticated AMH systems themselves. The issues of Table 1 are driving even the largest suppliers to consider outsourcing and standardization between tool models to amortize costs over a larger base.
The in-tool AMH systems operating within the overall fab AMHS include multiple pod I/O ports, internal buffers, multiple internal handoff ports to the process side of the tool and, in some cases, pod buffering. This challenges the reach (working volume) and capacity (speed) of the robotics. It seems to favor multiple high-performance polar coordinate robots with overlapping workspace versus the now classic robot on x/y rails common to stockers.
Intrinsic Equipment Efficiency (IEE) analysis identifies in-tool handling as another area ripe for improving fab productivity. Application of industrial engineering concepts began under the names of Overall Equipment Efficiency (OEE) and Total Productive Manufacturing (TPM). These efforts have led to the identification and management of sources of "overhead" or "non-value-added" time, including wafer handling within the tool and in the fab using the new metric of IEE. A new standard, SEMI E79-0299, gives basic metrics and methods for measuring equipment productivity and its key components (Fig. 2).
The in-tool wafer handling robot market has been divided between atmospheric and vacuum applications. "Wet" applications is now emerging as a major new segment, driven by the high growth in demand for CMP and copper processing tools. New robots designed "from the ground up" appear to meet these needs better than "hardening" of existing designs. 1
Automation software
|
Table 1 Wafer Handling Expectations and Approaches
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Manufacturing Execution Systems (MES), essentially big databases interfaced to simulation models2 of the fab and planning systems, keep track of all material, process recipes, equipment status, setup parameters, operators and other critical factors in the fab. MES systems are critically needed to help management run what has to be among the most complex of all human endeavors -- integrated, automated semiconductor chip fabrication.
Underlying the MES is the Material Control System (MCS, Fig. 3). Tracking wafers through their intended handling steps is one level of concern and complication. Exception handling is a far greater software task. It is not uncommon for the "routine" software to comprise about 20% of the code and 80% of the exceptions and maintenance aspects. People, being very creative by nature, tend to add to process variability in any enterprise. An AMHS, including robust MCS, tends to remove a lot of this uncertainty making it possible for fabs to maintain better control.
Sophisticated, integrated software is the essential ingredient for implementing most of the future enhancements. This is further evidenced by the recent business integrations of the industry's major AMHS and factory control system suppliers. This is where resources have to be invested. From discussions with automation suppliers, it appears that getting the functionality and reliability needed for wafer handling automation software will likely consume 90% of suppliers' development staff-hours budgets.
Automation's future
End-users have come to expect aggressive standards for wafer handling equipment performance. To meet these expectations, suppliers are finding they need to enhance their wafer handling system designs. Some of these expectations and supplier responses are listed in Table 1. Future fab automation may encompass these and more:
Sealed transport carriers (SMIF or FOUP) from
~25% today to ~100% (300 mm wafers) with pods inert-purge compatible
-
Tool-to-tool automation
-
Overhead drop-down style transport
-
Automated reticle storage, handling, transport and tool-loading
-
3D-volume-optimized fully automated fabs?
-
Extensive microlan parameter monitoring
-
Single wafer lots with automated transport (by conveyor or equal)
- No-moving-parts-contact handling/transport (MAGLEV or equal)
Two areas are addressed further -- single wafer handling and monitoring.
Single-wafer handling
What is the future of single-wafer handling? Some companies are said to be seriously considering switching from the traditional 50/25 wafers-per-run lots to single-wafer lots and single-wafer handling. The pending migration to 300 mm wafers forces a step in this direction --reducing lot size to 13/25 wafers for ergonomic and risk management considerations. Going to single-wafer lots and true single-wafer handling has the added potential advantage of dramatically reducing WIP and throughput-time (the balance of the wafers in the lot are not sitting waiting while each wafer is being processed in single-wafer tools).
|
| Fig. 3. This pyramid chart depicts how the various components of the factory control system, AMH system and process tools interact to control wafer handling. |
The counter issue is that full-process-flow single-wafer-lot processing and handling puts tremendous loading on the fab's automation systems. It is not clear that anything like current AMH system approaches would have the capacity for volume single-wafer-lot fabs. Much research will be needed to prove the capacity of existing approaches or to develop new ones. Right now, real single-wafer activity seems to be limited to means of getting monitor and test wafers to metrology stations without taking all the rest of the wafers too.
Low-cost monitoring
How will low-cost microlan-based data automation impact the number of parameters monitored? The functional integration, miniaturization and cost reduction enabled by semiconductor technology can be reinvested into the more productive manufacture of semiconductors through extensive parameter monitoring automation. Microlan technology makes it mechanically feasible and cheap. This could be the enabler for moving fab automation, including wafer handling automation, to the next level of intelligence, a level of sophistication implied by the expectations of Table 1.
|
| Fig. 4. Strong growth is seen for fab automation through 2003, exfluding the in-tool market. (Source: VLSI Research Inc.) |
Microlans are based on small sensors capable of sensing and digitally communicating just about any process or tool performance parameter, from position and temperature to voltage, current, pressure and humidity. Two hundred or more sensors can be connected via 2000 ft of telephone type wire and as many as 26 microlans can be connected per PC serial port. That's over 10,000 monitor points and 50,000 ft of wiring per typical two-serial-port PC. Thus, 200 parameters can be monitored on a process tool for an incremental OEM cost of as little as $1000. A fab can independently monitor 50 parameters on 200 process tools and/or pieces of support equipment for an estimated monitor equipment cost of only $100K (not including installation labor). That is fab-wide independent monitoring for about 0.01% of the cost of the fab and its tools.
In addition to helping implement the requirements of Table 1, this microlan technology could provide other benefits as a bonus. While monitoring temperature and current flow to the motors of robotic systems to warn of potential system degradation or fault condition, it can also monitor and alarm for potential fire situations. In addition, energy consumption of subsystems could easily be monitored continuously to document energy conservation results.
Conclusion
Fab automation is becoming a major market segment: it is projected to be
$2 billion per year by 2002 (Fig. 4). Not only has fab automation
survived the economically turbulent '90s, it has emerged with more
robustness and far greater maturity. 
References
1. Discussions with Anthony Bonora, Dick Dexter and Ray Martin of Asyst Technologies Inc., March 1999.
2. Semiconductor Online, 2/15/1999, "Simulation Used To Model and Test Improvements in 300 mm Fab Throughput" by Philip L. Campbell, IBM and Michael R. Norman, AutoSimulations.
|
Charles VanLeeuwen has provided semiconductor industry consulting services through CAVLON Associates since 1985. Formerly fab equipment automation manager for Intel Corp., he has BS degrees in Solid State Technology and Digital Circuit Design from Oregon State University. Phone: 505-892-3319 Fax: 505-891-1785 E-mail: cavlon@highfiber.com |
|
Jim Irwin is principal of Irwin Consulting (Austin, Texas), an affiliate of VLSI Research Inc. Mr. Irwin spent four years at SEMATECH working on fab automation and earned his BS Degree in Industrial Technology from Texas A&M University in 1969. Fax: 512-502-0797 E-mail: irwin@kdi.com |
