Automation and control for 300-mm process tools
Mitchell A. Drew, Brooks Automation Inc., Chelmsford, Massachusetts
Michael G. Hanssmann, Dan Camporese, Brooks Automation (Canada) Corp., Vancouver, British Columbia
Working through industry consortia, the world`s semiconductor device manufacturers are driving the development of 300-mm process tools. They have begun to provide concrete process performance and cost-of-ownership (COO) benchmarks, as well as detailed mechanical automation and software control specifications against which they will evaluate equipment suppliers for 300-mm fab suitability. The level of detail provided by these unified worldwide groups is unprecedented.
The automation systems designed for these tools must not only meet the physical, mechanical, and sophisticated software control requirements of 300-mm wafer processing, but also be more cost-effective and reliable than ever before due to the increased costs of downtime, misprocessing, and scrap. The shift to 0.25- and 0.18-?m device geometries will coincide with the 300-mm life cycle, increasing the sophistication and magnitude of change required in tool automation and control.
The I300I group is targeting 300-mm tool throughput:footprint ratios that are equal to those of existing 200-mm tools, at 1.3? the current cost. Due to the high costs of raw wafers (50 unprocessed wafers can cost $75,000-$125,000) and processed wafers (worth up to $250,000 each), I300I is targeting 99.995% equipment process yield . As another cost saving measure, device manufacturers are requesting tools able to handle and process both 200- and 300-mm wafer sizes, allowing use of 200-mm wafers for R&D, qualification, and test purposes.
Groups such as I300I, SELETE, J300, and SEMATECH`s 300-mm Equipment Control Standardization (ECS) Requirements Focus Team are basing a number of their performance and evaluation criteria on existing SEMI standards, particularly those under development in the areas of Wafers & Carriers Interfaces, ARAMS, Cluster Tool Communications, and Factory Integration and Controls. A list of related standards to date is provided in Table 1. Guidance documents are being prepared and circulated to equipment suppliers to facilitate implementation by projected 1997 demonstration dates and 1998 pilot line start-up schedules. Both I300I and the 300-mm ECS Requirements Focus Team have recently released preliminary guidance documents regarding specific evaluation criteria for process tool performance and equipment control.
Manufacturers are using the 300-mm transition as a catalyst for taking complete factory automation and sophisticated process control techniques to a much higher level. Development of this completely new type of integrated factory control system is expected to take place using 200-mm wafers before implementation for 300-mm wafer production.
This article surveys proposed 300-mm fab requirements for process tool automation and control, summarizing the key requirements for automation of wafer carriers and interfaces within a cluster tool. These requirements include incorporation of advanced process control techniques, recipe management, factory-wide communications for data collection and wafer/lot ID tracking, and increased equipment utilization.
Figure 1. Relative footprint sizes of 300- vs 200-mm cluster tool platform.
Wafer carriers and interfaces
Loadlock configurations. Initially, cluster tool transport, cassette, and process module designs will be scaled up to accommodate 300-mm wafers, resulting in larger overall footprints (Fig. 1). However, to meet emerging throughput/unit of floor area expectations, equipment manufacturers are beginning to evaluate nontraditional cluster tool footprints and loadlock configurations for next-generation tools. These new tool configurations will require automation platforms with more flexible loadlock interfacing capabilities.
In 200-mm tools, typical loadlock offsets (the angle between the operator access to the loadlock and the loadlock`s interface with the transport chamber) were either 0? for rectangular loadlocks, or 30? for traditional cluster tool platforms. For nontraditional 300-mm cluster tool platforms, loadlocks will be required to function within a range of 0 to 20-32?. One approach to this type of multiangle flexibility, used by Brooks Automation, is a new loadlock indexer that moves both horizontally and vertically to accommodate a range of loading angles.
Single wafer handling. To handle the increased size and weight of 300-mm wafers, robot arm and end-effector designs will have to be stronger, stiffer, and heavier. All designs developed for 300-mm wafers will be required to handle 200-mm wafers as well.
Unprocessed 300-mm wafers (125 g) are about 2.25? heavier than 200-mm wafers (56 g), and the larger cluster tool footprints translate into 50% longer arm extensions. An average six-sided 200-mm cluster needs a 735-mm extension, whereas a similar 300-mm tool will require a 1050-mm reach. Robot arms will thus be expected to travel further with more weight, while maintaining throughput comparable to 200-mm tools.
A recent design using a "time optimal trajectory" technique, which helps to offset the weight:speed:length issues, has resulted in throughput increases of up to 23% over conventional designs. This technique allows the robot arm to calculate the trajectory for each movement in advance, and apply the velocity and acceleration that will maximize transfer speed, while remaining within the breakaway acceleration limit for the application and maintaining smooth end-effector movement. Smooth movement reduces the risk that the end effector will abrade the wafer backside, generating particles.
The longer, heavier arm also makes maintaining a straight trajectory more difficult. Despite these limitations, the SEMI standards committee increased vertical motion allowances and process module entry slots by only 4 mm, from 6 to 10 mm, and 46 to 50 mm, respectively . These dimensions were minimized due to conductance and contamination concerns.
I300I is accepting a 3-mm edge exclusion for wafer handling for equipment evaluation purposes . Still, to reduce particle contamination from backside abrasion, a 2-mm zone for both front and back surfaces while in transport has been discussed . Although each individual transport function can be designed to meet these requirements, the hand-off and pick-and-place of wafers between modules becomes more difficult because two modules (e.g. a transfer arm and an aligner) must meet the requirements simultaneously. Thus, the entire cluster tool automation system -robot arm handling, aligner spinning, and so on - becomes more complex. Designing this capability into new tools will require a significant capital investment, so tool suppliers are evaluating alternative approaches that could achieve the required particle performance without major design changes.
Wafer alignment. Current 200-mm wafer aligners use a notch on the wafer as a reference point, as will initial 300-mm tools. As device features shrink, however, the exact alignment of the wafer between processes will become even more critical. Fabs are considering a move to notchless wafers, with marks on the surface to provide alignment information that will be read by an optical device. The marking of the wafer would also allow wafer IDs to be read for tracking purposes - another requirement for 300-mm fabs. Wafer aligners would thus have to be modified to include optical readers and sensors.
For more reliable hand-offs, new 300-mm automation tool designs will use sensors to read the actual position of the wafer, reducing blind placement and allowing arms to adjust for such inconsistencies as warpage in plastic cassette boxes.
Cassette handling and load ports. Proposed fab designs for 300 mm allow several additional configuration possibilities for the space in which the cassettes or box/pods are handled and delivered to the load ports. Initial 300-mm wafer cassette designs are for 13 wafers - 12 product and one test wafer - but I300I is requesting equipment demonstrations with 25-wafer cassette capabilities. Fabs are expected to use 13-wafer cassettes for initial pilot lines, then switch to 25-wafer cassettes for volume production.
Wafers may be transported in open cassettes, boxed cassettes, or pods, currently defined by SEMI as either bottom or front-opening . They will be delivered to the load ports either manually by person or cart, or automatically via overhead track systems or automated guided vehicles (AGVs). While 13-wafer cassettes are still light enough to be handled by factory personnel, 25-wafer cassettes are quite heavy, and will require carts or automated systems for transport. SEMI is defining a mechanical standard and transfer mechanism for cart-to-tool docking applications to ensure repeatable alignment between the transport cart and the tool without manual operator intervention. Meeting all of the configuration possibilities will require a variety of load port options and communications interfacing between factory automation systems (Table 2) .
Kinematic couplings are replacing the former H-Bar specification for cassette-to-tool loading. The kinematic coupling, specified as three rounded pins mating with three V-shaped grooves for pod-to-tool, cassette-to-pod, or cassette-to-process chamber floor interfacing, increases the accuracy of manual or automatic cassette and pod positioning the load port. Table 3 shows the advantages of the kinematic coupling design vs. H-Bar . This design change will need to be incorporated by both tool automation platform and cassette/pod suppliers.
Both front-opening and bottom-opening pods will require a mini-environment in the loadlock area in front of the cluster tool to maintain the pod environment and minimize contamination. With the front-opening pod design, the cassette may be an integral part of the pod itself: the pod is lowered down onto a load port, then a mechanical door opening mechanism (as defined by front-opening interface mechanical specification) opens the front of the pod for automatic unloading of wafers into the cassette elevator. A bottom-opening SMIF pod design, in contrast, requires either removal of the top of the pod, or lowering the cassette through the open bottom. The pros and cons of bottom-opening vs. front-opening pods are currently being debated; the direction that the industry takes will depend on which format the first 300-mm semiconductor fabs choose to order. Although bottom-opening pods have some track record for 200-mm wafers, neither pod format has been proven for 300-mm production .
The Factory Integration Working Group conducted an informal survey during one of its meetings in the summer of 1996. Most device manufacturers say they will protect wafers with containers throughout the fab. They will most likely start with open or boxed cassettes, then move to front-opening cassetteless pods when technologies are field-proven. Significant time and capital investment are needed before fabs will feel comfortable with pods and mini-environments. The most recent I300I guidance document specifies a front-opening unified pod configuration for demonstration purposes.
Equipment control strategies
The high dollar value of each 300-mm wafer heightens the risks associated with process steps and wafer movements. To minimize misprocessing and breakage, operator error must be significantly reduced through improved automation and software control, more controlled recipe management, and tighter factory integration. Errors in processing or process drift must be detected and reported in real-time and corrected within the next process step or wafer, rather than a few days and 1-10 cassettes later. More in-situ sensors and tighter process control methods will be needed to obtain consistent process results across a wider, more complex wafer surface.
Figure 2. MIMO control techniques use multiple zone sensors and heaters to improve temperature control over a larger wafer surface.
Over the past decade, advanced techniques such as run-to-run process control and statistical process control have been studied and implemented to a limited extent within production fabs. The process yield requirements of 300-mm wafers, combined with 0.25- and 0.18-?m device geometries, will force the mainstream implementation of these techniques. Two primary groups within SEMATECH - the 300-mm Equipment Control Requirements Focus Team, and the Control Systems Standardization Working Group - are attempting to set baseline requirements and evaluation standards for 300-mm fab equipment control and factory interfacing.
Temperature, uniformity, and timing requirements. For the first 300-mm tools, process chambers will be modified to perform the same processes qualified at 200 mm. This scale-up will have little effect on software control systems. Still, obtaining the same uniformity results across the larger wafer may be difficult because the thicker 300-mm wafers are more sensitive to temperature variations, and cannot be heated and cooled as quickly as smaller wafers. Heating from one side of the wafer may create stresses, and temperature gradients across the wafer will require better thermal management.
Multiple input/multiple output (MIMO) control and zoning techniques offer one solution for such temperature sensitive processes as RTP and epitaxy. In this type of control implementation, the wafer surface is divided into concentric zones (Fig. 2) over which multiple sensors are distributed. Pyrometers, backside thermocouples, or fiber-optic and acoustic techniques, are used to measure temperature. The wafer heaters are also separated into zones that can be powered independently. In the past, independent single input/single output (SISO) controllers have been used to adjust heater power, but the larger wafer surface area and greater temperature sensitivity require more zones, sensors, and actuators. The coupling between zones will be too great for SISO techniques to achieve the temperature control needed. Instead, a MIMO controller developed using system identification techniques can determine the heater power for each zone by considering all temperature sensor inputs.
The scaling of deposition and etch processes for 0.25 and 0.18 ?m requires better time resolution by the equipment controller. Data sampling intervals of 100 msec for 0.5-?m processes may need to be reduced to 10 msec for the narrower line widths and resulting shorter processing times. Existing controllers may not directly scale as the increased polling frequency creates more CPU overhead. Ultimately, flexible control architectures will support a wide range of time resolutions, depending on the specific control parameters required.
Statistical process control. Statistical process control (SPC) techniques will continue to reduce 300-mm wafer production costs by detecting sudden process shifts that, if passed undetected, could lead to poor yield results. If a tool is not operating within predefined process windows, an alarm will allow operators to take immediate corrective action. To further improve the predictability of process results, to monitor equipment health, and to reduce the reliance on test wafers, 300-mm tools will need to increase monitoring capability via equipment controller improvements and advanced in-situ sensors.
By selecting and measuring events and equipment operations in real-time, tool operators will obtain increased data on chamber and wafer conditions. Users may choose to measure target hours, number of times a system has been pumped and vented, gas actuators, etc., within a process chamber. For transport subsystems, they may monitor the number of rotations of a robot arm, or wear of subsystem components that may contribute to vacuum integrity and particle generation.
Application of SPC techniques to this enhanced data set can allow earlier detection of equipment failure and more accurate scheduling of preventive maintenance.
Advanced process control. Existing process recipes assume that incoming wafers have consistent properties, and that fixed machine settings yield stable process results. As 300-mm wafers and reduced line widths reduce acceptable process variance, these assumptions fail, and static machine settings will no longer suffice. When in-situ measurements of wafer parameters are not feasible, run-to-run control can attempt to correct for inevitable process drift by adjusting the recipe. If wafer parameters can be measured during processing, then real-time closed loop control should yield highly repeatable process output. Advances in in-situ sensors are still needed, so real-time closed loop control will probably not be available throughout the fab until after 2000.
In order to implement run-to-run control, parameters must be identified and models developed for each individual process and module. The controlled parameters should be strongly related to measured process output. For example, resist thickness and exposure time in photolithography strongly affect the critical dimension of the patterned wafer. If too many parameters, or parameters with weak influences, are chosen, an inefficient control model results. A controller that tries to make a small correction in process output will then need to make a large change in a weakly influencing parameter. These large parameter shifts are likely to have undesirable, unmodelled side effects. Current run-to-run controllers typically use less than five parameters.
Implementation strategies for run-to-run control of a cluster tool are discussed among members of the Control Systems Standardization Working Group; to date no clear path or standard has been identified. Run-to-run control at the cluster tool level is a new domain, and no one is sure who will take ownership of its integration - the fab, OEM, or control supplier. One option would use an independent controller to feed data from the metrology tool to the process tool via the cluster tool controller. Another option would place the run-to-run capability within a cluster tool by including a metrology module within the cluster. The run-to-run controller could be embedded in the cluster tool controller itself, using the intra-tool communication infrastructure to collect metrology data and update recipe parameters.
Figure 3. Proposed run-to-run control architecture using model-based recipe changes.
Recipe management. Advanced process control techniques will initially transform recipes from static to dynamic entities that are changed automatically from wafer to wafer to correct for machine variances. Sensors that measure process state parameters in real time allow closed loop process state control in which recipes specify process state targets rather than machine state targets (Fig. 3) . Though these process state parameters are more closely coupled to the resulting wafer output than machine parameters, these more advanced recipes may still require adjustment from a run-to-run controller to achieve stable wafer state results.
Ultimately, process recipes will specify "effects" on the wafer such as "deposit 2 ? of Ti on wafer within ?2% error." Advanced model-based controllers will translate these tool-independent recipes into instructions for a particular tool. "Effects" recipes will be particularly critical for 0.18-?m devices.
In 200-?m cluster tools, recipes are currently "owned" by, and reside in, individual process module control systems. As part of the 300-mm factory vision incorporating fully integrated wafer flow tracking systems, the status of individual wafers will be known at the factory host, cluster tool, module, and process step levels. In this environment, recipe ownership and editing capabilities move from process tools to other levels within the factory control network. Recipe management will be distributed to module, CTC, or factory host levels. Ultimately fab operators will assign different recipes to each individual wafer or lot ID; a capability particularly useful in R&D or ASIC fabs that use multiple recipes.
To support advanced process control and allow for recipe parameters that could be altered for each wafer, the process module must be able to validate, execute, and track who created or modified a specific recipe for a particular wafer. The current format of recipe data - typically in binary format that can be stored and downloaded by the factory host, but read and edited only at the tool level - is a significant barrier to distributed recipe management. One attempt to address this problem is the Generic Equipment Model (GEM) Formatted Process Program, defined as an ordered sequence of command codes with their associated parameters. GEM assumes one recipe/cassette, however, so it can`t deal with the complex recipe hierarchies needed in cluster tools.
All 300-mm tools will link into a factory-wide automated network for wafer ID and lot tracking, data collection, and recipe verification. Every controller will have to time stamp specified events associated with individual wafer IDs in order to record a complete unified database of wafer events. The controller`s internal clock system must therefore be synchronized with a factory clock to maintain accuracy and consistency.
As discussed earlier, a method of identifying specific wafers, such as a dot matrix or barcode pattern that can be scanned by optical sensors at each tool or event recording station, must be selected, standardized, and implemented by all equipment vendors. It is still unclear which methodology will be recommended and where the wafer ID information should be stored or verified.
Integrated factory-wide data collection can help determine why a wafer became defective and correlate defects with specific cleanroom events, and can also provide valuable data for further development of automated control models and strategies. However, no consistent data model guidelines presently state what format the data should be stored in, what data should be available, or where this data should reside. Industry consortia must resolve these issues so that control developers can build these capabilities into their systems and fabs can perform meaningful cross-tool comparisons.
Current guidelines recommend that all control systems integrated into the 300-mm factory network have the ability to communicate using GEM and high-speed messaging (HSMS) specifications. Some early implementations of communication among these automation components have used either SECS or proprietary communications over a serial RS-232 interface.
Figure 4. Proposed control hierarchy from Standards Roadmap following SEMATECH CIM Framework and OBEM standards.
These highly sophisticated advanced process control and recipe management features are not expected in the initial 300-mm tool offerings, but chipmakers hope to see them in the next generation 300-mm volume production tools. The complexity and breadth of software applications will force a reexamination of factory and tool software integration techniques. At present, the control system is often the last component added onto a process module or cluster tool. In-situ measurement and advanced control of process states require concurrent engineering of the control systems, as an integral component affecting process reliability and predictability on a run-to-run basis. Control models must be tightly coupled with process drivers, recipes, and the tool`s ability to measure and obtain the necessary in-situ data for analysis.
At the factory level, the current approach - using existing control systems and adding on more pieces to obtain functionality - becomes even more costly to manage as software becomes more complex. Continued cooperation among 300-mm consortia members is needed to develop guidelines for a cohesive comprehensive factory-wide control solution. Additional cooperation and partnering among equipment and factory control suppliers is also required. Some control functions will be pushed upwards to the factory host, down to the tool level, or distributed among factory-wide controllers. For example, more sophisticated tools will request preventive maintenance as they detect excessive process drift. Should downtime of this type be scheduled by the tool or the factory host? The SEMATECH CIM Framework and Object-Based Equipment Model (OBEM) standards efforts are relevant to these issues (Fig. 4) .
Equipment utilization improvements
The cost of downtime will unquestionably increase with 300-mm wafer production. COO measurements have shown in the past that return on investment is most affected by machine uptime and effective equipment utilization. To ensure cost-effective production, 300-mm fabs are determined to shorten the amount of time it takes to ramp-up a new tool, and to reduce unscheduled downtime due to software and equipment failure.
New equipment today, on average, achieves only 30% utilization within the first 1.5-2 years of installation. To meet 300-mm fab requirements, utilization will need to increase to 60-65% within a shorter time period. Consortia members are recommending implementation of standardized operator user interfaces, on-line help, and on-line user documentation features as part of the tool control system to reduce the learning curve on a new tool and improve its immediate productivity.
Forty percent of production tool downtime is currently attributed to control software failure, including a direct software failure, selection of a wrong recipe, and other operator interfacing errors. The 300-mm production facilities cannot afford this level of software performance. SEMATECH`s controls groups are imposing mean time between failure (MTBF) specifications that must be achieved during evaluation in order to qualify for 300-mm processes. They are also requiring use of certified software development procedures, based on Software Engineering Institute guidelines, which include error reporting and correction methods . To prevent invalid operations, more software interlocks are provided on control and automation platforms.
To reduce unscheduled downtime, preventive maintenance techniques will use on-line data gathered from equipment components via the tool controller to track life cycles of components and replace them on an as needed basis, or when notified of a problem. Still, time-based scheduled maintenance will continue as part of routine operations.
Proposed computer-aided field engineering and remote troubleshooting techniques can help support and maintain the equipment. With advanced control capabilities, tool vendors could dial into 300-mm fab sites and view tool operations on-line, or retrieve stored data records of recent tool events, greatly reducing the MTTR and further increasing tool uptime and utilization.
For the first time in the history of the industry, semiconductor manufacturers are clearly setting goals and guidelines for equipment suppliers to ensure that qualified tools are in place for volume 300-mm wafer production. Although much progress continues to be made, consensus has not been reached on all specifications, placing a large burden on equipment suppliers who are striving to bring suitable 300-mm tools to market in a timely, cost-effective manner. Until the industry as a whole decides which unified direction it will take, leading suppliers are expected to provide several different configuration variations and the flexibility to handle multiple integration scenarios, depending on the individual fab`s chosen path.
Some automation and controls vendors have forged ahead and begun implementation of consortia guidelines and SEMI standards, and are already delivering platforms to OEMs for integration in their new 300-mm tools. Equipment demonstration target dates are less than two years away and the risk for suppliers this time is high. Suppliers are making significant investments in tool designs for a committee of worldwide device manufacturers who will recommend tools for 300-mm production.
The authors wish to thank Andy Lintz, Brian Hardegen, Chris Hofmeister, Peter van der Meulen, Gerry Morgan, Mike Jones, and Joe White for their valuable contributions.
1. International 300-mm Initiative, "Equipment performance metrics," September 30, 1996.
2. 300-mm Interfaces & Carriers Informational Forum, SEMICON/West, San Francisco, July 17, 1996.
3. Factory Integration Working Group, Meeting Minutes, June 4-5, 1996.
4. Mitchell A. Drew, "Flexible tool interfaces increase responsiveness of process equipment suppliers," Solid State Technology, Vol. 39, July 1996.
5. "Intelligent Factory Tutorial," presented by Dr. Costas Spanos, University of California, Berkeley, SEMATECH AEC/APC Workshop VIII, Santa Fe, NM, October 27, 1996.
6. "Control system requirements for semiconductor process applications," presented by James Moyne, University of Michigan, SEMATECH AEC/APC Workshop VIII, Sante Fe, NM, October 28, 1996.
7. 300-mm ECS Factory Requirements Focus Team, preliminary "Top ten needs" list.
MITCHELL A. DREW received his BS degree from Southern Illinois University and his MBA from New Hampshire College. He is senior product manager at Brooks Automation for 200- and 300-mm wafer transport, factory interface, and thermal conditioning modules. Brooks Automation, 15 Elizabeth Drive, Chelmsford, MA 01824; ph 508/262-2400; fax 508/262-2500.
MICHAEL G. HANSSMANN received his BSc degree in electrical engineering from the University of British Columbia. He serves as director of technologies for Brooks Automation (Canada) Corp. (formerly Techware Systems Corp.). Hanssmann has more than 12 years of experience in providing equipment control solutions to chip manufacturers, equipment suppliers, and R&D facilities in the semiconductor industry.
DAN CAMPORESE received his PhD degree in electrical engineering from the University of British Columbia in 1986. He is currently R&D manager at Brooks Automation Canada. He has been instrumental in developing a natural language approach to equipment control for thin-film processes at Techware Systems Corp.