Simulation Used To Model And Test Improvements In 300 Mm Fab Throughput
Michael R. Norman, Senior Simulation Analyst,<%=company%>, Bountiful, UT
Microstocking of work in process (WIP) on semiconductor process tools has proven to be an effective means to improve fab throughput. Whether the microstocker takes the form of multiple tool load ports, or automated internal tool buffers, these WIP queues will be loaded by Automated Material Handling Systems (AMHS) in future 300 mm fabs. A full-featured simulation model of an IBM 300 mm fabricator was created to investigate the interaction of microstockers with the algorithms that drive the various types of transport systems. "Push" material transport rules yielded an improvement in microstocker performance and thereby fab performance.
Contents
Factors of Investigation
Main Model Features
Discussion of Simulation Results
Summary
For more information
Author Information
Many factors influence the effectiveness of an automated material handling system implementation. This paper investigates three areas of current interest: microstockers, push / pull lot dispatch, and transporter track system layout. To explore these topics in a robust simulation environment, the Factory Technology Integration group at IBM East Fishkill developed a full fab model that simulates a 300 mm fab. The tool set and routing from a DRAM process served as the data set for the model.
According to the I300I factory guidelines, AMHS is an integral component of all upcoming 300 mm fabs. Ergonomic risk factors force 300 mm fab designers to deal with not only the traditional interbay automation systems, but also with direct delivery of product to the process tools, better known as intrabay automation.
A new intrabay technology called over-head transport (OHT) may emerge as the dominant automation system for 300 mm. OHT consists of a ceiling-mounted monorail vehicle with an onboard hoist system that delivers a 300 mm front opening unified pod (FOUP) directly to a front opening interface mechanical standard (FIMS) load port on the process tool. OHT performs the intrabay (tool-to-tool) material transport function. The traditional interbay systems (bay-to-bay) that ply the main aisles of most major 200mm fabs will continue to be the backbone of the 300 mm fabs. The new 300 mm interbay systems have been reengineered to accommodate the larger FOUP. The OHT monorail track is mounted on the ceiling directly above the load port of the process tool. The OHT vehicle typically has four axes of motion as well as a grip axis. The vertical axis is typically a kinematically-stabilized servo-controlled hoist. There is a side shift axis, and a rotary axis for load port alignment.
Once the line is crossed from interbay automation into intrabay automation, keeping the work in process (WIP) flowing to the process tools becomes far more complex than simply managing boxes of wafers in a WIP stocker. The semiconductor industry has adopted "continuous operation" of the process tools as a means of increasing overall fab throughput. One of the physical manifestations of this strategy is the microstocker. This device stores a small cache of WIP on the front of the process tool so that the idle time between product wafer runs is limited to several seconds instead of several minutes. Simulation results have shown an improvement in overall fab performance when tool idle time is reduced. According to the theory of constraints, "time lost on the constraint tool is time lost for the whole fab." Typically the tool designated as the constraint changes dynamically through the day due to unscheduled down time and other "lost time factors." Therefore keeping WIP in the microstockers is the key to gaining benefit from the microstocker. Maintaining the proper level of WIP in the microstocker is the task of the AMHS and its material control system (MCS) software.
Factors of Investigation (Back to Top)
IBM used the Automod 8.5/AutoSched 5.5 Simulation suite to research three axiomatic factors influencing the design of an interbay and intrabay AMHS. The three areas of concern included: microstocking (MStk), push vs. pull transport dispatching, and two types of track layout—a discrete interbay/intrabay system (Inter/Intra) linked through a stocker, and a unified point-to-point transport system (Pt-Pt) with turntables linking the bays. The scenarios were implemented with data files; no code changes were required to explore the various permutations.
Eight experiment scenarios were considered, encompassing three factors with two choices per factor:
- Inter/Intra, Pull, No MStk
- Pt-Pt, Pull, No MStk
- Inter/Intra, Pull, with MStk
- Pt-Pt, Pull, with MStk
- Inter/Intra, Push, No MStk
- Pt-Pt, Push, No MStk
- Inter/Intra, Push, with MStk
- Pt-Pt, Push, with MStk
The microstocker device can store several FOUPs right the tool. Microstockers of this type are integrated into the footprint at the front of the tool. The typical microstocker hardware is comprised of a SCARA-type robot arm mounted to a servo-controlled vertical axis tower. Storage buffer bins are attached next to the tower. The OHT may have a dedicated load port on this microstocker. The operator will have I/O ports at 900 mm. A dedicated automation port is preferred because the operator and the automaton are kept in different areas. The OEM microstocker is beneficial for tools that have short process times, high throughput, or are the constraint equipment in the fab.
The simulation code models the microstockers on the process tools as a "Storage," combined with a load port "Resource." For clarity, in the AutoMod / AutoSched environment, a process tool is called a Station. Each process tool (AutoSched station) has an AutoMod Resource that represents the input / output port of the tool. The Resource capacity corresponds to the actual number of SEMI E-15.1 loadports. If a tool has four I/O ports, then it has a resource of capacity equal to four. The industry has realized that multiple I/O ports are a very effective form of Microstocking. In the case of four I/O ports, the process tool has a Storage of capacity four.
Some process tools in the model have OEM microstockers mounted on their front, with the ability to store up to 20 FOUPs. This type of microstocker is modeled with an I/O resource of capacity two (these units have two intrabay I/O ports) and a storage of capacity 20. (In AutoMod terms, this is equivalent to an ASRS of capacity 20 with 2 P/D stands). For a FOUP to enter a tool both a load port resource and a storage must be available. OEM microstockers typically have more storages than port resources.
Two general types of material moves are made in the model—leaving a tool after processing (where to go?), and moving to a tool that has task-selected the material for processing. Routines were developed for the two movement system layout options: interbay/intrabay, and point-to-point. The two types of transport system dispatching are called "Push", and "Pull".
In pull moves (pull with respect to the tool doing task selection), a specific lot has been task-selected by a tool and now needs to be moved from wherever it is to the requesting tool. In this model, all lots not currently being processed on a tool are stored in some type of storage. The storage may represent a stocker, a microstocker or a loadport. A move is constructed based on the storage index and other information maintained in the model (i.e. the bay that the lot is currently in) and the location of the tool that has selected the lot. All of the information required for AutoMod to make the move (movement system locations, queues) is saved in a set of load attributes and written to a data file. A call to another AutoMod subroutine carries out the move based on the stored information.
There are four possible "types" of pull moves:
- From a microstocker or loadport in the same bay as the destination tool
- From a stocker in the same bay
- From a microstocker or loadport in some other bay
- From a stocker in some other bay
Push moves (with respect to the tool having just completed processing) occur when a lot has completed processing on a tool and must now be removed and stored somewhere. This model contains routines to handle this choice in two different ways.
The first method is used if capacity is not available at the next logical route step. A lot is pushed to a traditional, end-of-bay stocker after processing. The stocker of choice (based on the next routing step) is maintained in the routing file, along with an alternate choice to be used if no space is available in the primary stocker (a third fallback is provided in case both choices are unavailable). The move is constructed in a similar manner to the "pull" moves described above and the same subroutine is called to execute the move.
The second type of push move involves a more detailed algorithm developed to chose a "best" push destination based on available capacity at specific tool loadports or microstockers. The algorithm used in this model chooses a tool from the next step's family that has the most storage space available (Batching steps use the opposite criteria in order to facilitate the formation of batches). After ensuring the tool is not currently in a PM or down state, the microstocker or loadport of the tool becomes the destination of the move to be constructed and executed as before.
There are four possible "types" of push moves:
- From the current tool to a microstocker or loadport of another tool in the same bay
- From the current tool to a microstocker or loadport of another tool in a different bay
- From the current tool to a stocker in the same bay
- From the current tool to a stocker in a different bay
When lots are pushed to local tools, the AutoSched task selection rules must select lots already located in that tool's microstocker or loadport set (or possibly in-route to the tool, having already claimed microstocker space or a loadport). This selection preference is required to prevent unnecessary material handling moves and also to prevent deadlocks. A deadlock could occur between lots the tool might select that could not get to the tool, due to unavailable microstocker or loadport space.
A pure filtering mechanism does not work because there are times when lots not currently at the tool should be selected based on other criteria (i.e. due date) and there is space available locally. Therefore a ranking algorithm prioritizes lots for selection based on these factors.
A generic AutoMod subroutine executes the AMHS moves constructed. The module consists of the 8 types of moves described above. Each move systematically checks for space and port availability, claims each as necessary, and executes the "move into" and "travel" commands associated with the AutoMod move. Statistics are captured for move duration, and counts by process step and by type. A from/to table is also maintained for move rates within and between all bays.
Layout Configurations (Back to Top)
Two transport system designs were built into the simulation model. The choice of layout options is limitless, but the ones that were chosen represent two major differences in philosophy of design. The first design combines the classic interbay transport system with an intrabay system servicing the process tools (stations). The intrabay system, in this case, is an OHT. The large end -of-aisle WIP stocker is the link between the interbay and the intrabay systems. The second AMHS investigated is a linked interbay and intrabay vehicle system. At the head of the process aisle, a turntable connects the OHT intrabay track with the interbay track. The interbay and intrabay systems use the same vehicles and can go anywhere in the fab, hence the name point-to-point. The WIP stocker is not the link between the interbay and intrabay in this design, it just functions as a distributed queue in the overall system.
Main Model Features (Back to Top)
The model that functions as the backdrop of the simulation experiments is a full semiconductor fabricator with a complete process tool set. Forty-four tools have OEM microstockers, and the rest of the tools have multiple I/O loadports. There are more than a dozen WIP stockers, and a process route with several hundred steps.
- The model has three basic modes of operation:
- Detailed AutoSched using AutoMod subroutines (not AutoSched!) to build and execute AMHS moves, move history and delay times saved
- AutoMod only using move history from first mode (with delays or not)
- AutoSched only with delays for AMHS moves from first mode
- The model has the ability to "beam" lots from tool to tool without use of AMHS movement system to test non-AMHS related factors.
- Interbay/intrabay, Point-Point, and Operator movement systems.
- Push and Pull mode of transport dispatching.
- Microstockers modeled as storages and resources.
- Turntable kinematics in point-point mode (ability to turn off).
- Extensive custom reporting, including threshold values for reporting, stocker shelf and robot statistics, vehicle statistics (individual and cumulative), move duration and rate statistics (by type and process step), turntable use statistics, loadport and microstocker use statistics, "wait for transportation" statistics, lot start/completions, cycle times, and X-theoretical ratios.
- Batching—by piece (delay based on max batch size), or by a filtering and ranking mechanism in push mode (custom rule, batching function). The "wait no longer than" time is based on half the processing time of the batch.
- Custom yield calculation to attain predetermined overall process yield.
Discussion of Simulation Results (Back to Top)
The factors investigated in this study (Microstockers, Push/Pull, Layout design) are not just a technical challenge to implement in a fab, but are also major policy decisions that need to be made before the facility is designed.
The fab throughput in this model becomes limited by the transport system if the wrong combinations of factors are chosen. There is an 89% difference in lot completions from the worst case to best case. The overall interbay and intrabay moves per hour differ by 110% from worst case to best case, and intrabay moves change by 47% from worst to best.
The addition of OEM microstockers on 21% of process tools and installation of a minimum of 2 load ports on the rest of the process tools resulted in a significant improvement in all key fab throughput performance parameters (lot completions, WIP level and cycle time) across all experiments. The transport system utilization was higher to support the fab performance improvement.
Push transport dispatching yielded a significant improvement in all key fab throughput performance measures across all experiments, except for a negative impact to the overall transport system moves per hour rate.
The point-to-point transport track layout produced the best fab throughput and transport system performance across all experiments except for peak intrabay moves, which were higher.
The point-to-point system with push dispatching and microstockers on the process tools attained the overall highest fab performance. The second best system was the inter/intra with push and microstockers. The worst performance was produced by the inter/intra system with pull dispatch and no microstockers on the process tools.
As plans move forward for 300 mm fabs, it is clear that microstocking will play a role in improving overall fab throughput. Combining microstocking with "push" material dispatching will ensure that WIP is where it is needed, when it is needed. The point-to-point transport system provided the best throughput performance. The discrete interbay/intrabay systems are easier to implement, however, and give almost the same throughput when combined with push technology and microstockers.
The Automod/AutoSched simulation environment is a powerful and flexible engineering tool. The ability to operate eight different automation scenarios in the same fab gave insight into the best combinations of factors.
For more information (Back to Top)
Campbell, Philip L., "Overhead intrabay Automation and Microstocking - a virtual fab case study," IEEE ASMC '97
Campbell, Philip L., "On Track with intrabay Automation," Industrial Automation, Integration and Control Conference '96.
Campbell, Philip L., "On Track with intrabay – A Case Study," AutoSimulations Symposium ‘95.
Stanley, Timothy D., Ph.D., "300mm Wafer Fab Design," AutoSimulations Symposium ‘97.
Weckman, Jerry, "Simulation Applied to the Analysis of a 300mm Fab AMHS," AutoSimulations Symposium ‘97.
Author Information (Back to Top) AutoSimulations, Inc., a Brooks Automation Companyhilip Campbell is currently an Advisory Engineer with International Business Machines Corp., Microelectronics Division in East Fishkill, NY. His present assignment is with the Factory Technology Integration group implementing new 300 mm fab capacity. Over the past sixteen years, he has held various engineering positions in process equipment engineering, automation systems development, and mechanical systems design. His current professional focus is on automated material handling systems (AMHS), manufacturing logistics, process tool automation and discrete event simulation of semiconductor manufacturing. He holds a MS degree in Mechanical Engineering from Rensselaer Polytechnic Institute (NY), a BS in Mechanical Engineering from Polytechnic University (NY), and a BS in physics from St. John's University (NY).
Michael Norman, Senior Simulation Analyst, has been with AutoSimulations since 1997. He has over 14 years experience in electronics and aerospace manufacturing and industrial engineering. Michael has 10 years experience in simulation modeling using the AutoMod and AutoSched products. He holds a BS in Industrial Engineering from the University of Washington, is a licensed Professional Engineer in Industrial Engineering, and is a Senior Member of IIE.
For more information: AutoSimulations, 655 East Medical Drive, Bountiful UT 84010, USA. Telephone: 801-298-1398. Fax: 801-298-8186.