MILLENNIUM SERIES: 300mm factory layout and automated materials handling
12/01/1999
Robert Wright, Carl Cunningham, SEMATECH, Austin, Texas
Kamal Benhayoune, Elizabeth Campbell, Venkat Swaminathan, Rick White, I300I, International SEMATECH, Austin, Texas
 |
To help the industry's transition to 300mm wafer production, the modeling team in the productivity and infrastructure department of Sematech's International 300mm Initiative built a simulation model of a generic semiconductor factory. While work continues on this model, which uses AutoSched discrete-event simulation software, its output so far can provide IC manufacturers guidance in making decisions about future factory layouts and insight into factory performance.
Collaborative efforts by global semiconductor manufacturers have paved the way for the transition from 200mm to 300mm wafers by driving standardization of equipment and interfaces and by demonstrating 300mm equipment readiness. The next step in transitioning to 300mm is to understand a variety of high-volume-manufacturing layout scenarios and their relative impacts on equipment count, packing density, factory size, and automated material-handling systems (AMHS).
 |
Figure 1. I300I's vision of a 300mm factory AMHS showing automatic guided vehicles (AGVs), overhead hoist transport (OHT), rail-guided vehicles (RGVs), and person-guided vehicles (PGVs).
Several detailed factory models with a common base-level architecture were developed by International Sematech's International 300mm Initiative (I300I). These models were used to forecast the effect of 300mm equipment placement in a future (early 2000s) mature production environment.
Layouts
In the general layout envisioned for 300mm fabs (Fig. 1), equipment is located in several bays that stem from a central aisle. An interbay stocker (i.e., storage buffer) resides at the junction of the main aisle and each bay. Each bay also has its own aisle with space for equipment on either side. This type of layout lends itself to an AMHS format that consists of two different types of transportation:
- An interbay transport loop, which spans the central aisle and connects to all bays at their respective stockers, moves wafers from one bay to another. Common in today's factories, interbay transport uses standard wafer carriers to move wafers between factory bay stockers by way of an overhead rail system.
- Intrabay transport moves wafers between equipment within a single bay. One example is an overhead hoist transport (OHT) that travels on a monorail loop around a bay, carries standard wafer carriers, and lowers the carriers to an interface in front of the appropriate equipment. OHT and the other types of intrabay transport shown in Fig. 1 interface with bay stockers.
Conceiving a factory
To model future factories through simulation, we first conceived three different 300mm factory layouts to support performance requirements of overall factory throughput and cycle time: farm (pre-compressed and compressed), hybrid, and modified hybrid. We then built AutoSched models for each — factory simulations starting at 20,000 300mm wafers/month with a single 180nm Sematech process flow and I300I equipment performance metrics. Our goal was to determine tool count, inventory level, process cycle time, and equipment availability.
The pre-compressed farm layout placed all like tools together in the same or adjacent bays, leaving enough space in each area for approximately 10% tool expansion. The compressed farm layout reduced the overall factory footprint to evaluate the minimum base-level factory. We used two methods to perform compression: 1) allowing adjacent functional areas to share bordering bays, and 2) absorbing the expansion areas reserved for future tools.
The hybrid layout was a fairly simple extraction of the compressed farm layout. We distributed metrology tools to the functional areas they could best support. This allowed for further compression because smaller metrology tools could better fit in the corners or remaining expansion space of the layout. It also decreased the number of interbay moves necessary and reduced intrabay congestion by eliminating the metrology bay.
 |
The modified hybrid layout was produced from a process-centric perspective. We dissected and categorized the process flow to distinguish logical process sequences of four to six steps that were repeated several times. These sequences defined specific tool sets that we grouped together within a bay. For our assumed wafer-start volume, we needed multiple bays with the same tool set. We defined these as functional areas for the modified hybrid layout. By limiting the tool types in a set, some bays were less than optimally compressed based on the total number of tools required to meet the overall factory throughput. This left some space for future tool expansion. There were not, however, enough of several types of metrology tools to populate each bay, so 11 metrology tools were added and distributed to appropriate bays. The additional tools helped to fill ancillary space.
We considered three main factors in laying out the relative position and adjacency of functional areas. The layout should:
- promote efficient use of utilities and facilities connections,
- maintain vibration isolation for photolithography equipment, and
- support general process flow and reduce material movement distance.
Modeling approach
To distribute 20,000 wafer starts throughout each month, nine lots of 25 wafers each were started every 8 hr in our AutoSched model.
Every tool was modeled as a single station with station storage. All single-tool stations within a bay selected lots from the same stocker at the end of the bay to which they were returned for storage after processing.
If tools from the same family had to be placed in two different bays, appropriate delays were added to reflect interbay moves.
With the hybrid and modified hybrid layouts, tools were distributed among some of the bays. These tools were not, however, limited to serving lots coming from other tools within the same bay, as if, for example, metrology tools located in the photolithography bay and dedicated to that bay could only serve lots leaving tools in that bay. Instead, the factory was modeled so that lots would travel to the closest available tool. This means that a lot would sometimes pass by a busy tool in its current bay and travel to a similar tool in another bay.
Each bay has a stocker modeled as a storage unit and defined as both the station family storage for the families within that bay and as the outstorage for the stations in that bay. All necessary batching is performed at the stockers.
 |
Since lots must move between tools and stockers both within a bay and between bays (requiring two different types of AMHS), and because station families may sometimes be split between bays, several different types of moves were defined. These were combined with interbay and intrabay delays to model lot movement.
The model uses intrabay delays when a lot moves from a stocker to a station, or vice versa, in the current bay and when a lot moves from a stocker to a station in an alternate bay in conjunction with an interbay move.
When moving from bay to bay, a lot has to leave the stocker at the bay it is departing and travel to the stocker at its destination bay. A lot incurs an interbay delay when the lot moves from a stocker to a station that is a member of the next step's family if that family is not in the same bay as the current station. Interbay delays are also used in conjunction with intrabay delays when a lot is moving from a stocker to a station in the current step's family if that station is in an alternate bay.
Layout simulation
Our four 300mm fab layouts are depicted in Figures 2-5, each roughly to scale. The table summarizes data from comparable simulation runs for each layout; all values are normalized to the pre-compressed farm layout. Each simulation run lasted 900 simulated days and achieved a steady state. Tool count, average utilization, cycle time, and storage and inventory are all simulation results. Factory size and area/wafer starts/week are results of the layout exercise after tool count and factory size had been adjusted for each layout configuration.
The farm layout was the easiest to create. When constructing the initial factory, additional room was built in to provide for expansion, making it the easiest of the three configurations to expand because tool additions are straightforward. The performances of pre-compressed and compressed farm layouts were identical. The only differences occurred in factory size; the compressed farm layout resulted in a smaller factory due to the method of modifying the pre-compressed layout to form the compressed layout.
 |
The hybrid layout has advantages that include performance, inventory, smallest factory area/wafer start, and easy expansion. The only pieces of equipment distributed among the bays are metrology tools, which tend to be smaller than an average process tool. Therefore, metrology tools can be easily distributed to the other process areas of the factory while maintaining the relative size of these areas.
The hybrid layout factory performance showed a 13% reduction in cycle time over the farm layouts. Unlike the farm configuration, many of the process steps followed by a metrology step only require lots to travel on the intrabay system. Therefore, delivery times and interbay system requirements are significantly reduced. In addition, work in process (WIP) inventory is 12% lower than that of the farm layout.
The modified hybrid layout has the best cycle time performance — 14% lower than the farm layouts. It also has the lowest average tool utilization, the lowest interbay movement, and the least inventory buildup over our 900-day run. The average factory WIP level is 14% lower than those with farm layouts. It results in the second largest factory, however. Only the pre-compressed farm layout needed a larger footprint. Because additional metrology tools are necessary, the modified hybrid layout has 11 more tools than the other layouts, but there is still space allocated for expansion.
The hybrid layout configuration performs with a reduction in cycle time over the farm layout configuration and results in the smallest footprint, fewest bays, and lowest factory area/wafer starts/week. It shares honors for the best tool count and tool utilization with the farm configurations.
If a semiconductor manufacturer is planning a high-volume factory with a limited number of processes where no significant changes will be made in tools, layout, etc., then the modified hybrid configuration could provide an extra margin of efficiency.
Future work
Admittedly, our modeling work is a highly idealized view of a generic 300mm factory. Future work on this model will focus on adding more complexity and flexibility. For example, our AMHS time delays used to model interbay and intrabay movement will be replaced by multiple AutoSched AGV systems to study the effects of congestion, queuing, and availability of the transportation system on factory performance.
In addition, we are improving our model to increase the rate at which model layouts and AMHS strategies can be altered and to facilitate comparisons between factory scenarios. The improved model is being built on a standard grid with personnel and AMHS layouts running in parallel in every bay. Ample equally spaced control points for equipment placements will be put on both transportation systems. Therefore, to move between scenarios, a simpler and more rapid change in equipment assignments can be performed instead of moving the equipment graphically on the layout. In the improved factory model, the main concerns are high-level cycle time, tool quantity, and WIP level. Graphically representing equipment status is secondary to model validation.
To understand the effect of AMHS on factory performance, more focus on the details of lot handling is necessary. Detailed stocker operations, interbay and intrabay AMHS operations, and interfaces between the AMHS and other equipment have to be defined. Since stockers are the bridge points between interbay and intrabay systems, it is important to model stockers in detail. In the improved model, each stocker will be modeled as storage with multiple resources that represent I/O ports and internal robot arms. Interbay and intrabay transport systems will be modeled as independent, one-directional loops with vehicles that move continuously until they reach a control point where a lot is ready to be moved. The interface between the transport systems and tools will be modeled with capacitated storages. The capacity of each interface will be set to the average buffer requirement of the tool type to which it is attached. Each of these storages will have two associated control points on the transport system. The control point dedicated to unloading lots from vehicles onto the tool will be downstream from the point where lots are loaded onto vehicles to prevent them from being dropped off and picked up by the same vehicle.
Conclusion
The vision for the future of our modeling activity is to create one model that will integrate factory and AMHS modeling, both interbay and intrabay. This new flexible model will be used to perform a variety of experiments to measure the sensitivity of base-model dynamic parameters. We have not finalized a specific list of experiments, but could include exception lot-handling options, multiple products and processes, technology diversity (e.g., Al to Cu or 180nm to 130nm), 13 vs. 25 wafer carriers, OHT strategies, buffer adjustments, and computer integrated manufacturing messaging. More information on this project can be found in the proceedings of the "Workshop on 300mm Manufacturing" held in conjunction with the 1999 International Symposium on Semiconductor Manufacturing (ISSM).
Acknowledgments
The authors thank these study group members for their support throughout this project and for providing data for the models and reviewed simulation results: Kishore Potti, Advanced Micro Devices; Philip Campbell, Karl Gartland, Ashwin Ghatalia, and John Konopka, IBM; Alan Allan, Edward Bass, Devadas Pillai, and Timothy Quinn, Intel; James Ammenheuser and Randal Goodall, International Sematech; Michael McEwen, Lucent Technologies; Mathias Schulz, Siemens; Leonard Foster, Texas Instruments; and C.N. Wang, TSMC. Special thanks to Eddy Bass and Tim Quinn for the great amount of help they provided. AutoSched is a trademark of AutoSimulations.
Robert Wright received his BBA in management and is currently completing his MS degree in industrial technology at Southwest Texas State University. He has a financial planning background and is working as a productivity analyst, modeling with simulation products for Sematech, 2706 Montopolis Dr., Austin, TX 78741; ph 512/356-3500, fax 512/356-3135, e-mail [email protected].
Carl Cunningham received his BS in chemical engineering from Clarkson University and his MS in material science from Columbia University. He has spent 15 years in IC-manufacturing development and productivity analysis at IBM.
Kamal Benhayoune received his BS in aerospace engineering and his MS in operations research and industrial engineering from The University of Texas, Austin. He works at DuPont Photomasks Inc.
Elizabeth Campbell is working on her PhD in the operations research and industrial engineering department at The University of Texas, Austin. She is an intern working with the factory-modeling team in International Sematech's I300I division.
Venkat Swaminathan was an intern on Sematech's I300I factory-modeling team. He now works for Sabre.
Rick White received his BS in electrical engineering and is currently completing his MSE in operations research and industrial engineering at The University of Texas, Austin. He is with Dell Computer Corp.