Simulation modeling for 300mm semiconductor factories
10/01/2000
SPECIAL REPORT: State-of-the-art manufacturing
Elizabeth Campbell, Robert Wright, Joshua Cheatham, International Sematech, Austin, Texas; Mathias Schulz, Semiconductor300, Dresden, Germany; James L. Berry, Motorola, Austin, Texas
overview
A generic factory is the root model that can be easily altered to perform experiments to evaluate 300mm fab layouts and operations, in this case a factory running a single 180nm logic process flow and starting 20,000 wafers/month. Experiments performed have included looking at the effects of various factors on factory productivity as defined by cycle time, work in process inventory, and tool utilization. These experiments include studying the effects of AMHS equipment downtime, multiple process flows, the number of stockers, and the method of material transport. The simulation models show that the published International Sematech vision of a fully automated factory would sufficiently support the production requirements of a 300mm factory.
Previous reported work from the modeling team in the productivity and infrastructure department at International Sematech revealed some preliminary results about 300mm fab tool counts, tool utilization, process cycle time, factory size, and work in progress (WIP) for several different factory layouts [1]. The previous article, in the Dec. '99 issue of Solid State Technology, reflected the complexities involved with 300mm fab simulation and modeling and the ongoing nature of our work where we use discrete-event simulation to develop an understanding of factory operational issues associated with fully automated 300mm wafer semiconductor manufacturing.
To review briefly, our generic factory model can be easily altered to perform experiments. Its base layout supports a factory starting 20,000 wafers/month in a 180nm logic process flow. We assume the general layout envisioned by International Sematech, developed under the I300I program, for 300mm factories [2] where equipment is located in several bays that stem from a central aisle (Fig. 1). A storage buffer or stocker resides at the junction of the main aisle and each bay. Each bay also has its own central aisle with space for equipment on either side of the aisle. This type of layout lends itself to interbay and intrabay automated material handling system (AMHS) formats. All like tools except metrology tools are located in the same or adjacent bays. Metrology tools are placed in the same bays as the equipment they support.
Additional experimentation
We have now performed more detailed experiments that compare the effects of various factors on factory productivity as defined by AMHS equipment downtime, factory layout, the number of process flows, and the number of stockers/bay. These experiments also compared the effects on factory productivity as defined by cycle time, WIP, and utilization. Each of our modeling experiments changed one factor at a time and compared the results of the experimental model to the results of a base model in which all the parameters listed above were set at base levels [3, 4].
We are reporting here the effect of changing the levels of AMHS downtime, changing the factory layout, including multiple products, and changing the number of stockers/bay.
Our model, which uses an extensive list of realistic assumptions (see "300mm factory modeling assumptions," on p. 96), is a highly idealized view of a generic 300mm factory. Therefore, the model output should not be viewed as comparable to any real production factory. The value, however, is in the relative performance of the simulated factory as a function of input parameter selections.
AMHS downtime
The guidelines for 300mm AMHS and production equipment interfaces [5] specify that the mean time between incident (MTBI) for both vehicles and stockers should be at least 500 hrs. Thus, in our base model we set the MTBI to 500 hrs and the repair time for a vehicle to 5 min 80% of the time and 30 min 20% of the time. A 5-min repair represents a failure in which a short action (e.g., a technician giving the vehicle a push to get it moving again) will repair the vehicle. A 30-min failure represents a more extreme failure that requires an action such as replacing the vehicle with a working vehicle. In addition, we assumed that on each interbay or intrabay system, the number of vehicles does not change during the simulation period. Our simulated stocker-repair times are always 30 min.
To determine the effect of different levels of AMHS downtime, we ran the model in two configurations. In the first, all vehicles and stockers failed with a MTBI of 500 hrs. In the second, the MTBI was 250 hrs for all vehicles and stockers. Each model was replicated five times.
 |
Table 1 shows the results of the modeling. Because this experiment was performed before we finalized the base model, hot lots were not included. However, two nonproduct wafer lots (NPWs) were included in the model every 24 hrs. For each of the two modeling experiments (and for the other experiment discussed below), we averaged the cycle time for each replication, determined the standard deviation over the set of replications, and calculated a 95% confidence interval (CI) for the average cycle time.
The same values were calculated for the WIP inventory. We also determined the average factory tool utilization and the idle time tools.
 |
The modeling data in Table 1 show that the average tool utilization and average percent of time spent idle by tools in the factory are the same in both models, and the average cycle time and WIP in the factory are only slightly higher when failures occur closer together. Failures that occur at twice the frequency as the recommended minimum frequency have little effect on factory performance. We may perform further experiments to test the effect of failures that occur closer together. It is expected that at some point the failure frequency will cause the model to be unstable with constantly increasing WIP and cycle time.
Multiple products
Our base model and all other experiments that have been run so far have had a single process flow. Therefore, we designed a modeling experiment to study the effects of multiple process flows on factory performance, altering the base model to include two process flows—the logic flow used for all other experiments and an International Sematech DRAM flow.
 Figure 2. Stocker operations. |
The logic flow has a raw process time of 8.9 days and 316 process steps; the DRAM flow has a raw process time of 8.35 days and 238 process steps. We ran two models with different product mixes. In one model, 33% of the lot starts were logic lots and 66% were DRAM lots; in another test the percentages were reversed. The results for this experiment are tabulated in Table 2.
The data show that the model with 33% logic and 66% DRAM has a higher average WIP and cycle time. The two process flows use the same tool set, but each requires a different number of visits to each type of tool. For instance, the logic process flow requires a wet bench tool 18 times while the DRAM flow uses a wet bench 28 times. The equipment in the factory was chosen for the logic process flow. Thus, when DRAM products make up more of the WIP, the different equipment requirements cause congestion and the WIP and cycle time to increase.
 |
Single stocker
Each bay in the base model has two stockers: an in-stocker is used by lots arriving at the bay and for lots that will possibly be sent to another tool in the same bay as their previous step. An out-stocker is used by lots that have completed processing within a bay and will definitely be sent to a tool in another bay because no tools appropriate for the next processing step are present in the current bay. Figure 2 shows the operation of a stocker. Vehicles traveling along both the interbay and intrabay systems have access to the stocker in and out ports. A robot inside the stocker transfers lots between the ports and the storage locations inside the stocker.
 |
Because in-stockers are used in more situations, in all modeling results the in-stocker in each bay is used more than the out-stocker (see Table 4). Accordingly, we designed an experiment to test whether the out-stocker is necessary at all.
First, each bay was given a single stocker with all other modeling factors held at their base levels. This model never reached stability, indicating the need for an additional stocker in at least one bay. So, we gave the bay with the highest stocker utilization two stockers (one in-stocker and one out-stocker) that performed as in the base model. The results of this model, which did reach stability, are given in Table 3.
The data in Table 3 show that factory performance for the two models is very similar. Average tool utilization and idle time percentages are identical. Average WIP and cycle time are also essentially the same. Table 4 gives the average WIP for each stocker in both models.
As expected, the average work in progress in all in-stockers in the one-stocker model is greater than the WIP in the corresponding in-stocker in the two-stocker model. Despite having a very low average amount of work in progress in its out-stocker, the fourth bay requires the second stocker. The utilization of one of the fourth bay's in-stocker I/O ports is above 90 percent when both stockers are present. The load port could not handle the additional demand placed on it when the out-stocker was removed.
Intrabay transport modeling
The base model uses overhead hoist transport (OHT) systems for intrabay transport. To test the effect of using a different type of intrabay transport, a second model uses person-guided vehicles (PGVs) for intrabay transport. In both the models with OHT intrabay and PGV intrabay systems, the interbay transport system is a one-directional overhead automated material handling system. The track forms a loop through the center of the factory. Empty vehicles move continuously until blocked by another vehicle or they reach a point where a FOUP is ready to be transferred onto a vehicle. Each vehicle on the interbay track has a capacity of one FOUP. Each bay has its own independent, intrabay transport system. When overhead hoist transport systems are used, each bay has a one-directional loop spanning the bay's central aisle. Vehicles constantly travel the loop in search of work. There is not enough space between stocker I/O ports or tool load ports for more than one vehicle to access a stocker or tool at a time. Each intrabay overhead hoist transport vehicle has a capacity of one FOUP.
 |
When person guided vehicle systems (Fig. 3) are used for intrabay transport, each bay has a set of PGVs that can only be used within that bay. Thus, person guided vehicles are not shared between bays. It is assumed that there are enough operators to control the person guided vehicles whenever they are needed. In addition, there has been no attempt to model human behavior. Person guided vehicles can move in either direction through the central aisle in a bay, but the aisle is only wide enough for two PGVs, whether they are moving or stopped at tools. There is not enough space between stocker I/O ports or tool load ports for more than one vehicle to access a stocker or tool at a time. When there is no work for a person guided vehicle to do, it is moved toward one of the parking spots located at each end of the bay. Person guided vehicles have a capacity of two FOUPs.
To compare the performance of overhead hoist transport and person guided vehicle intrabay systems, we built two models. One of these models is the base model with OHT intrabay transport and the other has all the characteristics of the base model except that it uses PGVs for intrabay transport instead of OHT systems. The tool counts and locations were the same and interbay transport is a one-directional overhead loop in both models. The results of this experiment are tabulated in Table 5.
 Figure 3. PGV transport. |
As modeled in this project, the mode of intrabay transport appears to have little effect on tool performance as indicated by the average tool utilization statistics. The method of transport does have a slight effect on overall factory performance as measured by cycle time and WIP. Person guided vehicles travel at slower speeds than overhead hoist transports vehicles and thus have slightly longer delivery times than OHT vehicles, causing an increase in cycle time and thus WIP. On the other hand, PGVs can carry two lots while OHT vehicles can carry only one, causing a need for about 1.5 times as many OHT vehicles than PGVs to support this factory.
Assuming that enough operators are available for each person guided vehicle to be constantly available and that the operators move according to the logistics presented here, the results of this project indicate that person guided vehicles would support the number of lot movements required in this factory. In addition, person guided vehicles could be used to support a bay if the overhead hoist transport system in that bay were unavailable.
Conclusion
In addition to testing the effects of various factors on factory performance, the modeling experiments described here have another purpose. They show that the vision of a 300mm factory as described in ISMT documents, such as I300I Factory Guidelines: Version 4.2 [2] will support the move demand required by the process flows used in these experiments.
To continue our work, we are looking at modeling experiments that examine multiple lots in a pod, NPWs with shorter routes that only go through the lithography and etch bays, and modeling more than two process flows. Another set of experiments to be performed includes changing vehicle capacity, modeling the intrabay and interbay systems as one combined transport system, and modeling the interbay and intrabay systems as conveyors.
Acknowledgments
AutoSched and AutoMod are trademarks of AutoSimulations.
References
- R. Wright, et al., "300mm Factory Layout and Automated Materials Handling," Solid State Technology, pp. 35-42, December 1999.
- A. Ghatalia, "I300I Factory Guidelines, Version 4.2," Technology Transfer 97063311F-ENG, International Sematech, Austin, TX, Nov. 1999(http://www.sematech.org/public/docubase/abstract/ 3311geng.htm).
- J. Ammenheuser, E. Campbell, "300mm Factory Layout and Material Handling Modeling: Phase II Report," Technology Transfer 99113848A-ENG, International Sematech, Austin, TX, Nov. 1999.
- E. Bass, T. Quinn, "300mm Factory Layout and Material Handling Modeling, Phase I Report," Technology Transfer 99023688A-ENG, International Sematech, Austin, TX, March 1999.
- E. Bass, P. Jai, "Metrics for 300mm Automated Material Handling Systems and Production Equipment Interfaces: Rev. 1.0," Technology Transfer 97123416B-TR, International Sematech, Austin, Texas, Dec. 1998.
Elizabeth Campbell received her BIE (industrial engineering) from Georgia Tech and her MS in engineering from the University of Texas at Austin. She is a simulation analyst at International Sematech, 2706 Montopolis Dr., Austin, TX 78741; ph 512/356-3069, fax 512/356-7848, e-mail [email protected].
Robert Wright received his BBA in management and his MS in industrial technology at Southwest Texas State University. He has a financial planning background and is the project manager for productivity analysis, modeling with simulation products for International Sematech.
Joshua Cheatham received his BS in civil engineering from the University of Texas at Austin and is currently completing an MS there. He is a simulation analyst at International Sematech.
Mathias Schulz received his PhD and his graduate degree in engineering from Technical University Dresden. He is a system specialist working in the automation-CIM group at Semiconductor300.
James L. Berry received his BA in computer science from St. Edwards University, Austin. He is a manufacturing analyst and senior staff scientist at Motorola's Advanced Products Research and Development Laboratory, Austin, TX.
300mm factory modeling assumptions
Operational "assumptions" used by the modeling team in the productivity and infrastructure department at International Sematech:
- A 300mm factory runs 24 hrs/day and 7 days/week.
- The single process flow used is ISMT's 180nm aluminum process flow.
- There are 20,000 wafer starts/month: nine 25-wafer lots released every 8 hrs, two single-wafer nonproduct wafer (NPW) lots released every 24 hrs, one 25-wafer hot lot released every 72 hrs.
- Hot lots follow the same routing as regular lots and are given priority over every other lot type at all tools.
- NPWs follow the same routing as regular lots without priority at any step.
- Equipment is located in 22 bays.
- Each bay has two stockers. All lots arriving in a bay enter through a designated "in-stocker." A lot that has completed processing at a tool within the bay and is possibly being routed to another tool within that bay returns to the in-stocker. If this lot is rerouted to a tool in another bay, a vehicle on the interbay system picks it up at the in-stocker. A lot that has completed processing at a tool within the bay and is being routed to a tool in another bay is sent to the "out-stocker" in the bay.
- Processing times are modeled without variability.
- Hot lots always have priority over all other lot types. If no hot lots are present at the in-stocker, then implant and lithography areas pick the oldest lot in the in-stocker that requires the same setup as the last lot processed by the tool (same setup rule); all other areas dispatch lots from stockers to tools first-in-first-out basis.
- Batches are formed at in-stockers and must be made up of lots from the same previous process step. No lots will be sent to a tool until at least a minimum batch is formed at the stocker.
- Automated wet cleans are processed in batches of two like-lots/batch. Hot lots may be processed alone if no other lots are available.
- Furnaces process batches of 2-4 like lots. Once a batch of two lots is present at the in-stocker, the tool will wait for up to half the process time for up to two more lots to become available at the stocker. Hot lots may be processed alone if no other lots are available.
- Tool failures and preventive maintenance, stocker failures, and vehicle failures are modeled using exponential distributions.
- The interbay system is assumed to be a one-directional, single-loop overhead system. Vehicles are capable of carrying one lot at a time.
- The intrabay transport system is assumed to be a one-directional overhead hoist transport (OHT) system. Each vehicle is capable of carrying one lot at a time.
- On both the interbay and intrabay transport systems, idle vehicles move continuously while looking for work.
- Due to the size of tool loadports, only one vehicle is allowed access to a tool at a time. Vehicles that want to pass a tool or gain access to a tool's loadport must wait until the loadport is free of vehicles before moving to the tool.
- No rework and yield assumptions are implemented.
- Reticle handling is not simulated.
 |
Time for simulation
As 300mm semiconductor manufacturing facilities become more automated to increase productivity, understanding the fab's operational issues becomes exponentially difficult. These challenges have caused semiconductor manufacturers and organizations like Sematech, Motorola, and Semiconductor300 to turn increasingly to the technology of simulation. Interestingly, the use of discrete-event simulation in the semiconductor industry is still in its formative years compared to its long-term pervasive use in heavy manufacturing, warehousing, and nonsemiconductor materials handling applications.
Simulation continues to prove its worth in this space, however, as illustrated in the ISMT project. It provides a reliable, cost-effective, virtual world test bed where new concepts and ideas can be tested, refined, and proven without disrupting the productivity and profitability of the fab.
Mike Thompson, president of AutoSimulations, a Brooks Automation company, Bountiful, UT