In this chapter, we will discuss the experiment. We have described the concept of TCS, EAP, and equipment sequence process in Fig 2.8. Fig 3.1 shows the flow of the two main operations of a semiconductor production flow by which the circles represents the operation.
Fig. 3.1 Part of the operations in a product flow.
The first circle in Fig. 3.1 on the left hand side is an internal buffered operation, labeled I1. In this operation, all the lots to be processed are run in a batch, or group. The
batched lots are removed from their FOUPs and the FOUPs are kept in the shelves inside the equipment as in Fig. 2.5 and the wafers are loaded into a furnace. In the original process flow, the wafers are moved back to their original FOUPs from the furnace tube after step E13 of Fig. 2.8. These FOUPs are then unloaded from the equipment to the stocker and wait for the next operation. We assign a smaller circle labeled W in Fig. 3.1 to
24
represent the “idle” or “waiting” status of these lots. The next operation on the right hand side indicates a FOUP exchange operation, labeled X1. In this operation, the wafers are required to be transferred from a certain type of FOUP, say Before Metal FOUP, to an After Metal FOUP due to metal contamination constraints. We simply call the types of the FOUPs B-type and A-type respectively. After the X1 operation, the B-type FOUP is emptied, it is then transferred to a FOUP cleaning equipment for further use; the A-type FOUP will then contain the wafers of the lot.
In section 3-1, we will describe what modifications of software we have done to make it work using the SEMI E94-1107 standard as well as the problems that may encounter. In section 3-2, we will discuss the solutions to 3-1 and their trade-offs. Finally, the results of the experiment in section 3-3.
3-1 Modifications
Elaborated from figure 3.1, node I1, we will look closer inside the TCS-equipment sequence and discuss some more detailed events from the equipment that we may capture to make IPFE function possible.
In current TCS control job creation, the content of attribute “MtrlOutSpec” is left blank and the EAP software takes the content “SourceMap” and “DestinationMap” as the same value. These two contents provide the FOUP ID to let the equipment know where to take the wafers from and where to place the finished wafers back after process. The first modification takes place while TCS creates the control job. We provided the whole content of the “MtrlOutSpec” according to the SEMI specification with both “SourceMap”
and “DestinationMap” the same data since we have not decided the new empty FOUP yet.
This modification is in step T4 of Fig. 2.8. After all the wafers of the same batch lots are
25
loaded into the furnace tube and the process started as step E10 of Fig. 2.8, EAP will catch this event and signal TCS. This event gives us a new T6 transaction from EAP to TCS as shown in Fig. 3.2. We label the transaction such as M1 in the figure to indicate the transaction being sent from MES.
Fig. 3.2 Additional modifications in EAP event handling elaborated form Fig. 2.8.
Continue with Fig. 3.2, TCS will send a new transaction, T7, to MES to inquire a new empty FOUP when the furnace is loaded if the process time of the batch has more than an hour left. The inquiry input would be some basic information of the first lot in the batch such as type of FOUP required, lot’s product name, lot’s flow name, and lot’s operation number…etc. The one-hour-buffer is to be kept for any exceptions that would require human resources to take over. For example, we may expect some transportation issue due
26
to hardware failure and this buffer could be used for operators or engineers to recover the chance of the finished wafers having no FOUP to return after process since the original FOUP may not be on site. This problem will be discussed in section 3-2. MES, on the left side of Fig. 3.2, would calculate whether to return a new empty FOUP or the ID of the original FOUP and return the result in M1 as well as mark an transfer reserve record on the new FOUP in case of this new FOUP being selected by other operations in the FAB. TCS then examines if the returned FOUP ID is the same as what had provided in the inquiry transaction. If the FOUP ID remains the same, TCS will use the next lot in the batch for next inquiry. Otherwise, TCS will use “SetAttr” to modify the control job’s
“DestinationMap” content, T8. By receiving an update success, TCS will tell EAP to unload the old FOUP, T9, inside the equipment and ask MCS to make a transfer command, T9’, to carry this old FOUP away by means of MES; if “SetAttr” had failed, the
“DestinationMap” would remain the same and TCS cancels the transfer reservation made in M1 earlier and move to next lot in batch. When the old FOUP leaves the load port of the equipment, T10, EAP could sense this signal and notify TCS such that TCS can then ask MES to establish another transfer command to MCS for the new FOUP to be transferred, T10’. The reason of the asynchronous transfer rather than transferring both the old and the new FOUP at the same time will be discussed later in this thesis. When the new FOUP has successfully arrived and docked to the equipment load port and moved to shelf inside the equipment, the new FOUP is ready to be filled with finished wafers. The same cycle will be repeated for the same batch of products until no changes are required. This is the first portion of IPFE.
The second portion of IPFE is when the wafers have finished process. After the products have finished their process in the furnace, the wafers are transferred to the new FOUP while reporting an operation complete to MES. A new operation complete
27
transaction is introduced here because in this transaction, we need to call a FOUP exchange operation as well as a skip operation since the original operation X1 is no longer needed and the information of the lot should be at W2 of Fig. 3.1 instead of W1. In other words, the next operation of the LOT will be the one after the original FOUP exchange operation, status “waiting” at W2 of Fig. 3.1.
As we can see from Fig. 3.3, we have bisected the times needed for the original flow process.
Fig. 3.3 Time required for operation in original scheme.
In Fig. 3.3, the total time used in the original scheme from point S to point E would be TF (time for furnace operation) plus Ti (time wait for sorter operation, idle time) plus TX (time for sorter operation to complete). Tm in the figure is calculated in Ti because the average time required for Ti is calculated from the events of end of TF to the start of TX.
28
TSE , therefore, is the total time required in the original product scheme, which will be denoted as function:
TSE = TF + Ti + TX (Eq1)
On the other hand, Fig. 3.4 gives us what we did for the IPFE function. The shaded part W1 and X1 with the respective times required Ti and TX is eliminated and a new arc is drawn from the end point of operation I1 to the start point of W2.
Fig. 3.4 IPFE function acts as an arc over W1 and X1 operations.
We can tell from Fig. 3.4, since we have done all the work for X1 while the lots are in the process of I1. When the lot is completed with process I1, the lot status would be
“waiting” at beginning of W2. Therefore, by using the IPFE function, the new total time elapsed for I1 to end of X1 would be calculated as from point S to point E’ in Fig. 3.4.
This gives us a new time line, TSE’, in Fig. 3.4. This new time line is equal to that of the
29
lot being processed in operation I1 and would be:
TSE’ = TF (Eq2)
or
Tdiff = TSE - TSE’ (Eq3)
And as to the ratio of time reduction R using IPFE function, compared with the original scheme, we may obtain:
R% = (Tdiff) / (TSE) (Eq4)
The value of Tdiff may be different in the above formula since Ti may vary in the original scheme, which depends on the work in process (WIP) of the sorters. We will discuss more about the benefit obtained from IPFE function with statics in chapter 4.
Before we end this section, there is a main problem that we might encounter using the IPFE function. That is, while the old FOUP has been removed or unclamped from the equipment, we cannot definitely be sure if the new FOUP could be delivered. This may be caused by some unpredictable issues by exceptions that may occur at AMHS. In addition, this would extract a minor problem of what if the new FOUP to be delivered arrived later than the furnace process have finished. However, these problems did not occur during the experiment since we modified the software and assumed that the mechanical site of AMHS worked perfectly in simulation as well. We will discuss the solution to these problems in the next section.
30
3-2 Error handling
From section 3-1 we know that MES will provide a new FOUP if available in a nearby stocker for IPFE. However, if the nearby stocker is disabled due to power failure or any possible issues, the reserved new FOUP for transfer would be stuck inside the stocker and this new FOUP cannot be delivered. To solve this problem, the MES will have to check once more which new FOUP to be delivered while the old FOUP disconnects from the equipment. As long as the stocker of the new FOUP is available, no changes are required.
In contrast, if the stocker is unfortunately unavailable, the MES will have to shift the reserved new FOUP information to a new one from the other stocker and hence a new transfer job should be created. In the worst case scenario, if no more empty FOUPs are available, the reservation will be marked to the old FOUP that had just been carried away.
This is a big trade-off however, the time for the old FOUP to be carried to-and-fro from the equipment. This trade-off may be minor since the furnace operation usually last for hours and there would be enough time for engineers to recover the fault stocker as well as the to-and-fro transfers.
The second problem we have mentioned at the end of section 3-1 could be solved by adding timer logic in the TCS since TCS has the start time of the process mentioned earlier at beginning the of section 3-1. The TCS may decide whether to request MES for a new FOUP depending on the time the furnace process has elapsed. For example, if a furnace operation takes an average of four hours, the TCS may judge whether the IPFE function is required if the running batch process has already been run more than three hours. That is why we have said earlier if the IPFE function would take place if there is more than one hour left of the furnace operation. The trade-off of this solution would be that some lots may not apply the IPFE function since there are four LOTs in a batch. And for each lot to
31
apply the IPFE function, the time required for both old and new empty FOUPs to be carried in an asynchronous method would require more time for a complete cycle of the IPFE function. However, the cycle time of the lots that had applied the IPFE function before the three-hour limitation mentioned above may still be benefited.
To conclude this section, the key point of the errors and its handling depend on AMHS.
In chapter 4, we will study a few statistics from the concept of AMHS with the original flow process and then discuss whether the additional exception handling work is worthwhile using the IPFE function in chapter 5.
3-3 Test Result
In the simulation, we established a small process flow containing 5 operations by which the second operation is the furnace operation and the third operation is the sorter FOUP exchange operation. This is to simulate what we have described in section 3-1. The experiment is to test if the ideal case of the IPFE function can be applied to the automated systems in a 300mm FAB. For the simulation environment, we used a licensed IBM SiView4 on a IBM AIX5
4 SiView is a product registered to International Business Machines (IBM) Corporation. The term of use may be requested to follow the agreements within the software.
operating system for the MES site. Since the software and the hardware of MES are considered as company privacy, we may not describe the versions of the software in detail. We used the SiView solution to establish 100 lots in the simulation environment and made 25 batches of four lots at a time to execute the simulated furnace operation. On the TCS and EAP site, we used the software developed by Powerchip Semiconductors Corporation and made the modifications. Fig. 3.5(a) to Fig. 3.6(c) shows the flow diagram of the experiment.
5 AIX is an operating system registered to International Business Machines (IBM) Corporation. The term of use may be requested to follow the agreements within the software.
32
Fig. 3.5(a) shows the flow for the normal condition case. The batches go through from point S to point E with only five minutes of waiting time for the next operation, sorter operation to start. The three minutes waiting time is the ideal and average time in the actual production FAB for a FOUP to be delivered to the sorter. This experiment is similar to what we have described in Fig. 3.3. The time measured in this case is TSE.
Fig. 3.5(a) Flow chart for an ideal furnace and sorter operations.
The next figure, Fig. 3.5(b), is the flow diagram when we added the variable of various waiting time for “Wait for sorter operation” case. In Fig. 3.5(b), we can see that we used the word “vary” because the time waited for sorter operation is the randomized time taken from actual production FAB of 3000 lots in the similar furnace to sorter operation. The time measured in this simulation is TSEi. The “i" represents “idle” in this circumstance.
This experiment is also equivalent to the diagram described in Fig. 3.3 with only the difference of various waiting time added for the lots.
33
Fig. 3.5(b) Flow chart for furnace and sorter operations with various waiting time.
Finally, we applied the IPFE function to the simulation and the flow chart would be shown in Fig. 3.5(c). Please note that this experiment is the concept of Fig. 3.4, the time
waiting for sorter operation and the time for the sorter operation is by-passed, or neglect.
Therefore, the terminal node is denoted as “E’ ” in this case and the time measured is TSE’.
Fig. 3.5(c) Flow chart for furnace operation with IPFE function applied.
34
After knowing the three major cases of the experiment and applying the modifications mentioned to the related systems, we simulated one hundred lots on the short process flow.
The IPFE function worked properly and was as expected. In this experiment, we have not applied the actual transfer of the FOUPs but have controlled the delays of the time that for both old and new FOUP being transported to the equipment. We summarized the results of the three experiments mentioned above and plot the graph for TSE, TSEi, and TSE’ into a graph to see the difference in Fig. 3.6.
Fig. 3.6 Experiment result of Fig. 3.5(a) ~ Fig. 3.5(c).
We tested the experiments again with the IPFE function to the flow of Fig. 3.5(a) but set the condition that failed to apply the IPFE function, i.e., all the wafers still stayed in the same FOUP. The flow chart of this case is diagrammed in Fig. 3.7.
35
Fig. 3.7 Flow chart of IPFE function applied but failed to take place.
We can see that because the IPFE function failed to take place inside the furnace operation, the original operation, the sorter operation, must be performed. This is the worst case of what would happen even if the IPFE function is applied but failed to operate in the simulation. The time measured in this experiment would also be TSE’, and it is compared with the original experiment, case of Fig. 3.5(a), again. The comparison graph is in Fig. 3.8.
The purpose of doing the experiment of Fig. 3.7 is because we have to make sure that even with the new IPFE function applied to the FAB, the original scheme of the process must still work properly. As the result of Fig. 3.8 shows, the new IPFE function will not make a difference if it had failed to operate.
36
Fig. 3.8 IPFE function failed to operate compared with original case of Fig. 3.5(a).
Furthermore, since the function worked properly, we have added a few more checks and algorithms in the MES program such that the decision of whether IPFE is necessary.
This decision making function will be discussed more in detail in the next chapter. We can see that the improvement does not seem to be so effective when the stocker failure rate is high even with the IPFE function had implied but it does not affect the original process scheme.
In the next chapter, we will not only discuss the new MES algorithm but also how the IPFE function may benefit the traffic of the semiconductor manufacturing.
37