Study For Realizing Effective Direct Tool-to-tool Delivery

Study for realizing effective direct tool-to-tool delivery Hiroshi Kondo, *Mitsuru Harada Asyst Shinko, Inc. 100 Takegahana-cho Ise-City, Mie-Pref., Japan, [email protected] Abstract – Direct tool-to-tool delivery is one of technical topics for excellent fab performance. Automatic Material Handling System (AMHS) vendors supply unified OHT systems that can provide direct tool-to-tool delivery routes. But providing such routes is not sufficient to realize direct deliveries. The purpose of our study is to identify remaining issues required to realize direct toolto-tool delivery.

EFFECTS OF AMHS DELIVERY TIME ON LOT CYCLE TIME Our previous study (see [1]) shows effects of AMHS delivery times on fab performance. The model we used in our study had a control scenario among AMHS, tools and Manufacturing Execution System (MES) as shown in Fig.2. This scenario indicates typical ‘push-pull’ delivery policy. Fig. 3 shows detail of this policy.

Our conclusions are (1) AMHS cannot realize direct toolto-tool delivery by itself without any degradation of fab throughput, and (2) effective algorithms for lot scheduling / dispatching and variability reduction of tool process rate are required to realize it without such degradation.

MCS

Tool_A

MES

ReadyToUnload LP1 Dispatching to determine destination of this foup(foup#1)

transport command:　foup#1, from Tool_A:LP１,to Stocker#900 　

TU pick up foup#1

vehicle

INTRODUCTION

brings it to stocker#900

ReadyToLoad LP1

Recently, some IC manufacturers are asserting that direct tool-to-tool delivery is one of the key technologies required to achieve excellent fab performance, as it seems to reduce lot cycle time (Figure 1).

transport command:　foup#3, from Stocker#1, to Tool_A:LP１　

Dispatching to determine next lot for Tool_A

TL set down foup#3

vehicle

transport complpeted event

Intra-bay and inter-bay unified OHT system

Fig.2 Control scenario Stocker

Stocker The lot is completed its process on the tool

This lot is delivered from Tool-A to Tool-B directly without passing or staying any stockers or buffers. Tool-B

Are there any empty loadports of the tools for next process step of the lot?

The loadport become empty

Yes

Are there any lots wait for the tool that has the loadport? Yes

No

Tool-A 装置

The lot is delivered to the stocker corresponded with next process step and waits for one of the loadports to become empty. (Push logic)

Fig. 1. Direct tool-to-tool delivery.

No

The lot is delivered to one of the loadports. (Direct tool-to-tool delivery)

One of the lots is selected and delivered to the loadport. (Pull logic)

Exit

AMHS vendors have already developed inter-bay and intra-bay unified OHT systems that can provide direct tool-to-tool delivery routes, but these systems alone have not been sufficient to realize direct tool-to-tool delivery in a fab.

Fig. 3. Flow charts to describe typical push-pull delivery policy.

The purpose of this study is to identify remaining issues required to realize direct tool-to-tool delivery. It is important to note that if there were only one lot in a Fab and there were no tool failures, the fab would realize direct tool-to-tool delivery spontaneously. So, the issue worthy of study is what is needed to realize direct tool-totool delivery without degrading fab throughput.

We defined two kinds of delivery times in the model. One was named TU , that was defined as time duration from unloading request to unloading completion. The other was named TL , that was defined as time duration from determination of lot that should be delivered to the tool to completion of loading. CT denoted tool cycle time and

0-7803-9144-6/05/$20.00 ©2005 IEEE.

Exit

21

LP denoted number of loadports on the tool. TU , TL CT were assumed constant for simplification. Our previous study shows that if TU + TL ≤ (LP − 1) • CT ,

Intra-bay and inter-bay unified OHT system

and

AMHS delivery time does not add any additional waiting time to lot cycle time. This conclusion was extended in order to apply to whole fab in the paper [1].

Stocker

Considering actual values for tool cycle time and AMHS delivery time, and the above conclusion, as long as buffers or stockers are located appropriately, that is, TL is small so that the inequality TU + TL ≤ (LP − 1) • CT is satisfied, AMHS delivery time does not add additional waiting time to lot cycle time.

Tool-B1

When the lot is completed,Stocker if there are no empty load ports at the tools for its next step, the lot will wait on current load port.

Tool-B2 Tool-B3

Tool group for next step of the lot

Tool-A 装置

Fig. 4. Mandatory direct tool-to-tool delivery policy.

To confirm our conclusion, we perform simulation study. Because unfortunately we don’t have any tool, process and process flow data for any semiconductor fab, we use data for a LCD TFT line, which has 90 tools, 33 tool groups and 2 process flows. Graph 1 shows the result. The simulation runs are done with AMHS and without AMHS (i.e. delivery time is set to zero). We can see that existence of AMHS does not affect lot-waiting time from this result.

The lot is completed its process on the tool

Are there any empty loadports of the tools for next process step of the lot?

The loadport become empty

Yes

Are there any lots wait for the tool that has the loadport?

No

Yes

No

One of the lots is selected and delivered to the loadport. (Direct tool-to-tool delivery)

The lot stays on the current loadport until one of the load ports to become empty.

Lot cycle time (relative)

3.5 3.0

The lot is delivered to one of the loadports. (Direct tool-to-tool delivery)

0.17

Exit

2.5 2.0

1.82

1.82

1.5

transporting w aiting in process

Exit

Fig. 5. Flow charts to describe mandatory direct tool-to-tool delivery policy.

1.0 0.5

1.00

1.00

w ith A M H S

w /o A M H S

We modify the previous simulation model for the LCD TFT line in order to accommodate above ‘mandatory direct tool-to-tool delivery policy’ and perform simulation runs. The result is ‘dead lock’ because existence of loops of process flow and this policy build up dead lock. To eliminate possibility of dead locks we use a process flow that has no loop.

0.0

Graph 1. Comparison of average lot cycle times obtained from the simulation studies of models for a fab with and without AMHS models.

This result suggests that lot waiting time is not caused by AMHS. Lot waiting time for tools means the need to use buffers. So, we can conclude as follows: “AMHS can prepare routes for direct tool-to-tool delivery, but cannot realize direct tool-to-tool delivery by itself without any degradation of fab throughput.”

Lot cycle time (relative)

3.0

MANDATORY DIRECT TOOL-TO-TOOL DELIVERY POLICY What will happen if we select another delivery policy that forces direct tool-to-tool delivery instead of current pushpull policy? Fig 4 shows outline of this new delivery policy. Flow charts in Fig.5 describe logics of this policy. We will call it ‘mandatory direct tool-to-tool delivery policy’.

2.5 2.0 m andatory push-pull

1.5 1.0 0.5 0.0 0.0

22

0.2

0.4 0.6 0.8 Throughput (relative)

1.0

1.2

Graph 2. Lot cycle time versus throughput curves for two delivery policies: typical push-pull policy and mandatory direct tool-to-tool delivery policy.

Graph 2 shows the simulation result. It shows mandatory policy has no effect to reduce lot cycle time.

delivery may go up spontaneously as cycle time goes down. How can we reduce lot cycle time? If we cannot reduce raw process time and delivery time, we must reduce lot wait time. Then, what are the causes of waiting lots? Possibilities are 1) inefficient lot scheduling / dispatching algorithms and 2) variability of process rate (wafer/hr) of a tool (see [2]). First we consider variability of process rate. By using simulation runs we can see how reduction of process rate variability decreases lot cycle time and increases the probability that lots are delivered directly. To see them we developed simple simulation model that has one tool and one stocker. Process rate variability is caused by tool down in the model. We can generate different levels of the variability by varying MTBF 0.5h, 6h, 12h, while keeping tool availability constant (i.e. we vary MTTR proportionally to MTBF). Graph 3 and 4 show simulation results. Graph 3 shows relationship between MTBF and lot cycle time under condition that tool utilization is uniform.

USAGE OF LOCAL BUFFERS Another way AMHS can do for direct tool-to-tool delivery is redirection of a lot to a local buffer near the original destination tool when the destination tool has no empty loadports. The local buffer holds the lot temporarily. Then AMHS delivers the lot to the original destination tool when one of its loadports becomes empty (Fig.6). Intra-bay and inter-bay unified OHT system

Stocker

Tool-B

This lot is delivered temporarily to a local buffer near to the destination tool.

Lot cycle time (relative)

Local buffer

Stocker

Tool-A 装置

Fig. 6. Usage of local buffers.

It may not be suitable to call this scenario ‘direct too-totool delivery’ because it violates definition of ‘direct toolto-tool delivery. Regardless of whether we call it so nor not, it is worthy to investigate whether this scenario has some advantages or not. If we assume the size of the buffer is infinite, to use such a local buffer is equivalent to move a stocker near to the tool. But both the original and the new location of the stocker satisfy the inequality TU + TL ≤ (LP − 1) • CT . Therefore both lot cycle times are represented lot cycle time in case of no AMHS, plus, AMHS delivery times. So, if there is not so much difference between delivery times for both delivery routes (previous production tool Æ stocker Æ destination tool, and, previous production tool Æ local buffer Æ destination tool), usage of local buffer near a tool has little advantages.

12 10 8

Tool Utilization =94%

6 4

Tool Utilization =80%

2

Tool Utilization =90%

0 0

2

4

6 8 MTBF (relative)

10

12

14

Graph 3. Lot cycle time versus MTBF under condition that tool utilization is uniform.

Graph 4 shows relationship between lot cycle time and probability of direct tool-to-tool delivery. It gives us examples that show how reduction of lot cycle time increases probability of direct delivery.

P robability of achieving direct tool-to-tool delivery (%)

100

DIRECT TOOL-TO-TOOL DELIVERY IS NOT A REAL GOAL These results suggest that the efforts of AMHS to realize direct tool-to-tool delivery have little advantages and the necessity to reconsider the original idea, ‘direct tool-totool deliver is one of the key technologies required to achieve excellent fab performance’. Realizing direct toolto-tool delivery is not a goal by itself. The goal should be reducing lot cycle time while keeping fab throughput uniform. The probability of achieving direct tool-to-tool

14

90 80

Tool Utilization =80%

70 60 50

Tool Utilization =90%

40 30

Tool Utilization =94%

20 10 0 0

5

10 15 Lot cycle time (relative)

20

Graph 4. Probability of direct tool-to-tool delivery versus lot cycle time under condition that tool utilization is uniform.

23

DEVELOPMENT OF STANDARDS FOR IMPROVEMENT OF FAB PERFORAMCE

(2) Effective algorithms for lot scheduling / dispatching and reduction of variability of process rate of tools are required to realize effective direct tool-to-tool delivery.

Let’s go back to the consideration of causes of waiting lots. We listed following causes there.

(3) Standardized ways to gather data from tools that could affect fab performances is needed to facilitate improvement of fab performances.

1) Inefficient lot scheduling / dispatching algorithms. 2) Variability of process rate (wafer/hr) of a tool.

REFERENCES

As for item 1), TOC (Theory Of Constraints, see [3]) is a promising method to minimize cycle time while keeping maximum throughput. TOC asserts that we should place a safety inventory buffer in front of the bottleneck tool group in order to exploit capacity of the bottleneck because the bottleneck determines fab throughput. If we apply TOC to semiconductor fab line, we will give up to realize direct tool-to-tool deliveries to the bottleneck tool group and make our efforts to increase ratio of direct deliveries to non-bottleneck tool groups. To perform TOC it is important to identify the bottleneck tool group.

[1] H. Kondo, M. Harada, “Estimation of influence that AMHS delivery time has on 300 mm fab productivity,” Proceedings in International Symposium on Semiconductor Manufacturing – ISSM, Sept. 2004, pp.87-90. [2]

As for item 2), from the ‘Queuing theory’, we can turn out that the greater the variability of lot inter-arrival time to a tool group and the variability of process rate of a tool, the longer a lot waiting time. The variability of lot interarrival time is mainly caused by the variability of process rate of previous production tools, therefore it results in variability of process rate. So, it is required to reduce variability of process rate of tools in order to improve ratio of direct delivery.

Wallace J. Hopp, Mark L. Spearman, “Factory Physics – Foundations of Manufacturing Management SECOND EDITION” New York: McGraw-Hill, 2000, especially Section 8.6.7

[3] Eliyahu M. Goldratt and Jeff Cox. “The Goal” New York: North River Press, 1992. AUTHOR BIOGRAPHY Hiroshi Kondo joined NEC Corporation in 1982 and had been engaged in development of Factory Automation System (especially real-time Dispatching system) for semiconductor Fabs. In 1998 he joined Selete (Semiconductor Leading Edge Technologies, Inc.) and developed AMHS simulation model and common communication specification between AMHS and MES. In 2002 he joined Asyst Shinko, Inc. and now is in charge of AMHS simulation researches.

In order to reduce the variability it is required to improve stability of production tools. But before doing so, we must establish means to measure such variability. Furthermore in order to identify the bottleneck tool group measurement of raw process times of tools is required. Therefore, it is preferable to standardize ways to gather such kind of data automatically from production tools in a fab. So, an activity is expected that investigates whether some of current SEMI standards (e.g. E58 ARAMS, E90 STS, E116 EPT etc.) are useful or not to gather such data for fab performance improvements. If we will find current SEMI standards are not useful enough to prepare standard way to gather such data, we should either modify some of current SEMI standards or develop new SEMI standards. This activity would facilitate to develop software tools to measure factors to degrade fab performances and help us to improve them.

Mitsuru Harada joined Asyst Shinko, Inc. in 2003 and now is in charge of AMHS simulation researches.

CONCLUSIONS Our conclusions are as follows. (1) AMHS cannot realize direct tool-to-tool delivery by itself without any degrading fab throughput.

24