Transcript Talk - DDDAS.org

Symbiotic Simulation Control in
Semiconductor Manufacturing
Authors: Heiko Aydt, Stephen J. Turner, Wentong Cai,
Malcolm Yoke Hean Low, Peter Lendermann, and Boon Ping Gan
Presented by Heiko Aydt
Parallel and Distributed Computing Centre
Nanyang Technological University
Singapore
Outline





Background
Symbiotic Simulation and DDDAS
Symbiotic Simulation Control System (SSCS)
Wet Bench Tool Set (WBTS)
Conclusions and Future Work
2
Background: Semiconductor
Manufacturing

Semiconductor Manufacturing is
 Highly complex: several hundred processing steps and up to
three month for production [1]
 Asset intensive: a single tool can cost up to US$ 2 million [2]
 Ongoing improvement of the manufacturing process
is crucial to stay competitive
 Integrated automation solutions [3]
 Investment range between US$ 130 million and US$ 180
million
 Include real-time control of equipment
3
Background: Challenges and Need for
Symbiotic Simulation
 Decisions regarding tool configuration is made
by engineers and carried out by workers
 Problems:
 Mostly based on experience
 Workers don’t always strictly follow the instructions
 Demand for fully automated control solutions
 Symbiotic simulation control is such a solution
4
Symbiotic Simulation and DDDAS
Physical System
Implement
Measure
“What if” experiments
simulate
Output
Analysis
simulate
Optimizer
simulate
5
Symbiotic Simulation and DDDAS

Symbiotic simulation and DDDAS are very similar
paradigms which are overlapping to some extend

Symbiotic simulation
 Close relationship between a physical system and a simulation
system
 Concept of what-if analysis is essential in symbiotic simulation
 Multiple simulations (of what-if scenarios) are dynamically datadriven
 Control feedback aims to modify the physical system

DDDAS
 Not necessarily limited to dyamically data-driven simulations
 Control feedback aims to steer the measurement process
6
Symbiotic Simulation Control
System (SSCS)

Symbiotic Simulation Control System (SSCS)



An agent-based generic framework for symbiotic simulation
systems has been developed



An SSCS evaluates decision alternatives (e.g, alternative
configurations) by means of simulation
The ‘best’ decision is directly implemented in the physical system
using corresponding actuators
Reference implementation using Jadex/JADE
Applicable in application scenarios of DDDAS/Symbiotic Simulation
Framework provides various functional components



Implemented as agent capabilities [4,5]
Standard implementations are provided
Allows flexible design of application-specific symbiotic simulation
systems
7
Symbiotic Simulation Control
System (SSCS)
Various Capabilities (x-C):




S-C: Sensor
WORC-C: Workflow Control
SCEM-C: Scenario Management
SIMM-C: Simulation Management



SIMA-C: Simulation Analysis
DECM-C: Decision Management
A-C: Actuator
8
Wet Bench Tool Set (WBTS)

A wet bench is used to clean wafers after certain fabrication
steps

A wet bench consists of a number of baths with chemical
liquids

Wafers lots are processed strictly according to recipes

Several wet benches with different setups are typically
operated
WetBench
Bath 1
Bath 2
Batch 1
Lot
Lot
Bath 3
Batch 2
Lot
Lot
Bath 4
Batch 3
Lot
Lot
9
Wet Bench Tool Set (WBTS)

Particles of contaminants are introduced into the baths
during the cleaning process [6]
 Some recipes introduce particles faster than others
 We therefore distinguish between ‘clean’ and ‘dirty’ recipes



Wafers are processed in different wet benches
depending on whether they are ‘clean’ or ‘dirty’
Thus, a wet bench operates either in ‘clean’ mode or
‘dirty’ mode
Switching the operation
 From ‘clean’ to ‘dirty’: does not require any activity
 From ‘dirty’ to ‘clean’: requires a complete change of liquids
10
WBTS: Control Approaches

Performance depends on product mix and operation modes of wet
benches


Reconfiguration required if product mix is changing
Common practise approach:


Engineers make decisions based on experience: difficult to model
Following heuristic used:
1.
2.
3.
4.

If the number of pending lots is exceeds limit: determine the critical recipe
Identify all wet benches which are capable of processing the critical recipe
but which are not configured yet
Reconfigure one of the wet benches and wait some time (settling-in
period)
If, after the settling-in time, the situation did not improve: repeat
SSCS approach:



Observes queues and trigger what-if analysis if number of pending
lots exceed limit
Create and evaluate a number of alternative configurations
Determine best configuration implement it in the physical system
11
WBTS: Experiments
 Emulator was used instead of a real physical
system
 Paced simulation
 Runs 3600 times faster than real-time
 Two kinds of experiments performed
1. Alternating product mix
2. High workload
12
WBTS: Experimental Results for
Alternating Product Mixes

Two different product mixes were used with constant load

The product mix is changed every 2 weeks
Figure 1: Cycle Time of the WBTS over a period of 20 weeks using the SSCS
(left) and the common practise control approach (right).

SSCS produces a more homogeneous performance with



Less variability
Lower mean cycle time
Disadvantage of the common practise approach:

Needs several attempts to find a stable configuration
13
WBTS: Experimental Results for High
Workload

At high load, it is necessary to oscillate between configurations


Only two wet benches have a bath setup which provides many recipes
At high loads, these two wet benches become a bottleneck
Figure 2: Cycle Time of the WBTS over a period of one month using a load of
1000 (left) and 1200 (right) lots per day.


Common practise approach becomes unstable when using 1200
lots per day
SSCS can handle both loads without problems
14
WBTS: Experimental Results for High
Workload
Figure 3: Cycle Time of the WBTS over a period of one month using a load of
1500 (left) and 1700 (right) lots per day.


The SSCS can handle 1500 lots per day but becomes unstable
for 1700 lots per day
Critical loads



For common practise approach: 1000-1200 lots per day
For SSCS approach: 1500-1700 lots per day
Performance improvement of 25-70% when using the SSCS
approach
15
Conclusions and Future Work
 A dynamic data-driven application, more particularly a
symbiotic simulation system, has been used to for real-time
control of semiconductor manufacturing equipment
 Our results show that using the SSCS yields a notable
performance improvement over common practise
 Symbiotic simulation is therefore a promising candidate for
integrated automation solutions
 Future work will be concerned with the application of an
SSCS to an entire semiconductor factory
16
References
1.
2.
3.
4.
5.
6.
Potordi, J., Boon, O., Fowler, J., Pfund, M., Mason, S.: Using simulation-based
scheduling to maximize demand fufillment in a semiconductor assembly facility. In:
Proceedings of the Winter Simulation Conference. (2002) 1857-1861
Scholl, W., Domaschke, J.: Implementation of modeling and simulation in
semiconductor wafer fabrication with time constraints between wet etch and
furnace operations. In: IEEE Transactions on Semiconductor Manufacturing.
Volume 13. (August 2000) 273-277
Gan, B.P., Chan, L.P., Turner, S.J.: Interoperating simulations of automatic
material handling systems and manufacturing processes. In: Proceedings of the
Winter Simulation Conference. (2006) 1129-1135
Busetta, P., Howden, N., Rönnquist, R., Hodgson, A.:Structuring BDI agents in
functional clusters. In ATAL. Volume 1757 of Lecture Notes in Computer Science,
Springer (1999) 277-289
Braubach, L., Pokahr, A., Lamersdorf, W.: Extending the capability concept for
flexible BDI agent modularization. In PROMAS. Volume 3862 of Lecture Notes in
Computer Science, Springer (2005) 139-155
Gan, B.P., Lendermann, P., Quek, K.P.T., van der Heijden, B., Chin, C.C., Koh,
C.Y.: Simulation analysis on the impact of furnace batch size increase in a
deposition loop. In: Proceedings of the Winter Simulation Conference. (2006)
1821-1828
17