International Journal on Engineering Performance-Based Fire Codes, Volume 7, Number 4, p.

155-173, 2005

EVALUATION OF DOWNWARD DESMOKE SYSTEM IN A CLEANROOM

Chen-Wei Chiu, Chun-Chieh Wang and Chiun-Hsun Chen

Department of Mechanical Engineering, National Chiao Tung University, HsinChu 30056, Taiwan

(Received 1 April 2004; Accepted 16 June 2004)

ABSTRACT

This study adopts the SFPE performance-based design procedure to evaluate the performance of upwards and
downwards desmoke systems, respectively, in a wafer fabrication zone. The tools used for this purpose include
Fire Dynamics Simulator (FDS) and Simulex. The design fires are 800 kW (scenario 1) and 3 MW (scenario 2),
respectively. Both desmoke systems in scenarios 1, 2 and 4 meet the performance criteria, whereas the
downward system in scenario 3 cannot satisfy the property protection requirement. However, the downward
desmoke system can still be utilized, provided it complies with some restrictions in fire protection design, for
example materials used and the provision of an acceptable emergency plan. For occupant evacuation, the
Simulex result shows that the total evacuation time is 191 s. FDS simulations involving the installation of smoke
curtain confirm that the evacuation time is less than the smoke layer descending time, indicating that the
occupants can safely evacuate in the event of a fire.

1. INTRODUCTION technology industries, such as TFT-LCD and LED.

Therefore, these industries have expended
1.1 Background tremendous attention and effort on developing an
advanced calamity strategy to reduce the risk of
Recently, the technical development of the fire.
Taiwanese semiconductor industry has accelerated
significantly, creating substantial profits. However, Some special features of cleanroom now used in
the 24-hour operating process of semiconductor advanced technology factories include the
manufacturing represents a significant potential following.
liability to the owner, operator and general public
in terms of life safety and preservation of assets, y Semiconductor manufacturing uses numerous
due to heavy use of hazardous production materials toxic and explosive chemicals, as listed in
(HPM). Additionally, the wafers and process tools Table 2. Most of these chemicals are so
are highly sensitive to smoke, particulates and active that they can easily cause fires or other
water. Therefore, a cleanroom is needed to severe damage in the event of accidental
maintain the required air quality, along with the leakage.
necessary humidity and particulate number to
prevent the wafers from suffering pollutant y The flow field in a cleanroom differs from
contamination. These factors raised the risks that in other buildings. The conditioned air
involved in operating such factories, particularly flow is drawn downwards vertically from the
the fire incident. In the event of a fire, the Filter Fan Unit (FFU) to the fabrication area,
resultant damage and property loss is much higher through the perforated raised floor, and
than in conflagrations in other types of buildings. eventually to the subfab, as illustrated in Fig.
For example, the cleanroom fire at Winbond 1. In the event of a fire, the movement of
Electronics Corporation [1] caused significant smoke in a cleanroom is expected to differ
property losses of approximately 6 billion NT, from that in other buildings. Additionally,
although nobody was injured, as listed in Table 1 cleanrooms contain no vertical and horizontal
[1]. Table 1 also lists the losses associated with fire-endurance compartments because of
another two severe fires in wafer fabs. production line allocation and the air flow
design mentioned previously.
However, the fire protection designs and the
corresponding equipment of these factories fully y Process tools inside cleanrooms are highly
complied with the local building [2] and fire codes sensitive to contamination from smoke/
[3], regarded as the minimum requirements for fire particulates and water.
safety. Apparently, the requirements by law
enforcement cannot fully meet the real needs for y The zone area of cleanroom for such factories
the fire safety design of the semi-conductor is very large, but the occupant density is low
industry or the other so-called advanced due to high automation. Furthermore, the

155
 International Journal on Engineering Performance-Based Fire Codes

distribution of production lines has become codes [2,3]. Therefore, this study introduces the
increasingly complicated owing to space fire safety engineering (FSE) design methodology
limitations. Therefore, occupant evacuation to provide a so-called substituted design
plans become extremely important and it may equivalency, and this methodology usually is
change occasionally due to the retrofit and superior to that required by the prescriptive codes
renovation of facilities. to resolve the fire protection situation.

Because the structure of cleanroom is regarded as Therefore, engineering tools, such as CFD models,
an enclosure, where there are no openings under are applied to design and evaluate the calamity
normal operating condition, it is required to equip system used in a cleanroom in an advanced
with the Desmoke System on the ceiling by the technology facility. This study evaluates a new
prescriptive fire code of Taiwan [3]. According to concept of downward desmoke system,
the code, the minimum desmoke capacity is set to incorporating a decrease of FFU air supply rate in
be 2 m3/min per m2 floor area. From the fire zone, the appropriate distribution of sprinklers
descriptions of the special features of cleanroom and an intensive trained evacuation plan, etc., to
mentioned previously, it is easy to understand that check whether such a design can achieve
fire safety design for semiconductor factories, and equivalent performance to the traditional upward
particularly for cleanrooms, has considerable desmoke system in meeting the local fire code.
difficulty in fully complying with the prescriptive The calamity system is described in detail later.

Table 1: Accident fires of semiconductor factories in Hsinchu, ROC

Winbond Electronics United Integrated Circuits Advanced Microelectronic

Corp. [1] Corp. [1] Products Inc. [1]

Time 1996/10/14 1997/10/03 1997/11/11

Casualty N/A N/A One man injured

Loss Property 6 billion NT 12 billion NT 3 billion NT

Fire Occasion Wet bench Fume exhaust system Wet bench

Table 2: Semiconductor manufacturing process gases

Manufacturing Process Using Gases

Epitaxy
SiH4, SiHCl3, SiH2Cl2, SiCl4, AsH3, AsCl3, PH3, B3H6, H2, N2

Diffusion AsH3, AsCl3, PH3, POCl3, B2H6, BCl3, BBr3, SiCl4, SiH2Cl2, Ar, NH3, N2

Chemical Vapor SiH4, SiHCl3, SiH2Cl2, NF3, CF4, CHF3, C2F6PH3, B2H6, NH3, N2O, O2, SiF4,
Deposition H 2, N 2

Etching HF, HCl, CF4, CHF3, CClF3, C2F6, PH3, B2H6 SiH4, C3F8, SiF4, SiF6, CBrF3,
C2ClF5, Cl2, PH5, CClF2, C2ClF2, C2Cl2F4, CCl4, C3H8, CH4NF3, Ar, He

Sputtering Ar, O2, NH3

Implant PH3, PF3, PF5, SiF4, AsH3, B2H6, BF3, BCl3

156
International Journal on Engineering Performance-Based Fire Codes

Dimension 1:270

Fig. 1: The construction of cleanroom

1.2 Literature Review workstation, where the surrounding fluid is from

vertical FFU flow. Hu suggested that turbulent
Kathy and William [4] conducted full-scale intensity effect should be considered in simulating
measurements in an aircraft hangar with a nominal the flow field when the processing tools are set up.
30.4 m (100 ft) ceiling height. Gas temperatures
resulting from an approximately 8250 kW Heskestad and Lutton [6] experimentally
isopropyl alcohol pool fire were measured both investigated the reduced scale cleanroom fire (the
above the fire and on the ceiling. They found that room size is 4.8 m x 6.0 m x 2.44 m in height).
the detector on the 30.4 m ceiling could not Specifically, Heskestad and Lutton measured
respond to the high HRR, 8250 KW, from the ceiling temperature when using a CH4 burner (600
ground pool fire from 0.75 m2 fuel pan. It is kW) to simulate a cleanroom fire.
verified by comparing the predictions from the
DETACT-QS, FPETOOL, LAVENT and Cheng et al. [7] applied the commercial code of
FLOW3D models with the experimental data. The STAR-CD to study cleanroom airflow uniformity.
field fire model used the volume heat source (VHS) The k-ε turbulent model was adopted, and a
and the k-ε turbulent model. However, Kathy and perforated raised floor was simulated as a porous
William concluded that the LAVENT model medium. The parameters were the height of the
corresponding to temperature and flow fields subfab and the permeability coefficients of the
provides a better agreement with the experiment porous floor, respectively. Cheng et al. found that
data.
increased subfab height (0.6 m − 1.5 m) and
decreased permeability coefficients (0.1 or 0.25) of
Hu [5] measured flow field and turbulence
the porous floor can optimize airflow uniformity.
characters, such as velocity vectors, turbulence
intensity and turbulence kinetic energy, by using
Nam [8] used the CFD program REFLEQS
an ultrasonic anemometer in the cleanroom. He
developed from NASA to forecast the extent the
investigated the turbulent intensity near the
smoke cloud from a cleanroom fire. The CO2

157
 International Journal on Engineering Performance-Based Fire Codes

concentration (34.5 ppm) was used as a marker for venting, air conditioning systems, fire suppression,
smoke concentration. Nam also used the volume electrical power and evacuation alarm systems.
heat source (VHS), 250 or 500 kW, and altered the The main codes adopted include Uniform Building
flow rate (0.1 ~ 0.91 ms-1) of FFU to make a Code (UBC) and Uniform Fire Code (UFC).
comparison with the experimental data of
Heskestad and Lutton [6]. Nam claimed that the 1.3 Scope of Present Study
analytical results are accurate with the
experimental data on a cleanroom fire. This study is based on Fire Safety Engineering
(FSE) guidelines to design calamity strategies for
Huang [9] used the CFD software FLUENT to high technology facilities. The main purpose is to
compare different combustion models, and meet the requirements for preservation of life and
confirmed the model accuracy through comparison property via performance-based design, based on
with the measurements of Heskestad and Lutton the assumption that present building and fire codes
[6]. He adopted three combustion models, are difficult to completely comply with. The basic
including the volume heat source (VHS), eddy design tool adopted is Fire Dynamics Simulator
break-up and presumed probability density (FDS), a CFD program devised by National
function (prePDF) model. Additionally, Huang Institute of Standards and Technology (NIST) [13].
considered the radiation effect on global radiative This tool is used to simulate design cleanroom fires
heat exchange via the discrete transfer method and thus obtain the corresponding fire and smoke
(DTRM). developments. An egress model, SIMULEX
software [14], then is used to evaluate the
Chang [10] used the CFD software PHOENICS to evacuation time required for each design fire,
confirm the efficiency of the calamity strategy. The together with the information on facility layout and
conceptual method of modifying the airflow rate population density. Finally, the optimal calamity
(0.26 ms-1) of FFU to increase the pressure strategy is deduced from these evaluations.
difference between the fire-endurance zones was
proposed to limit smoke spread. Chang found that
the upward smoke exhaust strategy appears more 2. CASE STUDIES
effective than the downwards one for fire accidents
occurring in the FAB (fabrication) zone. On the This study evaluates the calamity systems, whose
other hand, the downwards exhaust strategy is client is a 12” wafer fabrication factory, whose
more effective for SUBFAB fires. Therefore, wafer manufacture process diagram is illustrated in
Chang recommended that the semiconductor Fig. 2. This study uses FDS and Simulex as the
industry should prepare two calamity strategies. engineering tools to simulate the fire and smoke
However, this recommendation is unworkable in development by assigned scenarios, and
real situations. subsequently uses the occupant evacuation route
and time required to reach the safe zone.
Chen [11] examined the characteristics of fires and
explosions in semiconductor fabrication processes, The case studies consider two types of calamity
and illustrated their common and peculiar features systems, namely the downward desmoke system
of fire and explosion in the semiconductor (Fig. 3) and the upward desmoke system (Fig. 4).
fabrication processes. Particularly, this study The so-called return air shaft desmoke system,
discussed and collated the fire risk via the exhaust which is an inappropriate design because its
system, which generally contained large quantities distance to exhaust is too long and overall exhaust
of flammable and corrosive gases and vapors. capability is too low. This study emphasizes
evaluating the effectiveness and performance for
From the perspective of fire safety codes in Taiwan, both upward and downward desmoke systems. The
Chiu [1] surveyed and investigated air-tight procedure for selecting an appropriate calamity
cleanroom fires, which cause severe damage to life system follows the SFPE Engineering Guide to
and property. Chiu intensively discussed how to Performance-based Fire Protection Analysis and
combine smoke control systems, fire (smoke) Design [15]. The performance-based design
zones, fire protection equipment, tool layout and procedure is detailed as follows.
space distribution to control fire and smoke
movement. 2.1 Define Project Scope
The first step in performance-based design is to
Acorn [12] conducted an intensive review, entitled
define the project scope. The present project
“Code Compliance for Advanced Technology
involves smoke exhaust design evaluation in the
Facilities”. This review described and discussed
cleanroom of a 12” wafer fabrication facility.
the various code requirements for advanced
facilities in terms of HPM storage, mechanical

158
International Journal on Engineering Performance-Based Fire Codes

Wet Etch

Drive In Deposition CVD PVD

Photo

IMP Dry Etch

Fig. 2: Diagram of wafer manufacture

Fig. 3: The downward desmoke system

159
 International Journal on Engineering Performance-Based Fire Codes

Dimension 1: 300

Fig. 4: The upward desmoke system

Employing an upward desmoke system enables 1.8 m by shutting down all of the FFU. However,
wafer fabrication to comply with local fire codes. shutting down FFU without any action of the
However, such a fire protection system will cause desmoke system in an emergency can be refused
problems for the production line design. Since by the management of a wafer fab. Therefore,
wafer sizes grow due to advanced manufacturing designing an acceptable evacuation plan is
technology, wafers now are transported by extremely challenging.
machines, for example Automated Guided Vehicle
(AGV), rather than manpower. An AGV system 2.2 Identify Goals
suspended beneath the ceiling is now extremely
common in 12” wafer fabrication facilities for The next step is to identify the fire safety goals,
transporting wafers among workstations. including levels of protection:
Apparently, this transportation facility will
intervene in the upward desmoke system and (a) Protect life: ensure all of the occupants can
penetrate the fire/smoke curtain/barrier and/or escape from the cleanroom without being
compartments. Therefore, the code-complied harmed by the fire.
desmoke system must be evaluated, and a
substitute/equivalence design may be required for (b) Protect property: restrict the spread of fire and
conflict resolution. smoke to adjacent compartments, and reduce
the damage in the zone of origin of the fire as
To prevent the above problems interfering with the much as possible.
performance of the upward desmoke system, a
downward desmoke system was considered by the 2.3 Develop Performance Criteria
12” wafer fabrication facility. Therefore, this The performance criteria are the further refinement
project was set up to evaluate its feasibility, of design objectives, and take the form of
determine the corresponding emergency operating numerical values with which the expected
procedure and demonstrate its equivalency to the performance of the trial designs can be compared.
code-complied system, with this last being the key Performance criteria may include two parts:
issue.
(a) Life safety criteria: life safety criteria address
However, it is very difficult to quantify the the survivability of individuals exposed to fire
prescriptive codes, particularly for making a and its by-products. Tables 3 and 4 indicate
comparison with the performance design. the life safety criteria, which are specified so
Therefore, this project adopts an occupant that the occupants have sufficient time to
evacuation time to provide a comparison basis. safely escape. This project adopts the
Since no standard procedure exists for the restrictions that along the evacuation route, the
evacuation, the time is commonly adopted as the temperature at the height of 1.8 m does not
required evacuation for a smoke layer descended to exceed 80°C, and the concentration of the

160
International Journal on Engineering Performance-Based Fire Codes

toxic carbon monoxide gas does not exceed wetbench fires in the fab-zone, while the second
1400 ppm at the same location [16]. cause is fume exhaust duct fire. A wetbench fire in
the fab-zone is adopted as the fire scenario here.
(b) Non life-safety criteria: in semiconductor
factories, wafers and process tools are highly The fire scenario in the Fab-zone is a wetbench fire,
sensitive to smoke contamination, and thus the and this fire is assumed not to spread to other
project selects the proper criteria, for example workstations. FDS simulation divides the
FM7-7/17-12 [17], a loss prevention data of fabrication area into eight compartments, each of
semiconductor fabrication facilities in which has an area of 450 m2 and a height of 5 m.
cleanroom, as the fire protection requirements. Additionally, the fire simulation should include the
In the fire-origin zone, the fire does not reach perforated raised floor (2 m high) and the subfab (5
the flashover [18], whose ceiling temperature m high), as illustrated in Fig. 1. Therefore, the total
exceeds 600°C. For smoke damage criterion, height is 12 m. The selected cubic grid size is
the smoke particles are not allowed to enter 0.125 m3 and the total grid number is 379,000. The
the neighboring zones. wetbench, which has a combustible area of 6 × 12
m, is made of PP (Polypropylene) and is located at
2.4 Develop Fire Scenarios and Design the center of the fire-origin zone. The floor is
Fire nonflammable steel and the ceiling is
nonflammable material, except for the FFU frame.
This step involves describing possible fire events, Fig. 4 illustrates the configuration of the cleanroom.
and involves fire, building and occupant Sprinkler installation complies with NFPA 13 [19],
characteristics. Next, several assumptions are and the sprinkler is selected as an early suppression
made regarding these fire scenarios: fast response with RTI (Response Time Index) of
45. The default activation temperature of the
y no arson is considered. sprinklers is 68°C.
y no explosion is considered.
y no destruction of building is considered. According to the FM test results [20] for different
y no human extinguishing of the fire is fire simulations on a wetbench, the CO2
considered. extinguishment system fitted inside the wetbench
can effectively reduce the heat release rate.
The cleanroom fire scenario is designed in However, this work neglects these effects by
accordance with the above restrictions. The F.M. assuming that the system fails, one of the worst
statistical data [17] on the cleanroom fires in conditions.
America from 1977 to 1997 identify two primary
causes of cleanroom fires. The first cause is

Table 3: Limiting conditions for tenability caused by toxic products of combustion

5 minutes exposure 30 minutes exposure

Chemical products
Incapacitation Death Incapacitation Death
Carbon monoxide 6000 ppm 12000 ppm 1400 ppm 2500 ppm
Low oxygen < 13% < 5% < 12% < 7%
Carbon dioxide > 7% > 10% > 6% > 9%

Table 4: Limiting conditions for tenability caused by heat

Symptom Exposure level

Radiation Severe skin pain 2.5 kWm-2

Conduction Skin burns 1 s contact (metal) 60°C
Convection Skin/lungs affected by hot gas in > 60 s 120°C
Convection Skin/lungs affected by hot gas in < 60 s 190°C

161
 International Journal on Engineering Performance-Based Fire Codes

Two fire scenarios are chosen for the Fab-zone fire. downward at a velocity of 0.5 ms-1. The desmoke
In one scenario the fire is ignited from combustible system manually activates 30 s after a fire is
vapors, evaporated from the wafer-cleaning liquid confirmed. The two smoke management strategies
IPA (Isopropyl Alcohol), in a wetbench in the are specified as follows:
etching area. The heat release rate of PP is
measured using a cone calorimeter test [21]. The (a) Downward desmoke system: reduce the FFU
heat release rate reaches 400 kWm-2 after 50 s and air velocity to 0.1 ms-1 in the fire-origin
800 kWm-2 after 217 s for this material. Notably, compartment and increase the velocity to 0.6
the burning area of the wetbench is limited to 1 × 1 ms-1 in the neighboring compartments. The
m2 at the floor surface. The total simulation time is exhaust capacity of the desmoke system in the
300 s. fire-origin compartment is 26.88 m3s-1, and the
system is not activated in other compartments.
To further verify the exhaust capability of
downward desmoke system, a higher fire-load is (b) Upward desmoke system: shut down the FFU
also chosen. This fire scenario results from a tool air flow in the fire-origin compartment and
fire caused by the careless usage of photoresist increase the velocity to 0.6 ms-1 in the
stripper in the photolithography zone. Because neighboring compartments. The exhaust
photoresist stripper is a flammable chemical liquid, capacity is 20 m3s-1.
the corresponding maximum heat release rate is set
as 3 MW. The fire growth adopts a t-squared 2.6 Evaluate Trial Designs
model [22,23] and is assigned ultra-fast, and the
heat release rate reaches 1 MW at 75 s. Table 5 For each trial design, the fire/smoke development
summarizes the fire scenarios and the two desmoke is simulated using FDS developed by NIST. The
systems. life and property criteria then are examined to
determine whether they meet the specified
requirement. Finally, the evacuation times, subject
2.5 Develop Trial Design to the former simulation conditions, are calculated
For the operation parameters, the initial condition using the SIMULEX program.
is for all of the FFU air streams to be flowing

Fab-Zone
Zone 3
Fire Damper Zone 5
Zone 1

Subfab-Zone
Return Air Shaft

Fig. 5: The simulation layout of cleanroom structure

Table 5: Trial designs

Number of Calamity
Fire Growth Model Fire Location
Scenarios Strategy
S1 800 kW (cone) Downward Wetbench fire in
Trial Design 1
S2 800 kW (cone) Upward zone 3
S3 3MW (t-squared model) Downward Photoresist stripper
Trial Design 2
S4 3MW (t-squared model) Upward fire in zone 7

162
International Journal on Engineering Performance-Based Fire Codes

3. SIMULATION RESULTS temperature. The variation initially increases

monotonically with time until the sprinkler is
3.1 FDS Simulation Results activated. However, the maximum CO
concentration, 113 ppm, occurs at t = 159 s, one
The FDS simulations comprise four scenarios second behind the activation of the sprinkler. This
involving both downward and upward desmoke value is significantly lower than the unattainable
systems. The respective scenarios are evaluated toxic criterion of 1400 ppm, indicating that the
based on whether they can meet the performance combustion is more complete. This occurs because
criteria of both life safety and property protection. the air is still supplied from FFU even if the fire
To compare the scenarios with the life safety occurs in this scenario.
criteria, three different points at the height of 1.8 m
in the fire-origin compartment are selected to Apparently, the activation of the downward
obtain the corresponding temperature and CO desmoke system in Scenario 1 can meet the
concentration. These three points are located at the required performance criteria specified in section
center of the fire source (point 1), and at 3 m (point 2.3. To further insure that the fire and smoke in
2) and 5 m (point 3) from the center, respectively. fire zone (zone 3) do not spread out to the
The last two points provide reference points for neighboring zones, such as zone 1 and 5, by the
checking the life safety criteria on the evacuation fire control method mentioned above, the
route. Finally, smoke movement is visualized corresponding temperatures and CO concentrations
using Smokeview, one of the FDS functions. at the height near the ceiling for these zones are
presented in Fig. 7b. It clearly shows that both the
Trial Design 1 temperatures and CO concentrations beneath the
For Trial design 1, the fire-origin source is situated ceiling in zone 1 and 5 are maintained at their
in Zone 3. A fire was started at the center of the original values, indicating that the fire and smoke
wetbench by simulation when t = 0, and the indeed are restrained in the fire zone.
environmental temperature was fixed at 24°C. The
air-conditioning continued to operate normally, Scenario 2 involves the case of a wafer fabrication
and the desmoke system was closed after the fire. area equipped with the upward desmoke system.
Fig. 6 shows that the smoke particles in the Notably, the FFU is switched off in the fire-origin
simulation started to descend at 21 s. At t = 30 s, zone because the fire is confirmed to occur in this
the desmoke system was manually activated. The scenario. Therefore, the sprinkler is expected to
FFU in the fire zone decreased to 0.1 ms-1, and activate earlier than that in Scenario 1. In Fig. 8a,
those in the other zones increased to 0.6 ms-1. The the highest temperature at point 1 reaches about
smoke particles were found to be restricted to the 157°C, lower than the corresponding one in
fire-origin zone when the desmoke system was Scenario 1. This difference occurs because the fire
manually activated 28 s later. At t = 110 s, the plume rises directly to the ceiling without
smoke particles were brought downward near the encountering any incoming flow. The temperatures
smoke exhaust vent. The sprinklers were activated in points 2 and 3 are maintained at around 24°C.
at about t = 158 s. At t = 300 s, the smoke particles
were almost completely extracted by the desmoke In Fig. 8a, the highest CO concentration at point 1
system. Also, no flashover occurred. is about 82 ppm, lower than the relative one in
scenario 1, owing to the local temperature being
In this scenario, Fig. 7a shows the temperature lower than criteria temperature. The temperature
history at point 1. Initially, the temperature rise is and toxicity do not exceed the life safety criteria.
almost linear with time, and begins to become Fig. 8b shows the temperatures and CO
oscillatory after t = 80 s. However, the trend is concentrations at the height near the ceiling for the
rising. The oscillation phenomenon results from fire zone (zone 3) and its neighboring zones (zone
the interaction among the rising fire plume, 1 and zone 5). Similar to the last scenario, the fire
downward air stream from FFU and suction of the and smoke do not spread out of the fire zone, in
desmoke system. The highest temperature, 200°C, other words, the neighboring zones are not affected
occurs just before the sprinkler takes action at t = by the fire occurrence.
158 s. After that, the temperature drops suddenly
as expected, then quickly recovers and Regarding property protection, since the desmoke
subsequently levels off. Apparently, the fire is not system is manually activated after 30 s with the
extinguished but is controlled by the action of pressure difference resulting from different
sprinkler system. On the other hand, the downward air velocities restraining the smoke
temperatures at points 2 and 3 are maintained at a particles to remain within the fire-origin zone. At t
constant 24°C. = 70 s, the upward desmoke system completely
exhausted the smoke particles. The sprinklers were
The variation of CO concentration at point 1 activated at approximately 149 s. By 180 s, the
shown in Fig. 7a almost synchronizes with that of

163
 International Journal on Engineering Performance-Based Fire Codes

device has effectively controlled the fire situation respectively. The highest CO concentration at
and no flashover happened. point 1 is 281 ppm (see Fig. 9a), which is still
below the unattainable one of 1400 ppm.
Trial Design 2 Apparently, both CO concentration and
Trial design 2 chooses a significantly stronger fire temperature meet the criteria for protecting human
of 3 MW in heat release rate. In scenario 3, using life. From Fig. 9b, which is the temperatures and
the downward desmoke strategy, the highest CO concentrations at the height near the ceiling for
temperature at point 1 is about 257°C (see Fig. 8) zone 1, 3 and 5, it indicates that the neighboring
at t = 60 s and then the sprinklers were activated at zones, zone 1 and zone 5, are not attacked by the
about 73 s. Meanwhile, the temperature maximum fire occurred in zone 3.
at points 2 and 3 are around 29°C and 27°C,

t = 24 sec

t = 30 s

t = 58 s

t = 110 s

t = 159 s

t = 300 s

Fig. 6: The smoke particles profiles (scenario 1)

164
International Journal on Engineering Performance-Based Fire Codes

Fig. 7a: The temperature and CO concentration measured variation at 1.8 m (scenario 1)

300 300
Temp Zone 3
Temp Zone 1
Temp Zone 5
CO Zone 3
CO Zone 1
CO Zone 5

200 200
CO ( ppm)
Temp

100 100

0 0
Time ( sec )
0 100 200 300

Fig. 7b: The temperature and CO concentration measured variation at the height near ceiling
(scenario 1)

165
 International Journal on Engineering Performance-Based Fire Codes

300 300

center
next to 3m
next to 5m
CO

200 200

CO ( ppm )
Temp

100 100

0 0

0 100 time ( sec ) 200 300

Fig. 8a: The temperature and CO concentration measured variation at 1.8 m (scenario 2)

300 300

Temp Zone 3
Temp Zone 1
Temp Zone 5
CO Zone 3
CO Zone 1
CO Zone 5

200 200
CO ( ppm )
Temp

100 100

0 0

0 100 200 300

Time ( sec )

Fig. 8b: The temperature and CO concentration measured variation at the height near ceiling
(scenario 2)

166
International Journal on Engineering Performance-Based Fire Codes

300 300

200 200

CO ( ppm )
Temp

100 100
center
next to 3m
next to 5m
CO

0 0

0 100 time ( sec ) 200 300

Fig. 9a: The temperature and CO concentration measured variation at 1.8 m (scenario 3)

300 300
Temp Zone 3
Temp Zone 1
Temp Zone 5
CO Zone 3
CO Zone 1
CO Zone 5

200 200
CO( ppm)
Temp

100 100

0 0

0 100 Time ( sec ) 200 300

Fig. 9b: The temperature and CO concentration measured variation at the height near ceiling
(scenario 3)

167
 International Journal on Engineering Performance-Based Fire Codes

However, Fig. 10 shows that the smoke particles Apparently, the fire is constrained in the fire zone
are restricted in the fire-origin zone within 50 s and it does not spread out of the zone 3. Therefore,
after the fire, and then pass across the boundary Scenario 4 successfully meets the performance
lines to enter the neighboring zones. Eventually, requirements.
the smoke particles are full of the fabrication zone.
Consequently, the smoke control strategy in 3.2 Evacuation Results
scenario 3 apparently fails to meet the property
protection requirement. From the above discussion, there appears to be no
problems in the safe evacuation of the occupants of
In scenario 4, using an upward desmoke system, the fabrication area in the event of a fire in the
the highest temperature at point 1 is about 250°C assigned fire scenario. However, the fire
(see Fig. 11a). Meanwhile, the highest simulations do not consider the installation of the
temperatures at points 2 and 3 are 27°C and 26°C, smoke curtain as explained in section 2.1.
respectively. The highest CO concentration at Nevertheless, the local fire code requires that a
point 1 is about 266 ppm (see Fig. 11a). smoke curtain be installed. Therefore, this part of
Apparently, scenario 4 complies with the life the simulation must prove that the evacuation time
safety criteria. The smoke particles are can still meet the provision in such design.
continuously restricted in the fire-origin zone
because the upward desmoke strategy can bring the The fabrication area contains 351 occupants, and
high temperature smoke away directly. Also, Fig. has eight exits. The width of these exits is 1.2 m.
11b shows that the temperatures and CO Fig. 12 shows the production line configuration
concentrations near the ceiling at zone 1 and 5 are and the distribution of occupants in the wafer
still maintained at their original values and not fabrication area.
affected by the fire taken place in zone 3.

t = 60 s

t = 73 s

t = 100 s

t = 150 s

Fig. 10: The smoke particles profiles within temperature (scenario 3)

168
International Journal on Engineering Performance-Based Fire Codes

300 300

200 200

CO ( ppm )
Temp

100 100
center
next to 3m
next to 5m
CO

0 0

0 100 time ( sec ) 200 300

Fig. 11a: The temperature and CO concentration measured variation at 1.8 m (scenario 4)

300 300
Temp Zone 3
Temp Zone 1
Temp Zone 5
CO Zone 3
CO Zone 1
CO Zone 5

200 200
CO ( ppm )
Temp

100 100

0 0

0 100 Time ( sec ) 200 300

Fig. 11b: The temperature and CO concentration measured variation at the height near ceiling
(scenario 4)

The resultant evacuation time is 67 seconds using The evacuation time then is 101 s. Importantly, 90
SIMULEX. Notably, the dynamic evacuation time s more must be added to consider the activation
should be multiplied by a safety factor of 1.5, time (30 s) for starting the smoke control system
which is a requirement of the Fire Administration.

169
 International Journal on Engineering Performance-Based Fire Codes

and the occupant response time (60 s). The total Fig. 13 shows the escaping movements at t = 30 s.
evacuation time is calculated as 191 s (see Table 6). The figure reveals that exits 2, 3, 4, 6, 7 and 8
display the slightly crowed condition. In summary,
Table 6: Total evacuation time since the wafer fabrication zone is very large,
occupant density (331/3600 = 0.09 persons/m2) is
Simulex Simulation very low, and these works are quite familiar with
Time (s) 30* + 60 + 101 (67 ×1.5) = 191 the environment, the evacuation is expected to
proceed smoothly. For example, the three
accidental fires mentioned in section 1.1 caused
just one injury and no fatalities.

Fig. 12: The facility configuration and occupants displayed in wafer fab

Fig. 13: The occupants escaping movements (t = 30 s)

170
International Journal on Engineering Performance-Based Fire Codes

3.3 Comparison between the Results and 3.4 Selected Design Meets Performance
the Prescriptive Code Criteria
The above simulated results obtained from the Table 8 lists the result of evaluation trial designs.
performance-based design procedure should be The simulation results confirm that the downward
compared with the corresponding results from the exhaust strategy can tolerate the situation of 800-
local fire code to ensure their equivalence and/or kW fire, but it fails owing to the fire intensity in
superiority. Since the prescriptive code does not the case of 3 MW. Therefore, the amount of HPM
mention any specific evacuation time, the present in the fabrication area must be controlled. This
study divides the Fab-zone into eight smoke provision is also suggested by Article 51 of UFC
compartments according to the fire code, which [25], NFPA 318 [26] and FM Loss Prevention
requires that each 500 m2 compartment in the Data 7-7/7-12 [17].
building should have at least one hour of fire-
resistance, enabling occupants to escape to a safe Finally, according to the trial design results,
place unharmed. Based on this regulation, this several recommendations for fire safety
study uses FDS to further evaluate occupant safety. engineering are proposed:

The FDS model uses two types of smoke curtain, y The wetbench should utilize metal material
with heights of 50 cm and 80 cm, respectively, instead of PP. In the event of a wetbench fire,
under two conditions, a successful and a failed heat release rate can be significantly reduced,
operation desmoke operation (the worst case), and the downward exhaust strategy then can
respectively. The designated desmoke capacity is control the fire.
15 m3s-1 in each smoke compartment, complying
y The wafer fabrication usually has an AGV
with the local fire code (Taiwan). Table 7 lists
wafer transport system to cut across the
comparisons of the results from the prescription-
compartment, and can seek a newly designed
based codes and the Performance-based design.
smoke curtain, which is hidden in the ceiling
The FDS smoke layer interface position is
normally and can be pulled downwards
determined by the N-percentage rule and the
automatically in the event of a fire.
corresponding value of N is chosen as 80,
recommended by NFPA 92B (suggesting N = y It is suggested that the variation range of the
80~90) [24]. In the operational case, the smoke FFU flow speed be maximized, to make the
descending time is 352 s using the 50-cm smoke resultant pressure difference more effective in
curtain and 387 s using the 80-cm smoke curtain. restricting the smoke spread.
In the failure case (or the worst case), the smoke
descending time is 281 s using the 50-cm smoke y In this study, the concentrations and spread of
curtain and 305 s using the 80-cm smoke curtain. corrosive and toxic gases, which are part of
combustion products, are not taken into
The simulated evacuation time (Simulex) for the consideration because the limited available
downward desmoke system is 191 s, less than the data in FDS itself. Therefore, a new project
is proposed recently to measure and evaluate
281 s obtained for the worst case, which complies
with the local fire code. It indicates that the the other combustion products of the
downward desmoke system can achieve the safety materials, especially the plastics, used in
semiconductor manufacture workstations by
goal in cases where occupant evacuation are
required. using Cone Calorimeter with FTIR. These
data will be applied in the future extension of
this study.

Table 7: Simulation differences based on prescription-based codes and the performance-based design
case

Smoke exhaust Smoke curtain Descending smoke Simulex

Venting time (s) evacuation time (s)
Prescriptive codes Yes 50 cm 352 -
(Taiwan) Yes 80 cm 387 -
Fail 50 cm 281 -
Fail 80 cm 305 -
Performance-based design Exhaust capacity - - 191
(Downward)
26.88 m3s-1

171
 International Journal on Engineering Performance-Based Fire Codes

Table 8: The evaluating result of trial designs

Performance requirement Trial Design 1 Trial Design 2

Scenario1 Scenario2 Scenario3 Scenario4

Life safety Temperature O O O O
criteria < 80°C

CO Concentration O O O O
< 1400 ppm

Non-life Restrict smoke O O X O

safety particles in the
criteria fire zone

“O”: passed “X”: failed

4. CONCLUSIONS occupant evacuation time using the criterion,

where the evacuation is completed before the
This study adopted the SFPE performance-based time of smoke layer descending to 1.8 m,
design procedure to evaluate the desmoke system may serve as a safe quantity for life safety.
in a wafer fabrication zone. The tools used are FDS
and Simulex, respectively. Upward and downward y In property protection, insurance companies
desmoke systems are set up in the wafer generally require significantly higher
fabrication zone. The design fires are 800 kW and standards of property protection. Advanced
3 MW. Both desmoke systems in designed fire 1 technology facilities should consider smoke
(800 kW) meet the performance criteria, while the damage as the first priority in the cleanroom,
downward desmoke system in a designed fire of 3 rather than no flashover in the fire-origin
MW do not satisfy the required criteria. However, compartment.
the downward desmoke system can still be utilized,
provided there is full compliance with various y Local fire authorities should establish a
restrictions in the fire protection design, for special fire code, similar to UFC Article 51,
example the materials used and the acceptable suitable for fire safety requirements for
emergency plan. Although the downward desmoke advanced technology factories.
system cannot completely protect assets, the fire
scenario (3 MW) passes the life safety
requirements by using the downward desmoke REFERENCES
system. The downward desmoke system thus
meets the local fire code requirements, but 1. Chen-Wei Chiu, “A study of fire safety design
insurance companies do not necessarily agree with factors used in semiconductor cleanroom
the fire protection system. facilities”, Master thesis, Central Police University
(1998).
Regarding occupant evacuation, the simulation 2. Building Technology Code, Fire Safety Code
result of Simulex shows total evacuation time of Compilation, National Fire Administration,
191 s. FDS simulations with smoke curtain Ministry of Interior, ROC (2003).
installation confirm that the evacuation time is less 3. The Domestic Code for Installments of the Fire
than the smoke descending time, indicating that the Protection Equipments for Buildings, Fire Safety
occupants can evacuate safely. Code Compilation, National Fire Administration,
Ministry of Interior, ROC (2003).
Presently, local fire codes remain inadequate for 4. D. Willam and A. Kathy, “The use of computer
advanced technology factories because suitable fire models to predict temperature and smoke
protection regulations currently remain unavailable. movement in high bay spaces”, Building and Fire
Therefore, several recommendations are proposed Research Laboratory Documents, NISTIR 5304
for local fire codes: (1993).
5. S.C. Hu, Y.Y. Wu and C.J. Liu, “Measurements of
y The governmental regulations basically are air flow characteristics in a full-scale clean room”,
considered the minimum requirements for Building and Environment, Vol. 31, No. 2, pp.
fire safety. Unfortunately, these regulations 119-128 (1996).
cannot quantify the safety level. Therefore,

172
International Journal on Engineering Performance-Based Fire Codes

6. G. Heskestad and J.C. Lutton, “Novel Association, Quincy, MA (1999).

smoke/flume control for cleanroom”, FMRC
Technical report, N.O. J.I. OB0J1.RU (1997). 23. G. Hestestad, “Rise of plume front from starting
fires”, Fire Safety Journal, Vol. 36, pp. 201-204
7. M. Cheng, G.R. Liu, K.Y. Lam, W.J. Cai and E.L. (2001).
Lee, “Approaches for improving airflow
uniformity in unidirectional flow cleanrooms”, 24. NFPA 92B, Guide for smoke management system
Building and Environment, Vol. 34, No. pp. 275- in malls, atria, and large areas, National Fire
284 (1999). Protection Association, Quincy, MA (2000).

8. Soonil Nam, “Numerical simulation of smoke 25. UFC Article 51, Semiconductor fabrication
movement in cleanroom environment”, Fire Safety facilities using hazardous production materials,
Journal, Vol. 34, No. 2, pp. 169-189 (2000). Uniform Fire Code (2000).

9. Yu-Lin Huang, “A study of the numerical models 26. NFPA 318, Standard for the protection of clean
in cleanroom fire of semiconductor factory”, room, National Fire Protection Association,
Master thesis, National Taiwan University (2002). Quincy, MA (2000).

10. Sheng-Kai Chang, “Discussion of calamity control

model in semiconductor factory,” Master thesis,
National Taiwan University (2002).
11. Jenq-Renn Chen, “Characteristics of fire and
explosions in semiconductor fabrication
processes”, Process Safety Progress, Vol. 21, No.
1, pp. 19-25 (2002).
12. W.R. Acorn, Code compliance for advanced
technology facilities, Noyes, New York, Noyes
Publications William Andrew Publishing, LTC
(1993).
13. Fire Dynamics Simulator (Version 3) user’s guide,
National Institute of Standards and Technology,
NISTIR 6784 (2002).
14. Simulex user manual evacuation modeling
software, Integrated Environmental Solutions Ltd.
(1998).
15. The SFPE Engineering guide to performance-
based fire protection analysis and design, Society
of Fire Protection Engineers and National Fire
Protection Association, Quincy, MA (1998).
16. Draft British Standard Code of Practice for the
Application of Fire Safety Engineering Principles
to Fire Safety in Buildings, Warrington Fire
Research Consultants (1993).
17. FM7-7/17-12, Property loss prevention data sheets,
Semiconductor fabrication facilities, pp. 28-36,
Factory Mutual (1999).
18. R.D. Peacock, P.A. Richard, W. Buckowski and V.
Babrausskas, “Defining flashover for fire hazard
calculations”, Fire Safety Journal, Vol. 32, No. 4,
pp. 331-345 (1999).
19. NFPA 13, Standard for the installation of sprinkler
systems, National Fire Protection Association,
Quincy, MA (2000).
20. FMRC Clean room materials flammability test
protocol, Factory Mutual Research Corporation,
Class Number 4910 (1997).
21. H.E. Mitler and K.D. Steckler, “Wall fires and the
approach to flashover in an enclosure”, Fire Safety
Journal, Vol. 24, pp. 71-78 (1995).
22. NFPA 92A, Recommended practice for smoke-
control systems, National Fire Protection

173