CLEANROOM ESTABLISHMENT AND PROCESSING IMPLEMENTATION FOR ELECTRON DRAG DISSERTATION Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University By Anthony J. Ragucci, B.S., M.S. ***** The Ohio State University 2004 Dissertation Committee: Thomas J. Gramila, Adviser John W. Wilkins Steven A. Ringel Charles Pennington Approved by Adviser Department of Physics
146
Embed
CLEANROOM ESTABLISHMENT AND PROCESSING ......ABSTRACT The complete specification, design, and implementation of a class 100 cleanroom is described in addition to the sample processing
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
CLEANROOM ESTABLISHMENT AND PROCESSING
IMPLEMENTATION FOR ELECTRON DRAG
DISSERTATION
Presented in Partial Fulfillment of the Requirements for
X. G. Feng, S. Zelakiewicz, H. Noh, T. J. Ragucci, and T. J. Gramila, “NegativeElectron Drag and Holelike Behavior in the Quantum Hall Regime,” Phys. Rev.Lett. 81, 3219 (1998).
The ability to accurately pattern, manipulate, or measure at a microscopic level
or smaller is requisite for many areas of experimental condensed matter physics. As
the dimensions of the system under study shrink, the need for a clean environment
to avoid the influence of particulate and chemical contamination during processing
increases. Equally important is the development and implementation of a detailed
and explicit method for processing samples under study.
This dissertation will discuss the complete specification, design, and construction
of a cleanroom environment and the associated processing for sample development.
These samples are needed for electron drag, a technique which is unique in providing
direct measurement of electron-electron scattering.
This novel technique provides a new and powerful means for addressing funda-
mental questions in electron physics. However, it is quite challenging technically,
involving ultra-low noise measurement, cryogenic environments, and state-of-the-art
sample growth and preparation. In particular, the difficulty and complexity of the
sample preparation techniques require dedicated instrumentation. Meeting that need
required the construction of a semiconductor processing cleanroom.
1
Although much of the processing required for electron drag measurements is spe-
cific to that technique, many of the requirements needed for the cleanroom, as well
as a number of new processing developments, are more broadly applicable.
Unfortunately, the availability of specific information regarding the development
of a highly functional cleanroom for an academic environment is limited. The con-
straints for a cleanroom in this setting, both financially and in terms of the limita-
tions of existing facilities, require a unique design approach. It is with these goals
and restrictions in mind that the cleanroom and process under consideration here are
presented.
The following sections will introduce the experimental technique of electron drag
and its implementation. Electron drag measurements will not be shown in this work
but an understanding of the target experiment aids in the explanation of cleanroom
and processing design goals. Although the concepts presented in this dissertation
are not difficult to understand, they are important for the realization of many ex-
periments. This author hopes that the information presented is of practical value to
future researchers establishing a new laboratory.
1.1 Electron Drag
The experimental technique of electron drag explores the behavior of interacting
two-dimensional electron gasses (2DEGs) through a direct probe of electronic scatter-
ing rates. Theoretical investigations by Pogrebinskii [1] in 1977 and Price [2] in 1983
indicated that carrier momentum could be transfered between two closely spaced con-
ducting films through interlayer electron interactions. Following work demonstrating
drag between two-dimensional (2D) and three-dimensional (3D) systems by Solomon
2
Figure 1.1: Drag schematic
et al. in 1989 [3], electron drag between two 2D systems was clearly observed by
Gramila et al. in 1991, using epitaxially-grown GaAs/AlxGa1−xAs heterostructures
to define a pair of 2DEG layers [4]. Since that time, electron drag has proved to be
a useful and illuminating tool for exploring interacting electron systems [5].
The fundamental bilayer drag measurement is schematically represented in fig-
ure 1.1, which has been described previously [4]. Two closely spaced but electrically
independent 2DEG layers are shown, labeled as “Drive Layer” and “Drag Layer.”
A small current, typically 100 - 200 nA, is driven through the drive layer. Because
of interlayer electron scattering, momentum is transfered from electrons in the drive
layer to electrons in the drag layer. The drag layer electrons are pushed towards
one end of the sample but are prevented from leaving the layer. Due to the charge
imbalance along the length of the sample, a potential difference is established which
creates a force on the drag layer electrons. This electrical force opposes the drag force
3
created by momentum transfer from the drive layer electrons. At equilibrium, these
two forces balance and the voltage difference across the drag layer is used as a direct
indication of the electron-electron scattering rate between the two layers.
The basic relation between the drag voltage measured and the actual scattering
rate between electrons in the two layers can be obtained using the Drude model of
resistivity. If the average interlayer scattering time is τd for electrons with effective
mass m∗ and drift velocity vd, then the drag force is the rate of momentum transfer
between the two layers,
Fd =m∗vd
τd
.
Opposing this is the electrical force due to the charge imbalance across the length L
of the sample,
Fe = −eE = −eVd
L,
where Vd is the measured drag voltage. In equilibrium, these two forces balance so
that
m∗vd
τd
= −eVd
L.
The current driven in the drive layer can be described as
I = nevdW,
so that the measured drag resistivity, defined as the ratio of drag voltage to drive
current scaled by the sample aspect ratio, is
ρd = −Vd
I
(W
L
)=
m∗
ne2τd
.
This relation is quite similar to the equation for Drude resistivity but note that the
scattering rate τ−1d is the electron-electron scattering rate between the two layers
4
and not electron-impurity scattering, as in the Drude model. Therefore, from this
relation, the measured drag voltage, Vd, is directly proportional to the interlayer
electronic scattering rate.
1.2 Infrastructure
There are three distinct and equally important aspects of making an electron drag
measurement: epitaxial growth, sample processing, and low-noise cryogenic measure-
ment. The drag bilayer heterostructure for this group is currently grown through
molecular beam epitaxy (MBE) by Loren Pfeiffer and Ken West at Bell Labs / Lu-
cent Technologies. There are excellent texts written on the topic of cryogenic mea-
surement, including vacuum technology [6], cryogenic techniques [7], and low-noise
measurement [8]. However, little is written on the topics of small scale but highly
functional cleanrooms for academic environments or on processing for electron drag
samples. The aim of this dissertation is to address these two topics.
Chapter 2 discusses a fundamental prerequisite to the processing of samples; the
specification, design, and construction of a cleanroom. Specification of the cleanroom
requirements are determined based on the needs of an electron drag process, however
almost all of the requirements could be generally applied to a wide range of semi-
conductor processes. From these specifications, the room design process is explained
including the room structure, air handling, and services. Last, the cleanroom con-
struction process is described, since the methods which must be used differ in some
respects from the methods used in a more typical construction project.
Chapter 3 examines the bilayer system, itself, and describes the process of making
an electron drag sample from wafer containing epitaxially-grown GaAs/AlxGa1−xAs
5
heterostructures. The structure of the bilayer sample is described and the scheme al-
lowing independent contact to the two layers is explained. Then, the sample process is
followed, from an original MBE-grown wafer through wiring of the completed sample.
A brief scheduling summary of the entire process is presented at the conclusion.
6
CHAPTER 2
CLEANROOM
This chapter presents a method for specifying, designing, and constructing a low-
cost, highly functional cleanroom facility in a space that was not designed to provide
the necessary supporting services. The cleanroom is an essential tool for process-
ing electron drag samples or almost any modern semiconductor device. The cost
of constructing a cleanroom with the full capabilities necessary can be prohibitively
expensive, however, as well as a daunting undertaking for an academic researcher
without the benefit of training in the design and construction of a microelectronics
processing cleanroom. Designing a cleanroom to function in a space ill suited for the
purpose only compounds these difficulties.
The cleanroom shown in figure 2.1 was built in Smith Laboratory, a 50 year old
building on the Ohio State University campus, to meet or exceed Class 100 cleanroom
conditions for a cost of $125,000, excluding the furniture and processing equipment
shown inside. This was accomplished using a combination of standard tools and
techniques used in cleanroom construction with more unusual methods and materials.
This approach lead to significantly reduced costs. Had the project been completed
by an independent cleanroom contractor, the total expense would have increased by
a factor of 4, based on one contractor’s proposal.
7
Fig
ure
2.1:
Com
ple
ted,op
erat
ional
clea
nro
om.
The
yellow
hue
isdue
tofilt
ers
over
the
ligh
tsan
dw
indow
sw
hic
hpro
tect
the
phot
osen
siti
vere
sist
suse
din
pro
cess
ing
from
unin
tenti
onal
expos
ure
.T
his
phot
ogra
ph
was
take
non
May
2,20
03.
8
The goal of this project was ultimately achieved but several pitfalls were encoun-
tered along the way which impeded progress. With the benefit of hindsight, this
chapter presents an outlined description of a slightly modified approach that avoids
these difficulties. Where instructive, some of the original courses of action are de-
scribed.
The following four sections of the chapter describe the realization of a cleanroom,
from the determination of necessary features to testing of the completed room. Section
2.1 outlines the design goals. This includes a determination of the conditions and
support services necessary for the sample processing tasks to be completed. The
cleanroom is then designed from these specifications in section 2.2. Section 2.3 then
describes the procedures used for clean construction, which differ from those used in
more traditional construction. Finally, section 2.4 presents a measurement indicating
the cleanliness of the completed room. In addition, a set of drawings describing
various aspects of the project can be found in appendix B, which are referenced
throughout the chapter as an aid to the reader.
2.1 Objectives
Before pursuing an actual design for the cleanroom, the requirements for the par-
ticular process to be used must be defined and quantified. The most basic determina-
tion to be made is how much space is required for processing equipment and personnel.
Ideally, space for some future expansion would also be included. This assessment will
likely change somewhat as the design evolves but establishing a minimum floor area
facilitates later design calculations. For the cleanroom under consideration, here, the
area is limited by the dimensions of room 4029 Smith Laboratory (4029) that houses
9
the cleanroom and a large air duct in the room that cannot be relocated. This limits
the dimensions of the main room to roughly 10’× 25’. Also, the ceiling height must
be 8’, at a minimum, to enable an evaporator bell jar to raise fully.
2.1.1 Air
Of all the aspects of the cleanroom, the air properties are by far the most impor-
tant. Air cleanliness is defined by classes, which establish the maximum permissible
volumetric density of particles at a range of sizes. This limitation on particle density
is what establishes a room as “clean.” To quantify the degree of cleanroom cleanli-
ness, there are literally hundreds of standards used around the world. The one most
commonly recognized, however, is the Federal Standard 209.
This standard was developed by a research team at Sandia Corporation, now
known as Sandia National Laboratories, who were studying contamination problems
in clean environments for weapons development in 1959. Willis Whitfield and others
determined that current, state of the art clean spaces were limited due to turbulent
flow in the room. Room contamination was diluted by incoming clean air but it was
not being actively pushed out. Their solution to the problem was the laminar flow
cleanroom. Instead of piping air into the room through diffusers, high efficiency par-
ticulate air (HEPA) filters were used to pass air uniformly over the work surface and
down through vents in the room floor. Using this scheme, particulate density levels
dropped by several orders of magnitude. In 1963, the Sandia team, in conjunction
with other industrial and governmental agencies, formed a group which produced the
first Federal Standard 209. The latest version, Federal Standard 209E (FS209E) [9],
was produced in 1992 and contains the classification scheme shown in table 2.1.
10
Cla
ssLim
its
Cla
ssN
ame
0.1µm
0.2
µm
0.3µm
0.5µm
5µm
Vol
ume
unit
sV
olum
eun
its
Vol
ume
unit
sV
olum
eun
its
Vol
ume
unit
sSI
Eng
lish
m3
ft3
m3
ft3
m3
ft3
m3
ft3
m3
ft3
M1
350
9.91
75.7
2.14
30.9
0.87
510
.00.
283
––
M1.
51
1,24
035
.026
57.
5010
63.
0035
.31.
00–
–M
23,
500
99.1
757
21.4
309
8.75
100
2.83
––
M2.
510
12,4
0035
02,
650
75.0
1,06
030
.035
310
.0–
–M
335
,000
991
7,57
021
43,
090
87.5
1,00
028
.3–
–M
3.5
100
––
26,5
0075
010
,600
300
3,53
010
0–
–M
4–
–75
,700
2,14
030
,900
875
10,0
0028
3–
–M
4.5
1,00
0–
––
––
–35
,300
1,00
024
77.
00M
5–
––
––
–10
0,00
02,
830
618
17.5
M5.
510
,000
––
––
––
353,
000
10,0
002,
470
70.0
M6
––
––
––
1,00
0,00
028
,300
6,18
017
5M
6.5
100,
000
––
––
––
3,53
0,00
010
0,00
024
,700
700
M7
––
––
––
10,0
00,0
0028
3,00
061
,800
1,75
0
Tab
le2.
1:Fed
eral
stan
dar
d20
9Eai
rbor
ne
par
ticu
late
clea
nlines
scl
asse
s.T
his
table
,re
pro
duce
ddir
ectl
yfr
omFS20
9E[9
],es
tablish
esth
elim
its
for
par
ticl
eco
unts
asla
rge
orla
rger
than
the
stat
eddia
met
ers.
For
exam
ple
,in
aC
lass
100
orC
lass
M3.
5cl
eanro
om,
the
max
imum
den
sity
ofpar
ticl
esas
larg
eor
larg
erth
an0.
3µm
is30
0/f
t3or
10,6
00/m
3.
Alt
hou
ghth
isst
andar
dhas
bee
nsu
per
seded
by
the
new
ISO
1464
4-1
clas
sifica
tion
stan
dar
d[1
0],FS20
9Eis
stillth
em
ost
com
mon
lyuse
dm
ethod
for
des
crib
ing
clea
nro
omcl
eanlines
s[1
1].
11
The size distribution of particles shown in the table does not correspond to any
particular situation but, within the range shown, smaller particles are generally more
numerous than larger ones, which are more easily filtered and drop out of the air
more readily. The concentration limits are approximately defined according to the
equation CM = 10M(0.5/d)2.2. For a metric class M environment, CM is the limiting
concentration of particles/m3 with a diameter greater than or equal to d microns.
This classification scheme was superseded by a new classification developed by the
International Standards Organization, ISO 14644-1, in 2001 [10]. The ISO 14644-1
classification can be described by an exponential relationship similar to the one defined
in FS209E; CN = 10N(0.1/D)2.08. Here, for an ISO class N environment, CN is
the limiting concentration of particles/m3 with diameter greater than or equal to D
microns. Although ISO 14644-1 has officially replaced FS209E, the latter is still the
most commonly used standard worldwide and its nomenclature will be used in this
chapter.
The class requirement necessary depends on the process to be run. The typi-
cal guideline used in wafer fabrication cleanrooms is that the maximum size particle
that can be tolerated in a process is one tenth the width of the smallest critical fea-
ture [12]. Contamination sources are examined from air, gases, water, and chemicals
used in processing, considering the integrated time of exposure to each of these at
various stages of processing. For a large scale process, a balance can be determined
between the cost of building and maintaining a clean facility of a given class ver-
sus total process yield. As lithographic critical dimensions continue to shrink, now
nearing the 100 nm level in production, the requirements for cleanliness have scaled
correspondingly. For a pure research facility, however, the requirements are somewhat
12
different and less directly quantifiable. Typically, the cleanest environment that can
be produced is specified, to minimize the effect of impurities on measured results. For
the electron drag samples, for instance, the smallest critical dimension is only 40 µm.
However, defects in the active region of the sample can drastically effect transport
measurements. Also, since one man-hour week is required to produce a single sample,
the cost of a contamination defect can be substantial.
The target level of cleanliness chosen for the 4029 cleanroom is class 100. To
obtain an environment cleaner than this would require the use of a raised, perforated,
platform floor to establish a complete, downward unidirectional flow pattern in the
room. However, the vertical space in room 4029 is insufficient to implement such a
design, and the associated cost of construction, if the vertical space was available,
would scale by roughly a factor of 1.3 - 2.5. Class 100 clean space is only required for
the most critical aspects of processing, namely lithography and wet processing, and
the specification in the rest of the room can be relaxed to class 1,000. In addition, an
anteroom is required to transition from the outside world into the clean space without
introducing a burst of particles into the cleanroom. Maintaining this transition space
one class below the cleanroom space that it opens into is sufficient to filter particles
brought into the ’airlock’ in a reasonable period of time. Therefore, the anteroom is
held at class 10,000 and opens into the class 1,000 space in the cleanroom.
In addition to the cleanliness of the air, the temperature and humidity have a
significant affect on the processes to be run in the cleanroom. Consider the graph
shown in figure 2.2, which indicates the as-spun thickness of a photoresist over a
range of temperature and relative humidity as measured by Mark Wirzbicki of Rohm
and Haas [13]. The resist used for this test, Rohm and Haas Megaposit SPR510-A
13
%R
H
4041424344454647484950
64 65 66 67 68 69 70 71 72
Temperature °F
Thickness (Å) <= 10675.0 <= 10700.0
<= 10725.0 <= 10750.0
<= 10775.0 <= 10800.0
<= 10825.0 <= 10850.0
<= 10875.0 <= 10900.0
> 10900.0
Figure 2.2: Environmental influence on the as-spun thickness of a diazo-based posi-tive photoresist∗ [13]. The resist tested here is Rohm and Haas Megaposit SPR510-A,which is functionally and compositionally similar to Rohm and Haas MicropositS1811, used in sample processing. In the target parameter space of 68 ± 1 F and45 ± 5% relative humidity, the resist thickness varies by only 2%. This degree ofcontrol is especially necessary for back-side sample processing, as discussed in section3.2.7. As can be seen from the contour plot, the resist thickness is much more sensi-tive to moisture content in the air than it is to the room temperature over the rangeof environmental conditions maintained in the cleanroom. However, precise roomtemperature control is also important. The anisotropic etchants used in processinghave an etch rate ∝ e−1/T and require a stable, known temperature to produce re-peatable, timed etches. Therefore, both temperature and humidity must be carefullycontrolled.
Table 2.2: Requirements for water used in the electronics and semiconductorindustry∗ [15]. The water types shown, except E-4, are used directly for processingsemiconductor components. Type E-4 water is used in the production of electronicgrade chemicals and indicates the purity degradation of cleaner grades after contin-uous exposure to the atmosphere in storage. The limit of pure water resistivity at25C is 18.18 MΩ-cm. Note that the units µg/L and ppb are equivalent.
∗Reprinted, with permission, from the Annual Book of ASTM Standards, copyright ASTM Inter-national, 100 Barr Harbor Drive, West Conshohocken, PA 19428.
18
would be required in a day of processing. Ideally, clean water would be available
in both hoods but implementation of this would require the installation of a recir-
culating water system running throughout the room which, in turn, presents many
design difficulties. Because of the design complications involved with this decision,
the choice of where to install clean water dispensing ports is deferred to the design
discussion in section 2.2.
2.1.3 Services
The last elements that must be specified relate to the other support services for
the equipment to be installed in the room. This includes dry nitrogen, used for the
nitrogen blow guns in the two hoods and for the mask aligner and rapid thermal
annealer (RTA), as well as forming gas for the RTA, and compressed air to actuate
mechanical systems throughout the cleanroom. Additionally, low level vacuum is used
for mechanical purposes on the mask aligner, spinner, and RTA and cooling water is
required for the RTA, thermal evaporator, and cryopump compressor for the thermal
evaporator. Lastly, electrical power must be distributed as needed both inside and
outside of the cleanroom. A comprehensive list outlining all required services for all
equipment in the cleanroom is helpful when deciding the layout of piping and conduit
in the room.
For the gaseous services, cleanliness concerns must be addressed and specifications
made accordingly. The dry nitrogen used should meet all of the specifications for clean
air in the room as well as additional constraints relating to gaseous composition. As
the name implies, the gas should be “dry,” meaning that the contained water content
is low, and it should also be free of hydrocarbons and oxygen. Water and oxygen
19
content present similar problems when nitrogen is used for the RTA. Oxidation at
the surface of the GaAs is known to degrade contact performance significantly and
elevated temperatures exacerbate this effect [16]. Carbon contamination, introduced
through hydrocarbons, also dopes GaAs and should be eliminated as much as possible.
As to the levels acceptable for processing GaAs, estimates vary. However, levels of a
few ppm of oxygen during annealing should be acceptable [17]. Similar, if not more
restrictive specifications exist for forming gas. Compressed air is not brought into
intentional contact with the sample, but some leakage from mechanical actuators and
stable tables will enter the cleanroom. Therefore, the air should be filtered of oil and
particulate content, so that its introduction into the cleanroom does not adversely
affect the process.
The total power needed for this 250 ft2 cleanroom exceeds the requirements of
a typical research and development laboratory space of similar size by a factor of
three [18]. The original electrical capability of the room is insufficient to support
the required air handling equipment and new power cabling must be installed to
provide the necessary current. The exact requirements are difficult to determine at
the specification stage without a design for the necessary air handling equipment.
Therefore, the total power requirement is decided recursively through the design
process. However, the power needed for processing equipment can be determined.
An effective way to sum the necessary power requirements is to break the process-
ing equipment into two categories: constant and intermittent loads. The constant
category establishes a base requirement for the absolute minimum power necessary
to operate the room. On final analysis, the average intermittent power requirements
for the 4029 cleanroom are small in comparison but, at this stage of estimation, quite
20
the opposite is true. When summing power requirements from equipment, “name
plate amps” will generally not provide a good estimate of the actual power required
on a continuous basis. Likewise, specifying that 10 outlets in the room have 20 A
service is not the same as specifying that 200A of current carrying capability must
be dedicated to duplex outlets in the room. Also, some equipment must be treated
separately from calculated averages. A good example for the 4029 cleanroom is the
RTA. On average, the power requirements for this unit are completely negligible for
ongoing load calculations. However, for approximately 3 seconds as the unit ramps
power to the arc lamp, the power demand for the RTA is greater than all the other
power demands for the cleanroom combined.
Summing power for the mask aligner, spinner, dry roughing pump, cryopump
compressor, and various smaller items such as diaphragm pumps and hotplates gives
an estimate of 4.0 kW of continuous load for the room equipment. Note that this
estimate does not consider the ’overhead’ load necessary to support the cleanroom
and this equipment, which will require many times this quantity of power. The two
predominant sources of intermittent load are the RTA, which draws almost 42 kW
of power at peak usage, and the thermal evaporator power supply which can require
as much as 17 kW of power. These estimates will provide a basis for a true power
requirement estimation later in the design process.
2.2 Design
Once the requirements are specified, the cleanroom design can begin. There are
published sources available to aid in cleanroom design but, for the most part, they are
geared towards the construction of large-scale, industrial, stand-alone enclosures and
21
not small, academic, built-in rooms relying on obsolete building infrastructure [19,
20, 21, 11]. Often, the solutions which apply to a 5,000 ft2 room are impractical or
economically unreasonable to implement in a 250 ft2 built-in space.
Typical small area clean solutions consist of either a laminar flow hood or a
portable, soft wall unit in which a set of filter units are held up by a scaffold and
the area under the filters is isolated with a surrounding, vinyl curtain. Slightly larger
versions replace the curtain with a hard, acrylic wall, although the operating principle
is identical to the one used by the soft wall unit. These “single-pass” solutions are
adequate for a multitude of applications but they do not address concerns of tem-
perature and humidity control and, most importantly, no vented space is provided.
Additionally, the practically attainable cleanliness with these solutions is limited to
around class 100, provided that the final air filters are replaced annually for a typical
ongoing airborne particulate load.
Another solution is to hire a cleanroom contractor to design a room to the given
specifications. However, typical cleanroom designs are expensive, and the provided
solution will likely reflect this fact.
The last alternative is to design a room to specification independently. One of the
greatest difficulties with this approach is the lack of publicly accessible, published,
specific information that is applicable to such an endeavor. The methods, tools, and
techniques used by companies are usually considered proprietary and not publicly
disclosed. The best sources of information, aside from discussion with professionals,
are often found on the Internet. Unfortunately, due to the dynamic nature of that
medium, many sites accessed during the design process of the 4029 cleanroom are
now nonexistent.
22
The goal of this section is to follow the design process used for the 4029 cleanroom.
This course was taken through trial and error, combined with an investigation of how
existing techniques function, i.e. what about a given solution makes it applicable for
a clean environment. Then, the process was either duplicated, where necessary, or
simpler, less expensive, or more readily available alternatives were used, instead.
David Sarge, the Facilities Manager for the Pennsylvania State University’s Elec-
tronic Materials and Processing Research Laboratory, was hired as a consultant to
establish a base design for the room. A basic layout was proposed, in which fan filter
units (FFUs) would be installed in a suspended ceiling and the room walls would
be used as air return plenums. Additionally, estimates of the required cooling power
and airflow were determined. This provided a base design from which the final 4029
cleanroom eventually materialized.
2.2.1 Layout
There are three main types of flow patterns in cleanrooms. Non-unidirectional
flow or turbulent flow, as it is sometimes called, cleans a room through dilution of
airborne contaminants. Clean air is introduced from discrete sources, mixes with air
in the room, and reduces the average contamination level as the combined air is vented
from the clean space. In a pure unidirectional flow cleanroom, clean air is introduced
uniformly from one entire room boundary surface, usually the ceiling, and vented
from the opposite room surface. Contamination is pushed out of the room actively
as incoming air collects particles and drives them directly towards the room vent like
a piston. The flow is considered to be unidirectional as long as an airstream from
the originating boundary maintains a velocity within ± 20% of the average velocity
23
at the originating boundary. This type used to be called laminar flow but that name
is falling out of favor since flow velocity has become recognized as a better indicator
of cleanroom performance than the Reynolds number. The third type is a mixed
flow, in which aspects of both unidirectional and non-unidirectional flow are present.
The definition of where the boundary between mixed flow and non-unidirectional flow
exists is subject to some debate. However, this issue will not be addressed, here.
In order to meet the requirements of a class 100 environment, the room cross-
section shown in figure 2.3 was chosen. Two cross sections are shown, illustrating
the intended flow pattern in two different sections of the room. In the ceiling, FFUs
draw air in from plenum space above the ceiling and force it through high efficiency
particulate air (HEPA) filters in the plane of the ceiling. These HEPA filters remove
99.97% of the particles in the air that are 0.3 µm or larger. The clean air is expelled
uniformly from the face of the FFU and flows down through the room. At the floor
level, the air is directed through vents to the wall space, where the air returns to the
ceiling plenum, mixes with new, processed air from the supply ductwork, and is drawn
through the FFUs, again. In the left half of the figure, where no hood is present, the
airflow is unidirectional over much of the vertical distance from the ceiling to the
floor and then is redirected near the floor. This unidirectional airflow is an important
aspect of maintaining the level of cleanliness in the room. On the right, in the vicinity
of a hood, the majority of air is drawn into the hood and exhausted at the roof. Some
of the air, however, does flow down the face of the hood and is redirected up the wall
as before.
This design was chosen because a narrow room provides unidirectional flow over
much of the vertical space from the ceiling towards the floor. A floor-ventilated
24
Figure 2.3: Airflow in cleanroom cross-section. The two diagram halves shown indi-cate the flow in different sections of the room. On the left, where no hood is present,clean air delivered by the HEPA fan filter unit (FFU) flows downward uniformly,pushing dirty air down and over to the wall vents near the floor like a piston of air.The unidirectional flow pattern shown is necessary to maintain class 100 or bettercleanroom conditions. The contaminated air is then pushed up the wall space intothe ceiling, where it mixes with new supply air and is drawn through the FFU, again.The same principle applies to the schematic on the right, except that the majorityof the locally-produced air is drawn into the hood and vented to the roof. Note thatthe flow patterns shown are for illustration, only.
25
cleanroom would be ideal for this purpose, but such a design could not be constructed
in the available vertical space in room 4029. In a large, “ballroom” type area, the
design shown would not be as effective, since air released in the center of the room
would have to travel a significant distance horizontally, carrying contamination, before
venting to the wall space. For this same reason, an alternate unidirectional flow
scheme in which clean air is introduced at one end of the long cleanroom and drawn
out at the other has significant drawbacks. The greatest of these is that contamination
introduced upstream in the room travels the length of the room before exhausting to
the plenum. For these reasons, the design shown was chosen as the best solution that
was workable in the available space.
Cleanrooms, in general, are often described in terms of either air changes per hour
(ACH) or downward airflow velocity in the room as an indication of room cleanliness.
The more appropriate descriptor depends on the airflow pattern in the room. If the
flow is truly unidirectional, then air velocity is the better indicator. A very tall uni-
directional flow room, for instance, may not have a high ACH value even though the
flow velocity, which drives the contaminants from the room, could be high. Higher
velocity is not necessarily better, since the flow pattern will eventually become tur-
bulent at high enough flow, in which case local contamination could move in the
opposite direction of the average flow velocity. For the values considered here, how-
ever, higher flow velocity generally indicates more efficient removal of contamination
from the room. In a non-unidirectional flow room, however, airflow velocity is a poor
descriptor due to the lack of a well defined flow pattern in the room. In this case, the
older, ACH designation is more appropriate. In a mixed flow room, both measures
can be used.
26
Class Airflow Average Velocity (ft/min) Air Changes per hour
Air Changes per hour =average airflow velocity (over the whole supply ceiling)
× room area× 60 min/hrroom volume
N = non-unidirectional flow; M = mixed flow; U = unidirectional flow
Table 2.3: Air velocities in cleanrooms∗ [22]. The values shown here are a recom-mendation provided by the IEST that is used as a guideline in the semiconductorindustry. Note the wide range of velocities and ACH values for each class. Moreclean air will be required to meet a class specification as the rate of contaminationgeneration in the room increases due to personnel and equipment.
∗Copyright 1998 IEST. IEST-RP-CC021.1 Considerations in Cleanroom Design. Used by permis-sion, www.iest.org.
27
Table 2.3 is a reference published by the Institute of Environmental Sciences &
Technology (IEST) that is used by the semiconductor industry for determining an
approximate level of airflow to meet class specifications [22]. Complete FFU coverage
of the 4029 cleanroom is unnecessary to meet class 100 conditions and not all of the
room needs to be class 100. Since FFUs will be used in the ceiling, the average cubic
feet per minute (CFM) of air produced per FFU determines the average flow velocity
with masonry fasteners. This provides a mounting point for rod and turnbuckle
supports for the ceiling grid as well as an anchor for piping and electrical conduit
to run along the ceiling. This mounting scheme removes the need to drill into the
concrete ceiling to mount services later in construction or minor expansion, reducing
the production of concrete dust.
Since each FFU weighs 63 lbs, the ceiling grid must be much stronger than typical
grid used in drop ceilings to support the weight. The grid should also utilize an
internal gasket system, so that the weight of the FFUs and lights will compress the
gasket and prevent leakage of contamination from the ceiling plenum space down
into the clean space. In the class 100 area of the cleanroom, the required density of
FFUs leaves no room for standard-size, gasket sealed 2’× 4’ lighting fixtures to be
installed. Therefore, there must be some way to mount low-profile lighting fixtures
33
Figure 2.4: Cleanroom shell before interior panel and ceiling installation. Severalof the structural design aspects of the cleanroom are shown here. C-stud framingsupports the plywood exterior panels shown on the left, exterior wall. The whitewalls and ceiling have been painted with epoxy to seal the block and concrete. Theduct distribution system is partially visible at the ceiling, and the two exhaust ductsare ready to be connected to the hoods after the room interior is installed. All pipingand electrical conduit run flush to either the Unistrut bars attached to the ceiling,or to the C-stud above the height of the interior ceiling. The epoxy floor shown herewas later replaced with vinyl sheet flooring. This photograph was taken in June of2000 by Tom Gramila.
34
on the grid, itself. There are grid systems designed to incorporate lights directly in
the grid members but such systems could not be easily used due to space restrictions
their normal clothes, will be reasonably comfortable. The 45% relative humidity
level is chosen as an optimal compromise between an extremely dry environment, in
which electrostatic discharge can damage samples and equipment, and a high humid-
ity environment, which is more corrosive to metals, promotes bacterial growth, and
degrades photoresist performance. To stay within these design parameters, the total
heat transfer to the cleanroom air from both the cleanroom and the incoming makeup
air must be considered.
First, the average sensible heat load applied to the system from the room must be
determined. Applied to the context of HVAC systems, the sensible heat is the heat
energy added to the air which raises its temperature, as opposed to latent heat, which
changes the energy in the air through water content, without altering temperature.
The total sensible heat from the room is the sum of continuous loads and the time-
average of intermittent loads. Continuous loads are considered first, because they
dominate the sensible heat load on the cooling system. Since the English system of
units is used in the United States HVAC industry, British Thermal Units pre hour
(BTUh), are used to describe power, where 3,414 BTUh = 1 kW.
FFUs: 14,900 BTUh The 14 FFUs used in the cleanroom account for almost half
of the total sensible heat load.
38
Lighting: 7,000 BTUh A common formula used to estimate lighting heat load is
to sum the wattage of all the bulbs used and then to add 25% of this value for
ballast heat.
Air handler motor: 5,100 BTUh Use the horsepower of the motor to determine
the heat generation. If the motor horsepower is unknown, a reasonable estimate
is that 1 horsepower will be required for every 1,000 CFM of air run through a
fan [21].
3 people: 1,000 BTUh 3 people approaches the upper limit on the expected room
occupancy, but the heat load estimation assumes moderate to low activity.
Equipment: 8,000 BTUh This estimate for both the continuous and intermittent
equipment loads is the most difficult to make. One estimation method is to
follow the processing procedure in terms of heat generation. Consider all the
equipment that will be used and determine an average heat load based on that
process.
Total: 36,000 BTUh Power for cooling units is typically described in tons, where
12,000 BTUh = 1 ton of cooling power. Therefore, approximately 3.0 tons of
sensible heat load are generated in the cleanroom on average.
The makeup air directed into the room must maintain the minimum face velocity
required for the hoods as well as keep the room pressurized above ambient air outside
the cleanroom. The minimum safe velocity at the face of the hoods, to ensure that
chemical vapors do not flow back into the cleanroom space, is 80 ft/min [14]. So,
enough new air must be introduced into the room to maintain this value. There
39
are two hoods in the room, each with a length of 6’, and the minimum sash height
required for working purposes is 15”. From these dimensions, a volume flow rate of
6’× 1.25’× 80 ft/min = 600CFM of makeup air is required for each hood.
Room pressurization is necessary so that clean air leaks out of the cleanroom space
and contaminated air is not drawn in. Typically, 0.02 - 0.05 inches of water column
(wc) of room pressurization is sufficient. In order to keep the room pressurized,
supplementary air must be delivered to the room. This pressure can be obtained by
introducing an additional flow of makeup air at a rate of 2 air changes per hour [21].
This quantity of makeup air will be adequate as long as there are no significant leaks
in the pressure barrier of the room. Seal all penetrations into the space with silicone
caulk and install gaskets around all four edges of doors, including lock-down gaskets at
the bottom edge. The room volume considered for the calculation of supplementary
air should be the total volume of the pressurized space and not just the volume
of the room interior. From the main room exterior dimensions of 24’× 11’× 11’ =
2,900 ft2 in addition to the anteroom exterior volume of 4’× 9’× 9’ = 300 ft2, the total
enclosed volume is 3,200 ft2. Therefore, at 2 air changes per hour, 100 CFM of air
must be added to pressurize the room. In total, then, at least 600CFM + 600CFM
+ 100CFM = 1,300CFM of makeup air must be drawn into the room.
Because the cleanroom is housed in a 50 year old building that was not designed
to support such a facility, processing the required air for the cleanroom is particularly
challenging. Because there is no direct source of outside air available in room 4029 and
no simple way to install ductwork to the outside air, the building HVAC processed
air must be used. The greatest advantage of this approach is that, because the
air is already cooled to a temperature of 52 F, less cooling power is required for
40
the cleanroom than would otherwise be necessary. However, there are substantial
disadvantages associated with this scheme.
The building HVAC system is designed to make the building environment comfort-
able and to provide some air for small hood exhaust systems. From the 10” diameter
circular duct supplying room 4029, it was designed to flow around 400CFM of air,
which creates a pressure loss of approximately 0.1”wc per 100’ of straight duct. For
the 1,300CFM needed for the cleanroom, however, this pressure loss increases by
a factor of almost 8. By drawing this much air out of the supply, the pressure in
the ductwork leading up to the cleanroom fan drops below the pressure of ambient
air. At this negative pressure, leakage points in the duct allow additional contam-
ination from the surrounding air to be drawn in which must be filtered out before
allowing the air into the cleanroom. This requires more filtration, more fan power,
and, therefore, more cooling capacity than would otherwise be necessary with positive
pressure airflow. To compound this issue, the cleanroom is completely reliant on the
function of the building HVAC system. If the building system fails in some respect,
the cleanroom could lose control of temperature, humidity, or even room pressuriza-
tion and hood face velocity. For these reasons, extreme caution should be exercised
when designing a cleanroom to rely on pre-conditioned air. At best, the cleanroom
functionality will be as reliable as the the air handling equipment which supplies it.
The parameters of temperature and humidity are intimately related and should
be considered concurrently in cleanroom design. All of the air entering the cleanroom
enclosure issues directly from a blower which mixes two air streams. One stream is
the makeup air which, in the summer, enters the room at 52 F and 100% relative
humidity and the other stream is the air which is drawn from the cleanroom in a
41
recirculation duct and passed through the cooling unit. A schematic showing the
ductwork layout can be seen in figure B.6. The combined airstream must have the
capacity to absorb the sensible heat load and have a dewpoint in the acceptable
range for the cleanroom while still providing enough air for the hood exhaust and
room pressurization.
The dewpoint specification for the air entering the cleanroom can be determined
directly from the dewpoint specification inside the cleanroom. In the summer, a
large fraction of cooling power must be devoted to removing water introduced from
the makeup air duct. Water produced inside the cleanroom through wet processing
and user respiration is negligible in comparison and is safely ignored, here. At the
limit of acceptable water content of 68 F, 50% relative humidity, assuming that the
temperature is held on target, the cleanroom air has a dewpoint of 48.7 F. All of
the active water removal occurs at the cooling coil in the recirculation duct from the
cleanroom. Air reaching this coil cools below its dewpoint to the coil exit temperature,
condensing excess water onto the cooling coil. The air then exits the coil saturated
at the new, lower temperature. Because both the makeup air and the coil exiting air
will be saturated with water and at different temperatures, some fog will be produced
when they combine at the mixing fan, as can be determined from a psychrometric
chart. To account for the corresponding slight change in temperature and humidity,
the mixed flow entering the cleanroom has a maximum dewpoint specification of
48 F.
However, cooling the room with saturated air at 48 F will not meet both the
temperature and humidity specifications. The room heat load must be balanced by
42
the heat capacity of this mixed, incoming air, according to
qroom = mCp(troom − tmixed) =f
vs
Cp(troom − tmixed),
where q is the sensible heat load, m the mass, Cp the specific heat at constant pressure,
f the volumetric flow and vs the specific volume of the mixed air, defined as the inverse
density. The required flow of mixed air in cubic feet per minute (CFM) can then be
determined from
36, 000 BTUh =
(f 60 min
hr
12.9 ft3
lb
) (0.245
BTU
lb F
)(68 F− 48 F) ,
where BTUh are British thermal units per hour. Solving for the flow, f = 1,580 CFM,
which means that flow across the cooling coil is only 1,580 CFM - 1,300CFM =
280CFM. In order to sustain this small of a flow and meet the necessary cooling
requirements, the exit temperature from the cooling coil would have to be 29 F,
in which case ice would form on the cooling coil, blocking the flow and potentially
damaging the cooling unit. The operational specifications must be defined to prevent
the coil from freezing. Therefore, a heater must be added after the cooling coil, so
that the flow across the coil can be increased.
To meet both the cooling and dehumidifying requirements, a 3 ton Liebert Mini-
Mate II cooling system was chosen with an integral heater after the cooling coil. This
is a common system used for computer room cooling applications in which the load
is almost entirely sensible heat. The maximum sustainable flow through the unit
is approximately 1,250CFM of air, as determined by the maximum face velocity of
the cooling coil. Although the unit nominally provides 3 tons of cooling power, the
actual cooling power provided depends upon the volume flow rate through the unit
and the temperature and humidity of the incoming air before it crosses the cooling
43
coil in the unit. For example, at a flow rate of 1,250 CFM of air at 50% RH, the total
cooling power increases with temperature at approximately 1.5%/F, reaching 3 tons
of cooling power with an incoming air temperature of 78 F.
The sensible heat ratio (SHR) also changes with the incoming air. Defined as
SHR = Qs/QT , where Qs is the sensible heat cooling capability and QT is the total
cooling capacity of the unit, the SHR increases with flow across the cooling coil. As
a result, even though the total cooling capacity is higher at higher airflow, the latent
cooling capacity drops. To maximize the amount of water removed from the air, the
flow must be drastically reduced below the cooling unit’s normal operating airflow
range.
Reduction of the flow velocity across the cooling coil means that the air exiting
the coil is colder and drier than it would be at higher flow rates. However, the mass
flow of air is, of course, reduced as well. The optimal flow for maximum water removal
depends on the specific relationship between latent heat removal and airflow across the
coil. This relationship is defined by the coil contact factor β = (hin−hout)/(hin−hc),
where h represents the enthalpy of the air. Air enters the coil with enthalpy hin
and leaves with hout. The coil, itself, is at the temperature of saturated air with
enthalpy hc. Perfect coil efficiency at removing water would be indicated with a
contact factor of 1, in the limit of infinite coil area and zero flow rate, in which case
the air temperature leaving the coil matches the temperature of the coil itself.
For the cooling unit used for the 4029 cleanroom, the highest water removal rate
occurs when the airflow is at the lowest sustainable level which does not freeze the
cooling coil. As the flow rate decreases, less heat is removed from the coil, as described
earlier, and the equilibrium temperature of the coil decreases as well. In practice, the
44
lowest stable exit temperature from the Liebert unit is approximately 38 F. Below
this temperature, fluctuations cause the coil temperature to drop below the water
freezing point of 32 F at which point ice forms on the coil. This, in turn, reduces the
flow further and lowers the coil temperature. The positive feedback continues until
the coil is completely blocked with ice and the system fails.
The required airflow through the coil can be determined if a specific exit temper-
ature is assumed. This calculation assumes that 1,300CFM of incoming makeup air
is introduced saturated at 52 F and combines with air exiting the coil saturated at
38 F to produce cleanroom air at 68 F and 50% RH. This condition corresponds to
a set of measured parameters for the room, although the degree of saturation of the
air exiting the cooling coil may be less than 100%, due to intentional reheating of the
air. This point will be addressed later in the discussion. It is worth noting that if the
exit air temperature were higher, e.g. 40 F versus 38 F, the required airflow would
be correspondingly larger, as would be the required cooling power.
The specific humidity of the cleanroom air, i.e. the ratio g of water mass to dry
air mass in a volume of air, can be described by the average of the specific humidities
of the makeup air and the exiting coil air, weighted by their respective volumetric
flows, f .
gr =fmgm + fcgc
fm + fc
where the subscripts r, m, and c refer to the cleanroom return air, makeup air, and
coil exit air, respectively. Therefore,
fc = fmgm − gr
gr − gc
45
which can be solved by using the g values for the stated conditions as determined
from a psychrometric chart.
fc = 1, 300 CFM0.00823− 0.00726
0.00726− 0.00480= 510 CFM
At this low of a flow across the cooling coil, the total capacity of the cooling unit is
drastically reduced. The total cooling power can be determined from the change in
the enthalpy of this air crossing the coil.
qremoved = mc(hr − hc) =fc
vs
(hr − hc)
qremoved =510 CFM 60 min
hr
12.6 ft3
lb
(24.2BTU
lb− 14.3
BTU
lb) = 24, 000 BTUh = 2.0 tons
So, the nominally 3 tons of cooling for the coil is reduced to only 2 tons under these
operating conditions. Importantly, no reheating power was taken into account in
this calculation, which may be required to meet the temperature constraint in the
room, rather than the humidity requirement. This is accomplished by reheating the
air leaving the coil which does not affect its moisture content but does yield a lower
relative humidity. However, reheating can exacerbate the difficulty introduced by the
reduced cooling capacity, thereby also influencing the choice of an appropriate cooling
unit.
If the flow exiting the coil and heater has a lower water content, then a lower flow
is required to maintain the measured humidity content in the room. As discussed
earlier, as the flow through the coil decreases, the total cooling capacity drops as
well. This illustration demonstrates the crux of the difficulty in maintaining the
cleanroom environment within specification during the summer months. Maintaining
temperature and humidity within tight tolerances with high exhaust flows requires
46
a tremendous amount of energy which must be applied appropriately to achieve the
desired results.
Three solutions can be used to combat this problem. The first and most direct
approach is to include a desiccant wheel in the incoming air duct, a technique widely
used in the semiconductor industry. A honeycombed wheel is slowly rotated through
the incoming air where it absorbs water into a silica gel desiccant material. As the
wheel rotates out of the air stream, hot exhaust air is passed through the wheel,
removing the water and regenerating the desiccant to pass through the incoming air,
again. The total enthalpy of the air stays fairly constant, but latent heat energy is
exchanged for sensible heat energy and the air emerges from the wheel hotter but with
less water content. This method is required for most cleanrooms that must maintain
year-round humidity levels below 30%. For smaller academic cleanrooms, however,
this solution is usually impractical. A large quantity of hot, dry air must be available
which can be exhausted out of the room. Additionally, the financial and space costs
of implementing this solution can be substantial.
A second approach is to use a dedicated dehumidification coil in the incoming air
duct. This solution can be nearly as effective as a desiccant wheel without the same
degree of overhead associated with installation and maintenance. However, a large
volume of space is still necessary, since the area of the coil face must be large enough
to obtain a high contact factor without increasing the impedance of the airflow to too
great an extent. The cost of this solution is almost invariably lower than the total
cost of installing a desiccant wheel. However, a complete cooling system must still be
installed in parallel with the main cooling system for the room.
47
The third approach is to simply use a larger evaporation coil in the main cooling
unit. With a larger coil face area, more air can be passed at a higher temperature
and dewpoint to maintain the environment in the cleanroom. This approach permits
greater control latitude with more efficient operation without major modification of
the existing design. This approach is, perhaps, the lowest cost, most compact, lowest
impact solution of the three suggested here if it is implemented at the design and
construction phase of a cleanroom project.
The difficulties discussed here are due primarily to the fact that the cooling unit
used was not designed to serve a cleanroom. Problems are likely to be encountered
when using a computer room cooling unit for a tightly controlled environment in which
a large quantity of air is being exhausted from the room. These units are typically not
designed for such applications and their adaptability to the task is limited, even with
considerable modification. In large, industrial cleanrooms, each individual element
of the HVAC system may be specified individually so that each part has a limited,
specific task to accomplish. Small, academic cleanroom environments may benefit
from a similar approach, even if the initial costs are higher. At a minimum, however,
the manufacturer of a cooling unit should be consulted to determine if a specific unit
will meet the requirements for the intended application. Both sensible and latent heat
removal capability at the designed flows should be specified. This important step will
likely prevent costly and frustrating modifications to the cooling and dehumidification
system after the cleanroom is constructed.
Feedback Control
The tight tolerances for temperature and humidity in the cleanroom, as well as the
necessity of maintaining room pressure and hood face velocity, require that several
48
active feedback controls be implemented to regulate these parameters. The details of
each feedback control loop will not be discussed here. However, it should be noted
that extra precautions are necessary to avoid instability due to the short and, often,
closely matched timescales involved with air circulation.
The feedback and control systems for airflow, temperature, and humidity are
shown on the flowchart in figure 2.5. In general, airflow follows the central line down
the flowchart, passing from the building air supply at the top to the roof exhaust
fan at the bottom. There are three main control loops which affect the airflow in
the cleanroom. The first loop controls the amount of makeup air introduced into the
room by sensing airflow with a pitot tube and controlling a baffle at the inlet duct.
Next, water is added to the incoming air as needed according to a humidity sensor in
the cleanroom. Last, the temperature is controlled through a feedback loop between
a temperature sensor in the cleanroom and a heater array mounted after the cooling
coil in the central cooling unit. Each of these loops will be discussed below.
The airflow from the building air supply is monitored using a simple averaging
pitot tube, made from two stainless steel tubes, as shown in figure 2.6. A set of
holes on one tube faces upstream and a matching set of holes in the other tube faces
downstream. The concept behind pitot tube operation is that the air velocity passing
the pitot can be determined from the pressure differential between two measurement
points. A short derivation from the Bernoulli equation illustrates this behavior. Along
any streamline in an incompressible gas flow, the fluid pressure and velocity are related
by
1
2ρV 2 + ρgh + p = C
49
Figure 2.5: HVAC air processing, monitoring, and feedback control flowchart. Ingeneral, the process flow is from the building supply air at the top of the chart to theroof exhaust at the bottom of the chart. Because the system draws more air thanthe building supply duct was designed to provide, the section of ductwork from thebuilding supply air to the blower and mixing box is at a negative pressure relative tothe ambient air outside of the ductwork. The HEPA bank which follows the blowerand mixing box is designed to filter out contamination drawn into the ductwork fromambient air in this negative pressure section of ductwork, rather than overtaxing thefiltering capabilities of the HEPA FFUs. Note that the original Liebert feedbackcontrol system has been replaced by independent feedback loops for temperature,humidity, and airflow control.
50
where the fluid of density ρ has velocity V at pressure p and height h. The acceleration
of gravity is g and C is constant along the streamline. Consider a streamline that
lies exactly on the boundary that separates the flows which pass on either side of
the pitot tube. This streamline reaches the pitot tube at a “stagnation point” where
the flow velocity stops. Comparing this point (s) to a point far upstream (u) on the
streamline,
1
2ρv2
u + ρghu + pu =1
2ρv2
s + ρghs + ps.
If the heights of these two points are the same and the velocity at the stagnation
point is zero, then
1
2ρv2
u + pu = ps
and
vu =
√2(ps − pu)
ρ.
Although the downstream port on the pitot tube is not at a point far upstream, it
does measure the static pressure, or slightly below it, in the airstream. Therefore, a
measurement of the pressure difference between the two tubes is proportional to the
square of the air velocity.
A long section of ductwork is usually needed for installation of a pitot tube. A
rough guideline is that at least 5 duct diameters upstream and 3 duct diameters
downstream of the pitot should be straight and unimpeded to allow the flow to sta-
bilize. Unfortunately, this space is not always available, but by using the averaging
pitot tube geometry shown in figure 2.6, a more stable average of the total flow is
obtained. The absolute value of the measured pressure differential is not important
for this application because the flow will be held at a constant value. The pitot tube
is calibrated by measuring the pressure differential at the required flow of 1,300CFM,
51
Figure 2.6: Averaging pitot tube installed in duct. This pitot tube is used to monitorthe incoming airflow to the cleanroom to provide feedback to the linear actuator shownin figure 2.7. Pairs of holes are drilled into the independent tubes directly opposingone another. One hole faces upstream and the other faces downstream. Hole positionsfrom one duct wall are shown in inches. The pressure difference between the two tubesis proportional to the average of the square of the airflow velocity at the position ofeach pair of holes. This averaging pitot tube provides a more stable indication of theaverage airflow than a single point pitot tube measurement. The averaging geometryis also less susceptible to changes in the incoming flow pattern.
52
Figure 2.7: Linear actuator used to regulate incoming airflow. The pressure differ-ential measured by the averaging pitot tube is monitored by a control circuit whichprovides time-proportional feedback to the linear actuator shown. The lighter imagesindicate full-open and full-closed positions of the control baffle
as determined independently using a hot wire anemometer, and the flow is controlled
to maintain that measured pressure.
Incoming airflow can then be regulated by using a linear actuator controlled baffle,
as shown in figure 2.7. The pressure from the pitot is measured by a low differential
pressure transducer which relays the measurement to a proportional, integral, differ-
ential (PID) feedback process controller. the controller actuates two solid state relays
(SSRs) through time-proportional control which drive the linear actuator shown to
extend or retract, adjusting the flow baffle. This system provides a stable, controlled
airflow into the cleanroom that automatically adjusts for pressure changes in the
supply duct.
The temperature inside the cleanroom can be most accurately controlled by main-
taining a constant level of cooling power and then applying heat to the air to raise
53
its temperature to the required level. The cooling power is provided by the cooling
coil and makeup air, as described earlier, and heat is applied through a heater array
in the main cooling unit. This heater is controlled by a feedback process controller
which takes the temperature as measured by a sensor in the cleanroom as input.
The controller then actuates relays which energize the heater in a time-proportional
manner. The 8 kW heater coil supplied with the Liebert Mini-Mate II is more than
sufficient to provide the needed heat.
Although large quantities of water must be removed from the air during the sum-
mer to maintain the humidity specification, water must be added during the winter
to prevent the cleanroom humidity from dropping below the specified range. In the
winter, the incoming air temperature is still 52 F, but instead of being saturated, the
water content can drop to a quarter or less of the amount supplied in the summer,
depending on the local weather. Typically, for the purposes of calculating water to
be added in the winter, the makeup air is assumed to be completely dry. Under this
assumption, the required water to be added can be calculated as
W =fm
vs
gr
where W is the rate that water needs to be added, fm is the flow rate and vs the
specific volume of the makeup air and gr the specific humidity of the cleanroom air.
Therefore,
W =1, 300 CFM
12.9 ft3
lb
(60
min
hr
)0.00653 = 40
lbs
hr
of water must be added to the incoming air. To meet this demand, a steam-generating
humidifier with a capacity of 44 lbs/hr is used. The water which feeds the humidifier
must be purified so as not to introduce contamination into the incoming air through
54
the steam delivery system. A humidity sensor in the cleanroom provides a signal
to the humidifier which, in turn, heats the water through time proportional control
of SSRs. the steam generated then flows through insulated tubes which terminate
in delivery wands which extend into the supply air duct. The work of installing this
system for the 4029 cleanroom was largely completed by Gokul Gopalakrishnan, later
aided by Yuko Shiroyanagi.
2.2.4 Type E-1 Water
The production of type E-1 clean water occurs in three main stages [15]. The first
stage removes most of the incoming contamination and prepares the water for further
purification. Rough sediment is removed and the water is filtered and deionized to
a water quality level of E-4. The second stage continues purification by removing
organic contaminants and further reducing ionic contamination. The resulting water
purity from this section meets E-3 purity standards and can still be stored in plastic
vessels, as long as appropriate measures are taken to inhibit the growth of biological
contamination. The third and final stage “polishes” the water to the level required
for use in the cleanroom, E-1. Traces of ionized, organic, biological and, finally,
particulate contamination are removed just before use.
Distribution of the clean water through the room requires careful choice of piping
materials and flow patterns to avoid degradation of water quality before it reaches the
point of use. There must be no sections of piping in which the water is not constantly
circulating, and the water must be continually polished due to the inevitable, rapid
degradation of water quality. The piping used must be made of a high purity plastic
such as PVDF, with welded seams. Virgin polyethylene or polypropylene can also be
55
used successfully but more care is required with piping made from these latter mate-
rials because they lack the structural rigidity of PVDF and require more structural
support. An important point to note, here, is that these issues exist because the
water must be distributed to more than one point of use, i.e. both of the two hoods.
However, if clean water can be distributed from a single point of use, the problem of
piping processing water is largely eliminated. After careful consideration, it became
apparent that a recirculating water system would be difficult and expensive to install
and, ultimately, unnecessary for the purposes of the cleanroom.
A relatively simple, effective solution is to purchase a water purification system
designed for a single point of use (POU) and have water available in only one hood.
For the 4029 cleanroom, a Millipore system was chosen consisting of only three com-
Table 2.4: Summary of cleanroom particle count results. These particle counts, shownas number per ft3, were measured by Laboratory Certification Services, Inc. using aMet One 2400 Series optical particle counter on August 2, 2002. Three measurementswere taken at each point at a height of 38” above the floor, except for the two mea-surement points in the ceiling. The averages of these measurements are shown. Themeasurement conditions are “at-rest,” meaning that, at the time of measurement, thefacility was functional but not occupied by personnel working normally. Differencesin cleanliness between at-rest and operational states vary between cleanrooms but,typically, the particle density increases by roughly a factor of ten in the operationalstate, versus the as-built state [11]. Although these measurements are insufficientto formally classify the room cleanliness according to FS209E, an indication of thecleanliness level can be inferred by comparing these measurements with the class lim-its shown in table 2.1. There are strong indications that the levels of cleanliness inall sections of the room have surpassed the originally stated design goals.
71
as official validation that the room has met the designed cleanliness specifications.
However, if the measurements taken are an accurate indication of the average particle
distribution in the room, then the room has achieved a level of cleanliness well above
the design specification.
One probable reason for this is that all air coming into the cleanroom interior
space has passed through two HEPA filters in series, one in the ductwork outside
of the clean space and one in a ceiling-mounted FFU. This combination is more
than 99.99999% efficient at removing particles 0.3 µm and larger, which is better
than almost any single filter available. Ultra Low Penetration Air (ULPA) filters are
99.9995% efficient in removing particles 0.12 µm and larger, and tests have shown
that two HEPA filters in series remove more particles above 0.01 µm than a single
ULPA filter [21]. Therefore, the air being delivered into the cleanroom exterior is
extremely clean.
Continuous monitoring of the clean space ensures that the cleanroom conditions
stay within designed parameters. Cleanliness measurements are difficult to make on
a regular basis without having a particle counter on-site. However, an independent
company can perform spot check measurements on a regular basis to ensure that the
room continues to perform well. For the 4029 cleanroom, additional measurements are
made of the makeup air flow and temperature; room humidity, temperature, and pres-
sure; and temperature inside the cooling unit. The room temperature and humidity
measurements are continuously plotted on a recorder to reveal long-term trends and
to alert users to changes in the environment. Other important parameters which are
continuously monitored are the processing water resistivity and TOC content; cool-
ing water flow rate, resistivity, and temperature; and the temperature and humidity
72
of room 4029 outside of the cleanroom. Continuous monitoring of these indicators
enables typically quick diagnosis when a problem arises and promotes confidence in
the room conditions under normal circumstances. Adherence to cleanroom protocols,
continuous monitoring and maintenance of the cleanroom environment, and replacing
the prefilters and HEPA filters as necessary should keep the system performing well
for many years.
73
CHAPTER 3
PROCESSING
There are two main sections for this chapter. The first discusses the drag sample
in general, including an overview of the bilayer growth, selective contact to one of a
pair of closely spaced 2DEGs, and how these techniques can be used to enable a drag
measurement. The second section explores the steps taken to process a sample from
a bilayer grown wafer through wiring. Finally, a summary timetable of the processing
steps is presented at the end of the chapter.
3.1 Drag sample
The first sample which demonstrated the electron drag effect was developed by
Solomon et al. in 1989 [3]. A field effect transistor (FET) geometry was used in which
an induced 2DEG channel defined one system and a GaAs gate was used as the other,
3D system. The following year, a technique enabling drag measurement between a
pair of 2DEGs was developed by Eisenstein, Pfeiffer, and West [28]. Pfeiffer and West
produced a GaAs wafer in which two 2D conducting layers, each with extraordinarily
high mobilities, were spaced only hundreds of angstroms apart from one another.
Eisenstein then devised the essential method by which those two layers could be
contacted independently from one another. The combination of these two technical
74
innovations formed the basis for producing bilayer electron drag samples. Other
techniques have since been developed for independent contact, notably in the area of
back-side gating integration [29, 30, 31, 32], although the functional operation remains
essentially the same as in Eisenstein’s original design. One notable exception is the
work of Sivan et al. [33], who developed a sample structure reminiscent of Solomon
et al.’s original 2D to 3D FET-like drag sample. In the experiment of Sivan et al.,
however, A 2D electron gas and 2D hole gas (2DHG) are induced on either side of a
barrier, enabling the measurement of 2D to 2D drag. The basic technique presented
here follows Eisenstein et al. [28] and later developments stemming from that work.
3.1.1 Molecular Beam Epitaxy
The bilayer wafer was produced through the technique of molecular beam epitaxy
(MBE). The epitaxial growth process, itself, is beyond the scope of this work but
a conceptual understanding of the epitaxially grown structure is necessary for later
explanations of the processing procedure. Figure 3.1 shows the conduction band
diagram for the bi-layer system. The growth direction is horizontal in the figure and
only the area of interest for this explanation is shown. Two 200 A wide quantum wells
of GaAs are spaced anywhere from 200 A to 5000 A apart from one another, depending
on the planned type of drag measurement to be made. In a drag measurement, One
of these quantum wells will be the drive layer, and the other will be the drag layer.
Because the electrons are constrained to a 200 A wide well, only the lowest subband
in each well is occupied. This ensures that all electronic motion in each well is in a
plane which runs vertically and perpendicular to the page. The electron gas in each
well is quantum mechanically two-dimensional.
75
Figure 3.1: Conduction band diagram for the 2-DEG bilayer system. The MBEgrowth direction is horizontal in the above figure. Electrons in the two 200 A widequantum wells are in the lowest energy state in the confining dimension, ensuring twodimensionality. The electron mobilities in both wells are better than 3× 106 cm2/Vsand each has a density near 1.5× 1011 cm−2.
76
The wells are populated with electrons from remote Si sources per the established
technique of modulation delta doping [34, 35, 36]. By depositing the dopants far from
the quantum wells, the electronic scattering time is dramatically reduced since the
dopants, themselves, do not act as strong scattering sites. Today, the most common
application of this technique is the high electron mobility transistor or HEMT [37]. A
conventional HEMT, however, uses a single interface between GaAs and AlxGa1−xAs,
instead of a fully 2D-confining quantum well. The bilayer sample used for drag can
be visualized as two modified HEMT structures back to back, with a quantum well
substituted for the single GaAs/AlxGa1−xAs heterostructure. An inverted HEMT,
in which the Si dopants are deposited before the 2DEG, is followed by a barrier of
200 A to 5000 A and then a normal HEMT is grown on top of that, in which the
dopant deposition follows the 2DEG interface. Each quantum well is populated by
dopants on the side opposite the other quantum well, allowing the barrier thickness
to be limited only by the restriction that carriers from one well cannot tunnel into
the other well.
Although modulation doping had been applied to single layer systems for some
time, the mobility of inverted HEMT structures had always been lower than that of
normal HEMTs by a factor of 23 or more [38, 39]. Pfeiffer et al. determined that the
Si dopants deposited during inverted HEMT structure growth migrate towards the
2DEG during deposition, thereby presenting stronger scattering sites for the 2DEG
electrons than intended and lowering the mobility. By modifying the growth profile
to account for this effect, inverted HEMT structures could be grown with mobilities
as high as 2.4× 106 cm2/Vs at 4.2 K [40]. By extension, a drag structure was grown
in which the electron mobility in both layers exceeds 3× 106 cm2/Vs. In order to set
77
a scale for this mobility, consider the following comparison. At 4 K, for an electron
density of 1.5× 1011 cm−2, the electron-electron scattering time τee ≈ hEF /(kBT )2 =
10−10 s, whereas the impurity scattering time is τimp = µm∗/e ' 10−6 s, 10,000 times
longer.
3.1.2 Independent Contact
Figure 3.2 illustrates the fundamental selective contact mechanism used in the
drag samples discussed here. This simplified schematic shows a single bar of MBE
growth that has been isolated from the GaAs it was grown on through an etch. The
2DEGs are shown in gray and the GaAs is transparent to illustrate the back gate.
In order to make electrical contact to the two 2DEGs, contact material is deposited
on top of both layers and the sample is heated in a controlled manner to diffuse the
contact dopant into the wafer. In order to make contact to only one of the two layers,
two Schottky gates are deposited on the sample as well, one above both 2DEG layers
and one below both layers. By applying a negative voltage to one of these gates,
the electrons in the region below the gate are locally depleted, breaking the electrical
continuity between the two contacts in that layer.
For a typical original drag sample, the GaAs chip on which the MBE growth was
deposited would be thinned through a chemo-mechanical lapping/polishing process
using a bromine-methanol solution. The original wafer thickness of roughly 500 µm
would be reduced to only 50 µm across the entire chip. Then, gates could be deposited
on the back, lapped side of the sample which would allow selective contact to the top
layer by locally depleting electrons from the bottom layer.
78
Figure 3.2: Independent electrical contact to two closely spaced 2DEGs: gating. Fol-lowing Eisenstein et al. [28, 41], this schematic illustrates the selective contact methodused for drag. The 2DEGs, indicated by gray bands, are electrically connected to eachother and the outside world through the annealed contacts. Current, indicated byarrows, will flow through both layers simultaneously unless the conduction electronsare blocked by sufficiently energizing a gate. The top view shows the front gate overboth of the layers, and the back gate under both of the layers. Cross section A il-lustrates selective contact to the bottom layer by energizing the top gate and locallydepleting the top layer electrons. Cross section B illustrates the same effect with thebottom gate, allowing selective contact to the top 2DEG. The back gate is ∼ 50 µmfrom both 2DEGs, while the front gate is only ∼ 0.5 µm away, which accounts for thesignificant difference in required gating voltages.
79
A typical top gate in this configuration would be about 0.5 µm from the top 2DEG
and could be actuated with a voltage of about −1.2V. The back gates, however, are
nearly 50 µm away from the bottom 2DEG and require a voltage of −120V or more
to deplete the bottom layer electrons. There is some danger in using a gating voltage
of this magnitude because of the potential for breakdown of the semi-insulating GaAs
wafer which acts as a dielectric between the gate and the 2DEG, itself. If breakdown
does occur, relatively strong current can flow directly from the gate to the 2DEG,
destroying the sample.
By extension of this core selective contact idea, a drag measurement equivalent
to that shown in figure 1.1 can be realized as shown in figure 3.3. In this figure,
the top and bottom 2DEGs are represented in green and gray, respectively, and the
contacts are shown in yellow. Back gates, in blue, control access to the bottom layer
and top gates, in red, control the top layer. In order to drive a current through the
bottom (gray) layer, contact is made through C1 and C2 and a voltage of −1.2V is
applied to top gates T1 and T2 while B1 and B2 remain grounded. To detect a drag
signal in the top (green) layer, contact the sample through C3 and C4 and apply a
voltage of −120V to B3 and B4 while T3 and T4 remain grounded. This is how a
drag measurement is configured. Typically, a fifth arm would be used to ground the
sample but this detail has been omitted for simplicity in the schematic shown.
The drag sample processing technique was later improved by adding a window
region to the back side of the GaAs chip, reducing the required gating voltage and
improving the overall structural integrity of the sample [42]. Instead of thinning
the sample to a thickness of 50 µm, the sample was thinned to 100 µm. Then, a
rectangular region on the back of the sample was etched to a depth of 80 µm, leaving
80
Figure 3.3: Implementation of drag configuration through selective gating. Drive acurrent through the bottom (gray) layer: Contact through C1 and C2, energize T1and T2, ground B1 and B2. Detect drag in the top (green) layer: Contact throughC3 and C4, energize B3 and B4, ground T3 and T4. A fifth arm, which would beused to define sample ground, is not shown here for simplicity.
81
the sample only 20 µm thick in the central, windowed region. The thicker sample
dimension around the perimeter made the sample more mechanically robust, overall,
reducing the likelihood of strain-induced fracture or, more commonly, breakage from
stress during handling. More importantly, the gating voltage was significantly reduced
from −120V to −50V or less, depending on the actual thickness of the sample in the
window region.
Figure 3.4 shows an overlay of the current lithography patterns used in processing.
All of the elements discussed above are represented here with the exception of the
overall top and back gates. These gates covering the central, active region of the
sample, control the density of the bottom and top layers independently. A constant
voltage is applied which adjusts the Fermi level in the corresponding quantum well.
Density adjustments within a factor of 2 are typically possible. Note that the degree
of parallel of the etch floor to the front face of the sample is critical in determining
the homogeneity of the bottom layer density. If the etch floor has a wedge-shaped
cross section, the 2DEG region which is closer to the overall back gate would have a
lower density, for a negative gate voltage, than the opposite end of the active region.
Several drag measurements depend critically on a uniform density in both quantum
wells, which emphasizes the importance of keeping the overall back gate smooth, flat,
and parallel to the bottom 2DEG layer.
The following section will discuss how these samples are currently produced, start-
ing from an MBE grown bi-layer wafer. The bilayer wafer is still grown by Loren
Pfeiffer and Ken West. Processing done in this laboratory typically begins with an
as-received cleaved portion of the original MBE wafer.
82
Figure 3.4: Opaque overlay of lithography patterns used in sample processing. Themesa etch mask (green) is used to define the mesa pattern where both of the 2DEGsexist. Both layers are removed outside of this region by etchant. Contacts (gold) aredeposited over the mesa and diffused in to contact both layers through an annealingprocesses. Front gates (red) are deposited over the mesa arms. After thinning, awindow is etched on the back of the sample in the region indicated, allowing backgates (blue) to be deposited within 20 µm of the 2DEG layers. The enlarged view atthe bottom indicates the 40 µm × 400 µm active region of the sample. The overallfront and back gates covering this region are used to control the density of each layerindependently.
83
3.2 Sample Processing
The sample processing procedure requires a minimum of six consecutive days to
complete, assuming that no complications are encountered. An error made late in
the step progression can be costly in terms of both resources and, especially, time.
Therefore, a strong effort has been made to reduce dependence on user skill and
experience in favor of a more robust process. If errors do occur, they are more
likely to be correctable, rather than fatal to the sample. For instance, the sixth
arm on the lithography overlay shown in figure 3.4 is redundant. In the event that
one contact should fail, the sample retains full functionality and can be used for
measurement. Additionally, the incorporation of several new techniques and methods
enables the production of samples which exceed the specifications met by earlier
sample generations.
The importance of maintaining a high degree of cleanliness throughout the process
cannot be overemphasized [43]. As with most lithographic processes, particulate
contamination can be highly detrimental. Although the smallest dimension used
in lithography is only 40 µm, well above the critical dimension for the equipment
and materials used, the central bar must be precisely defined and free of defects.
Irregularities in the bar width or point defects can substantially alter the transport
characteristics of the sample. Likewise, the chip must be kept clean of undesirable
residue from solvent, wax, resist, and other chemicals. This point is particularly
important for making contacts, as will be discussed in more detail, below.
The resists used for processing are primarily Rohm and Haas Shipley Microposit
S1800 series photoresists. Although the composition specifics are proprietary, this
positive photoresist is based on a common resist chemistry [44]. A novolac, phenolic
84
resin is formed by reacting formaldehyde with a mixture of cresol isomers in a ratio
greater than one. This novolac forms the base material of the photoresist. By itself,
novolac dissolves readily in alkaline aqueous solutions, such as sodium hydroxide-
based developers, at a rate of roughly 15 nm/s [45]. The photoactive compound
diazonaphthoquinone (DNQ), however, is insoluble in a basic solution. When the
DNQ and novolac components are combined, the resulting mixture retains much
of DNQ’s resistance to alkaline solutions, reducing the dissolution rate to less than
0.2 nm/s [46]. This mixture is then carried in a solvent, such as propylene glycol
monomethyl ether acetate (PGMEA), which controls the viscosity of the resist. The
more dilute the novolac resin is in solvent, the thinner the spun resist will be.
When the resist is exposed to ultraviolet light, the DNQ reacts with water to
form nitrogen gas and indene carboxylic acid (ICA), which dissolves very quickly in
alkaline solution. When the resist is immersed in a sodium hydroxide-based developer,
the exposed areas dissolve away at a rate of up to 100 – 200 nm/s [45], leaving the
unexposed resist in the desired pattern on the substrate.
Heating the resist at various points in the process is often necessary to vaporize
solvent, reduce interference effects, or to harden the resist for further processing.
However, care must be taken not to heat the resist beyond ∼ 125C. If the resist
is heated beyond this point, uncontrolled crosslinking occurs between the phenolic
resin polymer chains until a transition to a bakelite-like substance is reached. At this
point, the resist becomes hard and insoluble in most solvents and the substrate on
which it was used is typically lost.
Lithography on the resist is carried out using a SUSS Microtec MJB3 mask aligner.
The aligner uses a mercury short arc light source and optics optimized for the 350 nm
85
– 450 nm range. The power spectrum for the light incident on the sample should be
dominated by the 436 nm (g-line), 405 nm (h-line), and 365 nm (i-line) wavelengths
from the Hg arc lamp. Although the S1800 series resist is optimized for g-line expo-
sure, the resist is sensitive to exposure in the 300 nm – 450 nm range and the broad
spectrum exposure used is more than adequate to precisely resolve the lithographic
patterns.
3.2.1 Mesa Definition
The first steps in processing a sample are to cleave a chip from the (100) MBE
growth wafer and clean it. During the MBE growth process, the wafer is heat-sunk
to a temperature controlled block with gallium to ensure adequate thermal contact
between the block and the wafer. Using gallium, instead of another heatsinking
material, also prevents contamination of the MBE chamber. To process a cleaved
chip of the wafer, the gallium must be etched off the back side with hydrochloric acid
(HCl) while the front side of the sample is protected from the HCl. The sample is
waxed face down to a microprobe slide to protect the MBE growth.
The crystallographic orientation of the GaAs must be chosen so that the long axis
of the active region runs parallel to the [011] direction. The window which will be
on the back side of the sample results from an anisotropic etch. two opposite walls
have an obtuse wall-angle etch cross section and the other pair of opposite walls have
an acute wall-angle etch cross section. In other words, along the [011] direction, the
unetched surface and the etch floor are connected by ramps so that the back gate
pattern can run continuously from the unetched surface down to the etch floor. In
the [011] direction, however, the etch undercuts the GaAs and a lithographic pattern
86
cannot be made continuously over the edge. For this reason, the crystallographic
direction of the GaAs must be determined by a short anisotropic etch on the back
side of the wafer before lithography can begin.
After determining the [011] direction, the sample is removed from the microprobe
slide and remounted to a silicon heatsinking platform. Heatsinking is important in
order to keep the sample temperature low during the contact evaporation process, as
will be discussed below. Wax used for mounting the sample is pressed to a thickness
of 3 µm or less to minimize the thermal resistance between the sample and silicon
heatsink.
Using standard photolithographic techniques, the mesa pattern is imaged on the
GaAs chip and then etched. Care must be taken to avoid any defects visible on
the GaAs surface, as they may effect the transport properties of the sample. The
appropriate etch depth varies according to the barrier thickness used in the MBE
growth.
3.2.2 Contacts
Ohmic contact to the 2DEGs is made using a Au-Ge-Ni system: a highly reliable
and low resistance method originally introduced by Braslau et al. in 1967 [47]. Since
that time, Au-Ge-Ni contacts have been extensively studied [48, 49, 50, 51, 52, 16]
and used as a standard for ohmic contact to n-type GaAs [17]. A series of depositions
is made on the wafer surface, such as Ni-Ge-Au-Ni-Au, which diffuses dopants in the
deposition, Ge in this case, into the wafer. This diffusion path establishes electrical
contact between both 2DEGs buried beneath the surface and the top surface of the
contact deposition which is soldered to, later.
87
The primary difficulty with alloyed contacts in general, including the Au-Ge-
Ni system, is that a tremendous amount of process tuning is required in order to
repeatably produce low resistance contacts. Although alloyed contacts have been
investigated in great detail, the physical mechanisms that regulate the contacting
process are still not well understood [17]. To complicate matters further, the quantum
wells in our samples are much farther from the wafer surface than is typical for a
commercial HEMT device. In a typical HEMT structure, the 2DEG is about 1000 A
beneath the surface. In the drag sample, however, the top layer is 5000 A beneath
the surface and the bottom layer could be up to 5000 A below the top layer. Making
reliable contact to a 2DEG 1 µm below the surface is a difficult task. Almost 70
contacting attempts were made, at a cost of a day’s worth of processing apiece on
average, before reliable low-resistance contacts were established.
The lithographic process for contacts is, in itself, surprisingly non-trivial. After
resist is spun onto the chip and patterned, contact materials are deposited in a film
over the entire chip. Immersion in acetone then dissolves away the photoresist beneath
the deposited material that is outside of the contact region. This process is known
as “liftoff.” Because the contact deposition may be over 4000 A thick, the film has
considerable structural integrity. If the film is not discontinuous at the perimeter of
the contacts, then liftoff will fail.
The ideal resist cross section at a contact edge is a shelf structure. As long
as the bottom of the shelf is more than the distance of the contact film thickness
away from the wafer surface, a natural film discontinuity exists at the contact edges.
Several commercial products known as liftoff or multilayer resists are available for
this purpose. The simplest of the multilayer resists is the bilayer system. Two resists
88
are spun on the sample, one after the other, which have different dissolution rates in
developer. The bottom layer, which dissolves faster than the top layer and is relatively
insensitive to exposure, leaves an undercut profile in the overall resist. However, these
resists and the chemicals used for processing them are more difficult to remove than
typical positive photoresists. As the resist is developed, dissolved organic matter can
redeposit onto the exposed area of the sample in a process known as “scumming.”
This residual organic contamination may then interfere with contact performance.
Another alternative is to harden the top of a single layer of resist with chloroben-
zene. This creates a region of resist that dissolves more slowly in developer than
untreated resist, also resulting in a shelf structure. Although the resist profile result-
ing from this process is sufficient for successful liftoff, chlorobenzene is fairly toxic
and can damage the liver, kidneys, and nervous system. However, there is another
alternative.
Toluene can be used for a similar surface-hardening process with lesser health
risks. This method was tested and yielded successful results, as shown in figure 3.5.
The desired shelf structure is achieved, enabling successful liftoff of the contact film.
Temperature plays a critical role in making successful Au-Ge-Ni contacts. If the
contact film is overheated before the annealing process, the contact resistances are
often high or, worse, contact may not be made at all. One theory for this is that
the contact materials may alloy before the anneal, influencing their diffusion into the
GaAs [53], although this has not been conclusively established. The standard method
for depositing Au-Ge-Ni contacts is with e-beam evaporation. This method imparts
little heat load to the sample and high deposition rates are common. However, e-beam
89
Figure 3.5: SEM images of toluene-processed photoresist. A GaAs chip was cleavedafter processing it through exposure and development of a contact pattern. The over-hang on the resist profile ensures that the contact film deposited during evaporationwill be discontinuous at the contact edges, enabling successful liftoff. Note that theperiodic waviness evident in the overall pattern of the top two images is part of thecontact lithography pattern shown in Figure 3.4.
90
evaporators are quite expensive and retrofitting the thermal evaporator available in
the laboratory would be costly and a substantial undertaking.
Therefore, a number of modifications have been made to reduce the thermal load
on the sample. Extensive copper water-cooled shielding has been designed and in-
stalled around the evaporation sources to absorb radiated heat generated during
source deposition. Also, new evaporation sources are used which reduce the radi-
ated power from 650W, which was typical for the older sources, to 150W. Finally,
heat sinking the sample in the chamber provides a reservoir for the thermal loading
on the sample. These changes also drastically reduced the chamber pressure during
evaporation from, on average, 1 × 10−6 Torr to 1 × 10−7 Torr. The lower pressure
reduces the entrainment of contaminants during deposition of the contact film.
After allowing the sample to cool in the chamber, it is removed from the heat
sink and cleaned. The contacts are then annealed in forming gas in a rapid ther-
mal processor (RTP). The RTP anneals the contacts in a controlled and repeatable
manner by raising the temperature of the chamber hundreds of degrees to the desired
anneal temperature in approximately 2 seconds. The temperature is then maintained
to within 2C through feedback control from an optical pyrometer for the duration
of the anneal. Afterwords, the sample cools in an inert atmosphere for an extended
period of time.
Through this method, test contacts to an un-etched wafer chip are repeatably
formed with a specific resistance of 0.02Ωcm2 and a contact resistance of 20Ω.
91
3.2.3 Front Gates
Lithography for the front gates is almost identical to that used for contacts except
for the pattern, itself. The resist shelf must be high enough to ensure deposited film
discontinuity around the perimeter of the gate patterns. Each top gate traverses a
mesa step height, at the edge of the mesa etch pattern, which can be over 1.6 µm
high. After development, no resist can remain in the corner at the base of those mesa
edges, or the gate continuity will be broken there during liftoff.
1000 A of chromium is deposited through thermal evaporation, and the liftoff
process is used as before with contacts. Because the chromium film is relatively thin
and brittle, solvent easily penetrates fissures in the deposition over resist. Liftoff is
complete within seconds, whereas contact liftoff typically takes 10 minutes or longer.
3.2.4 Back Side Overview
The back side processing differs from front side processing in three important
ways.
First, the back surface of the GaAs resulting from the thinning process is topo-
graphically different than the surface of an as-received, polished (100) wafer. Most
importantly for the purposes of later processing are the roughness, flatness, and par-
allelism of the back surface. Roughness is defined in terms of deviations from a mean
surface height over a length scale commensurate with the surface feature size. For
instance, the parameter Ra = 1lm
lm∫0|y| dx is commonly used to define roughness where
y is the surface height relative to the mean between 0 and the measuring length lm.
Typically, lm is a few times larger than the surface feature size. Flatness is an indica-
tion of surface height deviations at a larger scale. Usually, flatness is specified in terms
92
of the number of interference fringes seen when placing the polished surface against
an optical flat and illuminating the interface with a monochromatic light source. For
the 546.1 nm line of a mercury spectrum, which is often used, one fringe corresponds
to 0.27µm of deviation. Lastly, considering the sample as a whole, parallelism is the
angle between the mean planes defined by the front and back surfaces of the chip. All
three of these parameters are important in defining the back surface of the sample.
Second, after the sample is thinned to 100 µm, the chip can no longer be handled
with tweezers. The force exerted by tweezers on the sides of the chip is more than
sufficient to fracture and destroy the sample. Therefore, the thinned chip must be
mounted to other backing substrates to allow handling for further processing.
Third, after the window is etched to a depth of 80 µm on the back surface of the
chip, the electrically isolated back gates must be deposited continuously from the
region outside the window, down the sloped window edge wall, and onto the window
floor. Producing micron-accurate lithography patterns in the bottom of an 80 µm
deep well is one of the most challenging aspects of sample processing.
3.2.5 Thinning
The most important part of the thinning process for determining parallelism is
the sample mounting to the grinding block [54]. This is achieved by careful selection
of mounting adhesive and by applying a constant, central force to the back of the
sample as it is bonded face-down to the grinding block. Care must also be taken not
to scratch the sample face while it is being pressed into the adhesive. Scratching will
likely result if the sample face is allowed to make direct physical contact with the
grinding block.
93
The target specification for the electron density uniformity in both of the 2DEG
layers is 1% across the entire 40 µm × 400µm active region of the sample. At this
level, the density irregularities due to gating effects are roughly equal to the native
irregularities in the 2DEG, itself. Thinning is the processing step that has the greatest
influence on this uniformity. The three parameters that control the parallelism of the
etched window floor to the 2DEGs are the flatness of the stainless steel grinding block,
the parallelism of the mounting adhesive, and the parallelism of the window etch floor
to the thinned surface of the GaAs. Of these factors, the mounting adhesive accounts
for half of the tolerance allowance. The thickness deviation of the adhesive should be
less than 1 µm across the entire sample.
This thinning process was originally done through a chemo-mechanical polishing
scheme using a bromine-methanol solution. Although this method was used success-
fully to reduce the thickness to 100 µm, the thinned surface would be domed, instead
of flat. Also, significant user skill was required. Lapping the sample improperly
would easily result in pitting or other irregularities in the surface. Because the pro-
cess required the use of bromine, a particularly toxic and reactive chemical, layers of
protective clothing and shields were required as well, exacerbating the difficulty of
properly executing the thinning technique.
The current thinning method uses a combination of grinding, lapping, and chemo-
mechanical polishing techniques. A lapping jig is used, which ensures that the plane
of GaAs material removal is parallel to the face of the grinding block. Almost all of
the material removal is achieved using SiC and Al2O3 fixed abrasives. Because these
materials do not chemically attack the GaAs, the importance of timing is substantially
94
removed from the technique, reducing both the need for operator skill and the likeli-
hood of errors in processing. Just before the target lapping depth is reached, a chemo-
mechanical polish removes scratches from the surface in preparation for the window
etch. If the surface is scratched or significant subsurface damage is present before the
window etch, the floor of the etched window will reflect the damage on the pre-etched
surface. Also, the etching rate through the damaged GaAs is not precisely control-
lable, an issue that affects the accuracy of the window etch depth. Although this
new thinning process is more accurate and controllable than the bromine-methanol
method, the abrasive-based method requires a day of work to complete, whereas the
bromine-methanol method would require only one to two hours.
After the sample is thinned and polished, it must be removed from the grinding
block for subsequent processing. Because the sample is now only 100 µm thick, it can
no longer be handled with tweezers. The sample is removed from the grinding block
by allowing the bonding adhesive to dissolve in a solvent. The sample falls from the
block, in the solvent, onto a disc of filter paper. This paper then acts as a carrier for
the sample until the chip can be mounted to a more substantial substrate for further
processing.
3.2.6 Window
In order for the back gates and density control to function, a window region must
now be etched in the back side of the sample. First the sample must be re-mounted
face down to a quartz puck to enable handling, again. Because the sample is now
thin, it cannot be press-mounted as before or the chip will fracture. Instead, a 0.5mm
diameter piece of wax is placed on the center of a quartz puck and melted. The thinned
95
sample is then slid off of the filter paper face-down onto the hot wax. If the quantity
of wax used is correct, the sample will now be remounted with a minimum of wedge
error in the wax. Capillary action draws the wax across the interface between the
chip and the quartz puck, securely anchoring the chip to the puck and protecting the
front face of the sample during later processing. A wax thickness deviation of 3 µm
or less across the chip is adequate for later processing.
For the window etch, resist is patterned on the chip to expose the window region.
Alignment to the front face of the sample is made using an infrared camera integrated
into the MJB3 aligner to see through the sample. The resist is hard baked after
development to withstand the extended window etch process. A citric acid / hydrogen
peroxide solution [55] is then used to etch the window over a period slightly longer
than 6 hours.
The agitation method used for etching the window significantly influences the etch
floor roughness, flatness, and parallelism relative to the front side. Figure 3.6 shows
two different samples, each measured with a profilometer across the entire well floor
along the central axis of the sample. One has been etched in a stagnant solution, as
samples have been processed in the past, and the other sample has been made in an
etchant of the same composition but with uniform agitation of the solution. As can
be seen from the two graphs, the agitated solution is much flatter and more parallel
than the stagnant solution. Note that the entire vertical scale of the agitated etch
graph is equal to 1 vertical division on the stagnant etch graph. The roughness of
the agitated etch floor is also improved by a factor of 3 over the stagnant etch floor.
The reason for the gross unevenness of the stagnant etch floor is not known but slow
96
(a) Etch profile from stagnant solution
(b) Etch profile from uniformly agitated solution
Figure 3.6: Influence of etchant solution agitation on etched window profile. Theseprofilometer measurements indicate the etch floor flatness, roughness, and parallelismrelative to the original, pre-etched GaAs surface. The vertical scale indicates etchdepth in microns and the horizontal scale spans the centerline of the entire etch floor.The scan is taken along the longitudinal axis of the back layer density control gate.Note the factor of 6 difference in vertical scale between the two graphs. The agitatedsolution improves the deviation in depth of the etch floor by a factor of 20 over theregion from 800 µm to 1200 µm, where the back layer density control gate is deposited.
97
convection currents in the unagitated solution may establish a preferred direction
during the etch which leads to the anisotropic profile.
The GaAs etching rate varies considerably with temperature for the citric acid /
hydrogen peroxide concentration used. However, if the environmental conditions are
precisely controlled, the etching rate can be highly reproducible. This is the process
that depends most critically on the room temperature control system discussed in
Chapter 2. Using an empirically-determined etching rate, the etch is allowed to run
continuously for the period of time needed to reach a depth of 80 µm. Then, the
sample is pulled from the solution and cleaned. Dummy samples that are processed
in parallel with the actual sample are used to determine the actual etch depth after
the etch is complete. The sharp edges of the etched window are then smoothed
slightly [56] in preparation for the back gate lithography steps.
3.2.7 Back Gates
Of all the lithography steps in the sample processing procedure, the back gate
lithography is the most difficult by far. The goal of the procedure is to coat the
entire sample with resist and then to completely remove that resist only in the areas
where the back gates are to be deposited while leaving the resist outside the gating
pattern intact. These two goals are in direct conflict with one another. If resist is
deposited and spun about the center of the sample, as usual, a meniscus develops
in the bottom corners of the window well that is almost 10 µm thick. As ultraviolet
light passes through the thickness of the resist, the intensity is attenuated according
to the relevant Dill Parameters [45]. If the resist is too thick, the bottom of the resist
will not be adequately exposed to dissolve in developer at a rate significantly higher
98
than the rate for unexposed resist. Resist remains in the bottom corners of the well
where the gating arm is supposed to be, creating a discontinuity in the gating arm.
Meanwhile, the resist at the top edge of the window well is vanishingly thin. Because
resist cannot easily flow over the top edge of the well, resist flows away from the
window edge during the resist spin, leaving a line of bare GaAs along the top edge of
the window. This resist pattern would result in a short between all of the back gates
along one side of the window.
In order to pattern the resist continuously and without shorting paths between
gates, the back gates are broken into two lithography sets, one for each of two opposite
window edges. A carefully programmed off-center spin is then used to meet the
lithography requirements.
After the lithography is complete, chromium is evaporated onto the back surface
in much the same way it was for the front gates. After the chromium is deposited,
the chromium film over the photoresist is removed through liftoff in acetone.
After one back gate set is complete, the process is duplicated on the opposite
back gate set. At the conclusion of the second back gate set, the sample is removed
from the quartz puck it was mounted on by dissolving the mounting wax in acetone
and allowing the sample to fall onto filter paper in the acetone. The sample is then
thoroughly rinsed. At this point, wet processing for the sample is complete. All that
remains is to wire the sample to a header.
3.2.8 Wiring
The final stage of processing is wiring the sample to a header for insertion into a
cryostat. Because the contact pads are less than 200 µm × 600µm, the wiring process
99
Figure 3.7: Soldering iron attachment for nitrogen flow. This attachment preventssurface oxides from forming during wiring by bathing the soldering area in a drynitrogen environment. A regulated, clean dry nitrogen source (not shown) must beported to the inlet. The nitrogen flow should be adjusted so as to displace oxygen inthe soldering area without substantially cooling the contact surface.
must be completed under a microscope. A soldering iron with a clean, untinned tip
is used, and the temperature setting is just sufficient to melt indium. Because the
sample is quite small and has a low heat capacity, contact between the sample and
soldering iron should be brief. Excessive heating will unsolder previously made solder
joints and could affect contact resistances as well.
The sample is attached to a header by 25 µm diameter gold wires, soldered to the
sample with pure indium. The use of flux is avoided because residual flux left after
soldering can attack the sample and the cryostat environment where the sample will
be measured. In order to solder effectively without using flux, the solder and the area
around the solder joint must be kept in an inert atmosphere to avoid oxidation [57].
Figure 3.7 illustrates the method by which this is achieved. A stainless steel tube
is bolted to the hilt of the soldering iron and surrounds the body of the iron tube
100
over most of its length. Clean dry nitrogen is then supplied to the base of the tube
and flows out around the soldering tip. This keeps the solder tip and the area being
soldered to under a nitrogen atmosphere and prevents rapid oxidation of the molten
indium. A flowmeter in-line with the nitrogen supply tube allows precise adjustment
of the flow. There should be enough nitrogen to locally displace the oxygen without
actively cooling the contact area. Insufficient nitrogen flow results in solder joints with
a dull gray appearance. Excessive nitrogen flow either requires more power delivery
to the soldering iron to overcome the nitrogen cooling or pointed contacts resulting
from a thermal gradient in the indium as the soldering iron is pulled away. When the
flow is adjusted correctly, the indium will wet the contact surface readily and leave a
smooth profile when the soldering iron is removed.
When the wiring is done, the sample processing is complete. The header is inserted
into a dipper stick and cooled to 4.2K in liquid helium. Then, contact resistances
and gating voltages can be tested. If the sample passes all the necessary electrical
tests, it can then be used for measurement.
3.3 Summary
In total, a minimum of six days are required to produce one drag sample from
an MBE grown wafer. There are several steps which require an extended period of
time during which supervision is unnecessary, such as pumping down the evaporator
chamber or allowing a sample to unmount in a solvent bath. By timing the process
appropriately, these periods can elapse overnight, utilizing the processing time most
effectively.
101
Step Time# Process work (hours)
1-9Cleave and clean sample, pattern and etch mesa, pattern con-tacts, put in evaporator and bake out sources.
9
10-17
Evaporate contacts, liftoff and anneal. Pattern front gates andput in evaporator. Evaporate front gates, liftoff, and mount togrinding block.
12
18-19 Grind, lap, and polish. Remove sample from block. 12
20-22Mount sample to quartz puck. Pattern and etch window. Pre-pare sample for back gate patterning.
10
23-29
Pattern back gate set 1 and put in evaporator. Evaporate set1 and liftoff. Pattern back gate set 2 and put in evaporator.Evaporate set 2 and liftoff. Remove sample from quartz puck.
11
30 Wire sample and test. 8Total processing : 6 days
Table 3.1: Optimized summary timetable for sample processing. Each row representsone day of processing requiring the time indicated in the right column.
Table 3.1 briefly summarizes the steps discussed in this chapter to process a sample
and breaks them into a set of tasks to be completed over a continuous 6 day period.
The table indicates the estimated processing time in hours required to complete each
day’s tasks, where completion of the last task of the day may mean initiating a
process that will continue overnight. This timetable represents an optimized schedule
for sample completion.
102
APPENDIX A
CLEANROOM CONSTRUCTION DOCUMENT
The following document was provided to the contractor responsible for construct-
ing the main cleanroom structure, including the stud walls, paneling, and ceiling grid.
The document has been slightly altered from its original form to reflect modifications
to the design made after the distribution of the original document. It illustrates the
type of information and level of detail necessary to bridge the experience gap between
more typical construction projects and a clean, pressurized enclosure. In addition to
this document, a brief session with the construction crew provided an opportunity
to discuss clean procedures and demonstrate a few necessary techniques. Using this
document, the structure was constructed as specified using clean protocols. Cor-
responding documents, not shown here, were given to other contractors for piping,
electrical, and HVAC installation.
103
Walls and Ceiling Installation Guidelines for 4029 Cleanroom
This document provides a procedure for constructing the walls and ceiling grid
of the cleanroom in 4029 Smith Lab, shown in the set of “4029 CLEAN ROOM”
AutoCAD drawings. Many of the techniques used follow the conventions in the
Prescriptive Method for Residential Cold-Formed Steel Framing 2nd Ed. (PM) and
Builders’ Steel Stud Guide (BG) available from the North American Steel Framing
Alliance. These two publications are referred to often as a basis. Modifications from
the methods they prescribe, to meet the cleanroom requirements, are indicated where
necessary.
There are two primary points which differentiate this project from typical con-
struction for a built-in room. First, the interior of the room will be held at 0.02” wc
higher air pressure than the outside of the room, with the exterior of the new walls
holding the pressure in. So, all studwork must be sealed together and the exterior
panels must be sealed to the studwork to create a watertight barrier at the exterior
cleanroom wall. In practice, this generally means that quite a bit of silicone caulk is
used.
Secondly, because this will be a “cleanroom,” precautions must be taken to mini-
mize the amount of dust and debris produced during the construction process. Use of
chalk lines (with chalk), for instance, would produce a great deal of dust that could
contaminate the room after it is completed. The necessary precautions will increase
as the room progresses towards completion. After the exterior panels are installed,
booties and gloves will need to be worn inside the cleanroom.
104
C-stud walls
The existing walls, ceiling, and floor should already be clean; free of ductwork,
lights, and electrical conduit; and sealed except for new plumbing, electrical, and
ducting holes before beginning. The existing walls and ceiling will also be epoxy
painted.
The new walls are constructed from 18 gauge (43 mil thick) steel C-studs. All the
studs are 2”× 6” except for the ones used in the south wall of the anteroom, which
are 2”× 4”. Because panel trim will be used on the inside of the cleanroom, the stud
pattern alternates between single and double (back-to-back “C”s to form “H”s) king
studs, with the double studs at the seams of the 4’ wide panels. Note that, since the
door frames are sealed, they will need to be installed as the framing is constructed.
The procedure below follows a modified “in-place” wall-raising technique (BG
pg. 13) but a tilt-up framing technique could also be used if that would be easier.
(Note the close proximity of the large air duct through room 4029 to the main south
wall of the cleanroom.)
1. Mark the stud positions
• Use a black-felt marker to mark the locations of the layout studs on the
track.
• Use a red-felt marker to mark the location of openings.
• Your layout will be more accurate if you mark both tracks at the same
time.
2. Caulk and attach the headers and footers
105
• Use a string and marker to layout the location of the walls on the floor.
• Use a plum-bob and marker to mark the location of the walls at the ceiling.
• Silicone caulk the bottom track into place: Run 2 continuous beads of
silicone caulk on the bottom of the track 1/2” from either edge. Invert
and press the track into place on the floor.
• When securing the bottom track to the floor, use two masonry fasteners
every 24”.
• Silicone caulk the top track into place: Run 2 continuous beads of silicone
caulk on the bottom of the track 1/2” from either edge. Invert and press
the track into place on the ceiling.
• When securing the top track to the ceiling, use two masonry fasteners
every 24”.
3. Caulk and attach the studs
• Twist studs into the track and attach them with #8 S-12 Low Profile
screws through the track into each flange of the studs.
• Using silicone caulk, seal the stud flanges to the top and bottom tracks.
• For “H” studs, caulk 2 continuous beads floor to ceiling 1/2” from each
edge of the back-to-back web mating surfaces. Install one stud first, screw
and caulk it, and then install the second stud of the pair.
• Fasten the two studs in each ”H” pattern together using two #8 S-12 fluted
point screws every 24”.
106
Corner Framing
PM pg. 43 but caulk as directed earlier.
Framing at Door Jambs
Details for corners and doors can be given if needed.
Unistrut brackets
The ceiling grid, ducting, plumbing, electrical lines, fan filter units, and ceiling
lights all suspend from Unistrut channel bolted to the existing ceiling. The 9’ long
Unistrut channels are directly above the positions of the ceiling grid members which
run North and South. The North end of the channel butts against the structural beam
that runs along the corner between the North wall and the ceiling. Each channel is
attached to the existing ceiling with 4, equally spaced, 1-3/4” masonry bolts.
Anteroom roof
The anteroom roof is made of 23/32” SIS YP plywood. Because of the tight space
between the top of the anteroom roof and the existing duct in the room (approxi-
mately 3/4” from the top of the anteroom to the lowest part of the duct) installing the
anteroom roof may be tricky. One approach would be to attach the headers for the
anteroom c-studs to the roof first, and then install the c-studs with the roof elevated.
The roof is attached directly to the c-studs in the main south wall of the cleanroom
with L-brackets. Because the roof is slightly longer than 8’, two 4’× 8’ panels will be
needed, with the seam between the panels as far to the west as possible (8’ from the
exterior East end of the cleanroom). The plywood roof panels should be joined with
107
wood glue to make a seal and help to hold the panels together. Note, lastly, that
2”× 4” C-studs are used for the South wall of the anteroom.
Exterior panels
Exterior panels are 1/2” - 4’× 8’ birch plywood finished with spar urethane. There
is no trim for the exterior panels, except for an aluminum H–Channel at the horizontal
seam between panels. Each 4’× 8’ panel is manually plumbed and centered to the
studwork. The panels are sealed to the studwork along each flange and track they
come in contact with using silicone caulk. A full plywood sheet is attached vertically
from the floor to it’s full height of 8’. At the horizontal seam (at 8’) between that
panel and the one mounted above it, a single aluminum H-Channel, cut to the width
of the panel, is used to join the two edges. A bead of caulk is run in each channel to
create a seal to the edge of each panel. A continuous 1/4” bead of caulk is applied to
the surface of each panel 1/4” from all edges. The panels are secured to the studwork
with #8, S-12 fluted-point screws every 18” around the perimeter and down the
centerline of the panel. Some panels have ductwork passing through them. Sleeves
are provided by Limbach Company to be passed through the necessary (upper) panels.
The exterior of these sleeves should be sealed to the plywood with caulk. At door
frames, the plywood butts against the frame edge and is caulked onto and screwed
into the stud behind the edge of the plywood. No additional trim is used.
Interior panels
At this point, all plumbing and electrical services should be in place and ready
for intrusion through the interior panel.
108
All interior panels are 4’× 8’ Citadel Panel 20. Only vertical trim will be used.
The backing trim is attached to the double studs first using #8, S-12 Low Profile
screws every 12” down the length of the backing trim. All cutouts for return air
grilles, electrical and plumbing services, pressure gauges, and temperature controls
should be made outside of the cleanroom. Silicone caulk is applied along the channels
of the vertical trim, the stud which runs along the centerline of the panel, and the
bottom track of the stud wall. The pre-cut panel is screwed into place with 6 screws,
one each 1/2” from the corners and two 1/2” from the top and bottom center of the
panel, and the mating trim piece is snapped into the backing trim. Care must be
taken to ensure that the top edges of the panels all align, as the ceiling grid will be
mounted directly on the top edge of the panels. Gaps between the bottom of the
panel and the floor are filled with silicone caulk. All panels should be left at their
full, 8’ length. At door frames, the panel butts against the frame edge and is caulked
onto and screwed into the stud behind the edge of the panel. No additional trim is
used.
Ceiling grid
The ceiling grid is suspended from the Unistrut on the existing ceiling at grid
junctions and fastened to studs at the wall perimeters using #8 screws through the
vertical member of the grid perimeter. The perimeter should also rest on top of the
interior panels, where a silicone caulk seal is made after the grid is installed. The
Gordon DS-20 grid is installed according to the manufacturer’s instructions, with the
primary structural members running along the short axis of the room on a 24-1/2”
spacing parallel to and directly below the Unistrut channel they are suspended from.
Ceiling grid elements can be cut with a miter box. The grid is installed starting from
109
the East end of the room and progressing to the West. The tops of the two hoods
towards the West end of the room protrude into the drop ceiling. When the position
of the hoods is reached, the hoods are moved into place and the grid should follow the
perimeter of the hoods as if they were part of the wall. The grid should be attached
to the side of the hoods in the same manner the perimeter of the grid is attached to
the C-studs.
110
APPENDIX B
CLEANROOM PLANS
The following schematics illustrate the steps and stages of construction for the
250 ft2 cleanroom built in room 4029 Smith Lab on the Ohio State University cam-
pus. Although the following pages are not intended to be a complete set of architec-
tural schematics for construction, which would be difficult to re-create in this format,
they are based on the drawings actually used for construction and indicate some of
the more important or unusual aspects of the project. The drawings are arranged
chronologically in the order of necessary completion with few exceptions. Interior wall
panels, for instance, should not be installed until all in-wall plumbing and electrical
lines have been completed. The first drawing shows the general planned layout of
the room, indicating the intended functionality of different room sections and some
salient points of the design. Then, in the following drawings, the construction stages
are illustrated, beginning with the cleared room and ending with the installation
of all monitoring, control, and processing equipment in and around the completed
structure.
111
Fig
ure
B.1
:G
ener
alla
yout
112
Fig
ure
B.2
:B
ase
modifi
cati
ons
113
Fig
ure
B.3
:U
nis
trut
suppor
ts
114
Fig
ure
B.4
:Stu
dpla
cem
ent
115
Fig
ure
B.5
:W
allpan
elin
g
116
Fig
ure
B.6
:H
VA
Cduct
wor
k
117
Fig
ure
B.7
:P
ipin
gte
rmin
i
118
Fig
ure
B.8
:W
allin
stal
lati
ondet
ails
119
Fig
ure
B.9
:E
lect
rica
lre
quir
emen
ts
120
Fig
ure
B.1
0:C
eiling
grid
layo
ut
121
Fig
ure
B.1
1:Fin
aleq
uip
men
tin
stal
lati
on
122
APPENDIX C
CRYOSTAT SUPPORT STRUCTURE
One important aspect of making low-noise cryogenic measurements in high mag-
netic fields is the reduction of mechanical vibrations in the system. The motion of a
vibrating, signal-carrying wire in a nonuniform magnetic field will generate an elec-
tromotive force on the line which could reduce the signal-to-noise ratio. Also, joule
heating from this effect can skew thermometry measurements and even frustrate at-
tempts to achieve low temperatures in the cryostat. This appendix describes the
construction of a cryostat support structure designed to reduce the transmission of
mechanical vibrations from a building into a cryostat.
Because of space restrictions, the Gramila low temperature laboratory had to be
constructed on the first floor of Smith Laboratory, over a basement, instead of in the
basement, itself. The building is constructed from concrete reinforced with rebar.
Columns and beams define the main structure, and reinforced concrete floors span
the distance between beams, which is roughly 9’ in the first floor of the building. The
vibration level on the floor region directly over structural beams is within 10 dB of
the level measured on the well isolated pit structure used for one cryostat’s previous
installation in the basement of a different building. However, on the floor region
123
between structural beams, where the cryostat needed to be installed in Smith Labo-
ratory, the vibration level was 30-50 dB higher than the level measured directly over
the structural beams.
To mitigate this problem, cryostat support structures were designed which mount
directly to the structural beams of the building and span the distance between beams
at a height of 1” below the concrete floor slab. Holes are cut in the floor to allow
a cryostat to mount on a support via an existing, aluminum tripod structure. A
drawing of one assembled cryostat support structure is shown in figure C.1.
The design approach chosen was to use a massive structure made out of a fairly
lossy material, wood, to dampen vibrations from the building. The wood is structural
glue laminated timber manufactured by Anthony Forest Products. Anthony Power
To quantify the actual effectiveness of the support after its installation, the vibra-
tion level was measured on the wooden structure before a cryostat tripod was installed
on it. The vibration level was reduced by 30-50 dB compared to the vibration level
in the center of the concrete floor directly above the structure. Also, the vibration
level measured on the structure was within 3 dB of the lowest vibration level mea-
sured in the room. These measurements indicate, therefore, that the cryostat support
structure is an effective isolator from building mechanical vibrations.
130
BIBLIOGRAPHY
[1] M. B. Pogrebinskii, Mutual Drag of Carriers in a Semiconductor-Insulator-Semiconductor System, Sov. Phys. Semicond. 11, 372 (1977), originally publishedin Fiz. Tekh. Poluprovod. 11, 637-644.
[2] P. J. Price, Hot Electron Effects in Heterolayers, Physica B+C 117-118, part2, 750 (1983).
[3] P. M. Solomon, P. J. Price, D. J. Frank, and D. C. La Tulipe, New Phenomenain Coupled Transport between 2D and 3D Electron-Gas Layers, Phys. Rev. Lett.63, 2508 (1989).
[4] T. J. Gramila, J. P. Eisenstein, A. H. MacDonald, L. N. Pfeiffer, and K. W. West,Mutual Friction between Parallel Two-Dimensional Electron Systems, Phys. Rev.Lett. 66, 1216 (1991).
[5] A. G. Rojo, Electron-drag Effects in Coupled Electron Systems, J. Phys.: Con-dens. Matter 11, R31 (1999).
[6] A. Roth, Vacuum Technology, 3rd ed. (Elsevier Science, 1990).
[7] G. K. White and P. J. Meeson, Experimental Techniques in Low-TemperaturePhysics, Vol. 59 of Monographs on the Physics and Chemistry of Materials, 4thed. (Oxford University Press, New York, 2002).
[8] H. W. Ott, Noise Reduction Techniques in Electronic Systems, 2nd ed. (JohnWiley & Sons, 1988).
[9] U. S. Federal Standard 209E, Airborne Particulate Cleanliness Classes in Clean-rooms and Clean Zones, Technical report, General Services Administration,Washington, DC, 1992; withdrawn 29 November, 2001.
[10] EN ISO 14644-1, Cleanrooms and Associated Controlled Environments, Part 1:Classification of Air Cleanliness, Technical report, International Organizationfor Standardization ISO, Geneva, May 1999.
[12] J. G. King, in Cleanroom Design, 2nd ed., edited by W. Whyte (John Wiley& Sons, 1999), Chap. 3, The Design of Cleanrooms for the MicroelectronicsIndustry.
[13] M. A. Wirzbicki, Quantifying the Effects of Coat Bowl Temperature and RelativeHumidity for an Advanced i-Line Photoresist Coating Process, Technical report,Rohm and Haas Electronic Materials, LLC, 1996.
[14] T. Ruys, in Handbook of Facilities Planning, edited by T. Ruys (Van NostrandReinhold, New York, 1990), Vol. 1, Laboratory Facilities.
[15] American Society for Testing and Materials, Standard Guide for Ultra Pure Wa-ter Used in the Electronics and Semiconductor Industry, D5127-99 (AmericanSociety for Testing and Materials, Annual Book of ASTM Standards, Philadel-phia, 2004).
[16] K. C. Lee, Degradation of GaAs/AlGaAs Quantized Hall Resistors with AlloyedAuGe/Ni Contacts, J. Res. Natl. Inst. Stand. Technol. 103, 177 (1998).
[17] C. Y. Chang and F. Kai, GaAs High-Speed Devices: Physics, Technology, andCircuit Applications (John Wiley & Sons, 1994).
[18] R. E. Marshall, in Handbook of Facilities Planning, edited by T. Ruys (VanNostrand Reinhold, New York, 1990), Vol. 1, Laboratory Facilities, Chap. 5.4,Electrical Systems.
[19] P. R. Austin, Austin’s Clean Rooms of the World: Case Book of 200 Clean Rooms(Ann Arbor Science, 1967).
[20] A. Lieberman, Contamination Control and Cleanrooms: Problems, EngineeringSolutions, and Applications (Van Nostrand Reinhold, 1992).
[21] R. K. Schneider, Practical Cleanroom Design, revised ed. (Buisness News, 1995).
[22] Institute of Environmental Sciences and Technology (IEST), IEST-RP-CC012.1:Considerations in Cleanroom Design (Institute of Environmental Sciences andTechnology, 1998).
[23] American Iron and Steel Institute, Prescriptive Method for Residential Cold-Formed Steel Framing, 2nd ed. (American Iron and Steel Institute, Washington,D.C., 1997).
[24] North American Steel Framing Alliance, Shear Wall Design Guide (North Amer-ican Steel Framing Alliance, Washington, D.C., 1998).
132
[25] North American Steel Framing Alliance, Builders’ Steel Stud Guide, PublicationRG-9607 (North American Steel Framing Alliance, Washington, D.C., 1996).
[26] American Iron and Steel Institute, Fasteners for Residential Steel Framing(American Iron and Steel Institute, Washington, D.C., 1993).
[27] F. W. Murray, On the Computation of Saturation Vapor Pressure, J. Appl. Me-teor. 6, 203 (1967).
[28] J. P. Eisenstein, L. N. Pfeiffer, and K. W. West, Independently Contacted Two-Dimensional Electron Systems in Double Quantum Wells, Appl. Phys. Lett. 57,2324 (1990).
[29] K. M. Brown, E. H. Linfield, G. A. C. Jones, D. A. Ritchie, and J. H. Thompson,Fabrication of a Novel Split-Backgate Transistor by in situ Focused Ion-BeamLithography and Molecular-Beam Epitaxial Regrowth, J. Vac. Sci. Technol. B 11,2493 (1993).
[30] R. J. Evans, M. P. Grimshaw, J. H. Burroughes, M. L. Leadbeater, M. J. Trib-ble, D. A. Ritchie, G. A. C. Jones, and M. Pepper, Double Two-DimensionalElectron Gas Structure Formed by Molecular Beam Epitaxy Regrowth on an exsitu Patterned n+-GaAs Back Gate, Appl. Phys. Lett. 65, 1943 (1994).
[31] H. Rubel, A. Fischer, W. Dietsche, K. von Klitzing, and K. Eberl, Observationof Screening in the Magneto-Coulomb Drag between Coupled Two-DimensionalElectron Systems, Phys. Rev. Lett. 78, 1763 (1997).
[32] M. V. Weckwerth, J. A. Simmons, N. E. Harff, M. E. Sherwin, M. A. Blount,W. E. Baca, and H. C. Chui, Epoxy Bond and Stop-Etch (EBASE) Technique En-abling Backside Processing of (Al)GaAs Heterostructures, Superlatt. Microstruc.20, 561 (1996).
[33] U. Sivan, P. M. Solomon, and H. Shtrikman, Coupled Electron-Hole Transport,Phys. Rev. Lett. 68, 1196 (1992).
[34] R. Dingle, H. L. Stormer, A. C. Gossard, and W. Wiegmann, Electron Mobilitiesin Modulation-Doped Semiconductor Heterojunction Superlattices, Appl. Phys.Lett. 33, 665 (1978).
[35] M. Ploog, M. Hauser, and A. Fisher, in Gallium Arsenide and Related Com-pounds 1987: Proceedings of the 14th International Symposium on Gallium Ar-senide and Related Compounds, Crete, Greece, 28 September - 1 October 1987,Vol. 91 of Institute of Physics Conference Series, edited by A. Christou and H. S.Rupprecht (Institute of Physics, 1988), pp. 27–32.
133
[36] G. Gillman, B. Vinter, E. Barbier, and A. Tardella, Experimental and TheoreticalMobility of Electrons in δ-Doped GaAs, Appl. Phys. Lett. 52, 972 (1988).
[37] T. Mimura, S. Hiyamizu, T. Fujii, and K. Nanbu, A New Field-Effect Transistorwith Selectively Doped GaAs/n-AlxGa1−xAs Heterojunctions, Jap. J. App. Phys.19, L225 (1980).
[38] L. Pfeiffer, K. W. West, H. L. Stormer, and K. W. Baldwin, Electron MobilitiesExceeding 107 cm2/Vs in Modulation-Doped GaAs, Appl. Phys. Lett. 55, 1888(1989).
[39] D. Kim, A. Madhukar, K.-Z. Hu, and W. Chen, Realization of High Mobilities atUltralow Electron Density in GaAs-Al0.3Ga0.7As Inverted Heterojunctions, App.Phys. Lett. 56, 1874 (1990).
[40] L. Pfeiffer, E. F. Schubert, K. W. West, and C. W. Magee, Si Dopant Migrationand the AlGaAs/GaAs Inverted Interface, Appl. Phys. Lett. 58, 2258 (1991).
[41] J. P. Eisenstein, L. N. Pfeiffer, and K. W. West, Field-Induced Resonant Tunnel-ing between Parallel Two-Dimensional Electron Systems, Appl. Phys. Lett. 58,1497 (1991).
[42] H. Noh, Ph.D. thesis, The Pennsylvania State University, 1999.
[43] B. Kanegsberg and E. Kanegsberg, Handbook for Critical Cleaning (CRC Press,2001).
[44] G. S. May and S. M. Sze, Fundamentals of Semiconductor Fabrication (Wiley,New York, 2004).
[45] F. H. Dill, W. P. Hornberger, P. S. Hauge, and J. M. Shaw, Characterization ofPositive Photoresists, IEEE Trans. Electron. Devices 22, 445 (1975).
[46] F. H. Dill, Optical Lithography, IEEE Trans. Electron. Devices 22, 440 (1975).
[47] N. Braslau, J. B. Gunn, and J. L. Staples, Metal-Semiconductor Contacts forGaAs Bulk Effect Devices, Solid State Electron. 10, 381 (1967).
[48] M. Ogawa, Alloying Behavior of Ni/Au-Ge films on GaAs, J. Appl. Phys. 51,406 (1980).
[49] N. Braslau, Alloyed Ohmic Contact to GaAs, J. Vac. Sci. Tech. 19, 803 (1981).
[50] M. Heiblum, M. I. Nathan, and C. A. Chang, Characteristics of AuGeNi OhmicContacts to GaAs, Solid-State Electron. 25, 185 (1982).
134
[51] T. S. Kuan, P. E. Batson, T. N. Jackson, H. Rupprecht, and E. L. Wilkie, ElectronMicroscope Studies of an Alloyed Au/Ni/Au-Ge Ohmic Contact to GaAs, J.Appl. Phys. 54, 6952 (1983).
[52] H. Goronkin, S. Tehrani, T. Remmel, P. L. Fejes, and K. J. Johnson, Ohmic Con-tact Penetration and Encroachment in GaAs/AlGaAs and GaAs FET’s, IEEETrans. Elect. Dev. 36, 281 (1989).
[53] E. D. Marshall and M. Murakami, in Contacts to Semiconductors: Fundamentalsand Technology, Materials Science and Process Technology Series, edited by L. J.Brillson (Noyes, 1993), Chap. 1.
[54] K. W. Torrance, J. McAneny, and M. Robertson, in The 1999 International Con-ference on Compound Semiconductor Manufacturing (GaAs MANTECH, 1999).
[55] G. C. DeSalvo, W. F. Tseng, and J. Comas, Etch Rates and Selectivities of CitricAcid/Hydrogen Peroxide on GaAs, Al0.3Ga0.7As, In0.2Ga0.8As, In0.53Ga0.47As,In0.52Al0.48As, InP, J. Electrochem. Soc. 139, 831 (1992).
[56] D. W. Shaw, Localized GaAs Etching with Acidic Hydrogen Peroxide Solutions,J. Electrochem. Soc. 128, 874 (1981).
[57] Indium Corporation of America, Soldering to Gold – Alloy Choice and Lim-itations, http://www.indium.com/documents/applicationnotes/97743.pdf,2004.
[58] B. Knight, Effects of Surface Treatments on Adhesion to Metals, Epoxyworks22, (2004), http://www.epoxyworks.com/22/pdf/Ew22 Effects.pdf.
[59] International Conference of Building Officials (ICBO), ITW Ramset/Red HeadSelf-Drilling, Trubolt Wedge, and Multi-Set II Concrete Anchors, ER-1372, Tech-nical report,http://www.icc-es.org/reports/pdf files/ICBO-ES/1372.pdf.
[60] International Conference of Building Officials (ICBO), ITW Ramset/Red HeadEpcon System, ER-4285, Technical report,http://www.icc-es.org/reports/pdf files/ICBO-ES/4285.pdf.