Fabrication of Microelectronic Devices Final - 2 PDF
Fabrication of Microelectronic Devices Final - 2 PDF
Fabrication of Microelectronic Devices Final - 2 PDF
1.0 Introduction
The second part of the last century is known as ‘silicon age’, and the backbone of this era is
precise manufacturing of circuit elements on a single crystal Si wafer. This particular
technology of semiconductor processing is known as planar technology and makes possible
integration of large number of electronic components on a single wafer [1]. The
miniaturization of individual device to make large scale integrated circuits would not have
been possible without the advances in processes, techniques and equipments. The availability
of cheap electronics today is a testimony to these advances that took place in last 50 years.
Advances in microelectronic device fabrication have spun-off new fields like micro-
electromechanical-systems (MEMS) [2]. In microelectronic circuits there is no physical
displacement between components, while MEMS processing is done to allow physical
displacement of components. To a very large extent, today’s nanofabrication technology also
has its lineage in microelectronics fabrication as many tools developed to work at micro
range have been improved and enhanced to provide nano-level observation and manipulation.
If we wish to build a very complicated part, first step is to break it down in various
monolithic components, cast or machine them, finally join them using a mechanical or
metallurgical joining technique. Microelectronics fabrication is done in such a way that
joining between components is done by directly building one component on the other. In
other words, joining is at atomic level (nano-scale) between interfaces of two components.
The interface itself is a component, e.g. in p-n junction; therefore, we will see later that it is a
requirement that we have ultra clean joints between two materials. To attain this precision
clean room environment is required for microelectronic fabrication.
Metal electrodes
n Si
n-type
p-type Si
Metal electrodes
To illustrate the working principle of microelectronic processing, let us see the part in fig. 1
that we wish to fabricate; it is a simple p-n diode in silicon. To this junction, we will make
ohmic contacts (means the resistance of the contact is zero) using metal junctions on both
sides. In this chapter, all examples are taken from Si technology, because it has most
advanced processing technology and it also has large share of integrated circuit market. The
processing steps to make this diode are shown in fig. 2.
iii SiO2
Si wafer n- type PR
ii SiO2 UV Exposure of PR
Si wafer n- type through mask
SiO2
iv p-type
PR
Si (n- type)
After developing in
iii SiO2
SiO2 developer solution
Si wafer n- type
v SiO2
p-type PR
Si n- type
iv SiO2 Etching of SiO2
SiO2 Si wafer n- type
SiO2
vi p-type
Si(n- type)
v SiO2
Photoresist removal
Si wafer n- type
Metal
SiO2
vii SiO2
p-type
p-type
Si(n- type)
SiO2
viii SiO2
p-typep-type
p-type
Si(n- type)
(a) (b)
Fig. 2. (a)Processing steps in the fabrication of a p-n junction diode: (i) Single crystal
silicon wafer of required doping; (ii) Deposition of a layer of silicon oxide (SiO2); (iii)
Photolithography to open a window; (iv) Introduction of p-type dopant in silicon
through the window; (v) Once again, deposition of SiO2 layer; (vi) Photolithography to
open another window; (vii) Metal deposition; (viii) Photolithography to define contacts
to p and n region.
We start with single crystal silicon wafer of required doping, which is most likely made by
pulling a single crystal from a melt and then cut and polished to right size [3]. On this Si
wafer, approximately 500 microns thick and anywhere between 50-300 mm in diameter, first
we deposit a layer of silicon oxide (SiO2); this is generally done by thermal oxidation or
chemical vapor deposition (CVD). The thickness of this layer is decided by the requirement
of the subsequent processes. A window is opened in the oxide. This process is called
photolithography and is subdivided into many sub-processes as shown in fig. 2(b). In
photolithography, we spin-coat a photosensitive polymer (photoresist) on the substrate.
Typically we bake the wafer before and after spin coating (pre-bake and post-bake). This is
followed by exposure of the coated wafer by UV-light. However, the photoresist experiences
exposure to UV light only in the desired areas because it is protected by a patterned mask (for
example, a glass plate on which chromium is coated in a pattern, preventing the UV light to
pass through). Following development of the photoresist, the surface that needs to be
removed is exposed and the one we need to retain is protected by PR, we use an etching agent
that will remove the undesired part of SiO2. Finally PR is also etched away from top of the
layer and we get the net shape desired on the SiO2. In the next step, p-type dopant is
introduced in silicon through the window using diffusion or ion implantation process. The
thickness of the masking oxide should be enough to ensure that dopants are not introduced in
the covered portion of silicon wafer. The p-n junction is created by this process. Once again
we deposit SiO2 layer; next we open two windows in this as shown in (vi) using previously
explained photolithography process. Thin film of metal electrode is deposited (sometime
additional processing may be needed to ensure ohmic contact) to make contact to n and p
portion of the Si wafer. The metal film will again need to be patterned using
photolithography. Now the device is ready, in practice another passivating layer is deposited
and patterned to protect the devices. The final device in fig. 2 may not look the same as in fig
1 from geometry point of view, but it has exactly same electrical characteristics. For
microelectronics that is the important part, not the physical appearance. Another interesting
thing in this transformation of the component shape is that we can take both contacts from top
of the surface, this makes integration very easy. In integrated circuits, millions of such
devices need to be fabricated and connected on a single wafer. All fabrications steps are in a
single plane – hence it is called ‘planar technology’. After chip fabrication, it is tested, wire-
bonded, and packaged; these are also very important parts of microelectronics fabrication,
which we will not discuss in this chapter.
4
9
10 Intel Itanium 2 processor (9MB cache)
Intel Itanium 2 processor
8
10 Intel Itanium processor
Intel Pentium 4 processor
7 Intel Pentium III processor
10 Intel Pentium II processor
No. of transistors
Intel Pentiumprocessor
6 Intel486 processor
10
Intel386 processor
5 Intel286
10
8086
4
10 8080
8008
4004
3
10
2
10
1970 1975 1980 1985 1990 1995 2000 2005
Year of introduction
In summary, we have created a component by adding the material and patterning it. The two
processes that add a new material to an existing component are thin film deposition and
diffusion/ion-implantation (this modifies the existing material). Processes that pattern the
components are photolithography and etching. Each of them will be discussed separately in
subsequent sections. Since this information is meant for engineers/researchers not working in
the field, we have kept the level of details to minimum for sake of clarity.
Advances in microelectronics and optoelectronics have driven the development of thin film
science and technology, and vice-versa. To add a new material, we use thin film deposition
where atomic flux, in vapour phase, of the material is brought to the surface and is deposited.
(There are also liquid phase based techniques, but these are used in special cases.) There are
two main categories here, physical vapour deposition (PVD) and chemical vapour deposition
(CVD). We can deposit almost any material, metals, insulators and semiconductors, as
needed in the device structure.
Substrate holder
Crystal monitor
(DTM)
Thermocouple
Bell jar
Material
Voltage supply
Temperature controller
During thin film deposition process – vapour phase precursors (growth species) are created,
generally deposition techniques are named after the method used for creating the growth
species, e.g. thermal evaporation, e-bam deposition, chemical vapour deposition. Growth
species could be atoms/ions/molecules/clusters; they are characterized by their composition,
flux and energy. These are then transported to the substrate where they will be deposited due
to processes taking place at the surface – accommodation, adsorption followed by surface
diffusion and incorporation in the film or re-evaporation. Here, it is important to mention
that use of vacuum is required during deposition to maintain the cleanliness of the surface on
which film is deposited, and to minimize the impurity incorporation in the film. Processes
occurring at the surface are important in deciding the properties of the film, hence, substrate
cleanliness, and temperature are important. The thickness of the film is monitored in real
time using a deposition rate monitor (DTM). It is based on piezoelectric properties of a
quartz crystal. While film is deposited on the substrate, it is also deposited on the crystal. By
measuring resonance frequency of the quartz crystal it can be inferred how much material is
deposited on it. After calibration, this information is used to monitor film thickness on the
substrate.
A typical system is shown in fig. 4 consisting of vacuum chamber (bell jar, vacuum pumps
and pressure gauges), substrate holder and heater that holds the substrate (it can also rotate
the substrate), thermocouple to monitor the temperature of the substrate, source of growth
species, and DTM. Vacuum is helpful in all three steps of film deposition – generation of
vapour phase (e.g. evaporation rate depends on the level of vacuum in the chamber), mass
transport from source to substrate (line-of-sight from the source to the substrate is possible in
absence of scattering by ambient gas molecules), and low contamination of the substrate
surface.
Importance of vacuum: Using kinetic theory of gases, we get the expression, Eq.1, for the
rate of arrival of atoms/molecules (F) on the substrate surface from the gaseous environment:
P
F= [1]
2π Mk BT
Where P is partial pressure of the gas species having atomic/molecular weight M and
temperature T, and kB is Boltzman constant. Assuming all incident atoms stick to the surface,
time to deposit one mono-layer of atoms on the surface can be calculated. If we are at normal
7
pressure, it will take less than a micro second to contaminate the surface. To avoid this
contamination, film deposition is done in vacuum. Also, from the point of view of joining of
the two materials which are at the atomic level, if small amount of contamination is at the
interface it will lead to poor adhesion between the layers. If the partial pressure of the growth
flux is much higher than the base pressure, we will have less contamination of the film during
growth. Therefore, the desired purity of the thin film determines the required vacuum level
in the chamber. For example, ultra high vacuums (< 10-8 bar) are desired for depositing
silicon, whereas medium level of vacuum (~10-6 bar) is sufficient for metal layer.
Surface processes: Once the growth flux arrives on the surface, the surface processes
decide the film formation, see fig. 5. For incident atom to deposit on the surface, the incident
atom must transfer all its kinetic energy to the substrate. Lateral or horizontal component of
the velocity will keep the atom mobile on the surface. If incident atom is not accommodated
it is reflected /bounced back. This is determined by the incident energy of the growth
species, thin film and substrate materials. This phenomenon takes place in very short time
and is called thermal accommodation. Once accommodated, the atom can move around on
the surface due to surface diffusion. Eventually atom gets adsorbed on the surface – either
physical or chemical adsorption. In chemisorptions, electron transfer takes place between
film and substrate atoms, whereas in physisorption there is no electron transfer, van der Waal
type of bond exists. An adsorbed atom does not necessarily become part of the film, it can
thermally desorb after some time. Relative rates of these surface processes are important in
deciding the film structure.
incident
growth species desorption
reflected
surface diffusion
thin film
substrate
adsorption
thermally
accommodated
Thermal evaporation: The two common physical vapour deposition processes used in
semiconductor industry are thermal evaporation and sputtering. The difference exists in the
method of producing growth flux from the source material. In case of thermal evaporation,
source material is heated to a temperature at which there is enough partial pressure generated
to get the desired deposition rate as per Eq.1. The source material can be heated by
resistance/arc/induction/e-beam. In resistance heating the source is heated by a resistive
filament, sometimes material to be evaporated is kept directly on the filament or in a crucible
heated by the filament. In arc heating, high current discharge is used for heating. In e-beam
deposition, electron beams from thermionic gun or plasma gun is used to heat the material.
There are several issues with thermal evaporation - temperature control, uniformity and
surface coverage that varies with distance and angle from the source, and contamination from
crucible or other components in the system. Temperature control can be attained using a
thermocouple and more accurate feedback loops to the filament. Uniformity can be improved
by rotating the wafers, using large evaporation source and proper design of the deposition
chamber.
Ar+ anode
cathode
argon gas
vacuum
pump
The extra energy provided by the accelerating ions to the sputtered atoms results in a growth
flux with higher energy. The additional surface processes taking place during sputter
deposition of a film due to presence of energetic flux are shown in fig.7. As we increase the
energy of the incident flux, higher fraction of atoms are deposited as the film, if this energy
becomes too high, energetic growth flux starts sputtering the substrate instead of deposition
(This etching capability is used in dry-etching in section 4). The implantation of the incident
atom below the surface is also possible and this leads to altered sub-surface layers and
enhanced diffusion. For the right range of the energetic bombardment, it is observed that
film density is improved in sputtering as opposed to films deposited by thermal evaporation.
incident particle
with energy E1 E2 E3
E1 sputtering
enhanced surface of film
diffusion & reaction
near surface
adsorbed
region
damage implanted
species
E1 E2 > E1 E3 >E2
Fig. 7 Schematic of additional surface processes due to presence of high energy growth
flux in sputtering.
10
It would seem from fig.6 that only conducting materials can be used as target, but that is not
true. Insulating targets with radio frequency (rf)-power supply can be used for creating the
plasma. Another variation of sputtering is magnetron sputtering, where plasma can be
sustained at lower pressures and electric field with the help of magnetic field.
Chemical vapour deposition: In chemical vapour deposition (CVD) the source material is
in gaseous form and is brought in contact with the surface, where it adsorbs and reacts to
form the film, and by-products of this reaction are desorbed and exhausted out of the
chamber. The deposition system is simple compared to the PVD techniques as shown in
fig.8(a). The deposition rate depends on the availability of the precursor gas at the reactant
surface and reaction rate. At high temperature, reaction rate is fast and film deposition is in
flow controlled regime. In this region, temperature dependence is small. At low
temperatures, deposition is reaction rate controlled and exhibits a strong dependence on
temperature as shown in fig. 8(b).
gas
outlet
deposition rate
gas adsorption desorption
inlet Mass flow Surface
control limited reaction
reaction regime limited
regime
1/T
(a) (b)
Fig. 8. Schematic of (a) a CVD system and (b) different regimes for CVD deposition.
There are many variations of this process. CVD can be done at atmospheric pressure
(APCVD) or at low pressures (LPCVD). LPCVD provides better uniformity in a batch-type
reactor where many wafers are stacked and processed together. Low pressure allows uniform
distribution of the reactants throughout the chamber. Also, we have plasma enhanced CVD
(PECVD) which is often used to increase the reaction rate and this allows film deposition at
lower substrate temperatures. Metalloragnic CVD (MOCVD) uses metallorganic precursors
during deposition that dissociate at lower temperatures to form the film. This allows film
deposition to take place at a lower temperature.
11
Thermal oxidation: Instead of depositing the material, we can also grow a film by
oxidizing the substrate, e.g. SiO2 films are grown on Si by thermal oxidation. These films are
free of contamination, have an excellent Si/SiO2 interface as it is never exposed to the
atmosphere. Additionally, it only requires oxygen gas (dry oxidation) or water vapour (wet
oxidation) and a high temperature furnace. This is the preferred process to deposit SiO2 over
electrochemically grown or CVD deposited oxides as it exhibits superior dielectric property.
Therefore, for gate insulator in a metal-oxide-semiconductor transistor, we use thermal
oxidation.
diffusion of
C0 oxidizing species
Oxide thickness
Cg
Parabolic region
Cs Silicon
SiO2
linear
Si/SiO2 region
interface
Fig. 9. Schematic of the SiO2 growth mechanism on silicon and thickness vs. time curve.
Fig. 9 shows the processes involved in oxidation of silicon. First the oxidizing species adsorb
at the surface, followed by diffusion through the oxide layer, and reaction at the Si/SiO2
interface contributing to growth of oxide thickness. In the initial period of oxidation, when
oxide thickness is negligible, the oxide growth is linear with time and controlled by surface
reaction rate. As oxide thickness increases, diffusion of oxidizing species through the oxide
becomes rate controlling step and growth rate is parabolic with time. The oxidation rate is
dependent on oxidizing species, orientation of the silicon, doping of silicon, flow rate of
oxygen and temperature.
Thin film growth models: Depending on the technique in use and processing parameters,
it has been observed thin film grow in three modes. These are referred as island like growth,
layer-by-layer growth and mixed growth model as shown in fig. 10. Island growth takes
place when interaction between film atoms is greater than the interaction between film and
substrate atoms. Layer-by-layer growth takes place when interaction between substrate and
12
film atoms is stronger. Mixed model is when first few monolayers are grown layer-by-layer
and after that influence of substrate wanes off leading to island type growth.
island
(a)
monolayer (b)
(c)
Film structure: Finally, irrespective of the technique used, processing parameters are
optimized to get desired film property and microstructure. One of the important properties is
film macro-structure – we need films having smooth morphology and free from voids and
pinholes. They should adhere well, and provide good step coverage; stress in the film should
be controlled to maintain adhesion. These defects are shown schematically in Fig. 11. In
addition to this, uniformity of the film thickness is an important parameter. For better
uniformity growth flux must impinge on the substrate from many directions. A large area
source and high surface mobility promotes this.
Non-uniform thickness
Voids
Roughness Poor step coverage
Pin hole
Substrate
Silicon as a semiconductor by itself has limited utility. Its power in devices is harnessed
when a p-n junction is created by locally modifying the conductivity using n- and p- type
dopants. Typical n-type dopants are P and As, whereas p-type doping can be achieved with
B. To appreciate the quantities in which the dopant atoms are present in Si, first consider that
the number of Si atoms in a cm3 is approximately 1022. Typical doping levels, amounts by
which the Si atoms in the lattice are replaced by the dopant atoms, are 1016-1020 cm-3; amount
at the lower end of the range is for creating junctions and at the higher end for realizing
ohmic contacts. This doping in microelectronic fabrication is carried out mostly by what is
known as diffusion or ion implantation, both of which we describe next.
Diffusion: The diffusion method of doping in reality is a combination of several steps. The
basic approach is (i) to expose the Si surface to a dopant source, (ii) then to carry out a low
temperature and short duration diffusion of dopant, incorporating it only in a thin surface
layer of Si, (iii) followed by removal of the dopant source introduced in the first step, and (iv)
finally driving-in or redistributing the dopant that was incorporated in the thin surface layer
during the second step deeper into Si by diffusion process under appropriate temperature and
time combinations.
The dopants are introduced at the surface using a solid sources or gas sources. A common
and practical solid source is a spin-on-glass, which as the name suggests is coated on the Si
wafer by spin coating. The dopant is mixed in an organosilane, which is a liquid of viscosity
suitable for spin coating. After coating a thin layer, a heat treatment converts the
organosilane to a SiO2 matrix, within which lies the dopant atoms. For example, to dope with
B, carborane may be used, which by heat treatment converts to B2O3 in SiO2 matrix. With
this type of source, boron concentration at the Si surface is established by this oxide reacting
at the Si surface according to 2B2O3+3Si = 4B+3SiO2.
Alternative solid, liquid or gas sources are possible. In most cases though, the dopant is
finally transported to the wafer in gaseous form. If the source were solid, it will be heated to
provide enough vapour pressure. Then the vapours of the dopant are transported with a
carrier gas to the Si wafer. Similarly, for a liquid source, for example, trimethylborate for B
doping, a carrier gas is bubbled through the liquid. This carrier gas saturated with the dopant
then forms the source of dopant. If gas source of dopant is available, then it obviously can be
14
transported in similar manner. The problem with this approach in contrast to spin-on glass is
that it is difficult to maintain uniformity in a long diffusion furnace in which hundreds of
wafers may be stacked from one end to another. Thus, a spin-on-source may be more
attractive.
The dopant brought in contact with the Si surface is then allowed to diffuse into Si wafer.
The governing diffusion equation for establishing concentration of dopant at any location and
time is
∂ 2C ∂C
D = [2]
∂x 2 ∂t
In this we have assumed that diffusion coefficient of the dopant, D, is a constant and its
concentration, C (x,t), varies with position measured from the surface, x, into the depth of
silicon wafer and time, t; that is, Si lies in x>0. Step (ii), the pre-deposition step, represents a
case of diffusion in which the surface concentration of the dopant is fixed. In practice the
diffusion source is so high in dopant concentration that the solubility of the dopant species in
the silicon, Cs, determines the surface concentration. Hence, for the boundary condition,
C(0,t)=Cs and C(∞,t)=Co and initial condition C(x>0,0)=C0, the solution to the diffusion
equation is
C ( x, t ) − C0 x
= erfc [3]
CS − C0 2 Dt
This, Co is the background concentration of dopant already present in the wafer, if any;
otherwise, it will be zero. Figure 12, shows the concentration profile for increasing time
scales.
C (x,t), log scale
C0 t= 0
x
15
∞ 1/ 2
Dt
Q0 = ∫ c( x, t )dx = 2CS [4]
0 π
When in Step (iii), the dopant source is removed, for example, by etching off the spin-on-
glass, Qo amount of dopant remains within small depth at the surface. In Step (iv), this
retained dopant is redistributed by another drive-in diffusion. In this case, the concentration
of dopant in Si changes spatially by diffusion, but in a manner that total dopant remains Qo;
hence the terminology used is re-distribution.
The solution of the diffusion equation for boundary condition changed to
∂c ( 0 , t )
=0
∂x (impermeable barrier at x=0) is a Gaussian as given below
− x2
' Q0
C ( x, t ) = e 4 D 't ' [5]
' '
πDt
If we started with a n-type Si wafer uniformly containing 1016 cm-3 phosphorous ([P]) dopant
uniformly, a p-n junction can be obtained by diffusing boron ([B]). The junction depth, xj, is
located where the [B] concentration also becomes 1016 cm-3. In Figure 13, Gaussian profile
of the diffused dopant B is shown at various times. For all time scales, the total amount of
dopant remains Qo, but longer the diffusion time, greater is the junction depth.
16
, , ,
t 3> t 2> t
t ,2
,
t 3
-3
16
10 [P ]
1 2 3
x J
x x
J J x
As shown in Figure 14, an ion beam is rastered on the surface of the wafer to cover the entire
area. Essentially it means only one wafer can be processed at a time. But, in a modern
implanter, with automatic loading and unloading, the wafer throughput can be speeded up. In
any case, demands of high process capability in integrated circuits which can only be met by
ion implantation at acceptable production rate has made ion implantation the work horse of
doping technology in integrated circuit fabrication.
Ion beam
Si Wafer
Fig. 14 The rastering of the wafer by ion-beam to achieve uniform doping across the
wafer.
The energetic ions that impinge on the wafer surface come to rest after approximately 10-100
nm depth. This penetration depth is referred to as range (R) of the ions. Range of a given ion
in a specific semiconductor primarily depends on the energy of the ions. Thus, by changing
the energy, it is possible to implant the dopants at a prescribed depth, unlike in diffusion
method which always forces highest concentration of dopant at the surface. This is first
advantage over the diffusion process.
Further, the ion flux can be controlled independently so that the amount of dopant introduced
is adjustable. If the implant time for ions of charge state Z is t, the dose of dopant (number of
I ×t
dopant atoms per unit area of the wafer) introduced is Q o = b , where q is
Zq × A
electronic charge and A is the implanted area. This dose, available in the range 1010-1017
ions/cm2, is distributed in the wafer such that peak concentration of dopant in the wafer can
be in 1014-1020 cm-3. It may also be noted that the control in achieving this concentration is
extremely good control, within a few percent.
18
The second major advantage over the diffusion method of doping is due to the fact that these
two independent “handles” to control the depth at which dopant is implanted and its amount
is available with extremely good control.
Wafer
Substrate holder
Ion beam
flux
To vacuum
A
A Ammeter or
A
Current integrator
∫ I (t )dt
0
number of dopant atoms implanted = [6]
q
∫ I (t )dt
0
Q0 = [7]
qA
19
Since monitoring the ion flux would be difficult, this current integrator provides a convenient
method for real time control of ion dose, or equivalently, the number of dopant atoms
implanted per unit area of the wafer.
The ions that impinge on the wafer are scattered by the lattice. The primary mechanisms are
nuclear stopping and electronic stopping. The former is due to collision of ions with the
target (wafer) atoms and electronic stopping is due to transfer of energy to electrons by
coulombic interaction. In the process, excitation or even electron ejection (ionization) is
possible. By either mechanism, as the energetic ions entering the substrate come to rest, the
distance an ion has traveled is called the range. Being a probalistic event, some ions undergo
more collisions than others and also recoil differently in direction perpendicular to the beam
direction. Hence, there is a distribution in positions where these ions stop. For example, in
fig. 16, we are showing several ions which have come to rest at different locations in the
wafer. Rather than the range, we are more interested in its projection in the –x direction, the
depth into the wafer. Corresponding to the way these ions stop, the concentration of dopant
(cm-3) in –x direction is also shown. The peak of dopant atoms lies at the projected range, RP,
and the standard deviation in dopant concentration is ∆Rp. Unfortunately, the same
randomness in process leads to lateral distribution of the dopant perpendicular to the implant
direction, that is, there is also a lateral straggle of the dopant. That distribution has a standard
deviation of ∆Rd.
20
y or z
y or z
∆Rd
∆RP
RP
Fig. 16. Projected range and lateral straggle of ions during implantation.
Fundamentally, this depth to which the ions are implanted, that is RP, is dependent on the
energy of the ions; it is greater for larger implant energies. Thus, in figure 17, the dopant
concentration profile marked A can be shifted deeper to position marked B into the wafer by
increasing the ion energy. The shape, especially in the tail region may also change. On the
other hand, changing the dose, by either changing the ion beam flux or the time for
implantation, concentration profile A can be shifted up to higher concentrations (marked C).
In all cases
∞
∫ C ( x)dx = Q
0
0 [8]
21
C(x) C
A B
RP RP’
x
Fig. 17. Effect of ion beam energy and dose on the implantation profile.
There are additional advantages of ion implantation over diffusion method of doping.
Because diffusion is mostly an isotropic process, the lateral diffusion is as much as in the
depth. In ion implantation, the lateral spread of dopant is much less. Apart from that,
diffusion is also a high temperature process, whereas ion implantation is performed at room
temperature. An important consequence of high temperatures in diffusion method is also that
the mask that protects the regions on wafer where dopants are not desired will have to be
protected by a hard layer such as silicon nitride or oxide. But, because of low temperatures in
ion implantation only a 2-5 µm thick photoresist is sufficient to stop the ions.
However, ion implantation does pose several problems. The matter of throughput has already
been discussed. But, most important is the safety issue. Radiation hazard, high voltages and
toxic gases require elaborate safety arrangements. Also, the equipment takes up a large area
and is expensive.
Apart from equipment related issues, there is an important process and materials related
issue. The implantation process leads to damage in the target crystal, the Si substrate. This
damage could be simple creation of point defects or dislocation loops and even cause
amorphization. Thus, it is necessary to anneal the material to repair the damage, which bring
back the high temperature process. However, since the objective is to only anneal the
damage, it is possible to carry it out for short duration. This annealing of Si at elevated
temperatures may not cause much problem, but in materials such as GaAs, where preferential
loss of As may occur, additional provisions to cap the wafer have to be made. Further, the
dopants that have been introduced tend to diffuse at annealing temperatures. And this
22
diffusion is often accelerated due to presence of excess defects, leading to what is known as
transient enhanced diffusion.
4.0 Photolithography/patterning
A semiconductor device on a Si wafer is fabricated by adding many layers in specific
patterns. These layers may be insulators, such as SiO2 or Si3N4, or metals of many kinds.
Unfortunately, although we desire a patterned addition of these layers, most processes are
designed to add layers on the entire area of the wafer. Thus, after adding a layer on a wafer,
portions of these deposited layers must be removed to achieve the desired pattern.
Two strategies are possible. More common is, first, an addition of a layer on the whole wafer
followed by developing a protective pattern of photoresist (PR) on the wafer and then etching
off the underlying layer by a chemical which attacks it where the protective PR is not present.
Ultimately the PR is also removed, with net result that a patterned layer has been added on to
the wafer. Since the job of PR is to protect the underlying layer from etching, the PR should
be less susceptible to chemical attack than the layer which is being patterned. Often this
requirement is met by PR. But, when etchant is likely to attack the PR also, then alternatives
are possible; those specific cases are not discussed here, but in general, this requires
depositing an additional hard layer which first is patterned by PR and then this layer plays
that same role as PR when underlying original layer is being patterned. In both these cases,
however, first a layer is deposited and then patterned, in contrast to the second strategy, as
below.
In some cases, a process called “lift-off” is also practiced in which order of layer deposition
and PR patterning is reversed. That is, instead we develop a pattern of PR on the wafer first
and then deposit the layer on the entire surface of the wafer. The trick is to ensure
discontinuities in the deposited layer at the edges of pre-existing PR pattern. The idea is that
when PR is dissolved in appropriate chemical, the discontinuities allows the chemical to
access PR; this chemical attacks PR until all of it is removed. What is then left on the wafer
is a patterned layer of material which was deposited.
In both of these strategies, photolithography plays an important role. This term “photo”
appears because a light source is used in the process. But, when pattern dimension becomes
smaller, below even wavelength of UV light, then the light source is replaced by e-beam or x-
rays. In those cases, terminology becomes e-beam lithography or x-ray lithography. Both of
23
these methods come with their own fabrication peculiarities, but fundamentally they are
similar to photolithography; therefore, here we discuss only photolithography.
SiO2
(a)
Si
Coat with PR
PR (b)
Soft Baking
(c)
UV UV
(d)
Develop
(e)
Strip PR (h)
The main photoactive material is either termed positive or negative. The reason for this
terminology will become apparent once their application is described. At this stage, let us
understand that once a film of positive PR is formed on a wafer and then the wafer is
subjected to a chemical called developer, PR will not dissolve. But, the photoactive
compound in positive PR is such that when it is exposed to UV light, the PR turns soluble in
same developer. An example of such substance is diazonaphthaquinones, though most
25
commercially available PR will not disclose the chemicals used. The matrix material by
itself may have been soluble in developer, but addition of the photoactive substance makes it
insoluble until exposed to UV light. In contrast, the photoactive agent, and the matrix
material, in a negative PR is normally soluble in a developer. But, when it is exposed to UV
light, the cross-linking reaction in photoactive substance, such as bis-arylazide, leads to a
network that makes PR film insoluble in a developer.
Let us now describe the use of photolithography process with help of an example in which we
desire a pattern of SiO2 on Si wafer as in Fig. 18 (h). We use the schematic in Fig. 18 to
illustrate the step by step process. (a) a layer of SiO2 is deposited on a Si wafer by one of the
several available methods, thermal oxidation, sputtering or chemical vapor deposition. This
is the layer to be patterned; (b) a PR film, negative or positive, approximately 2-3 µm thick,
is spin coated on the wafer. That is, a measured quantity of PR is dispensed on the wafer and
the wafer is spun fast so that the PR spreads out before the solvent dries off. In order to
achieve good adhesion to the wafer, it may be first necessary to bake the wafer to remove
adsorbed moisture and/or give a surface treatment. Negative PR usually provided good
adhesion, but as feature sizes have become smaller, positive PR has become more important
where much more care is necessary for good adhesion; (c) during spin coating most of the
solvent is already removed. But, in a soft bake process, by heating at a low temperature
around 100 oC, all of the solvent is driven off; (d) the wafer is then exposed to UV light
through a mask, parts of which is meant to block the UV light. Notice the opaque and
transparent regions of the mask, chosen according to the resist used. The mask could be a
glass plate on which the opaque region, for example, could be due to Cr coating. The
negative resist, which is normally soluble in a developer, when is exposed to UV light has
turned insoluble, whereas the positive resist, normally insoluble, has turned soluble wherever
it is exposed to UV light; (e) the PR is subjected to a developer which removes the soluble
portion of the PR. Prior to this step, sometimes another soft bake may be required. The
solvents used in developer are important. Negative PR at one time used environmentally
harmful chemicals such as toluene and xylene. Thus, a gradual shift had been towards
positive PR which can be developed by mild inorganic aqueous solutions. However, now-a-
days, PRs are available in both tones which can be developed by the same solvent; (f) Hard
baking of PR to provide rigidity before etching; (g) etching SiO2 by dilute HF aqueous
solution which is chemically insensitive to underlying Si. The etchant attacks both SiO2 and
the PR. The question is of selectivity. When the etch rate of PR is much less than that of
26
SiO2, the etchant is highly selective and process will be successful; (h) strip PR by chemicals
known as strippers or by plasma ashing. The result is that both by positive and negative PR,
same pattern has been achieved. Positive PR is so named because it generates positive image
of the mask. The same logic explains terminology in negative PR.
5.0 Etching
Etching is the process by which the part is machined. This is required to remove various
layers from selective areas that are exposed using photolithography as described in the
previous section. Sometimes, there is also a need to remove a layer partially/completely for
better interfaces. We can use either wet etching or dry etching. In wet etching an etchant is
used in which the material to be etched can be dissolved, but the masking layer (PR/oxide
layer) is not dissolved. If the part to be etched is amorphous or polycrystalline in nature, then
wet etching is isotropic. If it is single crystalline the differential etch rates of different planes
can lead to profiles depending on the orientation of the silicon wafer. For more details on this
part, which is more relevant to MEMS, reader is referred to reference [2]. Figure 19 shows
the etch profile development in an amorphous layer with time during wet-etching. Due to
isotropic etching, there is possibility of forming a ledge, this can be troublesome for later
processing. Also, the width of the etched trench changes with depth (see fig. 19 (b)) which
may not be desirable.
Etchant solution
HF: DI
photo resist
Mask layer
SiO2 SiO2 layer
Layer to be
patterned t Silicon
(a) (b)
Fig. 19. (a) Isotropic wet etching profile (b) selective etching of SiO2 using HF based
etchant solution.
Process parameters that can change the kinetics of dissolution are – etchant solution, its
concentration, temperature, stirring or aeration of the bath, and etching time. These are to be
carefully monitored in order to get the desired profile in the fabricated part. Additionally,
etching also has an effect on the roughness of the surface. It needs to be minimized;
27
otherwise it can deteriorate the adhesion with the next layer that will be deposited. High etch
rates lead to rough surfaces. Therefore, to improve surface finish lower concentrations of the
etchant and lower temperatures are used. If any material needs to be used in the
microelectronics industry, it is required to have well controlled etching process. For a very
long time, this was a problem for using copper metal in the microelectronics industry.
In silicon technology, selectivity of HF: DI solution to etch SiO2 is used to open windows in
the SiO2 mask. SiO2 has a very high etch rate in HF: DI solution compared to silicon this
becomes self terminating machining process, see Fig. 19. In the semiconductor industries,
we can wet-etch Si3N4, Si, and metals. More details about various etching chemistries can be
found in ref [6].
The problem of dimensional accuracy exists in the wet-etching. The dimension of the
opening at the silicon surface is not the same as opening of the windows. Amount of under-
etching and over-etching can also decide the final opening on the silicon surface. Therefore,
design rules are developed to correct for these parameters when designing the
photolithography mask.
Additionally, due to isotropic nature of wet etching, the machined walls are not vertical. Dry
etching eliminates this problem since it has highly anisotropic etch rate, as shown in Fig. 20.
It is like mechanical drilling using energetic directional ions. These energetic directional ions
remove the material to create a hole. This process is physical removal of material. These
ions can be of an ion-beam of an inert gas, such as argon. It is also well known as Argon ion
milling. Sometimes plasma of a reactant gas is used that additionally removes the material by
reacting with it, by-products are subsequently removed. This process is also called reactive
ion etching (RIE). Generally floro and chloro carbon gases are used in dry-etching; however,
there are other options available depending on the layers to be etched.
Dry etching processes involving physical sputtering are not highly selective of different
layers; hence the end-detection is necessary in deciding when to stop etching. This is often
done by monitoring the reflectance of the surface or chemical composition of gases in the
chambers which will change as the process proceeds from one layer to the next.
Additionally, the mask layer needs to have enough thickness to survive the full duration of
etching process.
28
Etching Plasma
Mask layer
Layer to be pattern
Advantages of dry-etching are in providing vertical walls in the etched holes and better
dimensional accuracy. Moreover, as dimensions of the holes are reduced, wet-etchant may
not reach the bottom of the hole, but gas atoms/ions/molecules having the dimensions on the
order of atomic scale are likely to reach there. Finally, dry-etching processes reduce the wet-
chemical waste from the microelectronics industries.
The disadvantage of dry-etching is in terms of damage caused by energetic ions to the layer.
This at times can change the device characteristics by introducing defects in the material.
The cleaning process to be used depends on the materials on the surface. For example, we
can use a HF containing process for cleaning silicon surface, but not silicon oxide surface as
HF etches the oxide surface. A typical cleaning process that is used in silicon processing is
RCA cleaning ( named after Radio Corporation of America). Following steps are used in the
cleaning process, the concentration and temperature are varied by the user depending on
requirement.
All cleaning processes are terminated with a final DI water rinse followed by N2 gas drying.
Seemingly simple process, but needs to be controlled as water marks can be reason for
defective device.
particle
patterned layer
Silicon
Fig. 21. Blocking of the patterned window by a particle of similar dimension.
In order to reduce the chemical and DI water consumption, industries are also developing
dry-cleaning-processes where the vapors of the cleaning agents are brought in contact with
the surface.
30
The development in this industry is also well organized. Semiconductor Industry association
of five different countries – USA, Taiwan, Japan, Korea, Europe sponsor an organization
called International Technology Road map for Semiconductors (ITRS). The chip makers,
equipment makers, consortia and universities of these countries are members of it. It gives
future requirement of the semiconductor industries for next 15 years and these tables are
updated every year. The road map is divided by different technology groups. This exercise
has been instrumental in realizing the prophecy of Moore’s law and has provided directions
for research and development world-wide.
REFERENCES
1. Andrew S. Grove. 1967. Physics and technology of semiconductor devices. John Wiley and
Sons, Singapore
2. S.M. Sze. 1994. Semiconductor Sensors. John Wiley & Sons, Inc., New York.
3. S. Wolf and R.N. Tauber 1986. Silicon processing for the VLSI era, Vol. 1 (Process
Technology. Lattice Press. California, USA.
4. S. Wolf. 1990. Silicon processing for the VLSI era, Vol. 2 (Process Integration). Lattice
Press, California.
5. James W. Mayer and S.S. Lau. 1990. Electronic Materials Science: For integrated circuits in
Si and GaAs. Macmillan publishing company, New York.
6. S.K. Gandhi. 1994. VLSI fabrication principles. John Wiley and Sons, Inc., New York.
7. B. G. Streetman. 1980. Solid State electronic Devices, Prentice-Hall series in solid state
physical electronics. Ed. N. Holonyak, Prentice-Hall, Inc., NJ.