Chapter 3

Defects in Semiconductors

3.1 Introduction
In an ideal crystal lattice, each atom is at its designated position and deviations from
this perfect structure are called imperfections or defects. These defects may introduce
electronic energy states into the semiconductor band gap, which can be placed into
two categories: shallow levels and deep levels. Shallow levels are located near their
related band edges (valence band for acceptors and conduction band for donors) i.e.
~0.1 eV from the band edge, thus these levels are thermally ionized at room
temperature. The ionization energy of a shallow level can be approximately described
by a modified hydrogenic model [1]. For example, a shallow donor resembles a
hydrogen atom with a positive nucleus binding an electron. Impurity elements which
are used as dopants in semiconductors normally create these shallow levels which are
ionized at room temperature and provide free carriers to form p-type or n-type
semiconductor. Deep levels are those defects positioned deeper in the band gap than
the dopant levels and are found to bind the carriers much more strongly into highly
compact, localized states. The deep levels have higher ionization energies, therefore
contribute very little to the free charge carriers. Defects with deep levels in the band
gap are often referred to as, ‘traps’, ‘recombination centers’, or ‘generation centers’.
Deep levels are important in semiconductors since they modify the properties of the
semiconductors and therefore, those of the devices fabricated thereon. Deep levels are
desirable in some applications, e.g. in fast switching devices, where they can be
exploited as recombination centers which quickly remove minority carriers,
enhancing the device’s switching speed thereby increasing efficiency [2,3]. Deep
levels may also be a nuisance if present in semiconductors that are used for
photovoltaic applications since they reduce the cells’ efficiency by allowing created
electron-hole to recombine. Thus deep level study is of paramount importance in the

24
semiconductor device industry so that those deep levels which are useful can be
deliberately added and those that are deleterious can be reduced or eliminated. This
chapter will outline the common properties of deep levels (i.e. structure, charge states,
formation and migration mechanisms) in silicon and germanium.

3.2 Primary Defects

Defects in semiconductors can be divided into two main categories; point defects and
extended defects. Point defects are not extended in space in any dimension and this
implies that the perturbation of the lattice is localized about a lattice site and involves
only a few nearest neighbours. There are two kinds of point defects of great interest in
semiconductor crystals, intrinsic (e.g. vacancies or self-interstitial) and extrinsic point
defects (e.g. impurity atoms occupying substitutional or interstitial lattice sites. Small
agglomerations of several point defects like divacancies, vacancy-impurity
complexes, vacancy-donor etc are also generally considered as point defects.
Extended defects are extended in nature (such as, grain boundaries, dislocations or
stacking faults). The discussion in this section is focussed more on point defects
which are more relevant to the work covered in this thesis.

Vacancy Foreign Interstitial

Frenkel
pair

Self-Interstitial Substitutional atom

Fig. 3-1. A diagram showing the vacancy, self-interstitial, substitutional defects.

3.2.1 Vacancy Defect

If an atom is removed from its regular lattice site the empty lattice site is called a
vacancy defect (V), and is shown in Fig. 3-1. The vacancy in some semiconductors
(e.g. in Ge and Si) can have up to five charge states, V++, V+, V0, V- and V=. In order

25
to form a vacancy by removing an atom from its lattice site four bonds are broken in a
diamond crystal structure as shown in Fig. 3-2a [4,5]. The broken bonds can form
new bonds depending on the charge state (i.e. which is just the number of electrons
occupying the dangling bonds) of the vacancy (Fig. 3-2b and c). This causes small
inward and outward displacement of neighboring atoms, which either preserves the
local symmetry (relaxation) or alters it (distortion). The amplitude of the
displacements depends on the charge state of the defect. Another geometric
configuration for describing a vacancy is the ‘split-vacancy’. The ‘split-vacancy’
results when an atom resides at the bond center between the empty sites [6], Fig. 3-2d.
The ‘split-vacancy’ is often important primarily to help describe the transition state in
vacancy migration [7]. The lattice relaxation depends on the charge state of the point
defect (Jahn-Teller effect). Jahn-Teller effect is simply a geometrical distortion which
occurs when the electronic state is degenerate, in which case the nuclear state is
unstable. Atomic displacements always exist which by lowering the symmetry, split
the degenerate level.

(a) (b)

(c) (d)

Fig. 3-2. The vacancy configuration in diamond lattice. (a) Four bonds are broken in
order to create the vacancy. (b) When there is one electron per dangling bond (i.e.,
for the neutral vacancy V0) they form two new bonds leading to local distortion. (c)
When an electron is missing (i.e., for the positive vacancy V+) one of these two bonds
is weakened since it contains only one electron. The distortion is thus different from
that in the case of V0. (d) The ‘split-vacancy configuration, redrawn from ref. 4.

26
It should be noted there is general agreement in defect modeling studies that the
vacancy formation energy in germanium (1.7 eV – 2.5 eV) is significantly smaller
than in silicon (~4.0 eV), for all charge states. When two neighboring atoms are
removed and also when two migrating vacancies meet and combine a divacancy is
formed. A divacancy can also exist in four different charge states in Si [8].

3.2.2 Interstitial Defect

Interstitials are atoms, which occupy a site in the crystal structure, which is not a
regular lattice site as shown in Fig. 3-1. An interstitial defect can be of the same
species as the atoms of the lattice (self-interstitial) or of a different nature (an
interstitial impurity). Interstitials are generally high-energy configurations. Once
again, the introduction of an interstitial induces a relaxation and distortion of the
lattice, which surrounds it. The type of configuration the interstitial assumes depends
on its ability to make bonds with its neighbors and therefore can change with its
charge state. A nearby interstitial defect and vacancy defect is called a Frenkel pair.

3.3 Secondary Defects

The isolated lattice vacancy (V) and self-interstitial (I) are primary defects produced
after high-energy particle irradiation in semiconductors. The primary defects are
mobile at low temperatures, e.g. in Si the vacancy becomes mobile above 150 K -
200 K and the interstitial is mobile even at 4.2 K [8]. Therefore, deep level transient
spectroscopy measurements of room temperature irradiated silicon will not reveal an
isolated vacancy or silicon interstitials. The V and I which survive the recombination
of simple defects can diffuse into the semiconductor and interact with other intrinsic
and extrinsic defects giving rise to complex room temperature stable defects, some of
which are depicted in the schematic diagram shown in Fig. 3-3. For example, in
silicon when a vacancy becomes mobile, it can be trapped by an oxygen atom to form
a V-O complex (A-center), or the doping impurity (e.g. P) to form a V-P complex (E-
center), or by another vacancy to form divacancies. Other complex defects are via the
mobile interstitial (I). It is expected that room temperature defect evolution in silicon
shown in Fig. 3-3 should hold similarly for germanium.

27
 V-V

V-O
V
V-P

Implanted ion
BI-Bs

BI-Cs
BI

BI-OI
I

CI-OI
CI
P-CI

CI-Cs

Fig. 3-3. Schematic of room temperature defect evolution in crystalline silicon.

3.3.1 The Divacancy

The divacancy is formed by the removal of two neighbouring atoms. Generally
divacancies can be created in semiconductors by particle irradiation either as a
primary defect, when collision cascade is dense enough, or as a secondary defect by
pairing of single vacancies diffusing randomly. The vacancy in germanium is
negatively charged in a broad interval of Fermi level position in the band gap,
therefore the formation of divacancies by pairing of single vacancies is suppressed by
the coulombic repulsion of the vacancies [9]. The divacancy in Ge has not yet been
identified by experimental techniques. Using density functional theory (DFT) cluster
calculations, Janke et al [10] estimated the energy barrier for migration and
dissociation of the divacancy. The dissociation energy consisting of the binding
energy between two vacancies and the migration of a single vacancy was found to be
between (1.5-1.7) eV whereas the migration barrier of the divacancy was found to be
1.1 eV. This corresponds approximately to a thermal stability of 420 K. The

28
divacancy in silicon is well known, and can appear in four charge states, V2+ , V20 , V2−

and V2= .

3.3.2 The E-center

The E-center can be described as a vacancy trapped next to a substitutional donor
atom. The E-center can be formed as a primary defect or when the impurity atom
captures a mobile vacancy. As to the formation of an E-center, although local atomic
strain effect may need to be considered, the key role is played by the coulombic
interaction between a positively charged antimony and a negatively charged vacancy
[11]. Let’s consider the E-center in phosphorus-doped silicon, V-P. In the neutral
charge state i.e. PV0, two of the three silicon atoms surrounding the vacancy pull
together to form an electron pair, leaving an unpaired electron in the orbital of the
third silicon atom, while two electrons with antiparallel spins are accommodated by
the phosphorus atom. When the Fermi level is above the E-center an extra electron is
accepted and becomes negatively charged, PV¯ . The formation of E-center in silicon
removes two electrons in the conduction band by converting a positively charged P
donor atom to a negatively charged V-P center. The formation of the E-center is
regarded as an intermediate step in dopant diffusion in silicon and germanium [12].

3.3.3 The A-center

The A-center (V-O) may be regarded as a vacancy trapped next to an oxygen atom in
an interstitial position. Similar to the E-center the A-center can be formed as a primary
defect or when an oxygen impurity traps a mobile vacancy. The A-center competes for
the vacancies with the E-center and its concentration is dependent on the relative O
impurity concentration in the sample. The A-center has also been found to be an
efficient recombination center [13], and therefore can be used as to control minority
carrier lifetime in silicon for fast switching device application.

3.3.4 Other Complex Defects

The mobile Si interstitial (I) diffuses and will replace either carbon (C) or group III
impurities e.g. boron (B) (depending on the relative concentration of the two species)
through the Watkins replacement [14,15] to form interstitial carbon (CI) or interstitial
BI respectively. The CI or BI are mobile at room temperature and will eventually form

29
defect complexes with other impurities {e.g. interstitial boron – substitutional boron
(BI-Bs), interstitial boron – interstitial oxygen (BI-OI), interstitial boron –
substitutional carbon (BI-Cs), interstitial carbon – interstitial oxygen (CI-OI) or
interstitial carbon – substitutional carbon (CI-Cs)} as shown in Fig. 3-3. For a
particular defect with a large concentration, it tends to aggregate as the temperature
increases from room temperature. In case of a divacancy, when mobile or after
dissociating, it can form trivacancies, quadrivacancies, pentavacancies, and higher
order defects. This behavior should also be true for self-interstitials and for any type
of extrinsic defects.

Most of the primary and secondary defects discussed in the previous sections are
electrically active and introduce deep levels in the semiconductor band gap. A deep
level may act as a minority carrier trap, majority carrier trap or recombination centre
depending on its position in the band gap and on relative capture cross-section of
minority and majority carriers. A majority carrier trap is an electron trap in n-type
semiconductor or a hole trap in p-type semiconductor. Conversely a minority carrier
is a hole trap in n-type semiconductor or an electron in p-type semiconductor. If a
majority- or minority- carrier lives a mean lifetime in the captured state and is
thermally ejected to the band from which it came, the center may be regarded as
majority carrier trap or minority carrier trap respectively. From defect spectroscopy
measurements such as deep level transient spectroscopy (DLTS) it is possible to
extract the defect properties such as the concentration, energy level, and capture
cross-section of defect level. The capture cross-sections, σmajority and σminority can now
be used to deduce whether the defect will act as minority carrier trap, majority carrier
trap or a recombination centre.

Recombination centers are deep levels with approximately equal capture cross-
sections for both electrons and holes and these centers are normally located near the
middle of the band gap. After capturing a majority carrier, if the majority carrier stays
trapped at the center long enough for the trap to capture a minority carrier, then
recombination takes place and the center is acting as a recombination center. Most of
the defect spectroscopy techniques measure the defect concentration, energy level,
and capture cross section for majority and minority carrier traps, e.g. when using
DLTS, all the detected defects are in a situation where they behave as majority or

30
minority carrier traps. Hence it is difficult to say which of the detected defects will be
a recombination center. There have been attempts to improve on the technique used to
distinguish traps from recombination centers. Markvart et al [16] developed an
improved version of DLTS, known as recombination DLTS that can be used to
identify defects that act as recombination centers. These centers act as “stepping
stones” for carriers and contribute to the current-voltage characteristics of rectifying
junctions at a recombination rate, U given by [3,17]

σ pσ n vth ( pn − ni2 ) N T
U= (3.1)
σ n [n − ni exp{(ET − Ei ) / kT }] + σ p [n + ni exp{− (ET − Ei )}]

where Ei, ET, NT, ni, n, p, σn, and σp are the intrinsic Fermi level, defect level, defect
concentration, intrinsic carrier density, electron concentration, hole concentration,
electron capture cross section and hole capture cross section respectively. From
equation (3.1) it is clear that the recombination rate is higher for larger ET (i.e. most
efficient recombination centers are those close to the middle of the band gap and with
similar capture cross-sections σn and σp).

3.4 Theory of Displacement of Atoms in Solids

When passing through matter high-energy particles are decelerated and in the process
transfer energy to the material. The transferred energy can modify the structure and
properties of the materials. In case of crystalline semiconductors, particle-induced
materials modification will occur as long as the projectile particles can transfer
energy, E, larger than the displacement energy, Ed, to the lattice atoms [3]. The
capacity of a solid to slow a projectile is called the stopping power, and is defined as
the amount of energy lost per unit length of trajectory in the solid. The stopping
power depends on the type and energy of the projectile and on the mass of the target
material.

3.4.1 Energy-Loss Mechanisms

When a particle enters a target material there are two main mechanisms that cause an
energy loss: (a) elastic collisions with the nuclei of the target material (nuclear
stopping) and (b) inelastic collisions with bound or free electrons (electronic

31
stopping). In electronic stopping, the term inelastic is used to signify that the
collisions may result both in the excitations of bound electrons of the medium and in
the excitations of the electron cloud of the ion. The relative effect of the two
mechanisms depends on the mass of the target material as well as the energy, mass
and charge of the incident particle. Fig. 3.4 is a “universal” diagram showing the
nuclear stopping (dε / dρ )n and electronic stopping (dε / dρ )e in terms of Thomas-
Fermi (TF) ion energy, ε and path length, ρ as a function of ε1/2 (which is proportional
to the velocity of the implanted ion) [3,18]. The parameters ε and ρ are dimensionless
quantities which can be expressed in terms of laboratory energy E and distance x,
respectively as;

4πε 0 aM 2
ε= E (3.2)
Z1Z 2e 2 ( M 1 + M 2 )

and

4M 1M 2
ρ = Nπ a 2 x (3.3)
(M1 + M 2 )

where M1 and M2 are the mass numbers of the incident and target atom respectively,
Z1 and Z2 are their atomic numbers, e is the electronic charge, N is the concentration
of atoms, εo is the permittivity of free space and a is the screening radius. The
screening radius is normally expressed by [19]

0.8853a0
a= (3.4)
( Z12/3 + Z 22/3 )1/2

where a0 = 0.529 Å is the Bohr radius. The universal curve shown in Fig. 3-4 enables
the approximation of nuclear stopping and associated quantities (e.g. damage
production and sputtering) for all particle-target combinations using a single curve.
The ion bombardment analysis can be divided into two distinct regimes.
(a) Nuclear microanalysis regime: High-energy ions are slowed down mainly
by electronic stopping. The contribution from the nuclear stopping tends to be

32
 small at high energies because fast ions have only short time to interact with
the target nuclei.
(b) Ion implantation regime: When the ion has slowed down sufficiently, the
collisions with the nuclei become more and more probable, and the nuclear
stopping finally dominate the slowing down process. In this regime nuclear
stopping reaches a maximum value (ε1) around ε1/2 = 0.6 and decreases
thereafter as shown in Fig. 3-4 and this corresponds to process associated with
low energy ions e.g. ion beam etching and sputtering.

Fig. 3.4. Nuclear and electronic stopping versus the reduced energy of an implanted
ion redrawn from ref. 3.

It is important to note that in contrast to the nuclear stopping (dε / dρ )n which

depends only on ε (i.e. independent of the type of incident particle and target atoms),
electronic stopping (dε / dρ )e can be expressed by [3]

(dε / dρ )e = kε 1/ 2 , for ε 1/2 < 14 (3.5)

where k is a function of M1, M2, Z1 and Z2, thus electron stopping does not exhibit true
ε-scaling and therefore may not be described by a universal curve.

33
 (a)

(b)

Fig. 3.5. (a) Basic parameters for an implanted ion. R is the total path length, Rp is
the projected range, ∆Rp and ∆RpL are the projected standard deviations in the
directions parallel and perpendicular to the incident beam, respectively. (b) N(x) is
the number of ions per cm3 at depth x.

The process of ion stopping is a statistical process and so some ions will undergo
many collisions, stopping in a distance shorter than the average value. The range R of
the projectile is related to its mean track length and for a projectile with initial energy
ε0 the range can be written as

0 1
R=∫ dε (3.6)
ε 0 (dε / dρ )
total

where

(dε / dρ )total = (dε / dρ )n + (dε / dρ )e (3.7)

If M 1 ≥ M 2 the projected range Rp is obtained by multiplying the range R with the

projection factor ~(1 + M2/3M1)-1. The average or projected range Rp is defined as the

34
projection of R on the incident ion beam direction, a projected standard deviation or
straggle ∆Rp is the statistical fluctuation along incident ion direction if the spatial
distribution of the implanted ions is approximately Gaussian as depicted in Fig. 3-5.
The implanted ions may also be scattered along the direction perpendicular to the
incident direction and the statistical fluctuation along this direction is the projected
lateral straggle ∆RpL. This lateral penetration of ions may limit dimensions in some
devices [20]. The ion concentration profile in the solid is related to the projected
range Rp, standard deviation ∆Rp and ion dose (fluence) Ф, (assuming Gaussian
approximation) by

Φ  (x − R p )2 
N ( x) = exp −  (3.8)
∆R p 2π  2 ∆R p2 

The profile given by equation (3.8) is often called Lindhard-Scharff-Schiott (LSS)

[21] profiles for implants in semiconductors. The assumption behind the purely
Gaussian stopping distribution described by equation (3.8) is that the implantation
takes place into amorphous material, which ignores the effects of channeling related
to the single crystal nature of the target.

Computer based simulation of implantation profiles are now possible using Monte
Carlo-based techniques such as Transport of Ions in Matter (TRIM) [22] and Stopping
Range of Ions in Matter (SRIM) [23]. These simulation codes should be treated with
caution when used for crystalline material since they assume an amorphous material
and ignore the effects of channeling, diffusion effects during and after ion
implantation and also do not account for annihilation of vacancies and interstitials and
thus overestimates the concentration of vacancies and interstitials produced by the
implantation.

3.4.2 Defect Production by Irradiation

It is important to describe the dynamics of collision because it is fundamental to
defect production in semiconductors. Assume that an incident particle with kinetic
energy E and mass M1 strikes a target atom of mass M2.

35
 M1 M2 θ
E
φ

Fig. 3-6. Illustration of the collision between an incident particle of mass M1 and
energy E with a target atom of mass M2.

The kinetic energy T transmitted to the target atom depends directly on the angular
deflection θ of the incident particle (Fig. 3-6). For purely elastic collision, (i.e. when
momentum and kinetic energy is conserved) the energy transferred is given by

M1 1 − η (θ )
T = 2E (3.9)
M 2 (1 + M 2 / M 1 )2

where E is the energy of the projectile and M1 and M2 have previously been defined
and η(θ) is a function implicitly given by

1 + (M 2 / M 1 )
cos θ = (3.10)
1 + 2( M 2 .M 1 )η + (M 2 / M 1 )
2

In the non-relativistic limit the maximum energy Tmax is transferred for θ = 0, i.e., for
η(θ) = -1, therefore

4M 1 M 2
Tmax = E (3.11)
(M 1 + M 2 )2

In the case where the mass of projectile is approximately equal to the target, the
expression of the transferred energy will simplify to

36
 Tmax = E (3.12)

For a neutron irradiation, M 1 << M 2 the energy transferred is now given by

2M 1
Tmax = E (3.13)
M2

In the case of electron irradiation, M 1 << M 2 relativistic corrections are required and
the energy transferred in now written as

2M1  E 
Tmax = E2+  (3.14)
M2  M 1c 2 

where c is the speed of light. Equation (3.14) can be approximately represented by

2148 2
Tmax = E (3.15)
Z

In this case the Z is the atomic number of the target atom, E is in MeV and Tmax is in
eV. In order for an atom to be permanently displaced from its lattice position, the
energy that it receives must be greater than the displacement energy. The minimum
energy necessary to displace an atom from its lattice position is called the threshold
energy (Td). Normally the threshold energy is assumed to be isotropic (i.e.
independent of the direction in which the atom is displaced in the lattice.

Table. 3.1. The threshold energies and corresponding minimum incident electron
energies for atom displacement in silicon and germanium.

Material Td (eV)a Emin (keV)b

silicon 21.0 370

germanium 27.5 424
a
Experimental values obtained from ref. 24.
b
Values calculated based on equation (3.14).

37
It should be pointed out that threshold energies for most materials is generally greater
than the formation energy of Frenkel pairs, because defect formation is a complex
multi-body collision process (e.g. a recoil atom can bounce back to its lattice position
or kick back another atom to its lattice position) [25]. Table. 3.1 summarizes the
threshold energies and the minimum electron projectile energy Emin necessary to
displace an atom in silicon and germanium. If an incident particle has energy much
greater than the threshold energy (Td), it will transfer energy to the target atom. The
displaced atom may in turn collide and displace other atoms, creating cluster damage,
until it eventually comes to rest, usually in an interstitial site and sometimes in a
vacancy site. Light energetic particles (such as Si, He, Ar, neutrons, electrons and
protons) tend to leave tracks of relative small defect concentrations. These ions
initially slow down mainly by electron stopping process with little displacement
damage until eventually nuclear stopping becomes dominant at the end of their range.

Fig. 3-5. Schematic of ion track in a solid, and associated damage for a light ion (top)
and a heavy ion (bottom), redrawn from ref. 20.

Therefore there is generally little lattice damage along the track except for the end of
the range. Heavy ions by contrast may create damage clusters along their track. A
comparison of lattice damage by light and heavy ions is depicted in Fig. 3-5. The
heavy ions may undergo relatively higher degree of nuclear stopping than light ions
even right from the surface. The volume of the crystal in which the ion energy is

38
deposited is usually larger than the volume in which the lattice damage occurs. When
the damaged areas start to overlap with increasing ion dose, an amorphous layer can
result, which implies that all the nuclei have been displaced from their lattice position
and the long range order which describes a crystal is no longer present.

The number of displaced atoms (Ndisp.) by irradiating ion can be estimated by [26]

En
N disp. ≈ (3.16)
2Td
where En is the total energy deposited in primary and secondary nuclear collisions.
The expression in (3.16) should be used with caution, because it overestimates the
number of displaced atoms since it overlooks the effects of (i) ion channeling and (ii)
vacancy-interstitial recombination.

3.4.3 Defect Annealing Mechanisms

Defect characterization techniques such as deep level transient spectroscopy (DLTS)
and photoluminescence (PL) cannot be used to probe the defect structure. The only
way to correlate the defect structure obtained from electron paramagnetic resonance
(EPR) to the DLTS and PL measurements is by thermal annealing studies. The
characteristic temperature at which a defect disappears is the defect’s annealing
temperature. This annealing temperature is the parameter that allows for results
obtained by different techniques to be compared.

The defects annealing mechanism can be classified into two main categories:
(a) Diffusion, as the temperature is increased defects migrate to sinks (e.g. by
moving to the surfaces or grain boundaries) or they are subsequently trapped
by other defects or impurities (e.g. direct recombination of the interstitial with
a vacancy, complex formation or hydrogen passivation) to form new defects.
The mean distance between the interstitial and the vacancy depends on the
energy deposited by the irradiating particles, thus the annihilation process of
the interstitial and a vacancy is mainly related to the irradiation energy [27].
The ability of a defect to migrate through the crystal is determined by the
thermal energy of the crystal and its charge state.
(b) Dissociation, which is the breaking-up of the complex defects.

39
Each of the process i.e. defect migration, recombination, and complex formation is
characterized by migration enthalpy or activation energy (Ea). It should be noted that
for the simple defects the migration energy for vacancies in a solid is much higher
than that for interstitials. At a fixed annealing temperature (Ta) the annealing kinetics
can be deduced by monitoring the decrease in defect concentration with time. The
defect annealing kinetics can provide information on the defect distribution, annealing
mechanism and hence their identity. Consider an irradiated sample with a given defect
level with concentration NT, the number of defects which anneal per unit time is
proportional to the number of defects NT(t) present at time t and thus can be written as
dNT
= − Kf ( NT ) (3.17)
dt

where K is the rate constant and if f ( NT ) = NT , then the annealing kinetics is said to

be of first order and if f ( NT ) = NT2 then it is of the second order. Solving the
differential in equation (3.17) gives annealing kinetics for first order as

NT (t ) = NT (0)e − Kt (3.18)

where NT(0) is the initial defect concentration at t = 0. The rate constant K given in
equation (3.17) and (3.18) has the form

Ea
−
K = K0e kT
(3.19)

where k is Boltzmann constant, K0 is a pre-exponential constant (which contains the

vibrational frequency associated with the process) and Ea is the associated activation
energy (which might be migration energy, dissociation energy etc depending on the
process). Experimentally the activation energy can be obtained from isothermal
annealing studies. The variation of NT versus time is measured at a constant
temperature T1 and a plot of ln(NT) vs. t is straight line for first order kinetics which
will give a rate constant K1 from the gradient. If the process is repeated for other
constant temperatures T2 and T3, then rate constants K2 and K3 are obtained
respectively. From equation (3.19) a graph of ln(K) vs. 1/T is an Arrhenius plot and

40
will yield the activation energy Ea for the defect annealing from the gradient and K0
from the vertical axes intercept.

It is interesting to note that most simple defects anneal out at between 200˚C and
400˚C in silicon, while higher order defects are introduced at higher temperatures
between 350˚C and 500˚C. The structure of these higher order defects is critically
dependent on the irradiation condition (i.e. irradiation ion energy and ion mass). It has
also been shown that the formation of a silicide phase at a metal-Si interface during
thermal annealing injects vacancies into the substrate. Thus the silicidation technique
can be used to remove interstitial – related defects in processed p-type Si [28].

41
References

[1] G.L. Miller, D.V. Lang and L.C. Kimerling, Ann. Rev. Mater. Sci. 7 (1977)
377.
[2] A. Hallen and M. Bakowski, Solid-State Electron, 32 (1989) 1033.
[3] F.D. Auret and P.N.K. Deenapanray, Crit. Rev. in Sol. State and Mater. Sci.,
29 (2004) 1.
[4] M. Lannoo and J. Bourngoin, Point Defects in Semiconductors I, Theoretical
Aspect Springer series in solid state science 22, (1981).
[5] W. Fank, Inst Phys. Conf. Ser. 22, (1975) 23.
[6] A. Antonelli, E. Kaxiras and D.J. Chadi, Phys. Rev. Lett. 81 (1998), 2088.
[7] S.A. Centoni, B. Sadigh, G.H. Gilmer, T.J. Lenosky, T.D. de la Rubia and C.B
Musgrave, Phys. Rev. B 72 (2005) 195206.
[8] G.D. Watkins, Mater. Sci. Semicond. Processing 3 (2000) 227.
[9] J. Coutinho, R. Jones, V.J.B. Torres, M. Barroso, S. Oberg and P.R. Briddon,
J. Phys.: Condens. Matter. 17 (2005) L521-7.
[10] C. Janke, R. Jones, S. Oberg and P.R Briddon, Phys. Rev B 75 (2007) 195208.
[11] A. Mesli, L. Dobaczewski, K. Bonde Nielsen, V.L. Kolkovsky, M. Christian
Petersen, and A. Nylandsted Larsen, Physical Review B 78 (2008) 165202.
[12 ] A.N. Larsen and M. Mesli, Physica B, 401-402 (2007) 85-90.
[13] V. Rianeri, G. Fallica and S. Libertino, J. Appl. Phys. 79 (1996) 9012.
[14] G.D. Watkins, in: Hulin (Ed.), Radiation Damage in Semiconductors, Dunod,
Paris, 1964, p97.
[15] G.D. Watkins, Phys. Rev. B 12 (1975) 5824.
[16] T. Markvart, D.P. Parton, J.W. Peters and A.F.W. Willoughby, Materials
Science Forum 143-147 (1994) 1381.
[17] W. Schockley and W.T. Read, Phys. Rev. 87 (1952) 835.
[18] J. Lindhard, V. Nielsen, M. Scharff, and P.V. Thompsen, Kgl. Dan. Vid. Selsk,
Mat. Fys. Medd. 33 (1963) 10.
[19] J. Lindhard and M. Scharff, Phys. Rev. 124 (1961) 128.
[20] S.J. Pearton, Solid State Phenomena, Vol. 1-2 (1988) 247.
[21] J. Lindhard, M. Scharff, and H. Schiott, Kgl. Dan. Vid. Selsk, Mat. Fys. Medd.
33 (1963) 14.

42
[22] J.P. Biersack and L.G. Haggmark, Nucl. Instrum. Methods 174 (1980) 257.
[23] J.F. Ziegler “The Stopping and Range of Ions in Matter” Vol 2-6, Pergamon
Press, 1977-1985
[24] M. Lannoo, J. Bourngoin, Point Defects in Semiconductors II, Experimtal
Aspect Springer series in solid state science 35, (1983).
[25] H. H. Andersen, Appl. Phys. 18, (1979) 131.
[26] G.H. Kinchin and R.S. Pease, Rep. Prog. Phys. 18 (1955) 1.
[27] A. Mesli, L. Dobaczewski, K. Bonde Nielsen, V.L. Kolkovsky, M. Christian
Petersen and A. Nylandsted Larsen, Physical Review B 78 (2008) 165202.
[28] Dong-Zhi Chi and S. Ashok, Mat. Res. Soc. Symp. Proc. 442 (1983).

43
 Chapter 4

DLTS and Laplace-DLTS Aspects

4.1 Introduction
Deep level defects can be detrimental to or enhance the operation of devices
fabricated on semiconductors as discussed in the earlier chapter. Therefore, it is
essential to develop a sensitive experimental tool for characterizing the deep level
defects in a semiconductor. The deep-level transient spectroscopy (DLTS) is a high-
frequency (1 MHz) transient capacitance technique, which has proved to be a very
useful tool to probe the defects close to the semiconductor surface since it was first
discovered by Lang in 1974 [1,2]. Recently, high-resolution Laplace-DLTS (LDLTS)
[3-4], which greatly enhances the resolution and spectroscopic nature of capacitance
based defect characterization tools has been developed. In this chapter the emission
and capture of carriers form deep level defects is discussed in section 4.2, the DLTS
and LDLTS theory is presented in sections 4.3 and 4.4 respectively. The electric field
effect on the deep levels is discussed in section 4.5.

4.2 Emission and Capture of Carriers from Deep Levels

The fabrication and development of efficient semiconductor devices require prior
knowledge of the properties of deep levels as carrier traps or generation-
recombination centers. As mentioned in the previous chapter, the generation-
recombination centers can reduce minority carrier lifetime and diffusion length which
may, for example, limit bipolar transistor performance [5], reduce the efficiency of
photovoltaic cells [6] or increase the switching speed in semiconductor switches. A
defect level can be defined as an electron trap if, when it captures an electron from the
conduction band the electron stays there until it is re-emitted back to the conduction
band. This may occur for an empty level when the electron capture rate cn, from the
conduction band is much larger than the hole capture rate cp from the valence band,

44
i.e. cn cp. Conversely, a recombination center is one for which cn and cp are almost
similar, i.e. cn ≈ cp. According to Shockley and Read [7] a deep level almost always
changes its electron occupancy via carrier transitions between the level and the bands.
Figure 4-1 shows the four common processes, neglecting transfer between the deep
levels. The figure shows a trap which may exist in either two states, negative or
neutral. Similar treatment is possible for other pairs of two states, e.g. neutral and
positive or negative and double negative. If a trap is neutral it may capture an electron
from the conduction band Fig. 4-1 process (a) or it may capture an electron from the
valence band process (d) leaving behind a hole (hole emission). Processes (b) and (c)
are electron emission and hole capture respectively.

Fig. 4-1. Schematic representation of transitions of carriers between deep states (ET)
and the valence (EV)- and conduction (EC)- bands, neglecting the transfer between
deep levels (a) electron capture, (b) electron emission, (c) hole capture, (d) hole
emission. The arrow indicates the transition of the electron in the process (redrawn
from ref. 7).

The kinetics which governs the charge transfer between the deep level and the bands
are fully described by the Shockley-Read-Hall [7,8] (SRH) model. The model is
developed assuming thermal (or near thermal) equilibrium and studying of deep level
by DLTS uses the perturbation of the occupancy of the levels and then monitoring the
return to equilibrium. The electron capture rate is given by

45
 cn = σ n v n n (4.1)

where σn, is the defect’s electron capture cross-section, n is the electron concentration
in the conduction band and v n is the average thermal velocity of free electrons

which is given by

v n = 3kT / m * (4.2)

where m* is the effective mass of the electron, k is the Boltzmann constant, and T is
the temperature in Kelvin. Similarly, the hole capture rate is given by

cp = σ p vp p (4.3)

where σp, is the defect’s hole capture cross-section, p is the hole concentration and
vp is the thermal velocity of the holes. A similar expression to that of v n in

equation (4.2) can be written for v p . The thermal emission rate en, of electrons from

traps to the conduction band is proportional to the Boltzmann factor exp(-ET/kT), and
can be expressed as a function of temperature by [5,6,9]

σ n vn N C  E 
en (T ) = exp − T  (4.4)
g  kT 

where ET, is the energy level below the conduction band minimum (also referred to as
the defect activation energy if one assumes that σ is temperature independent), g is the
degeneracy of the defect level, T is the temperature in Kelvin, NC is the effective
density of states in the conduction band given by

3/ 2
 2πm * kT 
N C = 2 M C  2
 (4.5)
 h 

here Mc is the number of conduction-band minima, h is Planck’s constant.

46
An analogous expression can be written for hole emission rate ep, to the valence band.
In equation (4.4), the terms v n is proportional to T 1/2 , and NC is proportional to

T 3/2 , while σn may or may not be temperature dependent, thus the product v n NC has

T 2 dependence. If en is measured as a function of temperature and ignoring the

temperature dependence of σn, an Arrhenius plot of log(en / T 2 ) against 1/T is a
straight line which yields ET from the slope and σna, the apparent capture cross-section
(from intercept at T −1 = 0 ). The parameters ET and σna are often referred to as the
‘defect signature’. If a temperature-dependent capture cross-section is assumed then it
usually takes the form [4]

 ∆Eσ 
σ n (T ) = σ ∞ exp   (4.6)
 kT 

where σ∞, is the capture cross-section extrapolated to T = ∞ and ∆Eσ is the thermal
activation energy of the capture cross-section (i.e. thermal barrier for carrier capture).
The cascade capture into shallow levels and multiphonon capture into deep levels are
some of the possible contributing factors to capture cross-section temperature
dependence [10]. The temperature dependence of the capture cross-section may be
determined from the plot of log(σn) vs. 1/T, where ∆Eσ, is extracted from the slope and
σ∞ after extrapolation to T = ∞. Thus, the modified activation energy for a deep level
which exhibits a temperature-dependent capture cross-section can be written as,

∆E a = ET + ∆Eσ (4.7)

The modified activation energy has two components; (i) the energy difference
between the trap level and the bottom of the conduction band ET, and (ii) the thermal
activation energy of the capture ∆Eσ, as depicted in Fig. 4-2. A more general
expression of the thermal emission rate is now given by,

σ n vn N C  E + ∆Eσ 
en (T ) = exp − T  (4.8)
g  kT 

47
Fig. 4-2. Configuration co-ordinate (CC) diagram depicting the energy level of the
defect below the conduction band ∆ET (=ET), the thermal activation energy of the
capture cross-section ∆Eσ, and the total energy an electron requires to escape from
the trap level to the conduction ∆Ea, redrawn from ref. 6.

In this study, the defect characterization is based on ET (the position of the defect
level from the band edge) and σ the apparent capture cross-section, therefore great
care should be taken when these thermal emission measurements are compared with
results from other techniques, e.g. optical measurements.
The physical meaning of ET is that it is the Gibbs free energy change for the ionization
of the state given by [11]

ET = ∆H − T∆S (4.9)

where ∆H and ∆S are the changes in enthalpy and entropy due to the change in charge
state of the level. Substituting equation (4.9) into (4.4) yields

σ n vn N C  ∆S   ∆H 
en (T ) = exp −  exp − (4.10)
g  k   kT 

Therefore, the Arrhenius plot yields the activation enthalpy of the deep level, and not
the free energy, which can only be determined from optical measurements [6,9].

48
4.3 Deep Level Transient Spectroscopy (DLTS)
The conventional deep level transient spectroscopy (DLTS) is a powerful high
frequency (MHz range) capacitance transient thermal scanning technique that is used
to probe the space-charge region of a p-n junction, Schottky diode or MOS device
structure. This technique is based on the transient capacitance change associated with
the thermal emission of charge carriers from a trap level to thermal equilibrium after
an initial non-equilibrium condition in the space-charge region. The DLTS technique
offers the following characterization features:
(a) High sensitivity and good resolution.
(b) Straight forward, easy analysis of spectra and rapid scanning.
(c) Capability of measuring over a wide range of depths in the forbidden gap
and detection of very shallow levels.
(d) Spectroscopic nature (i.e. signals due to different traps can be resolved
from one another) [1].
A DLTS scan reveals each trap by a positive or negative peak on a flat base-line
plotted as a function of temperature. The sign of each peak indicates whether it is due
to a majority- or minority- carrier trap and positions of the peaks are simply and
uniquely determined by the instrument rate-window and the thermal emission
properties of the respective trap [1]. It is also possible to extract the thermal emission
rate, activation energy, concentration profile and capture cross-section of each trap
from the DLTS measurements. Since all the experimental work on defect levels
studies in this study was based on the depletion region formed by the Schottky diode
on a semiconductor, the discussion below will be confined to capacitance transient
within the space-charge region of a Schottky barrier diode.

4.3.1 Capacitance Transient Processing

The relationship between capacitance and depletion width for a Schottky barrier diode
has been dealt with in detail in chapter 2. The depletion width W is given by

2ε s
W= (Vbi − Va ) (4.11)
qN D

49
where, ND is the density of ionized impurities due to dopants and other defects with
levels in the band gap, q is the electronic charge, εs is the permittivity of the
semiconductor, Vbi is built-in potential, and Va is an externally applied voltage. The
corresponding junction capacitance is

qε s N D Aε
C=A = s (4.12)
2(Vbi − Va ) W

where A, is the area of the junction. The capacitance of the depletion region depends
on the applied bias voltage and the dopant concentration as shown in equations (4.11)
and (4.12). It is the sensitivity of the capacitance to the change in charge in the
depletion region that is exploited and forms the basis of DLTS.

In the derivation of the depletion width equation, the depletion approximation has
been used, which assumes that the semiconductor can be divided into two distinct
regions, i.e. the bulk region which is electrical neutral and the space charge region
which is depleted of charge carriers. In a real junction, as depicted in Fig. 4-3, there is
a region λ, which lies between the truly depleted region and the bulk region. This
region is defined as the distance between the depletion region edge and the point
where the deep levels ET crosses the Fermi level EF. Fig. 4-3 also illustrates the
energy band diagrams and space charge for a metal-n-type semiconductor with deep
donors for (a) unbiased junction and (b) after applying a bias voltage Va. In
equilibrium, and under zero bias, deep levels in the region λ are below the Fermi level
and therefore filled with carriers (a). After applying a reverse bias Va the depletion
region increases, the space charge region is altered, thus decreasing the capacitance of
the depletion region. The deep levels NT and shallow dopants ND contribute to the
charge density ρ in the depletion region and only the shallow dopants ND are the
source of the charge density in the region λ.

50
Fig. 4-3. Energy band diagram, the region λ, and space charge for an n-type metal-
semiconductor junction with deep donor levels for (a) an unbiased and (b) after
applying a quiescent reverse bias of Va. For each condition the corresponding charge
density ρ distribution is also shown, after ref. 14.

If the concentration of holes and electrons trapped at the deep levels is altered, (by say
thermally stimulated emission of the carriers to the conduction or valence band), then
this change can be monitored by measuring the junction capacitance or capacitance
transient at constant applied bias voltage [6,12]. To ensure that the deep levels are
filled with charge carriers after thermal stimulated emission, continuous refilling of
the deep levels is achieved by application of a repetitive voltage filling pulse
superimposed on a constant reverse bias voltage. The variation of the depletion
region width and capacitance after the application of a voltage bias and a filling pulse
sequence for majority and minority carrier traps in n-type semiconductor (e.g. n-type
Ge) is depicted in Fig. 4-4 and Fig. 4-5 respectively. For simplicity, the bending of the
bands due to the electric field in the space charge region has not been indicated. Also

51
the lambda effect has been ignored, so it is assumed that the defect levels in the
depletion region are above the Fermi level and those deeper than the depletion region
are beneath the Fermi level.

Fig. 4-4. Variation of the depletion region width and capacitance after the application
of a voltage bias and a filling pulse sequence for a majority carrier (electron) trap in
n-type semiconductor, after ref. 6.

The capacitance transient resulting from the pulse sequence is shown in the center.
Under a quiescent reverse bias V and steady state, Fig. 4-4 part (1), the deep levels
under the Fermi level are assumed filled and those above are empty as governed by
the Fermi distribution function. The empty deep levels in the band gap are indicated
by empty squares. After applying a majority carrier filling pulse in part (2), the
depletion width is reduced, trapping electrons in those levels that are now below the
Fermi level, symbolized by the solid squares. It is assumed here that the pulse width tp
is long enough to allow the complete filling of the trap levels. There is a
corresponding step increase in capacitance because of the reduced depletion width.
Immediately after the pulse is removed and the quiescent reverse bias V restored, part
(3), the filled states lie within the depletion region, above the Fermi level, therefore
the levels will start emitting the trapped carriers with a characteristic rate to the

52
conduction band where they are instantaneously swept away by the junction electric
field, part (4).

Fig. 4-5. Variation of the depletion region width and capacitance after the application
of a voltage bias and a filling pulse sequence for a minority carrier (hole) trap in n-
type semiconductor.

The capacitance variation as the trapped carriers are emitted to the conduction band is
the so-called majority capacitance transient. The emission rate can be determined
from the time dependence of the capacitance transient. The density of occupied defect
levels at time t after removing the filling pulse is given by [6]

N (t ) = N T exp(−en t ) (4.13)

where en is the electron thermal emission rate and NT is the trap concentration
junction capacitance, if NT ND, can then be given by an exponential time varying
function as

C (t ) = C o − ∆C o exp(−en t ) (4.14)

53
here Co is the equilibrium reverse bias capacitance and ∆Co the change in capacitance
immediately after the removal of the pulse, i.e. at t = 0 as shown in Figs. 4-4 and 4-5.
When a large enough pulse is applied such that the junction is forward biased,
(minority carrier injecting pulse), Fig. 4.5 part (2) and part (3), minority carriers
(holes in this case) are trapped. After removing the pulse and the quiescent reverse
bias V restored, part (4), the trapped carriers are emitted to the valence band giving
rise to the minority carrier capacitance transient, which is of opposite sign to the
majority carrier transient.

4.3.2 DLTS Principles

The utility of DLTS is in the processing of the capacitance transients obtained after
repeated pulsing sequence discussed in section 4.3.1. The basic idea of DLTS method
can be represented by the illustration in Fig. 4-6. The system response occurs only
when the emission rate of the trap falls within the ‘rate window’. For a given rate
window, the system response is shifted to higher temperatures for a trap with higher
emission rate and hence the system can resolve signals from different traps as a
function of temperature, as shown in Fig. 4-6.

Fig. 4-6. A Schematic illustration of how a rate window produces a peak in its
response when the emission rate of the input signal matches the rate selected by the
window, redrawn from ref. 1).

54
In consequence, DLTS has the ability to set up an emission ‘rate window’ so that the
measurement system gives an output only when a transient with a rate within this
narrow window occurs. Since the emission rate is strongly temperature dependent, a
thermal scan can reveal the presence of different traps at characteristic temperature
when their emission rates coincide with the window. Most early DLTS systems used
the dual-gated (double boxcar) signal filter for determining the rate window and
averaging transients to enhance the signal-to-noise ratio (SNR) of the output, enabling
low concentration defects to detected [1,9]. As the temperature is scanned, the filter
takes samples of the transients at preset times t1 and t2 and produces an output
proportional to their average difference, as shown in Fig. 4-7(a). Thus, the rate
window is determined by the values of t1 and t2. The output signal changes from a
small response as the decay time constant (τ = 1/e) moves into the range detectable by
the filter, to a maximum, and then drops as the decay time constant again falls outside
the filter detectable range, giving rise to a DLTS spectrum depicted in Fig. 4-7(b).
The early analog filter design by Lang [1] had an intrinsic dc rejection mechanism,
which would give a zero output on the filter when no defect is detected. The
normalized DLTS signal S(T) is defined by

C (t1 ) − C (t 2 )
S (T ) = (4.15)
∆C (0)

where C(t1) is the capacitance at t1, C(t2) is the capacitance at t2, and ∆C(0) is the
capacitance change due to the filling pulse at t = 0 as described in figures 4-4 and 4-5.
The position of the peak on the temperature axis depends on the rate window, e.g. a
smaller rate window will shift a defect peak to lower temperatures. If DLTS spectra is
produced by using different rate windows, then a series of spectra are produced as
shown in Fig. 4-8(a). The emission rate at a maximum peak height is a uniquely
defined and can be calculated via the expression for the time constant, (τmax) at
maximum peak height

t1 − t 2
τ max = (4.16)
t 
ln 1 
 t2 

55
Fig. 4-7. A schematic diagram which shows how a rate window concept can produce
a DLTS spectrum. Part (a) shows the capacitance transients at different temperature
which after processing with a double boxcar, and thereby resulting in the spectra
shown on part (b), after ref. 1.

56
Fig. 4-8. A diagram showing the (a) DLTS spectra at various rate windows and (b)
the Arrhenius plots obtained from the spectra, after ref. 1.

For each spectrum, the temperature at the maximum peak height can be measured and
the emission rate e, for which the DLTS system shows a maximum response, is
calculated via equation (4.16). These points are then used to plot a semi-log graph of
log(e/T2) vs. 1000/T, (Arrhenius plot), shown on Fig. 4-8(b) from which the defect
activation energy ET and apparent capture cross-section σ, are extracted.

An alternative to the double boxcar weighting function used in the original Lang’s
work [1] is to use a lock-in amplifier (LIA) [13]. In a LIA set-up the rate window is
set by altering the frequency of the filter. The LIA response to this transient is the
integral product of the capacitance signal and the weighting function w(t ) given by

1 τ

τ∫
S (τ ) = C (t ) w(t )dt (4.17)
0

 2π t 
where w(t ) = sin   is a sine wave of fixed frequency. The LIA gives the same
 τ 
result to that of the double boxcar method. For an exponential transient, using a sine
 1 
wave weighting function the DLTS signal reaches a maximum when λ =  .
 0.423τ 

57
4.3.3 Defect Depth Profiling
The DLTS peak height is directly proportional to the concentration of a deep level,
therefore the concentration can be obtained directly from the capacitance change
corresponding to completely filling the trap with a saturation minority carrier pulse (in
case of a minority carrier trap) or the largest possible majority-carrier pulse (in case of
a majority-carrier trap). DLTS enables one to determine the defect spatial distribution
within the semiconductor and thereby other parameters such as the introduction rate.

The concentration of deep levels NT is often calculated by using simple expression

[1,14]

2∆C (0)
NT = ND (4.18)
C

where ND is the concentration of shallow impurities, C is the junction capacitance

under quiescent reverse-biased conditions, and ∆C(0) is the capacitance change due to
the pulse at t = 0 (i.e. just after removing the filling pulse). Equation (4.18) is only
applicable if the minority carrier pulse or majority carrier pulse is large and long
enough to completely fill the trap and when ∆C(0) C. The correct pulse for defect
concentration determination can be checked by making several scans with increasing
larger and longer pulses until the defect peak no longer increases in size. As has been
pointed out in the past [1,14] using equation (4.16) sometimes results in significant
underestimation of NT especially for thin films and at low reverse bias voltages. In
order to determine the corrected expression for NT one has to consider the region λ
(the so called λ effect), where the defect level crosses the Fermi level a distance λ
shallower than the depletion region edge as shown in Fig. 4-3(b). The traps in this
region are occupied and do not contribute to the change in capacitance when a filling
pulse is applied. The width of the transition is given by [14]

1/2
 2ε ( EF − ET ) 
λ= 2  (4.19)
 q ND 

58
where ε is the semiconductor dielectric constant, EF is the Fermi level and q is the
electronic charge. To obtain the deep level distribution profile, the deep levels in the
region to be profiled must be filled with carriers. The depth profiling technique used
in this study uses a fixed bias voltage and a variable filling pulse (fixed bias-variable
pulse method) [15]. In the fixed bias-variable pulse method the incremental change in
capacitance δ(∆C) is monitored as the majority carrier pulse Vp is changed by a small
amount δVp. The relative incremental change in capacitance due to the pulse
increment can be expressed by [14]

 ∆C   ε  N T ( x)

δ = δV p (4.20)
 C   qw N D
2
 N D ( x)

where x is the depth below the junction, ND and w are the ionized shallow impurity
concentration and depletion region width, respectively, corresponding to quiescent
reverse biased conditions. The shallow impurity profile ND(x) is obtained from C-V
measurements. The total signal due to the majority carrier pulse, is then determined
by double integration of the Poisson equation according to a detailed derivation by
Zohta and Watanabe [14], to give the corrected deep level concentration expression as

−1
2∆C (0) N D ( x)  x − λ   x p − λ p 
2 2

NT =    −    (4.21)
C  x   x  

here x - λ and xp – λp are the depletion region widths before and after applying a filling
pulse respectively, and λ is distance from where the deep levels cross the Fermi level
to the depletion region edge and λp is the value of λ during the pulse. In the limit and
low noise measurements, values of 10-5-10-6 for ∆C/C can be achieved and if the
shallow dopants concentration is ND ≈ 1016 cm-3, a low defect concentration of the
order of 1010 cm-3 is detectable.

4.4 Laplace-DLTS
Since its discovery almost 35 years ago, conventional deep level transient
spectroscopy (DLTS) has been a valuable tool in identifying deep level states in

59
semiconductors, thereby improving device efficiency. Unfortunately this technique
has limitations in the details of information on the identity of defects it can measure
due to its poor emission rate resolution. The double boxcar or lock-in-amplifier filter
used in DLTS exhibit good sensitivity but has poor time constant resolution. Due to
this poor time constant resolution, DLTS cannot be used to separate closely spaced
transients, and thus its inability to study defect fine structure. Over the past 25 years
there has been much effort applied to improve DLTS resolution by developing
different weighting functions [16]. These higher order filters showed an improvement
of resolution by a factor of up to 3 but at the expense of noise performance.

In 1990, Dobaczewski et al [3,4] developed an improved high-resolution version of

DLTS, called Laplace-DLTS (LDLTS). This new concept is an isothermal DLTS
technique and employs a regularized inverse Laplace transform instead of the
conventional boxcar analysis. This results in an order of magnitude improvement in
emission rate resolution in the studies of the thermal emission of carriers from deep
states. Consequently, LDLTS can separate closely spaced transients (with emission
rates differing by a factor greater than 2) when a number of defects with similar
emission characteristics are present, thereby overcoming the major deficiency of
DLTS. Apart from the remarkable sensitivity (ability to measure very low
concentrations of defects), LDLTS can probe very narrow regions of the
semiconductors (e.g. regions of shallow implants) and can also be used to study
selectively the active regions of devices.

4.4.1 Laplace-DLTS Principles

Generally, in DLTS there are two main classes of transient processing methods, which
are analog and digital signal processing. Analog signal processing is a real-time
process which involves extracting the capacitance transients as the temperature is
ramped. An Analog filter will then produce an output proportional to the signal input
at a particular time constant range. In order to increase the resolution and sensitivity
of DLTS, several different filters have been investigated which include, boxcar [1],
lock-in amplifier [13], exponential [17] and multiple boxcar [18]. The digital signal
processing involves digitizing the transient output of the capacitance meter, normally
done with sample held at a fixed temperature and averaging many of these digitized
transient to reduce noise. If the transient is digitized, then it is much easier to apply
60
signal processing tasks, even complex ones. The concept of digitizing capacitance at
constant temperature and extracting the time constant is the basis of high resolution
Laplace-DLTS. The extraction of all accessible time constants from the transients is
achieved by numerical algorithm. There is a well-documented problem associated
with the extraction and separation of multiple, closely spaced time constants. The
problem is finding a suitable choice of algorithm to use in the extraction of the time
constants. The problem is due to several factors, including that (i) the exponential
decay transient baseline is not known with any degree of precision; hence this
becomes an unknown variable in the analysis, (ii) both polarities of the transient may
be present simultaneously and (iii) exponential transients are not ideal (due to
dependence of the emission rate on electric field or due to inhomogeneous strain
which produces a continuum of emission rates for a particular defect). Several
algorithms that have been developed in an attempt to solve the transient extraction
problem have been reviewed [4,6]. The techniques considered include, “a method of
moments” by Ikossi-Anastasiou et al [19], “a Gaver-Stehfest approximation algorithm
to effect a Laplace transform” by Nolte et al [20], and “Tikhonov regularization
method” to separate the constituent exponentials in a photo-induced current transient
spectroscopy (PICTS) signal by Eiche et al [21]. The Tikhonov regularization method
uses a similar approach to the technique employed with the Laplace-DLTS work
discussed here. To develop an algorithm for transient processing, assume that the
recorded transient f(t) is a non-exponential transient, which is composed of a
superposition of exponential transients and is given by

∞
f (t ) = ∫ F ( s )e − st ds (4.22)
0

where F(s) is the spectral density function. The function f(t) given in equation (4.22)
is the Laplace transform of the true spectral density F(s). Thus, to find the true
spectrum of emission rates in the transient, an inverse Laplace transform for the
function f(t), should be performed using some numerical method. For such a
procedure, (assuming that all the decay transients are exponential and have the same
sign), a spectrum of delta-like peaks is produced for multi-, or mono-exponential
transients. All the numerical methods used in LDLTS attempt to find a spectral
function with the least possible number of peaks, which is consistent with the data and

61
experimental noise [4]. Although the problem has been described in a general way, it
should be noted that equation (4.22) does not have a general solution for any given
function f(t). For an analytical multi-exponential function such a solution exists and,
according to Lerch’s theorem [22] it is unique. However, if noise is superimposed on
this function the number of solutions can be infinite. Therefore, the problem is to find
the best estimate for F(s), and according to the prior knowledge about the system
being investigated and its boundary conditions, to exclude unphysical solutions and
choose only the simplest one, i.e. the one that reveals the least amount of detail or
information that was not already known or expected [3,6].

(a)

(b)

Fig. 4-9. (a) A majority-carrier capacitance transient and (b) the corresponding
spectra obtained from the transient with use of the three numerical routines.
62
For Laplace transform inversion in the Laplace DLTS system used here, three
numerical routines CONTIN [23], FTIKREG [24] and FLOG (which was specifically
developed for Laplace DLTS by Matulis [25]), all of them based on the Tikhonov
regularization method, however they differ in the way the principle for finding the
regularization parameters are defined. The Laplace card sets the sample excitation
parameters (or in some cases is used to trigger the external pulse generator, which
supplies the biasing and pulsing conditions to the sample). The Laplace software then
acquires the capacitance transient, shown in Fig. 4-9, graph (a) before the transient is
converted into the LDLTS spectra depicted in graph (b) using the three numerical
routines, i.e. CONTIN, FTIKREG and FLOG and from the spectra, the emission rates
and magnitude of the signal can be measured.
Spectral Density Function [arbitrary units]

G4 (Si:Au,H)
rate window 50/s

200 240 280 320

T [K]

gold acceptor

100 101 102 103 104 105

Emission Rate [1/s] at 260K

Fig. 4-10. DLTS and LDLTS spectra of hydrogenated silicon containing gold. The
conventional DLTS spectrum is shown as an insert at the top of the figure. The broad
peak centered at 260 K is attributed to electron emission from the gold acceptor G4.
The main spectra were obtained by the Laplace technique and clearly separate the
gold-acceptor level and the gold-hydrogen level G4 (redrawn from ref. 26).

63
The parallel use of three different numerical routines increases the confidence level in
the spectra obtained. The evolution of LDLTS in past few years has enabled the
theoretical limit of DLTS to be achieved. For relatively shallow states that emit at low
temperature, the reduction in line width is remarkable and an increase in resolution of
at least two orders of magnitudes is readily achievable but its sensitivity is about an
order of magnitude less than that of DLTS. For example, if a sample with a trap
concentration approximately 1% of the shallow dopant concentration and quiescent
capacitance of about 10 pF is studied by LDLTS then the signal-to-noise ratio (SNR)
of 1000 is readily achievable. This is necessary to separate defects with similar
emission rates with time constant of 2 i.e. (e1/e2) ≈ 2. An illustration showing a
comparison of LDLTS (full figure) and DLTS (inset) spectra obtained from the same
(Si:Au,H) sample is depicted in Fig. 4-10 [26]. DLTS shows a broad featureless
spectrum of DLTS whilst the LDLTS resolves the broad DLTS spectrum into two
peaks.

Finally, to completely characterize defect levels in the semiconductor by DLTS and

LDLTS, both the majority and minority carrier traps should be identified. If the
depletion layer is formed from a p-n junction then both majority and minority carriers
can easily be injected into the depletion region by applying a correct filling pulse.
However, when a Schottky diode is used to form the depletion region, a forward bias
filling pulse will not always inject minority carriers into the depletion region,
therefore minority carrier traps may not be observed. This is true for Si, but not
entirely true for all semiconductors e.g. Ge. For Ge, metal-semiconductor Schottky
barrier diodes with large barrier heights (in relation to the band gap) can be formed. It
has been shown that for a high barrier height, an inversion layer with high
concentration of minority carriers can be formed near the semiconductor surface
[27,28]. When a forward filling bias is applied to such a diode it results in a flux of
holes from the inversion layer to the semiconductor bulk. Thus the minority carrier
traps can be filled, making them visible to the DLTS and LDLTS techniques. The
minority carrier trap study using Schottky barrier diodes on Si is achievable if traps
are filled by optical means. Throughout this work, the Schottky barrier diodes (SBDs)
have been used because they are easy to fabricate, easy to use and quality diodes are
possible.

64
4.5 Electric Field Effect
The DLTS and LDLTS techniques employed for defect characterization probe defects
in the reverse-biased depletion region. Therefore the emission process takes place in
the presence of a high electric field. For an n-type semiconductor, the magnitude of
the electric field in the depletion region is given by

2qN D qN D x
E = (Vbi − Va ) − (4.21)
ε ε

where q is the electronic charge, ND is the shallow dopant concentration, Vbi the built-
in-potential, Va is the applied bias voltage and ε the permittivity of the semiconductor.

Fig.4-11. The Coulombic well and the three mechanisms of field enhanced emission;
Poole-Frenkel emission, phonon assisted tunneling and pure tunneling where Eth is
the position of the virtual level above the deep level, Ei is the ionization energy of the
deep level and δEi is the change in the ionization energy due to the electric field,
redrawn from ref. 30.

The electric field in the depletion region can reach average values of 105-107 V/m,
depending on the doping density and the bias voltage. Such high electric fields can
influence the shape of defect potentials and may, therefore, enhance carrier emission

65
rates from potential wells, by the Poole-Frenkel effect [29,30,31] or phonon-assisted
tunneling [6,32]. A potential well may be considered as a trap with significant spatial
extent and can be described by different models, e.g. a Coulombic, square or Gaussian
potential well. For a Coulombic potential well, the mechanisms of field-enhanced
emission are depicted in Fig. 4-11. The Poole-Frenkel effect is a mechanism in which
a carrier is thermally emitted over the top of the barrier, which has been lowered by
the application of an electric field. The emission rate in the presence of the electric
field F is now given by

 q3 
e( F ) = e(0) exp F 1/ 2  (4.22)
 πεkT 

where e(0) is the zero field emission rate, q is the electron charge, T is the temperature
in Kelvin and ε is the permittivity of the semiconductor. If a plot of log[e(F)] vs. F1/2
is a linear plot, then it is experimental evidence of a charge leaving a trap of opposite
sign. This implies a donor type defect in n-type material and acceptor type trap in p-
type material. The Poole-Frenkel mechanism is dominant if the potential well has
some appreciable spatial extent. The other mechanisms shown in Fig. 4-10 are
phonon-assisted tunneling and pure tunneling. In the phonon-assisted tunneling
process, a charge carrier absorbs thermal energy and is excited to a virtual state at Eth
above the deep level. The electron will then be able to tunnel through the barrier from
this virtual level to the conduction band. For any of the field-enhanced emission
mechanisms, the electric field is spatially varying and so emission occurring from
different positions in the depletion will be affected to a different extent, giving rise to
non-exponential capacitance transients.

66
References

[1] D. V. Lang, J. Appl. Phys. 45 (1974) 3023.

[2] D. V. Lang, J. Appl. Phys. 45 (1974) 3014.
[3] L. Dobaczewski, P. Kaczor, I.D. Hawkins and A.R. Peaker, J. Appl. Phys. 76
(1994) 194.
[4] L. Dobaczewski, A.R. Peaker and K.B. Nielsen, J. Appl. Phys. 96 (2004)
4689.
[5] P. Blood and J.W. Orton, Rep. Prog. Phys. 41 (1978) 11.
[6] F.D. Auret and P.N.K. Deenapanray, Crit. Rev. Sol. Stat Mater. Sci. 29 (2004)
1.
[7] W. Shockley and W.R. Read, JR, Physical Rev. 87 (1952) 835.
[8] R.N. Hall, Phys. Rev. 83 (1951) 228.
[9] G.L. Miller, D.V. Lang and L.C. Kimerling, Ann. Rev. Mater. Sci. 7 (1977)
377.
[10] C.H. Henry and D.V. Lang, Phys Rev. B 15 (1977) 989.
[11] V.P. Markevich, I.D. Hawkins, A.R. Peaker, K.V. Emtsev, V.V. Litvinov, L.I.
Murin and L. Dobaczewski, Phys. Rev. B 70 (2004) 235213.
[12] L.C. Kimerling, J. Appl. Phys. 45 (1974) 1839.
[13] F.D. Auret, Rev. Sci. Instrum. 57 (1986) 1597.
[14] Y. Zohta and M.O. Watanabe, J. Appl. Phys. 53 No.3 (1982) 1809.
[15] D.V. Lang, In “Thermally Stimulated Relaxation of Solids” (P. Braunlich, ed.)
pp93-133, Springer-Verlag, Berlin, 1979.
[16] A.A. Istratov,.J. Appl. Phys. 82 (1997) 2965.
[17] J.A. Borsuck and R.M. Swanson, IEEE Trans. Electron Devices 27 (1980)
2217.
[18] C.R. Crowell and S. Aliphani, Solid State Electron 24 (1981) 25.
[19] K. Ikossi-Anastasiou and K.P. Roenker, J. Appl. Phys. 61 (1987) 182.
[20] D.D. Nolte and E.E. Haller, J. Appl. Phys. 62 (1987) 900.
[21] C. Eiche, D. Maier, M. Schneider, D. Sinerius, J. Weese, K.W. Benz, and
Honerkamp, J. Phys. Condens. Matter. 4 (1992) 6131.
[22] G.A. Korn and T.M. Korn, in mathematical handbook , (McGraw-Hill, New
York) (1968).
[23] S.W..Provencher, Comput. Phys. Commun. 27 (1982) 213.

67
[24] J. Weese, Comput. Phys. Commun. 69 (1991) 99; ibid. 77 (1993) 429.
[25] A. Matulis, Z. Kancleris, Semiconductors Physics Institute, Vilnius, Lithuania.
[26] P. Deixler, J. Terry, I.D. Hawkins, J.H. Evans-Freeman, A.R. Peaker, L.
Rubaldo, D.K. Maude, J.C. Portal, L. Dobaczewski, K.Bonde Nielsen, A.
Nylandsted Larsen, and A. Mesli, Appl. Phys. Lett. 73 (1998) 3126.
[27] V.P. Markevich, A.R. Peaker, V.V. Litvinov, V.V. Emstev, L.I. Murin, J.
Appl. Phys. 95 (2004) 4078.
[28] E.H. Rhoderick, Metal-Semiconductor Contacts (Clarendon Press, Oxford,
1978).
[29] J. Frenkel, Phys. Rev. 54 (1938) 647.
[30] S.D. Ganichev, E. Ziemann, W. Pretll, I.N. Yassievich, A.A. Istratov, E.R
Weber, Phys. Rev. B 61 (2000) 10361.
[31] N. Zangenberg, J. Goubet, A.N. Larsen, Nuclear Instruments and Methods in
Physics Research B 186 (2002) 71.
[32] S. Makram-Ebeid and M. Lannoo, Phys. Rev. B 25 (1982) 6406.

68