You are on page 1of 11

Research Rotation Fall 2019 – Nemanich Lab

Jonah Shoemaker

December 2019

In this study, the electronic properties of silicon were examined in preparation for an anal-
ysis of surface transfer doping on diamond. The intent of the study was to use a layer of
vanadium oxide, V2 O5 , as an electron accepting layer. The vanadium oxide served to initi-
ate p-type doping in the diamond. However, previous studies have shown[2] that vanadium
oxide deposited directly onto diamond causes a build-up of resistivity and loss of carrier
mobility in the diamond layer. Thus, in this study, it was planned to deposit a layer of
hafnium oxide, HfO2 , between the diamond and vanadium oxide to serve as an interface
layer. Ideally, the hafnium oxide-diamond interface would result in lower resistivity and
better carrier mobility. However, in this paper, we will explore the results for a hafnium
oxide deposition onto a silicon wafer in conjunction with the theory driving these results
and the data analysis techniques needed to extract the results from the experimental data
collected from x-ray photoelectric spectroscopy. These data analysis techniques are pack-
aged together in various programs with much friendlier interfaces then the python code
used in this study, and so it was concluded that performing these data analyses on one’s
own is rather inadvisable unless done as part of a learning exercise.

X-Ray Photoelectric Spectroscopy


X-ray photoelectric spectroscopy (XPS) is the use of the photoelectric effect to obtain information
on the chemical composition and bonding present within a material. When the surface of a
sample of material is bombarded with photons of a known energy, electrons within the top few
nanometers of the sample can absorb an incident photon and gain enough energy to escape the
potential well of the atoms to which they are bound. If the binding energy B of an electron is
defined as the energy difference between the Fermi level of the material and the electron’s energy
level and the work function φ is defined as the energy difference between the Fermi level and the
energy level of the vacuum, then conservation of energy gives us the relation

hν = T + B + φ (1)
where hν is the energy of the incident photon and T is the kinetic energy of the ejected electron.
Solving for the kinetic energy gives

T = hν − B − φ (2)

This tells us that if the energy of the incident photon and the work function of the material are
known, then measuring the kinetic energy T allows us to determine the binding energy of the
electrons within the sample. The kinetic energy of the electrons can be directly measured by pass-
ing the electrons through a magnetic field that bends the path of electrons possessing a higher
velocity to a higher degree. By recording the electrons’ relative positions after passing through
the magnetic field, their relative velocities and therefore their kinetic energies can be determined.

Calculating the theoretical current density due to incoming radiation is relatively straightfor-
ward. We will use the derivation given in Pathria and Beale’s Statistical Mechanics[7] . We start
with the rate of ejection of electrons per unit area per unit time:

4πmkT ∞
Z
ln 1 + e(µ−z )/kT dz

R= 3
(3)
h V−hν

Here, µ is the chemical potential and V the strength of the potential well in the material. This is
assumed to be a constant value for simplicity. We can change variables to x = (z − V + hν)/kT

1
and define the quantity V−µ
kT
= hν0 = φ, where ν0 is the threshold frequency of light required to
eject the electrons and φ is the work function, to get

4πm(kT )2 ∞
Z   h(ν − ν ) 
0
R= ln 1 + exp − x dx (4)
h3 0 kT
h(ν−ν0
Now we can define one more quantity, δ = kT
to simplify our integral down to

4πmk 2 2
Z
ln 1 + e(δ−x) dx

R= T (5)
h3 0

Multiply equation 5 by the electron charge e to get the current density:

4πmk 2 e 2 ∞
Z
(δ−x)

J = T ln 1 + e dx (6)
h3 0

Integrate by parts to get



4πmk 2 e 2 4πmk 2 e 2
Z
xdx
J = T = T f2 (eδ ) (7)
h3 0 e(x−δ) + 1 h3
where fn (x) are the Fermi-Dirac integral functions, defined as
Z ∞ n−1
1 x dx z2 z3
fn (x) = = z − + − ... (8)
Γ(n) 0 z −1 ex + 1 2n 3n
For light with frequency much higher than the threshold frequency such that their energy dif-
2
ference is on the classical scale, h(ν − ν0 ) >> kT , we can approximate f2 (eδ ) to δ2 to get the
result

2πme
J ≈ (ν − ν0 )2 (9)
h
Thus we see no temperature dependence on the current density, and a parabolic dependence on
the frequency of incoming light.

However, when photons with frequency less than the cutoff, instead of seeing 0 current as we
would expect, we approximate with f2 (eδ ) ≈ eδ to get

4πmek 2 2 (hν−φ)/kT
J ≈ T e (10)
h3
Given that we have been working with a system assumed to have thermal fluctuations within
the material, it makes sense that we observe a current even for photons with frequency below
the cutoff. Electrons in the material occasionally absorb thermal energy from their environment,
which then helps some of them to escape the potential well regardless of the low photon energies.

At the threshold frequency, we get eδ = 1, which gives us

π 2 mek 2 2
J = T (11)
3h3
These theoretical results agree very closely with experiment[7] .

Analyzing XPS Data


An XPS spectrum of a pure silicon wafer from a monochromatic Al-Kα x-ray source with photon
energy 1482.35 eV is shown below:

2
Figure 1: XPS spectrum for the whole range of binding energies available to the XPS machine
used. The peaks around 100 eV are due to the presence of oxidized and unoxidized silicon atoms
in the sample. The peak around 530 eV is due to the presence of oxygen. The peaks around 1000
eV are due to Auger electrons in the oxygen atoms.

Analyzing electrons near the surface of the material is relatively straightforward. However, x-
rays are able to pass through the material and stimulate photoemission of electrons deeper within
the material, and these electrons are then more likely to undergo both elastic and inelastic scatter-
ing events with other electrons in the material as they attempt to pass through the material and
escape into vacuum. Inelastic scattering events become more likely as an electron passes through
thicker amounts of material, and so these electrons are likely to lose a lot of energy before they
finally emit from the material, if at all. These electrons from deeper within the material are likely
to have low kinetic energies relative to the electrons ejected near the surface, and so a typical
XPS spectrum will have a background of increasing intensity as the binding energy increases.

This increasing background can be modeled by a linear function and subtracted out from the data
to improve the signal-to-noise ratio. In order to improve the accuracy of the background fit, the
data can be divided into many small regions, and the median of each piece taken in order to give
the broad shape of the background data, without the peaks included. Then, again in order to
improve the fit’s accuracy, this resulting data can be divided into a few smaller pieces, anywhere
from three to ten smaller regions depending on the data in question, and a linear function fitted
to each of these regions before being subtracted. After doing so, many of the values drop below
zero, and so the entire result needs to be raised by the mininum quantity to calibrate the lowest
count value to zero. The result of this fit is given below:

Figure 2: Background-subtracted XPS spectrum for a pure silicon wafer. Note that the smaller
features near 0 are much more visible after this processing.

For rough qualitative analysis, one can remove the entire background using a linear background
fitting and excluding the peaks. However, when conducting finer quantitative analysis, it is best
to fit separate backgrounds to each peak using a Shirley background fitting.

3
Fitting a Shirley Background
The standard Shirley background requires choosing two points on either side of the peak of in-
terest. The construction of the Shirley background is an iterative process:

1. To find the first iteration, for each point between the two points chosen, take the integral from
that point’s energy to the energy of the right limit:
Z Eright
dE 0 I(E 0 ) − Iright

B1 (E) = k1
E

k1 and all subsequent kn are found with the formula

Ileft − Iright
kn = R Eright 
Eleft
dE 0 I(E 0 ) − Iright − Bn−1 (E 0 )
2. Find the next iteration by repeating the integral but with the first background iteration sub-
tracted out:
Z Eright
dE 0 I(E 0 ) − Iright − B1 (E 0 )

B2 (E) = k2
E

3. Repeat the process until whenever:


Z Eright
dE 0 I(E 0 ) − Iright − Bn−1 (E 0 )

Bn (E) = kn
E

As an example, take the XPS scan of pure silicon in the region from EB = 520eV to EB = 550eV
without any background subtraction:

Figure 3: Selected region of pure silicon spectrum, no background subtracted yet.

After running the Shirley background fitting algorithm, the background obtained around this
peak is shown:

Figure 4: Fitted Shirley background to the selected peak.

The difference on the right and left side is not very pronounced for the peak chosen, but it can
still be seen that the Shirley background fit better accounts for the difference in background level
on either side of the signal. After subtracting out this background and re-calibrating the lowest
value back to zero as we did for the general linear background fit, we obtain the ready-to-fit peak:

4
Figure 5: Selected peak with the Shirley background subtracted out. Note the difference in scaling
on the y-axis from the raw data.

Fitting a Gaussian-Lorentzian to XPS Peaks


The Gaussian and Lorentzian functions are

(x − E)2
G(x; F, E, h) = h · exp −4 ln 2 (12)
F2
h
L(x; F, E, h) = 2 (13)
1 + 4 (x−E)
F2

These look like:

Figure 6: Gaussian function with h=1,E=0,F=1.

Figure 7: The Lorentzian function with the same parameters.

As can be observed, the Lorentzian has a roughly similar shape to the Gaussian. However, while
the Gaussian quickly dies to 0 on either side, the Lorentzian has characteristic lapels on either
side. Due to the statistically random nature of the inelastic scattering events taking place between
ejected electrons and other electrons within the sample as they attempt to escape the material,
there is bound to be some spread in the kinetic energies recorded from electrons ejected from
a specific energy level. Hence, it makes sense that the lapels of the Lorentzian would fit more
accurately to recorded data.

It is most common to fit XPS peaks to some combination of these two. The Gaussian-Lorentzian
sum is a sum of the two with a ”mixing” parameter m added. This mixing parameter determines
what fraction of the fit comes from the Gaussian and what from the Lorentzian.

(x − E)2
 
hm
GLS(x; F, E, m, h) = h · (1 − m) exp − 4 ln 2 2
+ 2 (14)
F 1 + 4 (x−E)
F2

The Gaussian-Lorentzian product is along the same idea as the sum, but the mixing parameter m
is contained within the Gaussian exponential’s argument and the Lorentzian’s denominator:

5
(x − E)2
 
1
GLP (x; F, E, m, h) = h · exp − 4 ln 2(1 − m) · 2 (15)
F2 1 + 4m (x−E)
F2

The resulting fits for all four of these fit options are given in Figure 8.

Figure 8: Resulting fits for the selected peak from the Gaussian, Lorentzian, Gaussian-Lorentzian
sum, and Gaussian-Lorentzian product functions. Note that the Gaussian fit’s tails lie below the
real peak’s, while the Lorentzian’s lie above them. The GLS and GLP functions’ tails, however,
adhere very closely to the peak’s in comparison.

For this study, we utilized data analysis techniques acquired from the CasaXPS article ”Peak Fit-
ting in XPS”.

Overlayer Thickness Calculation Using XPS


After depositing a layer of a metal oxide onto another material, if the overlayer is not too thick,
then x-rays can penetrate the overlayer into the substrate and eject electrons from the substrate
atoms. These electrons make their way towards the surface, but have a chance of experiencing
inelastic scattering events with atoms in the overlayer that increases with the thickness of the
overlayer. The likelihood of electrons making collisions with overlayer atoms is described by a
parameter called the inelastic mean free path of the overlayer material. We will work through
the derivation of the thickness calculation formula as covered in Jablonski et al’s paper[4] . The
differential signal intensity due to a thin layer at depth z is, from Jablonski et al, given by

dσ 
dI = T De Fx A∆ΩM exp − z/λ cos α dz (16)
dΩ

T , De , Fx , A, and ∆Ω are all characteristic quantities of the XPS machine, dΩ is the differential
cross section, λ is the IMFP, and α is the collection angle. For our detector, the collection angle is
0 degrees, so this cosine term goes to 1. We can take the integral of both sides for the overlayer,
which gives us
 
 t
Il = Cl 1 − exp − (17)
λl
Here, Cl is the intensity from the infinitely thick layer and serves to normalize Il . Similarly,
the intensity of the substrate layer can be calculated by taking a similar integral, but different
boundary conditions give a slightly different result:
 
t
Is = Cs exp − (18)
λl

6
Cs is the intensity of the uncovered substrate and also serves to normalize. Since the substrate’s
electrons are traveling through the overlayer, the IMFP of the overlayer takes a role in both in-
tensity functions.

Taking the ratio of these two intensities gives us the result


 
1 − exp − t/λl
Il Cl
=   (19)
Is Cs
exp − t/λl

Solving for the thickness t in this equation gives us the result

Il /Cl
t = λl log (20)
Is /Cs
Hence, if we can measure the intensities from electrons ejected from atoms in both the substrate
and the overlayer, then we can plug the signal peaks’ normalized intensities into equation 20 to
get an estimate for the overlayer’s thickness. If the two peaks are measured in the same spectrum,
then their normalizations will be identical, and they will cancel, leaving us with
 
Il
t = λl log +1 (21)
Is
In this study, hafnium oxide was deposited onto a silicon wafer at an unknown rate using atomic
layer deposition, and thickness measurements were performed as outlined above to calculate the
average thickness deposited per ALD cycle. The peaks due to 4f5/2 and 4f7/2 electron orbitals in
the hafnium dioxide can be seen in Figure 9.

Figure 9: XPS peaks resulting from ejection of 4f5/2 and 4f7/2 orbital electrons in the hafnium
dioxide with GLP curves fitted. These curves gave binding energies of 17.5 ± 0.03 eV and 19.2 ±
0.04 eV. The NIST database gives a range of 16.7 to 18.1 eV for the 4f7/2 orbitals and a range of
17.8 to 19.8 eV for the 4f5/2 orbitals, both for oxidized hafnium in hafnium dioxide.

The peak from the silicon 2p orbitals is shown in Figure 10.

7
Figure 10: XPS peak resulting from ejection of 2p electron orbitals in silicon and silicon dioxide
with a GLP function fitted. This curve gave a binding energy of 99.1 ± 0.03 eV. The NIST database
gives a range of 98 to 100 eV for unoxidized silicon.

There are many resources for obtaining theoretical calculations for IMFPs. The NIST IMFP database
reports the IMFP of hafnium oxide for energy of 1482.35 eV to be λHf = 2.272 nm. Plugging
this along with the integrated peak intensities into the thickness calculation formula gives us a
hafnium dioxide overlayer thickness of t = 1.4 nm. This was after approximately 10 cycles of
ALD, giving us a deposition-per-cycle of 1.4 Å.

Energy Bands in Solids


Approximate the potential in a solid as a Dirac comb, or a series of Dirac delta functions of inte-
grated strength aV and spacing a, as done in the Kronig-Penney model derived in Sze’s Physics
of Semiconductor Devices:
n=∞
X
V (x) = aV δ(x − na) (22)
n=−∞

To solve Schrodinger’s equation for this potential, first we take a single electron in a single Dirac
potential. Label the region to the left of the potential spike as region 1 and to the right as region
2. In either region, the potential is 0, and the wave function solutions are plane waves:
 
Ψ1,2 = A1,2 exp ik0 x + B1,2 exp − ik0 x (23)
where ~k0 is the momentum of the electron:

~k0 = 2mE (24)
The wave functions need to match up at x = 0, so applying the boundary condition Ψ1 (0) =
Ψ2 (0) gives us

A1 + B1 = A2 + B2 (25)
Since the potential is modeled by a discontinuous function, the wave functions’ derivatives won’t
match up at x = 0, so we have to integrate Schrodinger’s equation across − to + and take  → 0:
+ Z +
~2 d2 Ψ
Z
− 2
dx + aV Ψ(x)δ(x)dx = 0
2m − dx −

~2 dΨ
 

− − + aV Ψ(0) = 0
2m dx + dx −
dΨ1,2
Since dx
= ik0 (A1,2 − B1,2 , we select region 1 for Ψ(0) (either works) to give the constraint

i~2 k0
− (A1 − B1 − A2 + B2 ) + aV (A1 + B1 ) = 0 (26)
2m
8
Bloch’s theorem says that, for a periodic potential,

Ψ(x + a) = eika Ψ(x) (27)


This gives us the additional constraints

Ψ2 (a) = eika Ψ1 (0)

and
dΨ2 ika dΨ1

=e
dx x=a
dx x=0
These lead to

A2 eik0 a + B2 e−ik0 a = eika (A1 + B1 ) (28)

A2 eiko a − B2 e−ik0 a = eika (A1 − B2 ) (29)


Solve equations 4, 5, 7, and 8 to get

V ma2 sin (k0 a)


cos (ka) = cos (k0 a) + (30)
~2 k0 a
V ma2 2mEa2
Make the substitutions P = ~2
, K = ka, and α2 = k02 a2 = ~2
to simplify this down to

sin α
cos K = cos α + P (31)
α
Since | cos K| is limited to 1, only certain intervals of α give solutions to this equation. Since the
~2 α2
energy is a function of α, given by E = 2ma 2 , then only certain intervals of energy are permitted.

These are called the energy bands of the solid.


Due to the Pauli exclusion principle, only a pair of electrons can occupy each energy level, or all
of them would settle into the ground state energy. The probability of finding electrons in each
state with energy E is given by the Fermi-Dirac distribution:

1
f (E) =   (32)
1 + exp (E − EF )/kB T
The quantity EF is the chemical potential of all of the electrons, also called the Fermi level.
The density of electrons is given by the integral of f (E) multiplied with the density of states
g(E):
Z ∞
N
= g(E)f (E)dE (33)
V −∞

Semiconductor materials are characterized by a Fermi level inside of a gap between energy bands.
For such materials, the electrons of the highest energy occupy the energy levels immediately be-
low the gap, and so the amount of energy required for electrons to move up in energy level is
much greater than is likely to be provided by random thermal fluctuations in the material. This
maximum occupied energy band below the gap containing the Fermi level is called the valence
band, since it is filled with valence electrons. Since the energy band states above the gap are
primarily empty in the material, electrons that can make it into this band are then free to move
within the sample. Conversely, such electrons will leave behind a vacancy, or hole, in the va-
lence band that can then facilitate the movement of electrons within the valence band as well.
Hence, the band above the gap containing the Fermi level is called the conduction band of the
semiconductor.

Surface Charge Transfer Doping


As per the treatment of surface charge transfer doping in diamond given by Ristein in Structural
and Electronic Properties of Diamond Surfaces, we start with Gauss’s Law:

d2 V 1
= − ρ(z) (34)
dz 2 0

9
where V (z) is the electrostatic potential, ρ(z) the electric charge distribution, both as functions of
the thickness coordinate of the crystal, and , epsilon0 the electric permittivities in the material
and in vacuum, respectively. We can use this to relate the electric charge distribution to the
electric potential energy with the relation U(z) = −eV (z) to get Poisson’s equation for the
electric potential energy:

d2 U e
= + ρ(z) (35)
dz 2 0
We can use Fermi statistics to derive the probability density, onto which we can multiply −e to
get the electric charge distribution. We can find the total probability density gT (z) for a given
potential energy U(z) with the integral
Z ∞
D3 (z, E)
gT (z) = U (z)+E−µ0 (z,T ) 
dE
−∞ 1 + exp kT

However, this density includes all of the states with zero electrostatic potential energy, that is, all
of the neutral state density. In order to get the charge density for only the U(z) nonzero cases,
we have to subtract from gT (z) the integral with U(z) = 0, that is, we have to calculate
Z ∞ Z ∞ 
D3 (z, E) D3 (z, E)
ρ(z) = −e U (z)+E−µ0 (z,T ) 
dE − E−µ0 (z,T ) 
dE
−∞ 1 + exp kT −∞ 1 + exp kT

where D3 (z, E) is the three dimensional density of states and µ0 (z, T ) the chemical potential of
the diamond prior to any surface charge transfer. If we assume homogeneity in the diamond,
then we can drop the dependence of the chemical potential and the density of states on z and
consequently express the charge density as a function of the electrostatic potential energy:
Z ∞ Z ∞ 
D3 (E) D3 (E)
ρ(U) = −e U +E−µ0 (T ) 
dE − E−µ0 (T ) 
dE (36)
−∞ 1 + exp kT −∞ 1 + exp kT

U is still a function of the thickness, but since it is the only term in ρ(U) that has any z depen-
dence, we can express the charge density as a function of U and tuck the z dependence away
under the hood, although it is still present.

If we can integrate ρ(U) twice and exert two different boundary conditions to determine the con-
stants of integration, then we can solve for the electrostatic energy as a function of the thickness
coordinate within the diamond in order to determine the shape of the valence band. However,
proceeding any further requires either numerics or approximation.

When a material with a high electron affinity, ie a large difference between the energy level of
the vacuum and the minimum energy of the material’s conduction band, is layered onto diamond
crystal, electrons are pulled away from the diamond into the layered material, creating a thin
layer of holes near the surface of the diamond. The concentration of these holes increases near
the surface. In reality, this increase in hole concentration isn’t perfectly abrupt, but if this transi-
tion region between no induced holes and the high concentration near the surface is thin enough,
then we can approximate the charge density with a Heaviside function.

With positive electrostatic potential energy, the electrons will be induced to move towards a
location with lower energy, and so we should expect not to see any net charge for U > 0. At
U = 0, the chemical potential is equal to the Fermi level. Thus, the concentration of valence band
holes is given by the difference between electrons and holes donated by atoms within the diamond
as a result of local regions within the diamond reaching charge neutrality. With the Schottky
approximation, we assume that this charge density is constant for all electrostatic energies less
than 0, ie

ρ(U) = −e(nA − nD )Θ(−U) (37)


Now we integrate to get

e2
Z
dU
= − (nA − nD ) Θ(−U)dx + C1
dx 0

10
Since we are assuming a sharp transition from U less than 0 to greater than 0, we can change the
argument of the Heaviside to (x − x0 ) where x0 is the thickness of the sample at the transition
to get
e2
Z
dU
= − (nA − nD ) Θ(x − x0 )dx + C1
dx 0
This integrates to give us

dU e2
= − (nA − nD )(x − x0 ) + C1 (38)
dx 0
Our first boundary condition is that the first derivative of U is 0 when U = 0, which gives us that
C1 = 0. Integrating again gives

e2
U(x) = − (nA − nD )(x − x0 )2 Θ(x − x0 ) (39)
0
The Heaviside has been inserted manually in order to enforce the boundary condition that U(x) =
0 after x0 has been reached.

Equation 18 shows that the valence band is parabolic, with a maximum occurring at some point
within the crystal.

Works Cited
[1] Chen, W., Qi, D., Gao, X. and Wee, A. (2009). Surface transfer doping of semiconductors.
Progress in Surface Science, 84(9-10), pp.279-321.

[2] Crawford, K., Qi, D., McGlynn, J., Ivanov, T., Shah, P., Weil, J., Tallaire, A., Ganin, A. and
Moran, D. (2018). Thermally Stable, High Performance Transfer Doping of Diamond using Tran-
sition Metal Oxides. Scientific Reports, 8(1).

[3] Denisenko, A., Aleksov, A., Pribil, A., Gluche, P., Ebert, W. and Kohn, E. (2000). Hypothesis
on the conductivity mechanism in hydrogen terminated diamond films. Diamond and Related
Materials, 9(3-6), pp.1138-1142.

[4] Jablonski, A. and Zemek, J. (2009). Overlayer thickness determination by XPS using the mul-
tiline approach. Surface and Interface Analysis, 41(3), pp.193-204.

[5] Maier, F., Riedel, M., Mantel, B., Ristein, J. and Ley, L. (2000). Origin of Surface Conductivity
in Diamond. Physical Review Letters, 85(16), pp.3472-3475.

[6] Nebel, C. and Ristein, J. (n.d.). Thin-Film Diamond.

[7] Pathria, R. and Beale, P. (2012). Statistical mechanics. Amsterdam [etc.]: Elsevier.

[8] Ristein, J. (2006). PHYSICS: Surface Transfer Doping of Semiconductors. Science, 313(5790),
pp.1057-1058.

[9] Ristein, J. (2005). Diamond surfaces: familiar and amazing. Applied Physics A, 82(3), pp.377-
384.

[10] Sque, S. (2005). A First-Principles Study on Bulk and Transfer Doping of Diamond. PhD.
University of Exeter.

[11] Strobel, P., Riedel, M., Ristein, J. and Ley, L. (2004). Surface Transfer Doping of Diamond.
ChemInform, 35(40).

[12] Sze, S. (1981). Physics of semiconductor devices. New York [etc.]: John Wiley.

11

You might also like