PHYSICAL REVIEW B 96, 174426 (2017)

Thickness and angular dependence of the magnetocurrent of hot electrons in a magnetic tunnel
transistor with crossed anisotropies

C. Vautrin, D. Lacour, G. Sala, Y. Lu, F. Montaigne, and M. Hehn

Institut Jean Lamour, CNRS, Nancy-Université, B.P. 70239, F-54506 Vandoeuvre lès Nancy, France
(Received 3 September 2017; published 20 November 2017)

We have studied the thickness and angular dependence of the magnetocurrent of hot electrons in a magnetic
tunnel transistor (MTT) with crossed magnetic anisotropies. In a first step, we show that the magnetocurrent
increases with ferromagnetic layer thickness as for MTTs with collinear magnetic configurations. The maximum
magnetocurrent value is obtained to be 85%, which is close to the theoretical maximum value of 100% for
MTTs with crossed magnetic configurations. In a second step, we demonstrate that we are able to reproduce
both current vs field direction and current vs field intensity measurements in a framework taking into account a
reduced number of magnetic parameters and a simple cosine dependence of the hot-electron current on the angle
between magnetizations.

DOI: 10.1103/PhysRevB.96.174426

I. INTRODUCTION with a rotating electromagnet that can reach an applied field

of 2 T. A current of electrons, denoted Ie , is injected from the
Combining hot-electron spin-filtering effects in ferromag-
Cu/MgO/Co magnetic tunnel junction into the Co/Cu/[Co/Ni]
netic materials with the energy selectivity in spin-valve tran-
spin valve. The hot electrons that conserve their energy above
sistors led Monsma et al. to demonstrate huge magnetocurrent
the Cu/Si Schottky barrier height, typically around 0.7 eV, are
(MC) contrasts at room temperature [1]. The high MC value
collected into the Si substrate. This current of hot electrons is
reached through the spin filtering of hot electrons motivated the
development of magnetic tunnel transistors (MTTs), in which denoted Ic .
the injection of hot electrons is done using tunnel barriers Since our crossed magnetizations geometry cannot provide
[2–5] instead of Schottky barriers [1,6] simplifying the device the antiparallel configuration, we define the magnetocurrent
fabrication. Only collinear magnetization configurations have contrast between the crossed and the parallel magnetic config-
been studied in the past. Very recently, we reported the urations as
study of a MTT with crossed anisotropies by engineering  P 
⊥ IC − IC⊥
the crystallographic structure and the magnetic properties of MC = 100 , (1)
the spin-valve ferromagnetic layers [7,8]. IC⊥
In this study, we take advantage of the crossed magnetiza-
tions geometry to check the cosine angular dependence of the with ICP and IC⊥ being, respectively, the collected currents in
hot-electron current on the angle between both magnetizations the parallel and in the crossed magnetic configurations. The
as stated by Applebaum’s 2011 review paper [9]. Furthermore, parallel configuration is obtained for a magnetic field equal
as for MTTs with collinear magnetic configurations, we to 20 kOe applied in the sample plane [see Fig. 2(a)] while
report the magnetocurrent as a function of the magnetic layer the crossed configuration is obtained by decreasing the field to
thickness that increases with layer thickness up to 85%, slightly zero.
lower than 100%, the maximum theoretical magnetocurrent
that can be reached in the crossed magnetizations configura- III. RESULTS
tion.
The collected current in the semiconductor has been
measured as a function of the applied voltage for the spin valve
II. EXPERIMENTAL DETAILS in the parallel magnetic configuration or in a crossed magnetic
configuration as shown in Fig 2(b). As expected, the current
The MTT structure used in this study is identical to is higher in the case of the parallel magnetic configuration.
the one used in Refs. [7,8]. It is based on a multilayer Indeed, a hot electron has a larger mean-free path if its
grown by sputtering using an Alliance Concept DP850 UHV spin is parallel to the magnetization of the ferromagnetic
machine on a HF-prepared low-resistance Si with stacking layer. An electron having a spin of opposite direction suffers
Si/Cu(5)/Ta(1)/Cu(5)/Ni(0.6)/[Co(0.2)/Ni(0.6)]N /Cu(3.5)/ inelastic scattering events and loses a part of its energy. As a
Co(3)/MgO(2.7)/Cu(20)/Pt(5) (sketched in Fig. 1, consequence, if the spin valve is in the parallel configuration,
thickness in brackets in nanometers). In order to check the the electrons with the spin oriented parallel to the direction
magnetocurrent, the samples are processed using clean-room of both magnetizations can cross the spin valve with only a
facilities to make electrical contacts to define the emitter on small energy loss. On the contrary, an electron with a spin in
the top Cu/Pt(5), the base on the metallic Cu/Pt/[Co/Ni]5 / the opposite direction will lose too much energy to overcome
Cu/Co, and the collector on the Si substrate. A schematic of the Schottky barrier. Thus, a 100% spin-polarized current can
the electrical connections is given in Fig. 1. The electrical be collected. If the spin valve is in a crossed configuration,
transport measurements were done using an ARS cryocooler an electron will lose a part of its energy regardless of the

2469-9950/2017/96(17)/174426(5) 174426-1 ©2017 American Physical Society

VAUTRIN, LACOUR, SALA, LU, MONTAIGNE, AND HEHN PHYSICAL REVIEW B 96, 174426 (2017)

of the Schottky barrier with values close to those reported

previously [7]. At 0.68 V, the MC⊥ reaches 200% calculated
using ICP = −0.2035 nA and IC⊥ = −0.0687 nA. Considering
Eq. (3) that we will be introducing in the following, this
value of IC⊥ is possible only if ICAP , the hot-electron current
in this antiparallel configuration, is positive for this specific
applied voltage. We carefully checked that the origin of this
high MC⊥ is not linked to a spurious effect like a leakage
current of the Schottky barrier (that would be detrimental to
the magnetocurrent ratio value). The origin of this particular
situation is still under investigation and is beyond the scope of
this paper.
Since the MC is linked to the difference of mean-free path
between electrons having their spin parallel or antiparallel
to the magnetization of the ferromagnetic layer, the MC⊥
should increase with the layer thickness as observed previously
in MTT with collinear magnetizations configurations. We
have kept the Co thickness constant to 3 nm and varied
FIG. 1. Schematic band diagram of MTT on N-Si for hot-electron N, the number of repetitions of the [Co(0.2)/Ni(0.6)]N
injection. Voltage across the MgO tunnel barrier controls the injection
multilayer. Inside the multilayer, the thicknesses of the Co
energy of the hot electrons.
and Ni layers have been kept constant. Since the stack
[Co(0.2)/Ni(0.6)]N has perpendicular magnetization for N
spin direction: The current collected is between the collected between 1 and 5 [8], the MC⊥ has been measured for
current in the parallel and antiparallel configurations. five Si//Cu(5)Ta(1)Cu(5)Ni(0.6)[Co(0.2/Ni(0.6)]1−5 Cu(3.5)
Then, the MC⊥ versus applied voltage is obtained by the Co(3)MgO(2.8)Cu(20)Pt(5) magnetic tunnel transistors. In
normalized difference of both curves using Eq. (1). As reported Fig. 3(a), the variation of the MC⊥ as a function of the number
in Fig. 2(c), MC⊥ decreases slowly from 87% to 75% when the of repetitions is shown for an applied voltage of −1 V. As
applied voltage decreases from −0.69 to −1 V. This trend is the expected, the MC⊥ increases with the number of repetitions.
one expected for hot electrons since the contrast between the up The highest value of 85% is measured for the MTT with five
and the down spin channels decreases when the electron energy repetitions of the multilayer.
increases well above the Schottky barrier height. However, and According to Ref. [9], the collected current can be expressed
surprisingly, a strong MC⊥ is obtained just after the threshold as (this will be confirmed in the last part of the paper)
I = I ⊥ (1 + MC⊥ −
C C
−→ · −
uCo u−−→),
Co/Ni (2)
in which IC⊥ is the current collected in a crossed configuration
of the spin valve and MC⊥ is the crossed magnetocurrent
defined in Eq. (1). − u→ −−−→
Co and uCo/Ni are the unit vectors
pointing in the directions of, respectively, the Co and [Co/Ni]
magnetization. The current in the crossed geometry is then
given by
 
IC⊥ = 12 ICP + ICAP . (3)
By rewriting Eq. (1), we get
 P 
⊥ IC − ICAP 1
MC = 100 , (4)
AP
IC 1 + ICP /ICAP
1
MC⊥ = MCCol , (5)
1 + ICP /ICAP
FIG. 2. Measurements done at 50 K on a Si//Cu(5)Ta(1)Cu(5)
Ni(0.6)[Co(0.2/Ni(0.6)]4Cu(3.5)Co(3)MgO(2.8)Cu(20)Pt(5) MTT.
where MCCol is the magnetocurrent in the collinear magneti-
(a) Variation of the collected current as a function of applied field, zations geometry when switching from parallel to antiparallel
with the field applied in plane and applied voltage of 1 V. (b) configuration. While MCCol can reach several thousand percent
Current collected in Si as a function of the voltage applied for [5,6], the MC⊥ saturates toward 100%. With a measured
the parallel magnetic configuration (red curve) and for the crossed maximum MC⊥ of 85% at applied voltage of −1 V for the
magnetic configuration (green curve) of the spin valve. (c) Crossed MTT with five repetitions of the multilayer, we are close
magnetocurrent as a function of the applied voltage calculated via the to the ultimate value of 100% which can be obtained with
curves of the hot-electron current as a function of the applied voltage our definition of the crossed magnetocurrent. This indicates
in the parallel and the crossed magnetic configurations of the spin a very large polarization efficiency through the high-quality
valve. spin-valve structure.

174426-2
THICKNESS AND ANGULAR DEPENDENCE OF THE . . . PHYSICAL REVIEW B 96, 174426 (2017)

FIG. 5. Ic vs Happ measured at 60 K and at 1 V of injection bias,

with an in-plane field (orange curve) and an out-of-plane field (blue
curve). Continuous lines correspond to fits performed using the model
under discussion.

of the sample plane (OOP geometry) and so ϕH app = 90◦

or with the magnetic field applied in plane (IP geometry),
and so ϕH app = 0◦ . Figure 5 reports the obtained results in
these two configurations. In both cases, the magnetizations
of each layer are parallel and aligned along the magnetic field
direction for a high field (2 T). The same value ICP = −110 nA
of the collected current is obtained as expected. When the
FIG. 3. (a) Crossed magnetocurrent MC⊥ as a function field is decreased to 0 T, the magnetizations of both layers
of the number of repetition of the [Co/Ni] multilayer at an are orthogonal and the same value of the collected current is
applied voltage of −1 V, for a Si//Cu(5)Ta(1)Cu(5)Ni(0.6) obtained again in both cases:IC⊥ = −65 nA. Thus, the MC⊥ is
[Co(0.2/Ni(0.6)]N Cu(3.5)Co(3)MgO(2.8)Cu(20)Pt(5) MTT. The determined to be equal to 70%.
temperature is 50 K. (b) Calculated variation of the collinear We can reasonably state that the shapes of the curves are
magnetocurrent MCcol and of the crossed magnetocurrent MC⊥ as a
directly linked to the demagnetization field of the Co layer in
function of ICP /ICAP .
the OOP geometry, and to the perpendicular anisotropy of the
These conclusions are based on Eq. (2). In the last part Ni(0.6) [Co(0.2)/Ni(0.6)]4 in the IP geometry.
of the paper, we will use the fact that our MTT has a To determine precisely these two values, it is required to
crossed magnetization geometry to check the validity of (1) Model the magnetic response of the system with respect
Eq. (2). Indeed, our samples offer a unique opportunity to to the magnetic field and thus to obtain values of −−→ · −
uCo u− −→
Co/Ni
deepen our understanding on the angular dependence of the vs Happ .
magnetocurrent since the relative angle between the magneti- (2) Use Eq. (2) to compute the collected current knowing
zations of the MTT layers can be controlled between 0° and the value of − −→ · −
uCo u−−→
Co/Ni .
90°. In the following, a Si[100](HF)//Cu(5)Ta(1)Cu(5)Ni(0.6) The value of uCo · −
−−→ u− −→
Co/Ni vs Happ in both geometry IP and
[Co(0.2/Ni(0.6)]4 Cu(3.5)Co(3)MgO(2.8)Cu(20)Pt(5) device OOP has been determined thanks to numerical computation
has been used and an injection energy of 1 eV has been used. supposing that
The angles notation is given in Fig. 4. (i) The magnetizations of both magnetic layers remain
In a first set of experiments, the collected current Ic has uniform regardless of the values and the directions of the
been measured either with the magnetic field applied out applied field,
(i) The magnetizations of each layer are independent from
each other (uncoupled magnetic system),
(i) Only a first-order uniaxial anisotropy is acting in each
layer.
In this framework, we divide our problem into two uncou-
pled systems having each a volume energy density, normalized
to saturation magnetization, of the form
EZ + EA + ED , (6)
FIG. 4. Definition of the angles using for the experiments −−→
performed in the last part of the paper. (X,Y) is the plane of the where EZ is the Zeeman energy equal to −Happ · u, EA
sample. corresponds to the first-order uniaxial anisotropy expressed

174426-3
VAUTRIN, LACOUR, SALA, LU, MONTAIGNE, AND HEHN PHYSICAL REVIEW B 96, 174426 (2017)

−→ TABLE I. Set of used parameters to reproduce our measurements.

as 12 Han sin2 (cos−1 ( Han
Han
· u)), Han being the anisotropy field.
Parameters extracted from the Ic vs Happ fitting procedure are
The last term taking into account the demagnetization field is presented in green; the parameters used for the experimental angular
expressed as HD sin2 (ϕu ), where HD is the demagnetization dependence fitting are presented in black.
field, and ϕu is the angle between the magnetization direction
u and the sample plane, as defined in Fig. 4.
Each layer magnetization has only two degrees of freedom,
denoted in the following: θ[Co/Ni] and ϕ[Co/Ni] in the case of
the [Co/Ni] multilayer having an out-of-plane anisotropy axis
and θCo and ϕCo in the case of the Co layer with an in-plane
anisotropy axis. We have computed, in a recursive manner, the
stable state of magnetizations for each value of the field and
in both geometries (IP and OOP).
Our best fits of the Ic vs Happ curves are presented in Fig. 5
by the blue and orange continuous lines. This fitting procedure accompanied by both the appearance of irreversible (hysteric)
allows us to obtain a value of 0.55 T for the perpendicular parts and the loss of the cosine shape.
anisotropy field of the [Co/Ni] multilayer, and a value equal By using the values of I⊥ , MC⊥ , HD , and Han previously
to 1.6 T for the demagnetizing field of the 3-nm-thick Co obtained, we attended to compute the angular measurements.
layer sandwiched between the Cu layer and the MgO tunnel The values of θ[Co/Ni] , ϕ[Co/Ni] , θCo , and ϕCo are determined
barrier. It is about 10% lower than 1.8 T, which is the expected by minimizing the energies of two independent subsystems
value neglecting perpendicular anisotropy originating from the for each value of ϕH app . This enables then the calculation of
MgO/Co interface. The obtained fits are well reproducing the Ic (ϕH app ) thanks to Eq. (2). Our fits of the experimental data
measured curves except for the hysteretic parts present at are presented by the continuous lines in Fig. 6. Table I presents
low fields. This is certainly a limitation of the used model the parameters used to reproduce our measurements.
that supposes that the magnetization of each magnetic layer This set of parameters does not contain any surprising
remains uniform during its reversal. This uniformity condition value. A slight misalignment between the sample anisotropies
is probably not respected experimentally since the reversal axes and the magnetic field direction has to be introduced
of the Ni(0.6)[Co(0.2)/Ni(0.6)]4 magnetization in the OOP in order to obtain a reasonable agreement between the
geometry is likely to take place by nucleation propagation. computed signals and the measurements (see nonzero values
From this first step, the values of I⊥ , MC⊥ , HD , and Han have of θH an , ϕH an , and θH app ). Again, the magnetic part of the
been obtained. model failed to reproduce quantitatively the irreversible parts.
Then the collected current Ic has been measured as a As previously, we attribute this deviation to the nonuniform
function of ϕH app , the angle between the sample plane and the magnetic reversals occurring experimentally. However, it is
applied field. Figure 6 present measurements done for different remarkable that the model well captures the shape variation
magnitudes of the magnetic field (200, 1000, 3000, 5000 Oe). of our angular measurement when increasing the applied
During these experiments, ϕH app is increased from −50◦ to magnetic field. Furthermore, the variation of the base current
+310◦ and then decreased from +310◦ to −50◦ . The collected with the field magnitude is also very well reproduced.
current is measured at 60 K for an injection bias of −1 V. At
200 Oe (blue curve), a reversible signal having mostly a cosine
signature has been recorded. The magnetic-field increase is IV. CONCLUSION
In summary, by taking advantage of the crossed geometry,
we demonstrate that we are able to reproduce both current
vs field direction and current vs field intensity measurements
in a framework taking into account a reduced number of
parameters and a simple cosine dependence of the hot-electron
current on the angle between magnetizations. Furthermore, as
for MTTs with collinear magnetic configurations, we report the
magnetocurrent as a function of the magnetic layer thickness
that increases with layer thickness up to 85%, slightly lower
than 100%, the maximum theoretical magnetocurrent that can
be reached in the crossed magnetizations configuration.

ACKNOWLEDGMENTS
This work has been done partially in the framework of the
FIG. 6. Collected current Ic as a function of the field angle SpinPress ANR Project. Experiments were carried out on IJL
ϕH app , measured at 60 K and for an applied voltage Ve = −1 V. Project TUBE-Davm equipments funded by FEDER (EU),
The magnitude of the applied field is 200, 1000, 3000, or 5000 Oe. French PIA project Lorraine Université d’Excellence (Grant
Continuous lines correspond to fits performed using the model under No. ANR-15-IDEX-04-LUE), Region Grand Est, Metropole
discussion. Grand Nancy, and ICEEL.

174426-4
THICKNESS AND ANGULAR DEPENDENCE OF THE . . . PHYSICAL REVIEW B 96, 174426 (2017)

[1] D. J. Monsma, J. C. Lodder, Th. J. A. Popma, and B. Dieny, Phys. [5] Y. Lu, D. Lacour, G. Lengaigne, S. Le Gall, S. Suire, F. Montaigne,
Rev Lett. 74, 5260 (1995). and M. Hehn, Appl. Phys. Lett. 103, 022407 (2013).
[2] K. Mizushima, T. Kinno, K. Tanaka, and T. Yamauchi, IEEE [6] R. Jansen, J. Phys. D 36, R289 (2003).
Trans. Magn. 33, 3500 (1997). [7] C. Vautrin, Y. Lu, S. Robert, G. Sala, O. Lenoble, S. Petit-Watelot,
[3] R. Sato and K. Mizushima, Appl. Phys. Lett. 79, 1157 X. Devaux, F. Montaigne, D. Lacour, and M. Hehn, J. Phys. D
(2001). 49, 355003 (2016).
[4] S. Van Dijken, X. Jiang, and S. S. P. Parkin, Appl. Phys. Lett. 83, [8] C. Vautrin, Ph.D. thesis, Lorraine University (2017).
951 (2003). [9] I. Appelbaum, Philos. Trans. R. Soc. A 369, 3554 (2011).

174426-5