Optimization of magnetoresistive sensors for high powerapplications
Mafalda Vieira Garcia de Oliveira
Thesis to obtain the Master of Science Degree in
Engineering Physics
Supervisor(s): Prof. Susana Isabel Pinheiro Cardoso de Freitas
Examination Committee
Chairperson: Prof. Pedro Miguel Félix BrogueiraSupervisor: Profa. Susana Isabel Pinheiro Cardoso de FreitasMember of the Committee: Profa. Diana Cristina Pinto Leitão
June 2018
ii
Resumo
Sensores magnetoresistivos oferecem uma alternativa interessante a metodos convencionais the medir
corrente quando solucoes simples e compactas sao necessarias. Devido as suas sensibilidades al-
tas, ampla resposta de frequencias e a possibilidade de serem microfabricados em grande volume,
reduzindo os custos, estes sensores revelam ser uma optima escolha para sistemas de medir e moni-
torizar energia.
O presente trabalho compila a microfabricacao de uma bolacha de 15 cm com 136 juncoes magneticas
de efeito de tunel de AlOx conectadas em serie sem recozimento e a sua aplicacao num sensor
magnetico de alta corrente incorporado numa Ponte de Wheatstone.
Foi obtido um valor medio de TMR= 36.02% com uma nao-uniformidade de 8.9% em toda a bolacha,
correspondendo a um rendimento de 93.97%. Sensores com um produto RxA de 115kΩ.µm2, sensibil-
idade S = 0.9%/Oe, altura efectiva de barreira Φeff = 2.28 ± 0.06 eV e detectividade 16.8 nT/Hz1/2 a
30 Hz foram obtidos nesta tese.
Usando 136 MTJs conectados em serie como elementos resistivos da Ponte de Wheatstone alimen-
tada com 1 µA, um sensor de corrente foi desenvolvido com sucesso, apresentando uma sensibilidade
de 3.02 mV/V/Oe, tensao a campo nulo de Voffset = 9.8 mV (1.7%) e uma regiao linear de 80 Oe.
Estes sensores mantem o seu bom funcionamento com tensoes superiores a 100 V .
Palavras-chave: Magnetoresistencia, Juncoes de efeito de tunel, Ponte de Wheatstone,
Sensor de corrente
iii
iv
Abstract
Magnetoresistive sensors offer an interesting alternative to conventional methods of current sensing
when simple and compact solutions are a requirement. Features like high sensitivity, wide frequency
response and the ability to be microfabricated in high volume, reducing the costs, make these sensors
an optimal choice for power monitoring systems and energy meters.
The present work compiles the microfabrication of a full 6 inch wafer with annealing free 136 AlOx-
based magnetic tunnel junctions elements connected in series and its applicability in a magnetic high
current sensor based on a Wheatstone Bridge.
An average value of TMR= 36.02% with a non-uniformity of 8.9% across a full wafer was achieved,
giving a 93.97% yield. Sensors with a RxA product of 115kΩ.µm2, a sensitivity S = 0.9%/Oe, an effective
barrier height ofΦeff = 2.28± 0.06 eV and a detectivity of 16.8 nT/Hz1/2 at 30 Hz were obtained.
Using series of 136 MTJs as resistive elements of a Wheatstone Bridge biased with 1 µA, a sensitivity
of 3.02 mV/V/Oe with an offset voltage of Voffset = 9.8 mV (1.7%) and a linear range of 80 Oe was
successfully achieved for the current sensor with an accuracy of 4.4%. These sensors can hold up
voltages in excess of 100 V without breaking down.
Keywords: Magnetoresistance, Magnetic tunnel junctions, Wheatstone bridge, Current sensor
v
vi
Contents
Resumo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 State-of-the-art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Thesis Outline and Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Theoretical Background 5
2.1 Magnetoresistive Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Anisotropic Magnetoresistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.2 Giant Magnetoresistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.3 Tunnel Magnetoresistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Magnetic Tunnel Junctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 MTJ Resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1.1 Simmons’ Tunnelling Model . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1.2 IV Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 Linear Magnetic Tunnel Junctions . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.3 Sensor sensitivity and voltage dependence . . . . . . . . . . . . . . . . . . . . . . 16
2.2.4 Noise sources in MTJs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.4.1 Thermal Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.4.2 Shot Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.4.3 Random Telegraph Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.4.4 1/f Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.5 Detectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Wheatstone Bridge sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Electrical Current sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
vii
3 Experimental Techniques 25
3.1 Fabrication Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.1 Sputtering Deposition Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.1.1 UHV-II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.1.2 Nordiko7000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1.2 Ion Beam Deposition Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.1.2.1 Nordiko3000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1.3 Photolithography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1.3.1 Vapour Prime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1.3.2 Coating and Development - SVG system . . . . . . . . . . . . . . . . . . 31
3.1.3.3 Direct Write Laser - Heidelberg DWL 2.0 system . . . . . . . . . . . . . . 31
3.1.4 Etching Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.4.1 Ion Milling - Nordiko3600 . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.5 Lift-off . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.1.6 Annealing Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Characterization Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.1 Profilometer - Dektak 3030 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.2 Ellipsometer - Rudolph Auto EL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.3 Magnetotransport characterization systems . . . . . . . . . . . . . . . . . . . . . . 37
3.2.3.1 Automatic Prober Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.3.2 140 Oe Manual Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.4 Noise Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4 Microfabrication of MTJ 41
4.1 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1.1 Microfabrication Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1.1.1 MTJ Stack Deposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.1.1.2 Bottom electrode definition . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.1.1.3 Pillar junction definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.1.4 Electrode insulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.1.1.5 Top electrode definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.1.1.6 Final passivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.1.1.7 Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.1.2 Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.1.2.1 Magnetotransport curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.1.2.2 Uniformity of TMR and RxA . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.1.2.3 IV Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.2.4 TMR dependence on Voltage . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.1.2.5 Voltage Breakdown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
viii
4.1.2.6 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5 Current Sensor 57
5.1 Sensors integration and Device layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 Wheatstone Bridge Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2.1 MR characterisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2.2 Wheatstone Bridge transfer curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2.3 Induced Magnetic Field calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 Final Sensor and Encapsulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3.1 Encapsulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.3.2 Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6 Conclusions 69
Bibliography 71
A Runsheet 77
ix
x
List of Tables
3.1 Al2O3 Oxide deposition conditions for UHV-II. . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Nordiko7000 set-point conditions for the metallization step. . . . . . . . . . . . . . . . . . 28
3.3 Read values for working conditions during etch in N3600. . . . . . . . . . . . . . . . . . . 34
4.1 Obtained values of some parameters for the sensors in study. . . . . . . . . . . . . . . . . 49
4.2 Summary of noise measurements for different sensor’s resistance and Vbias, before and
after annealing at 0 Tesla field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1 Results of the characterisation of the array sensors chosen for the Wheatstone Bridge. . . 58
5.2 Results of the characterisation of the array sensors chosen for the Wheatstone Bridge. . . 63
xi
xii
List of Figures
2.1 Change of resistivity of a NiCo alloy versus the induced external magnetic field. . . . . . . 6
2.2 Effect of current flowing through a trilayer formed by two ferromagnetic layers (FM) sepa-
rated by a non-magnetic (NM) metal layer, in the parallel and anti-parallel states. . . . . . 7
2.3 Schematics of a spin valve sensor structure. . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Schematic of the density of states of FM electrodes for both spins in the parallel and
anti-parallel states. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Simmons’ model diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.6 Linearisation method used in this thesis: shape anisotropy. . . . . . . . . . . . . . . . . . 15
2.7 MTJ sensor linear response to an external magnetic field. . . . . . . . . . . . . . . . . . . 15
2.8 Wheatstone Bridge circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.9 Different methods for measuring current. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.10 Magnetic field generated by the current line with different amplitudes. . . . . . . . . . . . 23
3.1 UHV-II system at INESC-MN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Uniformity of the deposition of Al2O3 for 1h at UHV-II, along the xx axis of a 6inch wafer. . 27
3.3 Nordiko7000 system at INESC-MN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 Nordiko3000 system at INESC-MN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 SVG coating and development tracks at INESC-MN. . . . . . . . . . . . . . . . . . . . . . 31
3.6 DWL 2.0 system at INESC-MN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.7 Schematics of the steps of the etching process. . . . . . . . . . . . . . . . . . . . . . . . . 32
3.8 Front view of Nordiko3600 at INESC-MN. . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.9 Schematics of Nordiko3600 assist gun used for Ion Milling etch . . . . . . . . . . . . . . . 34
3.10 Schematic of the lift-off process steps. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.11 Annealing at INESC-MN used for this work. . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.12 Dektak3030 Profilometer at INESC-MN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.13 Picture and scheme of the Automatic Prober Setup available at INESC-MN. . . . . . . . . 38
3.14 Automatic Prober setup at INESC-MN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.15 Noise measurement setup at INESC-MN. . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.16 Correspondent circuit for the noise setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1 3D schematics of the microfabrication process steps. . . . . . . . . . . . . . . . . . . . . . 42
xiii
4.2 AutoCad mask used for the microfabrication process of AlOx MTJs in this work. . . . . . 43
4.3 MTJ stack deposited in N3000. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.4 Schematics of bottom electrode definition. . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.5 Bottom electrodes definition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.6 Schematics of pillar junction definition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.7 Pillar junction definition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.8 Schematics of electrode insulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.9 Schematics of top electrode definition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.10 Top electrodes definition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.11 Schematics of the final passivation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.12 Picture from optical microscope of the final sensors, after final passivation. . . . . . . . . 48
4.13 TMR transfer curves of the sensors in study. . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.14 Wafermap of TMR of the full 6inch wafer for the 7474 MTJ series sensors over 37 dyes
measured in transport after microfabrication, not annealed. . . . . . . . . . . . . . . . . . 50
4.15 TMR versus RxA distribution of series sensors before and after annealing. . . . . . . . . . 51
4.16 Histogram of TMR and RxA distributions of AlOx 136 series MTJs. . . . . . . . . . . . . . 51
4.17 IIV curves for comparison between as deposited and annealed AlOx MTJ sensors. . . . . 52
4.18 IIV curves for comparison between as deposited and annealed AlOx MTJ sensors. . . . . 53
4.19 TMR ratio normalized (TMR/TMRmax) versus voltage. . . . . . . . . . . . . . . . . . . . 53
4.20 IV curves until breakdown of AlOx MTJs with and without annealing. . . . . . . . . . . . . 54
4.22 Noise level and Detectivity for annealed AlOx series sensors. . . . . . . . . . . . . . . . . 54
4.21 Noise level and Detectivity for annealing free AlOx series sensors. . . . . . . . . . . . . . 55
4.23 Minimum detectable variation in the wire current in function of the separation between the
current line and the sensor, for a detectivity of 67.9nT/Hz1/2. . . . . . . . . . . . . . . . . 56
5.1 Schematics of the PCB design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 Sensors mounted on the PCB and respective connections. . . . . . . . . . . . . . . . . . 58
5.3 R-H curves of the 4 array sensors chosen to mount on the Wheatstone Bridge. . . . . . . 59
5.4 Wheatstone Bridge voltage output as a function of the Magnetic field. . . . . . . . . . . . 60
5.5 Bridge voltage output as a function of Magnetic field, for various bias voltage. . . . . . . . 60
5.6 Schematics of the setup for characterising the current sensor. . . . . . . . . . . . . . . . . 61
5.7 Wheatstone Bridge voltage output as a function of the current passing the copper wire. . 61
5.8 Wheatstone Bridge voltage output as a function of the distance to the copper wire. . . . . 62
5.9 FInal current sensor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.10 R-H curves of the 4 array sensors chosen to mount on the new Wheatstone Bridge. . . . 64
5.11 New Wheatstone Bridge voltage output as a function of the Magnetic field. . . . . . . . . . 64
5.12 New Bridge voltage output as a function of Magnetic field, for various bias voltage. . . . . 65
5.13 Final encapsulation of the current sensor. . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.14 New Bridge voltage output as a function of the current oassingthrough the copper wire. . 66
xiv
5.15 Accuracy measurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
xv
xvi
Chapter 1
Introduction
1.1 Motivation
Thin film magnetoresistive (MR) sensors are of ultimate importance nowadays and widely used in var-
ious applications, ranging from ABS sensors [1] (in the automotive industry) and compasses in mobile
phones to detecting very weak magnetic fields (pT) at room temperature [2], which include biochips for
biomolecular recognition [3]. Tunnel Magnetoresistive sensors are driving the next generation of spin-
tronic devices and becoming a technology of choice due to their growing maturity, CMOS compatibility
[4] [5] and higher magnetoresistive ratios.
The current sensing technology has been addressed for many years and widely used, but magnetore-
sistive sensors offer an interesting alternative to conventional methods in applications where compact
and simple solutions are a requirement. This sensors have interesting features like high sensitivity, wide
frequency response and the ability to be microfabricated in high volume, which translates in a low cost
with optimal electrical characteristics alternative for power monitoring systems and energy meters [6],
for example.
The objective of this thesis is to develop a prototype of a Magnetic Tunnel Junction based current
sensor capable of measuring high currents in industrial applications.
1.2 State-of-the-art
Current measurements is of ultimate importance in the field of electronic instrumentation, being neces-
sary for many purposes such as control and regulation of any power converter, motor control, lighting,
DC power supplies, and so on [7].
There are several ways based on different physical principals of sensing a current. The most common
and simple one is with a shunt resistance based on Ohm’s law, which measures the voltage drop across
its terminals as proportional to the current flow. But as the shunt resistor is introduced into the current
path, this can generate a substantial amount of power loss and also restrict their use in high current
applications, which is the final aim of this thesis application [8].
1
An alternative to more conventional methods of sensing current is making use of magnetic field
sensors technology. The magnetoresistive sensors can be classified as anisotropic (AMR), giant (GMR)
or tunneling (TMR). GMR and TMR have been experiencing rapid growth due to their compatibility with
standard CMOS technology. These sensors offer higher resolution than Hall sensors and a sensing
direction in the chip plane. Magnetoresistive sensors bring great advantages, so there is already quite
some companies commercialising these solutions. However, Honeywell, Infineon and Tamura (some
of the most relevant competitors) still use Hall-effect devices. DowayTech is offering TMR technology
that can be used for current sensing applications. TMR technology can offer field sensitivities 104 times
higher, much lower power consumption, more compact chip and consequently package sizes, better
thermal stability and a much better detectivity when compared with Hall effect sensors [9]. The only
comparable limitation is the dynamic range, which in Hall-effect can go higher than 10000Oe while in TMR
only up to 1000Oe. The most used configuration of the magnetoresistive sensors is a Wheatstone bridge,
as it suppresses the temperature variation of the electric resistivity of the magnetoresistive material.
Usually, this type of current sensors have the current conductor already integrated with the sensors in a
single device to ensure stable geometry [10] [11], however, this way the sensor cannot be integrated in
an already designed circuit and can have some limitations when measuring high currents.
Most of the companies [12, 13, 14] offer open and closed-loop Hall-effect current sensors with very
different specifications. The current range can go as high as ±900A, accuracy is around 0.6% − 3.5%
and need a supply voltage of ±15V in the majority of cases. The price of these devices depends
hugely on the specifications wanted, but are usually around 10 − 60 euros. Currently available in the
market, the best TMR sensors that can be used in current sensing DowayTech is offering consists
of a full Wheatstone Bridge configuration with a voltage supply between 0V and 7V , sensitivities of
60mV/V/Oe, resistance ranging 8− 45kΩ, saturation field at ±11Oe, hysteresis below 0.8Oe and noise
of about 2nT/√Hz@1Hz. This device is being sold for approximately 80 euros. Other devices are also
available with similar characteristics as ours: 200kΩ resistance with 3mV/V/Oe sensitivity, hysteresis of
0.1Oe and saturation field of ±70Oe [15].
Other prototypes were previously developed at INESC-MN [74] [75], but with the current line incor-
porated on the PCB. This feature limits the usage of the current sensor in already mounted circuits or for
high power applications. Our sensor can be incorporated in any existing current line circuit and measure
high currents, which is a big improvement when these features are desirable.
1.3 Thesis Outline and Goals
The goal of this thesis is to fabricate a high current sensor for industrial applications using MTJ sensors
microfabricated at the clean room facilities of INESC-MN (class 10/100). A 6 inch wafer with MTJ
sensors is to be microfabricated and fully characterised. Four very similar sensors have to be chosen to
incorporate in a full Wheatstone Bridge configuration mounted on a PCB and characterised as a current
sensor. Finally, a 3D encapsulation is to be printed in order to protect the final current sensor and enable
its use in any wire for current sensing purposes.
2
This stated, the thesis is organised in the following way:
In Chapter 2 a theoretical background of magnetoresistive sensors and electrical current sensing is
presented.
In Chapter 3 the experimental techniques available at INESC-MN facilities and used in this work are
briefly explained.
In Chapter 4 the microfabrication process for the magnetoresistive sensors is demonstrated and the
results of the full wafer characterisation also presented.
In Chapter 5 the current sensor is designed, mounted and characterised.
In Chapter 6 the overall conclusions of the fabricated current sensor in this work are presented.
3
4
Chapter 2
Theoretical Background
2.1 Magnetoresistive Sensors
The magnetoresistive effect is the dependence of a material’s electrical resistance on the direction of the
magnetization, which can be changed by an external magnetic field. This effect is what makes it possible
to construct magnetic field sensors. By sweeping the magnetic field while measuring the resistance, its
maximum and minimum value, Rmax and Rmin respectively, allow us to express the magnitude of the
magnetoresistive effect. Usually expressed in percentages using the minimum resistance value as a
reference, magnetoresistance comes as:
MR(%) =Rmax −Rmin
Rmin× 100 (2.1)
There are different types of magnetoresistive sensors based on different physical effects, mech-
anisms and features, being the most relevant: the Anisotropic Magnetoresistance (AMR), the Giant
Magnetoresistance (GMR) and Tunnel Magnetoresistance (TMR). Through this chapter we will discuss
these types of magnetoresistive mechanisms.
2.1.1 Anisotropic Magnetoresistance
The Anisotropic Magnetoresistance (AMR) effect was discovered by Lord Kelvin in 1857 and is observ-
able in 3d transition metals and their alloys, due to the electron scattering in d orbitals [16]. This effect
consists in the change of electrical resistance with the orientation of the magnetization relative to the
direction of the electrical current in the material. For the majority of materials, the resistance is mini-
mum when the current flow and the magnetization are perpendicular and maximized when the current
flows parallel to the magnetization (because the probability of scattering is greater, hence maximum
resistance), as depicted in the example of Fig. 2.1.
Despite the early discovery of the AMR effect, it was not until a century later that it was used as a
magnetic field sensor in a reader unit of bubble memories. For this applications, the current direction
in the ferromagnetic alloy is always fixed while the direction of the magnetization is set by an external
5
Figure 2.1: Change of resistivity of NiCo alloy versus the induced external magnetic field [16].
magnetic field. As a result, the resistivity has a dependence on the angle Θ between current and
magnetization, being described as:
ρ = ρ⊥ + (ρ// − ρ⊥)cos2(Θ) (2.2)
with ρ⊥ and ρ// being the minimum and maximum resistivity respectively.
In thin films, the highest values for anisotropic magnetoresistance rarely go above 3%, although in
bulk specimens they can reach up to 6%. The AMR depends on many factors such as alloy composition,
film thickness, grain size, substrate and buffer layers and deposition rate and conditions [17].
2.1.2 Giant Magnetoresistance
Giant Magnetoresistance (GMR) is a quantum mechanical magnetoresistance effect that describes the
change of the electrical resistance of a magnetic multilayer structure when an external magnetic field is
applied. The physical origin of this effect is due to spin dependent electron scattering by layer interface
roughness and was first explained in 1988 [18] when Baibich first noted that the resistance of Cr/Fe/Cr
multilayers decreases more than 50% (at T = 4.2K )when the magnetizations of the Fe layers are par-
allel to each other. This is achieved by applying an external magnetic field to switch the magnetizations
that are naturally aligned anti-parallel.
The GMR effect can be observed when we have two ferromagnetic layers (FM) separated by a non-
magnetic metal layer (spacer). Depending on whether the magnetization of the adjacent FM layers are
parallel or anti-parallel to each other, a minimum or a maximum resistance will be obtained respectively.
Since scattering is dominated by collisions between electrons with opposite spin and similar energy,
when the electrons enter the ferromagnet, the scatter probability will rise for one of the spin orientations
and be suppressed for the other, depending on the relative orientation of the ferromagnet’s magneti-
zation. When the magnetization of the FM layers are anti-parallel, spin-up electrons and spin-down
electrons scatter at the same rate at each interface. When the magnetizations of the FM layers are par-
6
allel, the structure ceases to be symmetric, as both ferromagnetic layers will have a larger population of
one spin orientation on the interface. For this spin orientation, the scattering will be strongly enhanced,
contributing very poorly for the current flowing. On the other hand, electrons with the complementary
spin orientation will move comparably free in the structure due to a much smaller scattering rate, being
almost entirely responsible for the conduction current. In Fig. 2.2 we have a comparison of the two
situations described above, using a resistor model for the current passing through the trilayer structure:
a spin-up and a spin-down current.
Figure 2.2: Effect of current flowing through a trilayer formed by two ferromagnetic layers (FM) separatedby a non-magnetic (NM) metal layer, in the parallel (a) and anti-parallel states (b). The bottom of thefigure displays the respective equivalent circuits [19]
The magnitude of the GMR effect is translated by the magnetoresistance of the multilayer as pre-
sented before in Eq. 2.1. The typical value of magnetoresistance in GMR devices at room temperature
is around 10 − 20%, which is a major difference in comparison with AMR, despite the last having better
sensitivity as sensors. To improve this sensitivity issue of GMR sensors, in 1991 Dieny and co-workers
proposed a new type of GMR structure called Spin Valve (SV) sensor [20]. One basic example of a spin
valves is a four layer stack, with current flowing parallel to the plane of the layers, made of a non-magnetic
material sandwiched between two ferromagnetic layers together with an antiferromagnet adjacent to one
of the FM layers, as displayed in Fig. 2.3. The role of this anti-ferromagnet is to pin the ferromagnet
magnetization providing a fixed reference while the other FM is left free to align itself with the external
magnetic field. This configuration gives a lower resistance state when the magnetization of the pinned
and free layer are parallel and a higher resistance for the anti-parallel case. CoFe and NiFe are usual
choices for ferromagnets (due to their high MR and soft magnetic properties) and Cu is usually chosen
for the non-magnetic spacer.
2.1.3 Tunnel Magnetoresistance
Very similar to the spin valve structure discussed above, a Magnetic Tunnel Junction (MTJ) is composed
of two ferromagnetic layers separated by a thin insulating layer, usually AlOx or MgO. Thin enough
(typically 4 − 30A) for the electrons to be able to move from one FM electrode to the other, through
7
Figure 2.3: Schematics of a spin valve sensor structure. [21]
tunnel effect. This tunnelling process is a strictly quantum mechanical phenomenon, as it is forbidden in
classical physics due to the resistance of the barrier being infinite.
Again, the magnetization of one of the ferromagnetic layer is pinned to serve as a reference while
the other is free to move in order to align with an external magnetic field. The main functional difference
between MTJs and SVs is that in the former the electrons move across the tunnel barrier, i.e. the current
flows perpendicularly to the plane defined by the layers in the MTJ stack. When the magnetizations of
the two FM layers are parallel the resistance of the MTJ is low. When the magnetization of the free and
pinned layer are in anti-parallel state, the resistance is high.
In 1975 Julliere proposed the first model to explain the magnetoresistance for electrons tunnelling
between ferromagnets, based on two hypothesis [22]. The first is that all the electron quantum numbers
are conserved in the tunnelling process, since the wave function of the electrons is the same for both
electrodes. This states that the electrons are only able to tunnel between states of equal spin in the
two FM layers. The second hypothesis is that the probability of tunnelling is proportional to the product
between the density of states in the original and in the final ferromagnet, in agreement with Fermi’s rule.
The current density in the parallel (JP ) and anti-parallel (JAP ) configuration is therefore given by:
JP ∝ D1(↑)D2(↑) +D1(↓)D2(↓) (2.3)
JPA ∝ D1(↑)D2(↓) +D1(↓)D2(↑) (2.4)
where Di(↑) and Di(↓) represent the spin-up and spin down density of states at the Fermi level for
the ferromagnetic layer i = 1, 2, since only electrons near the Fermi level contribute for the conduction
process. Note that the ferromagnetic metals such as Ni, Co, Fe and their alloys have a strong spin
imbalance that results from the filling of the 3d band. This imbalance can be modelled by splitting the
bands for spin-up and spin-down electrons in energy, as can be seen in Fig. 2.4.
This imbalance in the density of states (DOS) at the Fermi level in the FM materials is normally
expressed as the spin polarization quantity, given by:
P =D(↑)−D(↓)D(↑)−D(↓)
(2.5)
From this definition we can rewrite the Eq. 2.1 for the magnetoresistance ratio as:
TMR(%) =2P1P2
1− P1P2× 100 (2.6)
8
Figure 2.4: Schematic of the density of states of FM electrodes and the resulting tunnelling currentintensity for both spins in the parallel state (left) and in the antiparallel state (right). [23]
This equation defines the TMR of an MTJ as a function of the polarization if the electrodes alone,
i.e. without any influence of the tunnel barrier. Therefore, this formula as been mainly used to infer the
polarization of different ferromagnetic alloys from the experimental data obtained for the TMR. Despite
the Julliere model being used as a fairly good description of tunnelling transport for many barrier mate-
rials, it is a very simple one. Over the years, refinements have been made in terms of accounting the
mutual influence of the FM electrode, which gives a dependence on the barrier parameters, and others
[24] [25]. However all models predict an increase of the TMR with barrier height and barrier thickness,
which is supported by experimental data.
In 1995, MTJ’s with amorphous Al2O3 tunnel barriers were fabricated and achieved a TMR ratio of
11.8% at T = 295K [26]. Being this value the highest reported to date for pseudo spin-valve structures,
it received a great deal of attention. Over the next years, the intense research on AlOx barrier based
MTJs was able to produce MR ratios at toom temperature (RT) up to 80% by optimizing the ferromagnetic
materials and the fabrication conditions of the aluminium oxide barrier [27]. Despite these advances, the
MR ratio values where still lower than the one required for several applications like MRAM cells, who
need to have ratios higher than 150% at RT. With this, AlOx based MTJ’s simply didn’t have MR ratios
high enough for the next generation applications. However, last year (2017), [28] reported a MR ratio
with AlOx of 220%, by testing nitrogen-doped amorphous CoFe as ferromagnetic free layers.
The theoretical prediction for MgO(001) based single crystalline MTJs showed the possibility of
extremely high MR ratios, because of the coherent tunnelling of electrons. The calculated optimistic
theoretical MR ratio in [29] is in excess of 1000% for an MgO barrier of about 20 atomic planes. The
reason for such a large difference of MR ratios is due to the fact that the moment of electrons during the
tunnelling through an amorphous barrier (like AlOx) is not conserved because of scattering within the
barrier, destroying any coherence or symmetry of conducting electrons. However, in a crystallized MgO
barrier, due to its electronic structure, the electron momentum parallel to the barrier is conserved and a
coherent tunnelling transport of electrons through the barrier occurs [29].
In 2004, two very similar studies reported magnetoresistive ratios at RT of 180% for Fe/MgO/Fe MTJs
fabricated by Molecular Beam Epitaxy [30] and 220% by sputtering methods [31]. ForCoFeB/MgO/CoFeB
MTJ structures, a 472% MR ratio at RT was reported in 2006 [32]. Up to date, the highest achieved value
found for MR ratio is 604% at RT, reported in [33] for a system of single barrier MTJs consisting of an
MgO barrier and CoFeB electrodes, annealed at T = 450C .
9
While TMR ratios have been increasing over the years reflecting improvements in the material depo-
sition, experimental results for MgO are still far from matching the theoretical predictions for the TMR,
indicating that large improvements are still possible.
2.2 Magnetic Tunnel Junctions
As introduced in the previous section 2.1.3, a typical structure of a MTJ stack consists of two ferromag-
netic layers, one free and one pinned, separated by an insulating barrier. The magnetization of the free
layer is set to rotate under the influence of an external magnetic field, while the magnetization of the
other layer is pinned by an exchange bias with an antiferromagnet that serves as the reference direction.
The reference layer can be either the bottom one (commonly called a bottom pinned MTJ) or the top
one (top pinned MTJ). Additionally, a buffer layer and a capping layer also compose the full MTJ stack.
The layers and respective functionalities will be presented in detail next.
Buffer When dealing with perpendicular currents relative to the plane of the stack, the main purpose
of the buffer is to provide a low resistance contact to the junction. Therefore it is primordial that the
buffer has a high electrical conductivity and low roughness, to not compromise the growth of continuous
thin films for the barrier. Ta and Ru alternate layers are usually used. Ta is an amorphous material with
resistivity of 154µΩcm when deposited by ion beam sputtering. This value is quite high, however the use
of a 50A Ta buffer layer reduces the resistivity of the NiFe and CoFeB layers [34]. The high resistivity
excludes using only Ta to provide a low resistance contact to the junction, therefore a Ru layer (resistivity
of ∼ 10µΩ.cm) is added to lower the total resistivity of the buffer layer. Note that a very low resistivity
material is only strictly needed for thin insulating barriers and very low resistance junctions.
Free Layer The free layer will provide the response of the sensor. Because of the rotation of its
magnetization with an external magnetic field, the resistance of the MTJ will vary and this variation is
what we aim to measure. Therefore, the properties of the free layer are primordial for achieving the
desired properties of an MTJ.
Ferromagnetic layers used as electrodes for tunnel junctions should be high polarization materials
in order to obtain higher TMR signals. The most common ferromagnetic materials used in magnetic
sensors are based on magnetic transition metals like Ni, Fe, Co and their alloys. These materials have
a low coercivity that allows for magnetization reversal in magnetic fields smaller than 100Oe.
Besides polarization, the response to a magnetic field is an equally important parameter since it
provides control over the signal response. Apart from low coercive fields, soft magnetic materials have
high permeability, allowing high sensibility to small magnetic fields. In this work, an amorphous CoFeB
alloy is used as it possesses spin polarization up to 65%, much higher than the values for Co and
Fe (43% − 45%) [35] or for CoFe (around 50%). Amorphous CoFeB ferromagnets has been extensively
exploited in spintronic devices and have been proven to show larger tunnelling magnetoresistance values
10
than those with crystalline CoFe [36]. NiFe is also used as part of the free layer electrode to improve the
magnetization reversal properties.
Insulating Barrier In tunnel junction depositions, the most challenging step is to deposit the insulating
barrier. The barrier must be a non magnetic oxide to prevent spin flip events in the tunnelling process
and form a stable oxide that preserves the integrity of the ferromagnetic electrodes. Also, the film needs
to be continuous and without pinholes while having a thickness below 20A to ensure current conduction
through tunnelling only.
To deposit the barrier, methods vary from bulk Al2O3 deposition or oxidizing a metal Al layer. When
using the former method, the properties of the barriers are usually worse than the latter and reproducibil-
ity is hard. On the other hand, results from oxidation of metallic Al films showed it is possible to produce
continuous insulating thin layers.
In this work, Al2O3 barriers are produced by depositing metallic Al films and exposing it to a remote
Ar-O plasma. The process was optimised in previous work [37]. The plasma is created away from the
sample with no acceleration energy provided to the ions, so that the oxidation occurs by impact of weak
kinetic energy oxygen atoms. Instead of producing the barrier by the deposition of a single Al layer
of the desired thickness followed by oxidation, in this work half of the desired Al layer thickness was
deposited and oxidized followed by another iteration of Al deposition and oxidation (called DSB - double
step barrier). This method was chosen as for equal nominal thickness, DSB showed higher TMR, lower
R × A and better MR (V) symmetry than for the corresponding single step barrier, as demonstrated in
[23]. [38] also states that double barrier are less prone to ageing than single barrier.
Pinned and Reference layers The signal of the magnetic sensor depends on the relative magneti-
zation direction of the two ferromagnetic layers. While one is the free layer as presented before, at the
other interface of the barrier lies the reference layer that should not rotate under the influence of external
magnetic fields. This can be achieved by stabilising the magnetisation of the reference layer in a given
orientation using one of two strategies: applying an exchange bias with an antiferromagnet or using a
synthetic antiferromagnet structure (SAF). In a SAF structure, instead of having a single ferromagnetic
(FM) pinned layer there are two FM layers separated by a very thin, smooth non magnetic spacer (in
our case a 6A Ru layer). Therefore, successive FM layers will be ferromagnetically or antiferromagnet-
ically coupled depending on the thickness of the non-magnetic metal. The Ru thickness was chosen
to maximize the antiferromagnetic coupling between the two layers. The pinned layer is coupled to an
antiferromagnet and the other FM layer is called reference layer.
To achieve a proper exchange bias, a good crystallinity of the ferromagnet at the interface with the
antiferromagnet is required, therefore amorphous CoFeB is to be avoided. For the pinned layer, a NiFe
alloy was chosen.
For the antiferromagnetic layer, MnIr was used despite its lower thermal stability compared to the
typical choice of MnPt for industrial applications, because it provides a good exchange bias with NiFe
without annealing.
11
Cap Layer The capping layer is a layer of a non magnetic metal that ends the tunnel junction stack,
preventing oxidation of the underlying metal layers. The oxide that results from the oxidation over time
of this last layer is removed by a soft sputter etch prior to top contact deposition, done in N7000. In our
case, the cap layer is composed of two layers of Ru and Ta. Ta is usually chosen as the last layer due
to its high reactivity with oxygen, allowing the formation of a self limiting oxidation at the surface that
prevents further oxidation of the underlying layers.
2.2.1 MTJ Resistance
As explained before, a TMR signal consists of the change in MTJ resistance, which occurs when the
angle between the magnetization of the electrodes changes due to an external magnetic field. To find
this value experimentally, the easiest way is to measure the resistance and its variation versus an applied
magnetic field. The absolute value of the junction resistance is inversely proportional to the junction area,
since the number of electrons tunnelling is proportional to the total electrode area, resulting in a higher
tunnelling current for the same applied voltage. Therefore, for the purpose of comparison between
different area junctions, the resistance area product R×A (Ω.µm2) is introduced.
2.2.1.1 Simmons’ Tunnelling Model
Nowadays, the model developed by Simmons in current flow between metal electrodes separated by a
thin insulating film [39] [40], is the most generally adopted in tunnel junction analysis.
The energy diagram, depicted in Fig. 2.5, shows that the insulator that separates the two metal
electrodes forms a barrier for the current flow, shown in Fig. 2.5.
Figure 2.5: Energy barrier diagram of tunnel junction for different electrode materials. [34]
The barrier height φ at each interface is the difference between the metal Fermi level and the con-
duction band of the insulator, and results from the work needed to move one electron from the metal
conduction band to the conduction band of the insulator. The barrier heights of each side (φ1 and φ2)
are different between them, except when the metal/insulator interfaces are exactly the same, which is
not easily achieved in practice.
When an external bias voltage is applied, conduction takes place, creating a potential difference
between the two Fermi levels.
12
2.2.1.2 IV Characteristics
A feature of MTJs is that they behave as non-linear resistive elements, as the current required to impose
a certain voltage across the MTJ does not increase linearly with the voltage. The tunnelling transport
across the barrier gives rise to a Current-Voltage (IV) relation that does not follow Ohm’s Law.
The relation between Current, I, and Voltage, V , (called the IV characteristics) can be described by
the Simmons’ model. For intermediate applied bias Voltage so that 0 < V (eV ) ≤ φ1,2, we have [39]:
I = k0A
t2
[(φ− V
2
)exp
[k1t
√φ− V
2
]−(φ+
V
2
)exp
[−k1t
√φ+
V
2
]](2.7)
k0 =e2
2πh, k1 =
4π√
2m
h
√e (2.8)
whereA is the junction area, t the barrier thickness, φ the effective barrier height, e = 1.6×10−19C the
electron charge, m = 9.11× 10−31kg and h = 6.63× 10−34Js2 the Planck’s constant. The barrier height
φ is an average value between φ1 and φ2. Theoretically, the values of φ calculated from the positive and
the negative branch of the IV curve should be the same if the interfaces are equal. However in practice,
because the two barrier interfaces are not exactly the same even if the ferromagnetic materials chosen
are the same, the obtained values of φ from both analysis can differ. This barrier asymmetry can be
translated in ∆φ = φpos − φneg.
By substituting the constants and converting units from the expression above, we obtain the following
equation:
I = 307.277A
t2
[(ϕ− V )exp[−0.71926t
√ϕ− V ]− (ϕ+ V )exp[−0.72363t
√ϕ+ V ]
](2.9)
with A being the junction area in µm2, V the voltage in V , t the barrier thickness in A, I the current
in A and ϕ half the barrier height φ in eV (ϕ = 2φ).
By fitting the experimental IV curve to equation 2.9, one can obtain the values for the effective barrier
height φ and the barrier thickness t. The equation shows that the tunnelling resistance depends expo-
nentially on the barrier thickness and height (R = V/I ∝ exp[t√φ]). Nowadays, one of the goals in tunnel
junction research is to decrease the junction resistance, since lower resistances allow better signal to
noise ratios and faster access times. This resistance can be lowered by reducing either the thickness
or barrier height. The first one is preferable, as it is easier to control the thickness of the insulator than
change the barrier height. [34]
Simmons also discriminates the very low bias voltage regime V ' 0, which assumes only tunnelling
conductance and simplifies the former expression to
I = k0k1AV√φ
2texp[k1t
√φ] (2.10)
In this particular case, I and V are directly proportional and the junction has an Ohmic response.
13
2.2.2 Linear Magnetic Tunnel Junctions
MTJs have several different applications and can have two different magnetic responses to an external
magnetic field: a square response, where the MTJ is either in the parallel or anti-parallel state and has
hysteresis, or a linear response, where the magnetization of the free layer rotates coherently with an
external magnetic field and the sensor has a linear change in resistance until it reaches the parallel or
anti-parallel state. In order to obtain this linear behaviour, the free layer must be orthogonal to the pinned
layer in the absence of an external magnetic field.
For distinct applications, different responses are in need. For instance, magnetoresistive random
access memory, MRAM, require a square response, while in field sensing applications a linear response
with no coercivity is needed.
There are two different possible configurations for the MTJ stack: parallel and crossed anisotropies.
In parallel anisotropy, both the free and pinned layer have their easy axis defined in the same direction,
while for crossed anisotropy, the free layer is deposited with its easy axis perpendicular to the one of the
pinned layer. In a normal deposition at INESC, the easy axis of all the FM layers is defined with the same
external magnetic field, which introduces an intrinsic anisotropy that leads to parallel magnetization and a
square response. In order to get the desired linear response, there are different linearisation techniques:
• Some deposition systems allow to change the applied magnetic field during the process between
the deposition of the free layer and the pinned layer, setting the easy axis orthogonal to one
another, inducing crossed intrinsic anisotropies
• Reduce the sensing layer thickness to the superparamagnetic limit [41]
• Use an antiferromagnetic (AFM) thin layer to weakly pin the sensing layer by a small exchange
bias set orthogonal to the reference layer [42] [43]
• Use integrated permanent magnets or current lines to introduce a bias field [44]
• Use sensor geometry with a high shape anisotropy that take advantage of the self-demagnetising
field of the free layer [45]
In this work, the used technique is shape anisotropy as depicted in Fig. 2.6.
Shape Anisotropy With MTJs shaped in rectangular form, we can use the self-demagnetising field
Hd to obtain the orthogonality of the magnetisations between the reference layer and the free layers.
Since we have a much larger width w and height h than thickness t in both layers, we can assume that
the free layer magnetisation is in the sensor plane and so it is true for it self-demagnetising field. From
[47], assuming w, h >> t, the demagnetising fields along its height Hhd and along its width Hw
d are given
by:
14
Figure 2.6: (a) Square response and (b) linearisation method used in this thesis: shape anisotropy. Thearrows represent the direction of the layer magnetization when no external magnetic field is applied.Adapted from [46].
Hhd =− 8Mh
s
4π
t√w2 + h2
w
hcosθ
Hwd =− 8Mw
s
4π
t√w2 + h2
h
wsenθ
where Mhs and Mw
s are the saturation magnetisation along h and w respectively. If we consider
the easy axis along its height, we can increase the aspect ratio w/h so that Hhd is more dominant,
to counteract the intrinsic anisotropy and obtain an orthogonal magnetisation. Higher aspect ratios
enhance the sensor’s linear operating range but lowers its field sensitivity. This method results in a MTJ
characteristic curve as shown in Fig. 2.7.
Figure 2.7: MTJ sensor linear response to an external magnetic field
15
2.2.3 Sensor sensitivity and voltage dependence
In a linear magnetic response normalized by the sensor’s minimum resistance value (Rmin), the MTJ
sensor sensitivity (Ssensor) is given by the slope of the linear transition and is expressed by:
Ssensor =1
Rmin× Rmax −Rmin
∆H=TMR
∆H[%/Oe] (2.11)
with ∆H being the field range where the behaviour is linear. Consequently, a bigger slope in the linear
response corresponds to a more sensitive sensor, which is able to considerably change its resistance
value for a small variation of the applied field.
One feature of MTJs that cannot be found in AMR or GMR devices is the dependence of the TMR on
the bias voltage applied across the device. Therefore, the TMR value changes with the applied voltage
V , it is almost constant (TMR0) for voltage values below 30mV but decreases quite drastically for higher
voltages. The decrease of the TMR value is almost linear up to the voltage value at which the TMR drops
to half of TMR0, labelled as V1/2. This value is often used as a comparison of interface quality of similar
structured MTJs. The magnitude of the TMR decrease depends on the quality of the interfaces, barrier
type and on the ferromagnetic materials [48]. In junctions with contaminated interfaces or doped barriers
where big part of the conduction happens through defect states, the decrease of TMR with bias voltage
is larger that in good quality films [49]. Typically, V1/2 is in the 400− 800mV range [45].
The evolution of the TMR depending on the applied voltage can be described by the following ex-
pression:
TMR(V ) = TMR0
(1− V
2V1/2
)(2.12)
Consequently, the sensor’s sensitivity also has a dependence on the bias voltage and is given by:
Ssensor(V ) = S0
(1− V
2V1/2
)(2.13)
where S0 is the maximum sensor sensitivity, obtained at low bias voltages.
This TMR decrease phenomena occurs due to the presence of defects in the insulating barrier. If
the voltage across a tunnel barrier is increased beyond a certain point, called Vbreakdown (≈ 1.5V ), an
electrical disruption of the dielectric will result in the destruction of the junction. At this point one or more
pinholes will be formed across the tunnel barrier providing a low resistance metallic conduction channel
through which most of the current will flow, resulting in the loss of TMR [34].
This voltage breakdown will occur when a electrical field across the junction exceeds a critical value
Ecritical and is given by:
Vbreakdown = Ecritical × t (2.14)
where t is the barrier thickness. Assuming a dielectric strength of 10 × 108V/m from [50], for a 14A
thick AlOx barrier the voltage breakdown is estimated to be 1.4V , which is a value in accordance with
experimental results performed in MTJs with high resistance. For very low resistance MTJs the analysis
16
is more complicated, as many of these barriers show a gradual decrease of resistance and a signifi-
cantly lower TMR. This gradual change in resistance at the breakdown point is seen in devices failing
extrinsically, which derives from process defects related to the deposition of the aluminium layer and
unoxidized metal, rather than as abrupt change in resistance which is a signature of intrinsic breakdown
(voltage stress-induced degradation of a well-formed oxide) [51].
2.2.4 Noise sources in MTJs
Noise is an inevitable physical phenomenon which is present in any electrical circuit. It is generated by
every passive or active component of circuits and is manifested by fluctuations in the voltage measured
across components and in the current flowing through them. So, in order to measure an interesting
magnetic signal, the amplitude of the voltage and/or current variations must be substantially higher than
the noise level introduced by the sensor in use, electronic equipment and surrounding environment.
In the case of an MTJ, the noise comes mainly from the electron tunnelling across the insulator barrier
and the magnetic fluctuations on the sensing layer. The noise level is, therefore, influenced by the quality
of the insulator barrier and the magnetic materials used. Several types of physical mechanisms give rise
to noise, so a brief description of the main noise sources present in a MTJ is given. These are thermal
and shot noise, 1/f electric and magnetic noise and random telegraph noise (RTN ).
2.2.4.1 Thermal Noise
Thermal noise (short term for thermal electronic noise), also known as Johnson-Nyquist noise due to
their first observation reported in 1928 [52], is the most basic form of noise that can be found in any
electrical device, as it is caused by the random thermal motion of electrons. Therefore, the noise power
is proportional to the device absolute temperature and resistance, vanishing as the temperature ap-
proaches 0K (no thermal electron agitation and random motion). In MTJs, thermal noise in independent
of frequency as it appears in both high and low frequency regimes, and can be considered as approxi-
mately white noise [53], which can be seen in the theoretical description by the Nyquist:
Vthermal =√
4kBTR [V/Hz12 ] (2.15)
where kB is the known Boltzman constant, T is the absolute temperature and R is the resistance of
the device. As drift velocities of electrons in conductors are smaller than their thermal velocities, thermal
noise does not depend on the current flowing through the resistor (bias current).
2.2.4.2 Shot Noise
Shot noise was discovered in vacuum tubes by Walter Schottky in 1918, where he demonstrated that
there is noise even when all extrinsic noise sources are eliminated. This noise results from current
flowing through discontinuities in a circuit. Therefore, elements without discontinuities, such as carbon
resistors, do not exhibit shot noise.
17
In MTJs, this phenomena arises from the electrons tunnelling across the insulating barrier, which
represents a circuit discontinuity. Each time a conduction electron goes through this discontinuity it will
carry an elemental electric charge e and a small contribution to the current flowing through the circuit.
As a result, the current is made of many elemental contributions initiated at random time instants and
therefore exhibits fluctuations on a short time scale.
Summing up, shot noise depends on the insulating barrier quality, increases with the bias current
and is independent of the frequency. In MTJ’s, it can be described by the following expression.
Vshot =√
2eIR2 [V/Hz12 ] (2.16)
being e the electron charge, I the sensor’s DC bias current and R the device resistance.
2.2.4.3 Random Telegraph Noise
Random telegraph noise (RTN) or the Barkhausen effect, was first systematically studied by Kandiah
and Whiting in 1978 [54], where they found that RTN was generated by the capture and emission of an
electron at a defect centre, therefore present in various types of semiconductor devices [53]. RTN in
magnetic tunnel junctions arises due to oxygen vacancies in the tunnel barrier. Such vacancies can trap
a conduction electron, increasing the resistance of the device, because it is more difficult for conduction
electrons to tunnel through a charged defect state. Being this a meta-stable state, when the system
returns to ground-state, the electron is released from the trap, leading to a decrease of the resistance
to its original value. Since the origin of this noise are defects in the tunnel barrier, it can be reduced
by the optimization of the fabrication process. Another source of RTN are thermal fluctuations of the
magnetization in the free layer, which can be eliminated by proper annealing, as the magnetic layers
become better crystallized resulting in a decline of magnetic fluctuations [55].
In MTJ’s, RTN was observed after eliminating magnetic noise by saturating the junction with an
external magnetic field and it became more dominant with the increase of the biasing current. At low-
frequencies, RTN is not always evident as it is shadowed by 1/f noise, which becomes domninant in
the low-frequency regime. Nevertheless, a magnetic transport curve exhibiting a step-like profile is an
evidence of the existence of RTN.
RTN increases with the bias current, so a low bias voltage and large R×A product are a strategy to
reduce the influence of RTN noise.
2.2.4.4 1/f Noise
A very critical noise source in MTJs is the 1/f noise, also known as flicker noise. Although 1/f noise
is universal and can be found in numerous transversal domains to electronics, a complete theory for
its origin is yet to be proposed. In electronics, the origin of the noise is attributed to charge trapping
electrons in barriers and between interfaces of tunnel junctions. When current flows through MTJs, some
charges become immobilized at the defects in the barriers, slowing down the mobility of the carriers.
18
This particles’ random release follows a probability amplitude that favours energy concentration at low
frequencies.
Subsequently, at low frequencies, the magnetoresistive sensor intrinsic noise is dominated by the
1/f component, which is responsible for the limitation of the sensor field detectivity, since its spectral
energy density decreases with frequency, following a 1/f power law [53]. Highly crystallized tunnel
barriers not only enhance TMR ratio but also mitigates the effect of 1/f noise in MTJs [56].
Besides the electric component of 1/f noise, there is also a magnetic one, that appears as fluctuation
associated with the magnetization alignment switching status at the interface between pinned and free
layer [57]. This magnetic noise is also inversely proportional to the frequency, and absent only in the
saturation states. In both types of 1/f noises, above a certain frequency value (called 1/f knee), the
thermal and shot noise become predominant. The empirical Hooge formula [58] is used to describe the
1/f noise spectral density:
V1/f =
√αH
fAIbiasR0 [V/Hz
12 ] (2.17)
where αH is called modified Hooge constant, A is the tunnel junction area (number of carriers is
proportional to A), f is the operating frequency, Ibias is the sensor bias current and R0 =(∂V∂I
)f
is the
sensor differential resistance at the operating point. The modified Hooge constant can be decomposed
in an electric and magnetic contribution (that depends on the applied magnetic field), associated with
the respective 1/f noise component:
αH = αelectric + αmagnetic (2.18)
The value of αH is used to parameterize the noise level of electronic noise and as an indicator to
compare different MTJ sensors, as for the same parameters, higher αH means higher intrinsic noise.
In MTJs, αH changes with the R × A product, increasing for a larger R × A. On the other hand, αH
decreases with the biasing voltage of the junction and with TMR. Generally, MgO MTJs possess a larger
electronic 1/f noise than of AlOx MTJs.
2.2.5 Detectivity
The sensor’s detectivity is defined by the lowest magnetic field variation that can be detected by the
sensor for a certain bandwidth and applied field, implying that below that value the signal cannot be
detected due to total noise level. Detectivity can be expressed by:
D =SV
∆V/∆H[T/Hz
12 ] (2.19)
where SV (V/Hz12 ) is the output noise of the sensor and ∆V/∆H(V/T ) is the sensor sensitivity. To
achieve lower detectivity levels, the sensor sensitivity must be the highest possible, which corresponds
to a TMR maximization and a linear field range ∆H minimization (i.e. increasing the linear response
slope). Also decreasing αH and increasing the sensor area A will improve the detectivity.
19
Beyond the aforementioned methods, one of the strategies that has been used for reduction of 1/f
noise is connecting junctions connected in series. In the following calculus we intend to predict how an
array of tunnel junctions will affect the detectivity levels.
The voltage noise in each individual junction can be expressed by the equation:
S2V (f) = 2eIr2coth
(eV
2kBT
)+αHV
2
A
1
f(2.20)
where r is the resistance of one sensor. All other variables and constants are described in the prior
sections. The first term represents the white noise (including thermal and shot noise) and the second
one the 1/f noise. When eV << kBT , the usual thermal noise relation is obtained, whereas for higher
voltage values, coth(
eV2kBT
)approaches 1, yielding the shot noise expression.
Considering now a device with N junctions in series, the noise spectral density is given by:
S2V (f) = N
[2eIr2coth
(eV
2kBT
)+αHV
2
A
1
f
](2.21)
where V remains to be the voltage of a single sensor. For the same configuration, the sensitivity for
N elements can be expressed as
∆V/∆H =
(∆R
R∆H
)RI =
(∆R
R∆H
)NrI = γNrI (2.22)
where γ is the TMR ratio per field unit.
The detectivity then comes as
D2 =S2V
(∆V/∆H)2=
1
N(γrI)2
[2eIr2coth
(eV
2kBT
)+αHV
2
A
1
f
](2.23)
If V << kBT and noting that rI = V , the prior equation simplifies to
D =1√N
1
γ
√4kBT
IV+αH
A
1
f(2.24)
From this formula, one can conclude that for N MTJs elements connected in series, the detectivity
levels decrease with increasing N by a factor of√N , enhancing the goal for better devices.
2.3 Wheatstone Bridge sensors
A Wheatstone Bridge is an electrical circuit popularised in 1873 by Sir Charles Wheatstone. It is com-
monly used in electronic devices to measure an unknown electrical resistance by comparing it with
well-defined resistances. The fundamental concept of the Wheatstone Bridge (see Fig. 2.8) is two
voltage dividers fed by the same input current and the circuit output is taken from both voltage divider
outputs. This voltage output V0 is given by the following expression:
V0 =
(R2
R1 +R2− R4
R3 +R4
)Vb (2.25)
20
Figure 2.8: Wheatstone Bridge circuit
The bridge is said to be balanced when the two voltage dividers have the exact same ratio, R1/R3 =
R2/R4, and no current flows in either direction being the output null. If the value of one of the resistors
changes, the bridge becomes unbalanced which allows current to flow through it. Even though a single
resistance can be used as sensing element, a Wheatstone Bridge is a recommended alternative so
that the change in resistance resulting from variations in strength or direction of the magnetic field is
perceived as a change in the output voltage.
For electrical current sensing applications, MTJ based devices are promising due to their high signal
output, small size, low cost and low power consumption. A Wheatstone Bridge configuration is usually
used as it nullifies the output signal when no sensing field is applied. Additionally it also improves the
thermal stability.
In this work, a full bridge configuration was used since it has a higher linear output signal when com-
pared with other configurations (Quarter and Half bridge configurations) and has an improved thermal
stability. In order to get this response, paired resistances need to be made active R1 = R4 = R + ∆R
and R2 = R3 = R−∆R, giving a voltage output of:
V0 =∆R
RVbias (2.26)
In this thesis, each resistive element of the bridge is composed of 136 MTJs connected in series.
Series sensors were used instead of single MTJ resistive elements to improve the electric robustness of
the device and the detectivity, as shown in the previous section 2.2.5.
2.4 Electrical Current sensing
Current sensing techniques can be classified based on its underlying fundamental physical principle [8]:
• Ohm’s law of induction (V = R× I), with shunt resistances
• Faraday’s law of induction, with coils
21
• Magnetic field sensors
Shunt resistances are the simplest way to measure current. The voltage drop across the shunt
resistor is used as a proportional measure of the current flow. But since the device is introduced into
the current conducting path (see Fig. 2.9(a)), this can generate a substantial amount of power loss that
restricts its use in high current applications. The Faraday’s law of induction makes use of a Rogowski
coil around a nonmagnetic core material around the primary current line as depicted in Fig. 2.9(b). This
type of sensor provides electrical isolation between the current one wants to measure and the output
signal. Another big advantage is it inherent linear response and that it does not saturate. But since the
basic principle is based on the detection of a flux change, without knowing the current at t = 0, it is
impossible to reconstruct the DC component.
(a) Schematics of the use of a shunt resistor. (b) Schematic of a Rogowski coil [8]
Figure 2.9: Different methods for measuring current
On the other hand, magnetic field sensors are able to sense both static and dynamic magnetic fields,
offering an attractive alternative to the techniques above. There are two main configurations for this type
of sensing: open-loop and closed-loop. In open-loop technology, the magnetic field sensor is placed in
close vicinity to the current wire and it assumes that the magnetic field around the conductor at a certain
distance is proportional to the current at all times. This configuration has the advantage of being simple,
inexpensive and very compact. One disadvantage is the need of in situ calibration to determine the
factor of proportionality between the magnetic field and the current, in order to achieve high precision.
Also, if the sensor is located close to the wire current, the skin effect (phenomenon that forces high-
frequency current to flow along the outer edges of the conductor, thus changing the magnetic field at
the sensor) can further reduce the accuracy of the measurement. But the most serious limitation is it
susceptibility to stray external magnetic fields that can disturb the accuracy.
In this work, an array of Tunnel Magnetoresistive sensors are used to detect the magnetic field
generated by an electric current flowing through a conductive wire placed above or bellow the sensor.
We can calculate the magnetic field created by the current by the general equation of the Biot-Savart
law:
~H(~r) =1
4π
∫V
~J(~r′)× ~r − ~r′
|~r − ~r′|3d3r′ (2.27)
Because the current line is a Cu wire and we can assume the length is much larger that the distance
22
between the centre of the current line and the sensor’s position given by d (l >> d), we can simply apply
Ampere’s law to calculate the expected generated magnetic field:
H(d) =µ0I
2πd(2.28)
where µ0 = 4π × 10−7T.m/A is the vacuum permeability constant and I is the electrical current
passing through the wire. Fig. 2.10(a) presents the calculated field in the x-direction created by the
current line in function of the distance to the sensor.
(a) Magnetic field in the x-direction generated by acurrent wire versus the distance between the linecentre and the sensor, for various values of current.
(b) Scheme of the magnetic field generated by the current lineand the position of the current sensor
Figure 2.10: Magnetic field generated by the current line with different amplitudes.
If our sensors are linear in a −30, 30Oe range then, in order to measure currents up to 200A, we need
the sensors to be distanced at least 1.3cm from the current centre in order to not saturate them. For
a current of 50A, this same distance generates a magnetic field of only 7.5Oe. This saturation occurs
when the linear range ends (see Fig. 2.7).
23
24
Chapter 3
Experimental Techniques
The experimental techniques used in the development of this work are briefly described in this chapter.
The microfabrication processes required are performed using the clean room facilities (class 10/100)
available at INESC-MN [59].
3.1 Fabrication Techniques
3.1.1 Sputtering Deposition Systems
Sputtering is a method based on a physical phenomena whereby transfer of moment between ions
of a plasma and a solid target of material, particles are ejected from the target and thin films can be
deposited.
In this work, thin film are grown by magnetron sputtering, a Physical Vapor Deposition (PVD) pro-
cess, in which a plasma is created by introducing atoms of an inert gas (commonly Ar or Xe) inside a
chamber, a negative bias voltage is applied to the target, while the shield around the target and cham-
ber are grounded. The created plasma is confined by a closed magnetic field on top of the sample
and positively charged ions are accelerated by a superimposed electric field towards the target, where
they collide with the negatively charged target atoms. The sputtered particles are then deposited on a
substrate. The magnetic field, created by a permanent magnet (magnetron) placed behind the target,
enhances the efficiency of the ionisation process, allowing to generate plasmas at lower pressures, re-
ducing the background gas incorporation in the thin film and the energy losses of the sputtered atoms
through gas collisions. The target consists of the material one wishes to deposit (in the cases of mag-
netron sputtering typically metals and insulating materials with specific optical and electrical properties)
and, since sputtered atoms are neutrally charged, they travel in direction of the substrate without being
affected by the magnetic field.
The target can be biased with a DC or RF power supply, depending on the material to be deposited.
RF is used with insulating materials, since the material does not allow a current flow. Consequently,
the accumulation of the negatively charged ions in the target will repel other incoming ions, interrupting
the deposition. Another option for sputtering deposition systems is, instead of grounding the substrate,
25
connecting it to a RF power supply, transforming the substrate in a target itself by attracting ions from
the plasma to the sample. This allows to remove deposited material from samples or cleaning metallic
surfaces from oxide residues. This process is called sputter etch.
3.1.1.1 UHV-II
UHV II is a manual sputtering system built in INESC-MN and installed in a class 10000 clean room. This
machine is dedicated to oxide deposition by sputtering from a Al2O3 ceramic target. For this work, this
system was used for the oxide Al2O3 layer deposition (as an insulator layer between two metals) in the
tunnel junctions.
The system, shown in Fig.3.1, consists of a single deposition chamber. The absence of a loadlock
implies that the deposition chamber needs to be vented each time a deposition is made in order to place
and remove the samples. The vacuum is made with a turbo pump so that a pressure of 3 × 10−7Torr
can be achieved, which can take approximately 10 hours. As the pumping speed cannot be controlled,
the only parameter that influences the process pressure is the gas flow.
Figure 3.1: UHV-II system at INESC-MN
The oxide is deposited from a target placed facing down under a 6 inch magnetron. The samples
are placed facing up in a cooled 6′′ diameter table under the target, therefore 6 inch Si wafers can be
processed in UHV-II as well as smaller area samples. The Ar plasma is created by an RF power supply.
The rate of deposition has a non-uniformity of about 11% as more material is deposited in the centre
of the table compared to the edges, as shown in Fig. 3.2(a). The uniformity test was performed by
firstly, drawing on the Si wafer equally spaced lines with a marker along the xx axis as depicted in 3.2(b).
Followed by the deposition of Al2O3 for 1h at UHV-II. Upon the deposition and cleaning carefully the
lines with acetone leaving the Si surface exposed. The deposited thickness of Al2O3 can be measured
with the profilometer in a step-like profile at the borders of the drawn lines.
The typical conditions for the oxide deposition in UHV-II are shown in the table bellow :
26
(a) Profile of the deposition of Al2O3 accross a 6 inch wafer inUHV-II.
(b) Schematic of the test performed for the deposi-tion rate of Al2O3
Figure 3.2: Uniformity of the deposition of Al2O3 for 1h at UHV-II, along the xx axis of a 6inch wafer.
UHV-II RF Power Pressure Ar Flow Deposition RateAl2O3 200W 2mTorr 43, 8sccm ≈ 12, 9A/min (at the center)
Table 3.1: Oxide deposition conditions for UHV-II
3.1.1.2 Nordiko7000
Nordiko 7000 (Fig. 3.3) is an automated machine composed of 4 process chambers, a dealer chamber
and a load-lock, where each chamber is pumped with a cryogenic pump and the load-lock with a turbo
pump. It was originally designed for 6 inch wafers but can also hold smaller batches using an adapter.
Wafer recipes allow the same wafer to be loaded into more than one module.
(a) Geometry of the chambers in N7000 cluster tool. (b) N7000 front view
Figure 3.3: Nordiko7000 system at INESC-MN
• Module 1 - Heating This chamber is used for rapid annealing processes. An array of lamps at
the bottom of the chamber, allows heating samples with temperatures up to 500 C. Although, this
module is not used anymore.
27
N7000 Module Power (W) Ar Flow (sccm) N2 Flow (sccm) Pressure (mTorr) Process time (s)
2 RF1 = 40RF2 = 60 50 − 3 60
4 2000 50 − 3 803 500 50 10 3 27
Table 3.2: Nordiko7000 set-point conditions for the metallization step
• Module 2 - Soft Sputter Etch This module can be used for material removal by sputter etch with
Ar ions (etch rate is ∼ 1 A/s) but since it is heated above 120oC in just 4 minutes, which is enough
to burn any photoresist used in the microfabrication process, it makes this process less suited
for microfabrication etching, and is therefore only used for soft sputter etch of the contact areas
prior to metallization steps, acting as a cleaning step to remove any oxides formed, enhancing the
area of ohmic contact of the film. The process uses two RF power supplies, one that biases the
substrate, accelerating the Ar+ ions towards it, and a second one on the bottom of the chamber
which is responsible for keeping the plasma stable.
• Module 3 - TiW Deposition This module is used to deposited the TiW(N) passivation layer. The
films are deposited by sputtering with Ar ions from a sintered 10 inch diameter Ti10W90 target.
During the deposition, a flow of N2 is introduced to incorporate 50% of nitrogen in the deposited
film. The wafer is loaded horizontally facing down and followed by a 90o rotation to face the vertical
target. The deposition rate is ∼ 5.5 A/s. A dummy wafer is loaded for target cleaning before the
sputtering process.
• Module 4 - AlSiCu Deposition For the deposition of the AlSiCu contact layer, sputtering from
a Al98.5Si1.0Ci0.5 target (also φ10 inch) by Ar ions is done in this module, using a DC power
supply. The wafer is loaded horizontally facing down and clamped to the substrate table on top
of the chamber, while the target is at the bottom of the chamber facing up. The deposition rate is
∼ 37.5 A/s and again, a dummy wafer is loaded for target cleaning before the sputtering process.
In this work, this machine is used for standard 3000A AlSiCu depositions with 1 minute soft etch prior
to the metallisation and a deposition of 150A of TiW layer after the metallisation to prevent the Al layer
oxidation. Table 3.2 presents the condition parameters for the metal deposition in the N7000.
3.1.2 Ion Beam Deposition Systems
Ion Beam Deposition (IBD) systems are used for deposition of thin films. In these systems, a highly
energetic ion beam is used to remove particles from a target that will deposit a thin film on a substrate.
The ion beam is created by an ion source called deposition gun, where a plasma is created in a vacuum
chamber inside a positively charged gun body. The power needed to ionize the gas atoms is provided
through an RF power supply. In order to extract the ions from the gun and focus the beam onto the
28
target, there is a set of three voltage biased grids that accelerate the positively charged ions as a
uniform collimated beam which will collide with the target. In this process, the ions hitting the target are
less energetic than in sputter processes, yielding in a lower deposition rate.
Besides the deposition gun and the assist gun (used for etching, described in section 3.1.4.1) there
are also two neutralizer guns inside the chamber pointed at the ion beams exiting the other two guns. The
neutralizers have two functions: the first is to guarantee the convergence of the beam, since the beam
changes shape and orientation over large distances due to electrostatic repulsion between the positively
charged ions. The second function is to prevent charge accumulation at the surface of insulator targets.
If the deposition gun is not neutralized when depositing an insulator, given enough time, the target will
become positively charged repelling the incoming ions from the target. In an extreme case, ions are
unable to reach the target and no deposition can take place. The same argument is valid for the assist
gun when ion milling an insulator.
The substrate table has a ring-shaped permanent magnet array mounted around it, which produces
a 40Oe magnetic field with the purpose of building anisotropy. This allows to define the easy axis and
exchange field directions during the deposition of magnetic materials. The substrate holder rotates
during deposition to assure a better surface uniformity throughout the sample. The target holder has
an hexagonal shape with different targets on each side, rotating according to the material that needs to
be deposited. The selected target is exposed to the ion beam while the other targets remain protected
from contamination by a shutter. Another shutter is also used to protect the sample until the assist and
deposition guns have stable ions beams, according to the set parameters.
There are two IBD systems installed at the INESC-MN class 100 cleanroom: Nordiko3000 and
Nordiko3600. The main difference between both is the size. Nordiko3600 is capable of processing
twelve 8inch wafers in a single batch, while Nordiko3000 can process eight 6inch wafers per batch. In
this work the N3000 was used to deposit the MTJ stack while the N3600 machine was used for etching
processes.
3.1.2.1 Nordiko3000
The Nordiko3000 Ion Beam System is a one module system with a loadlock separated from the main
chamber by a gate valve. Between the loadlock and the gate valve, a dealer, dedicated to handling the
samples, shares the same vacuum conditions as the loadlock.
In Fig. 3.4(a) a schematic diagram of the interior of the main chamber viewed from the back is shown.
As described in the beginning of this section, the system incorporates two ion beam guns (deposition
and assist gun), a substrate table and a target assembled vertically. The target drum can hold up to
six rectangular targets with dimensions of 15 × 10.4 cm. The deposition gun has a diameter of 10 cm
directed to the targets and the assist gun has a 25 cm diameter pointing directly to the substrate holder
[34].
The samples are placed facing down in the loadlock with the easy axis along a specific direction
marked on the sample holder. The holder is fitted with a permanent magnet array that provides a 40Oe
magnetic field to define the easy axis. To improve the uniformity both of the deposition and in the ion
29
(a) Schematic view of the main chamber in Nordiko3000.Adapted from [17]
(b) N3000 front view
Figure 3.4: Nordiko3000 system at INESC-MN
milling, the substrate table can rotate with frequencies up to 30rpm. The substrate table can also rotate
in order to change the angle between the substrate and the ion beams. A 0 pan angle corresponds to
a horizontal substrate table (loading position) while a 90 pan angle to a vertical substrate table (facing
the assist gun). Besides being used for ion milling, the assist gun can be used for reactive deposition,
assisted deposition, oxidation processes and ion beam smoothing. In N3000 the distance between the
deposition gun and the targets is relatively small (≈ 50cm), therefore even when no neutralizers are used
the beam remains focused.
3.1.3 Photolithography
Tunnel junction fabrication requires the patterning of micron sized devices, and to do so there is a need
to selectively deposit or remove material from a substrate, which is possible either by etching or lift-of.
In both cases, a mask must be transferred to the substrate in order to protect the structures that are not
to be removed. This can be achieved using an optical lithography process.
To transfer the pattern ti the substrate, the following steps need to be followed: mask design in
AutoCAD which is converted to a set of binary files stored in the lithography system hard drive, a vapour
prime step, photosensitive polymer (called photoresist) coating, lithography exposure and development
of the exposed photoresist (PR). Note that the masks can be converted as inverted or non-inverted
depending on whether the excess material is removed by etching or lift-off, respectively. When an
inverted configuration is used the laser exposes the areas outside the draw pattern, on the contrary,
when a non-inverted configuration is used, the laser exposes the areas inside the drawn pattern.
3.1.3.1 Vapour Prime
Vapour prime is a process used to improve photoresist adhesion on the sample’s surface, therefore, it
must be done before coating the sample with photoresist. The samples are placed inside an oven that is
pumped to vacuum and submitted to a cycle where HMDS (hexamethyldisilizane) is released at 130C
30
after a wafer dehydration step.
3.1.3.2 Coating and Development - SVG system
After the vapour prime, samples need to be coated with photoresist. There are two types of photoresist,
negative and positive. The positive photoresist (used for this work) has a photo-reactive component that
becomes unstable when exposed to light with a certain wavelength, which dissolves by developing. Due
to the possibility of inverting masks, the need of a negative photoresist is discarded.
In this work, using the standard procedure in the coating track (recipe 6/2), samples are coated with
a 1.5µm thick positive photoresist (PFR7790G27cP, JSR Electronics). The polymer is dispensed on
the substrate while spinning at 800 rpm for 1 minute to achieve the pretended thickness and uniformity,
followed by a 1 minute 85C soft bake, step to further improve uniformity and to evaporate solvents.
Once the sample is exposed, it is placed on the developer track, baked for 1 minute at 110C and
cooled down for 30 seconds in order to stop reactions after exposure. The developer is then poured
on the sample and left to dissolve the exposed areas for about 1 minute, not reacting with unexposed
areas. Finally, the residues are removed by a cleaning step consisting of washing with deionized water
and rotation.
Both tracks are integrated in the same system, shown in Fig. 3.5, and programed independently.
Figure 3.5: SVG coating and development tracks
3.1.3.3 Direct Write Laser - Heidelberg DWL 2.0 system
The optical lithography exposure with the DWL 2.0 system (Fig. 3.6) uses a 442nm diode Helium-
Cadmium Laser of 120mW (which can be adjusted for each exposure regarding the reflectivity of the
material’s substrate) to write the desired masks onto the samples photoresist which are placed on a
mechanical x-y stage and fixed by vacuum. The laser sweeps the sample in 200µm width stripes, has a
maximum resolution of 0.8µm (equivalent to the laser’s spot size) and an alignment precision of 0.1µm
(equivalent to the x-y stage resolution).
31
Figure 3.6: DWL 2.0 system
3.1.4 Etching Techniques
The pattern transfer process consists of two steps: lithographic resist patterning, described in the previ-
ous section, and the subsequent etching of the underlying material that is not protected by photoresist.
Finally, the substrate is submerged in microstrip, a resist stripper designed to remove the remaining
photoresist, leaving only the structures defined by the pattern, as depicted in Fig. 3.7. The resist pattern
can always be removed if found faulty on inspection, but once the pattern has been transferred on to
solid material by etching, rework is much more difficult, and often impossible.
Figure 3.7: Schematics of the steps of the etching process (thin films in yellow and green arrows repre-sent the etching). Not at scale.
There are two classes of etching: wet etching and dry etching (often called plasma etching). Wet
etching consists of a process to remove material from a wafer using liquid chemicals or etchants that
induce chemical reactions, consuming the original reactants. Usually a reduction-oxidation reaction
takes place, creating an isotropic etch profile. One big advantage of this process is that it can be highly
selective with very different etch rates for different materials. Also, it inflicts no damage on the substrate
and is considerably cheaper.
Dry etching refers to the removal of material from a wafer by exposing it to a bombardment of highly
energized ions created through a plasma. The particle beams with high kinetic energy are accelerated
towards the material, removing the unprotected parts by a physical sputtering process. In some film
removal processes, such as in magnetic head fabrication, dry etch has been replacing the wet etch, as
32
it shows improved process control and repeatability due to the facility in defining the parameters (mainly
time and angle)
At INESC-MN, Nordiko3000 and Nordiko3600 are used for dry etch processes and the LAM Rainbow
for reactive ion etching (RIE), which constitutes a mix of dry and wet etch. In this work, dry etching was
performed using the Nordiko3600.
3.1.4.1 Ion Milling - Nordiko3600
Ion Beam Deposition (IBD) systems, besides allowing deposition of thin films can also be used for ion
beam milling which consists of a dry etch process as stated before. This etching is based on the bom-
bardment of high energy Ar+ ions that are created in the assist gun and accelerated towards the sample
using an applied voltage difference at the beam exit grid, removing the unprotected stack material by
physical contact, in a highly directional and non-selective way.
Figure 3.8: Front view of Nordiko3600 at INESC-MN
Very similar to Nordiko3000 described in section 3.1.2.1, this machine (Fig. 3.8) is also composed
of 6 targets available for depositions, an assist and deposition gun. Besides offering the possibility
of loading 8′′ wafers, the N3600 ensures better uniformity in the process. The loadlock, dealer and
main chamber all have separate chambers, saving precious time in the pumping process. The main
chamber and the loadlock are pumped with turbo molecular pumps backed by rotary mechanical pumps,
in addition there is also a cryogenic pump connected to the main chamber. The deposition/etching
chamber can reach a base pressure of ∼ 10−8Torr, while the working pressure during an etch process
is around 10−4Torr.
The robot arm takes the loaded sample (facing up) from the loadlock to the substrate table, which
allows to define an angle between the ion beam and the sample. It is important to note the system’s
geometry, as the angle parameter inserted in the program corresponds to a +10 real angle, as shown in
Fig. 3.9. The sample is placed at a certain angle to reduce the probability of re-deposition of the etched
material. Also, to improve the uniformity of the etching, the sample is set to rotate at 30 rpm during the
33
N3600 RF Power Voltage+ Current+ Voltage− Current− Ar FlowEtch 200W 724V 104.4mA 344.8V 2.3mA 10.2sccm
Table 3.3: Read values for working conditions during etch in N3600
whole process.
Figure 3.9: Schematics of N3600 assist gun used for Ion Milling etch [60]
The etch profile is controlled by both incident angle and etch time, the latter is defined considering
the etch rate (∼ 1A/s for magnetic thin films) and the desired height to etch. The parameters used for
the etching process are presented in Table 3.3. The parameters correspond to the read values of the
assist gun, as it is the only gun used in ion milling processes (e.g. etch).
If the necessary etch time is higher than 220 seconds, cooling steps of 200 seconds need to be
added between each step. This prevents the photoresist from burning due to sample heating caused by
the continuous bombardment of particles.
During this work, etch processes were used to define the bottom electrodes and the pillars.
3.1.5 Lift-off
Contrary to the previous process, in lift-off, the material to be patterned is deposited on top of the
patterned photoresist (sacrificial layer). The sample is then immersed in a solution of Microstrip 3001
that acts as a photoresist solvent when heated at 65C, the efficiency of its action is improved when
also submitted to ultrasounds which facilitates the microstrip penetration into the photoresist. When the
photoresist under the deposited material dissolves, this material on top of it is lifted-off (see Fig. 3.10).
After the resist strip, the sample must be rinsed with isopropilic alcohol (IPA) with DI water and is
finally dried with compressed air.
3.1.6 Annealing Setup
All the work developed in this thesis was with top-pinned MTJ samples, which allows to get a final
working sensor without the need of annealing since the orientation of the magnetization of the layers is
easily defined in the moment of deposition. However, for bottom-pinned or MgO samples, an annealing
34
Figure 3.10: Schematic of the lift-off process steps (deposited thin films in yellow).
treatment is usually needed to obtain an exchange bias field between the antiferromagnetic and ferro-
magnetic layer adjacent to it. The annealing is used to define the alignment direction of the magnetic
moments to improve the magnetic properties of the films near the interface and to promote the diffusion
of the oxygen from the electrodes interfaces to the barrier, making it more uniform.
Nonetheless, some AlOx samples were annealed to compare with as deposited samples and to im-
prove some non-uniformities of the layers. The setup used for annealing at INESC-MN is presented in
Fig. 3.11(a) and it uses both thermal and magnetic processes for magnetic sensor optimization. The
sample is heated inside the oven at a constant rate of about 6C/min until the preset temperature is
achieved and maintaining it for a defined period of time. As the blocking temperature of the antiferromag-
netic layer Mn74Ir26 is around 250C [61], this is the maximum temperature allowed for this process.
At this blocking temperature, the exchange bias vanishes and the magnetic orientation of the AFM layer
is cleared. Then, the sample is moved and cooled down while exposed to a magnetic field created by
a permanent magnet of 1 Tesla until it reaches room temperature. The thermal cycle is represented in
Fig. 3.11(b). During this process, the system is kept in vacuum with a pressure below 10−6Torr.
(a) Annealing setup used for thermal treatment atINESC-MN.
(b) Thermal cycle for the anneal process. RT stands for roomtemperature, Tset is the preset temperature and t1 − t2 is thedurantion of the annealing plateau, after which the cool downoccurs naturally
Figure 3.11: Annealing at INESC-MN used for this work.
35
3.2 Characterization Techniques
In order to design and manufacture a high performance integrated circuit, the parameters of the man-
ufacturing process need to be carefully controlled: film thickness and material properties (resistivity,
optical) must be accurate, uniform and controlled, linewidths and edge profiles must fall within tight lim-
its, and the devices need to be free of defects that affect yield. So, metrology techniques are of maximum
importance not only after, but also during the microfabrication process, in order to optimize the stack and
fabrication.
During this work different characterization methods were used along the process of microfabrication
and are described in the following sections. The results are shown in Chapter 4.1.2.
3.2.1 Profilometer - Dektak 3030
Profilometry uses a contact diamond-tipped stylus sensor that allows to measure the topography of the
sample through a piezoresistive sensor. The sensor sweeps the sample surface for a defined range,
detecting any height changes such as pillars or holes in the film, causing it to move vertically. This
vertical movement is translated into electrical signals by a linear variable differential transformer, that
is mechanically coupled to the stylus. Profilometry has high sensitivity but certain limitations such as
slow measurement speed, contact between the probe tip and the sample and no thermal capability.
Nevertheless, this technique is very useful to control not only film thickness, but also the deposition
rate through a calibration sample, making possible to check if the processes are going according to
expectations and previsions.
A Dektak 3030 ST profilometer (Fig. 3.12) is used in this work. It uses a low force of typically 3mg
and has a vertical resolution of about 300A, the distance to sweep is defined by the user.
3.2.2 Ellipsometer - Rudolph Auto EL
Ellipsometry is used to determine the thickness and refractive index of thin films by measuring changes
in the state of polarization of collimated beams of monochromatic polarized light. The variations are
caused by the reflection from the surfaces of substances. At INESC-MN, the Rudolph EL ellipsometer
irradiates the surface of the sample at a known incident angle, using a Tungsten Halogen light source
giving a monochromatic ray of 405, 633 or 830nm wavelengths.
After being reflected in the sample, the beam has a different state of polarization, which can be
compared to the incident polarization state allowing to determine the angles ∆ and Ψ. Angle ∆ describes
the change in phase difference between the component p (parallel vector to the plane of incidence) and
the component s (normal vector to the same plane) and can be comprised between 0 and 360. Angle
Ψ describes changes in the amplitude ratio between the same two components, with values from 0 to
90. With this information and through a numerical model based on Fresnel equations, it is possible to
infer the refractive index and the film thickness.
Although this ellipsometer allows for other, more complex, types of measurement, for this thesis only
36
Figure 3.12: Dektak3030 Profilometer at INESC-MN
the most simple model of ellipsometry was used, where a dielectric must be deposited on top of a fully
reflective substrate with well known optical constants, such as Silicon. This method is used to control
the fabrication process, not only by measuring the thickness of deposited oxide (either Al2O3 or SiO2),
but also to ensure that the target is not contaminated by measuring the refractive index, which is a well
known constant for each material (1.62 for Al2O3 and 1.47 for SiO2).
3.2.3 Magnetotransport characterization systems
For the final characterization of the fabricated MTJ sensors, magnetic transfer curves (resistance of the
sensors as a function of magnetic field) can be measured with two different systems: a manual prober
setup commonly called the 140 Oe Setup and an automatic prober setup which allows measurements
on 150mm wafers.
3.2.3.1 Automatic Prober Setup
This setup uses two Helmholtz coils powered by a current source (Kepco BOP 50-4 D), which create
a static magnetic field from −140 to 140Oe between the coils. To perform electrical measurements,
the automatic prober is equipped with a current source (Keithley 220) and a multimeter (HP 34401A),
an optical microscope on top of the experimental setup for correct positioning and four micropositioner
tungsten needle probes, used to supply a constant electrical current and to measure the output voltage
at its terminals. One can choose to use either four probes (two apply current and two read voltage),
or only two probes used to apply current and measure voltage at the same time. The main difference
between the two methods is that in four-probe setup only the resistance of the sensor is being read,
37
while with two probes the measured resistance includes the contact resistance.
The Setup also has an automatic three axis stage (Thor Labs) which allows a maximum 220mm
range on the xx and yy axis, and 50mm on the zz axis. All the equipment is connected to a computer
and controlled by a software developed at INL. The program needs the input of the current, the intended
magnetic field sweep and an Excel file with the (x,y,z) coordinates of the points to measure. The output
is simply the number of the point in the sequence and the voltage obtained by the probes. In order to
get the magnetic transfer curve and TMR, one has to do a data analysis, converting the voltage using
Ohm’s Law into a resistance and current into magnetic field (Oe) using the calibration sheet available at
the computer.
Figure 3.13: Picture and scheme of the Automatic Prober Setup available at INESC-MN.
3.2.3.2 140 Oe Manual Setup
The 140 Oe setup (Fig. 3.14) is composed by two Helmholtz coils, four micropositioner Tungsten needle
probes with a spacial resolution of ≈ 10µm, a microscope, a Voltmeter and two Current Sources. One
of the current sources supplies current to the coils, creating a static magnetic field from -140 to 140 Oe.
Using the needle probes, a constant current is applied to the sensors and the voltage is measured. This
system is similar in everything to the automated prober setup except that the positioning of the needle
probes has to be done manually.
The computer software directly acquires the measured values and makes the initial data treatment,
providing a magnetoresistance curve aswell as some important parameters like sensitivity.
38
Figure 3.14: Automatic Prober setup at INESC-MN
3.2.4 Noise Measurements
The noise measurements are the last stage of the sensor characterization. In order to perform these
measurements the sample needs to be diced and wire-bonded to a chip carrier.
At the time the noise setup was used in this work, no magnetic field could be applied during the
measurements, therefore they were done at a zero Tesla field.
The chip with the sensor mounted on top is placed inside a shielded box, pictured in Fig. 3.15(a),
that protects the chip from any external noise. Note that the box is not connected to any source or power
grid. The correspondent box circuit is shown in Fig. 3.16. The circuit has one battery, typically with 9V ,
and two potentiometers to regulate the current that crosses the sensor, with the option to add one more
resistance in series for sensors with higher resistance values.
The output of the box is connected to an amplifier (SIM910 JFET) supplied by twenty 1.5V batteries.
The gain factor can be adjusted from 1 to 500, being the usual value a factor of 100 (40dB). With this
gain, the amplifier has a constant noise level of 4nV/√Hz.
For a complete noise measurement, two sets of two data acquisitions need to be obtained. Two made
in the 0-1 kHz range and the other two in the 0-100 kHz range, the first range yields a more detailed
measurement and consequently takes more time. For each range the sensor noise is measured with no
current and afterwards with applied current. The measurement with no current intends to gain data from
all noise related with other components of the circuit besides the sensor, acting like a calibration.
39
(a) Shield box to place the chip carrier with the sen-sor. This box has been actualized in the mean timefor a PCB with coils.
(b) Noise setup with the shield box inside the big black box con-taining also the batteries and amplifier
Figure 3.15: Noise measurement setup at INESC-MN
Figure 3.16: Correspondent circuit for the noise setup. Adapted from [62]
40
Chapter 4
Microfabrication of MTJ
4.1 Experimental Results
4.1.1 Microfabrication Process
The microfabrication process starts with the MTJ stack deposition followed by the 1st lithography to
define the bottom electrodes and an ion milling etch and resist strip. A 2nd lithography is used to define
the pillars and the stack is further etched until the stack antiferromagnetic layer is reached (outside the
bottom electrodes the stack is now completely etched until the substrate). Still with the patterned PR
from the last lithography, a layer of Al2O3 is deposited to isolate the bottom electrodes from the top
electrodes, and a lift-off process is done to remove unwanted oxides. A 3rd lithography to define the
top metal electrodes is performed and the aluminium is deposited and lift-off. Finally, the 4th lithography
is done to define the vias and passivate all the sensor by depositing a film of Al2O3. These steps are
shown schematically in Fig. 4.1.
41
Figure 4.1: 3D schematics of the microfabrication process steps. The bottom stack layer is representedin green, the Al2O3 barrier in black, the top stack layer in yellow and the photoresist in red. The z axis isnot at scale with the xy axis
42
The AutoCad mask is depicted in Fig.
Figure 4.2: AutoCad mask used for the microfabrication process of AlOx MTJs in this work. One dyecompiles 12 by 17 sensors, each with 136 junctions in series, and 3 test structures with a single junctionwith the same dimensions as the series sensors (40× 2 µm)
4.1.1.1 MTJ Stack Deposition
The first step in any top-down microfabrication process is to deposit the stack, consisting of all the layers
of the MTJ structure. In this work the Nordiko3000 was used. The deposition was done on top of a
6′′ Silicon wafer passivated with 100 nm of SiO2 and a stack (units in A): Passivated Si Substrate/
Ta(50)/ [Ru(150)/Ta(50)]x3/ Ni80Fe20(30)/ (Co70Fe30)80B20(30)/ Al(7)x2/ (Co70Fe30)80B20(30)/ Ru(6)/
Ni80Fe20(30)/ Mn76Ir24(180)/ Ru(150)/ Ta(50), as shown in Fig. 4.3. The barrier was deposited as
a thin metallic Al film in two equal steps and oxidized by a remote Ar − O plasma for 20 seconds
(7A+ 20s+ 7A+ 20s) [37].
The use of an Al2O3 insulating barrier is a simple and repeatable process that allows for a low
temperature anneal or annealing free functional devices [63] [23]. Therefore, the process becomes
cheaper and faster, which are factors to always keep in mind in microfabrication processes.
43
Figure 4.3: MTJ stack used for the AlOx sensors fabricated in this work, deposited in N3000. Thick-nesses in A
4.1.1.2 Bottom electrode definition
At this step, the bottom electrode is defined by photolithography followed by ion milling etch. In this
stage, the stack is not completely etched until the substrate because when the pillar structures are
etched, the bottom contacts will also be exposed. This results in an overall etch of the stack (excluding
the pillars) until the bottom electrodes are isolated as wanted. For this stack, the pillar must be etched
until the NiFe layer after the barrier so that the whole pillar has the same geometry, meaning that 520A
must be etched. As the whole stack has a height of 1170A, we need to etch 650A. In practice we always
overetch, as it is not critical and the consequences of under etching in the next step are critical. The
wafer is etched in N3600 by ion milling at a 45 angle in four steps of 175s each with cooling down steps
of 200s in between. An area of 130µm × 60µm is defined. Afterwards the sample is submitted to lift-off
to remove the photoresist.
44
Figure 4.4: Schematics of bottom electrode definition.
(a) Autocad mask of thefirst lithography for bottomelectrode definition.
(b) Picture from optical microscope after 1st lithography and etch(without PR) of the bottom electrodes defined
Figure 4.5: Bottom electrodes definition
4.1.1.3 Pillar junction definition
The 2nd lithography defines pillar junctions with an area of 40µm × 2µm and is again followed by an
ion milling etch step. In this etch, the material is removed until reaching the end of the NiFe layer after
the barrier (520A from the top of the stack) so that the entire free layer has the same geometry as the
junction pillar. In this step, it is critical that the barrier is fully etched or the sensor will be shorted, on
the other hand excessive over etch can also cause problems if the entire bottom electrode gets etched.
To avoid material redeposition around the barrier, the sample is etched in equally divided steps 130A
of alternating 70 and 40 angles, with etching rates of 1.05A/s and 1.15A/s respectively. Again, 200s
cooling steps were added between etch steps and 100s of over-etch was done at the end to ensure the
barrier and the free layer were etched through.
45
Figure 4.6: Schematics of pillar junction definition.
(a) Autocad mask of the2nd lithography step forpillar junction definition.
(b) Picture from optical microscope af-ter 2nd lithography with PR on the pil-lars
Figure 4.7: Pillar junction definition
4.1.1.4 Electrode insulation
After having the bottom electrodes and the junction pillars defined, an insulating Al2O3 layer of 1300A is
deposited using the UHV-II system. This layer is crucial to insulate the pillars and the bottom electrodes
from each other. This ensures that the electrical current flow happens through the barrier and not
through an alternative path. Afterwards a lift-off removes the excess of the photoresist together with the
oxide protecting the top of the pillars, leaving them uncovered to allow electrical contact with the top
electrodes.
The lift-off step is very critical because even when a single pillar in a series of sensors does not have
contact with the top electrode, the whole array sensor fails, as the current is interrupted mid-path and
cannot flow from one contact to the other. This phase of the process can take a long time (several days)
if the sample is placed in only microstrip and an ultrasound bath is used due to the high aspect ratio of
the pillars and the high step coverage of the oxide deposition on the PR walls.
46
Figure 4.8: Schematics of electrode insulation.
4.1.1.5 Top electrode definition
A 3rd lithography is done to define the junction top electrodes on top of the uncovered pillars, followed
by the deposition of 3000A of Aluminium plus 150A of TiW as protective layer in the N7000. Again, the
lift-off leaves only the metal of the contacts, with an area of 125µm× 56µm.
Figure 4.9: Schematics of top electrode definition.
(a) Autocad mask of the3nd lithography.
(b) Picture from optical microscope after depositionof Al and Lift-off
Figure 4.10: Top electrodes definition
4.1.1.6 Final passivation
This final step ensures that the whole sensor including the recently deposited top electrodes are pro-
tected from oxidation and physical damage. Firstly, a 4th and last lithography is done to pattern the vias,
then oxide is deposited on top (4000A of Al2O3) and finally a lift-off step opens pathways to the pads of
each sensor.
47
Figure 4.11: Schematics of the final passivation with opened vias.
Figure 4.12: Picture from optical microscope of the final sensors, after final passivation with vias open.
4.1.1.7 Annealing
Standard processes include a magnetic annealing step at the end of the MTJ microfabrication. In AlOx
devices this step is required for the exchange bias and the stabilisation of the barrier. Temperatures
and durations of the annealing can vary depending on the aim and the materials used. In this work, an
annealing of 30 minutes at 250C with cool down in a 1T field was performed in some AlOx devices. The
annealing treatment promotes the diffusion of the oxygen from the electrodes interfaces to the barrier,
improving the barrier uniformity and symmetry at the interfaces.
4.1.2 Characterization
4.1.2.1 Magnetotransport curves
Transfer curves were obtained for AlOx single sensors and for a sensor array of 136 MTJs in series,
both with and without annealing treatment. The results are presented in Fig. 4.13 and Tab. 4.1.
48
No anneal Anneal
Single Series Single Series
TMR (%) 36.3 36.0 50.8 43.5
Rmin(kΩ) 2.10 196 4.39 495
Sensitivity (%/Oe) 0.81 0.94 3.50 3.22
RxA (kΩ.µm2) 168 115 351 291
Table 4.1: Obtained values of some parameters for the sensors featured in Fig. 4.13 (AlOx single andseries, before and after annealing).
Figure 4.13: TMR transfer curves of the sensors in study (AlOx single and series, before and afterannealing).
Literature [64] [65] reports improvement of the TMR of 20% to 35% upon annealing which is also
observed in our devices, that have a TMR improvement of 20%− 40% after annealing and the absolute
TMR values are also competitive. The sensors show transfer curves with linear and non-hysteretic
behaviour and no discontinuities. The arrays present a slightly lower TMR value when compared with
the single sensors, being this difference more evident after annealing. This change in the TMR is mainly
due to the overall transfer curve of the array being a result from averaging each individual element.
Consequently, taking into account local deviations in the microfabrication process that can affect the
magnetic response of some particular elements, the saturation field of the series sensor can decrease.
4.1.2.2 Uniformity of TMR and RxA
In order to have information about the uniformity of finished sensors across the 6inch wafer, all the MTJs
were characterised using the AutoProber Setup. Fig. 4.14 shows a wafer map of the TMR values of
each sensor.
Only 6.03% of the total number of sensors on the 6inch wafer are not working. During the lift-off
process of the first passivation, some pillars still had photoresist on top after 4 days in microstrip bath.
The regions with a bigger concentration of PR observed with the optical microscope coincide with regions
of the Fig. 4.14 with more non-working sensors. Which is expectable, since pillars where the PR didn’t
lift-off cannot make contact with the top electrode and the series of MTJs get a discontinuity critical for
the array to function normally.
49
Figure 4.14: Wafermap of TMR of the full 6inch wafer for 7474 MTJ series sensors over 37 dyes mea-sured in transport after microfabrication, not annealed. Not-working sensors are marked in red, and thelack of response is mainly caused by residues of photoresist on the pillars during the lift-off step.
50
The data was also compiled in a plot and histogram to better analyse the results. Figure 4.15 shows
a plot of TMR versus RxA for the working AlOx sensors and also for 50 annealed sensors from the
wafer centre. An histogram of the same data is also presented for the distribution of TMR and RxA in
Fig. 4.16.
Figure 4.15: TMR versus RxA distribution for about 7000 MTJ series sensors over a 6inch wafer mea-sured in transport upon microfabrication before annealing and for 50 sensors after annealing.
(a) Histogram of TMR distribution (b) Histogram of RxA distribution
Figure 4.16: Histogram of TMR and RxA distributions of AlOx 136 series MTJs.
The results present an average value of TMR = 36.02% with a standard deviation of σ = 3.21%
and RxA = 169kΩµ2 with a σ = 62kΩ which translates in a TMR uniformity of 8.9% and RxA uniformity
of 36.6%. Comparing with the literature [66], the results are not fully satisfactory, as uniformities of
13.02% (RxA) and 2.36% (TMR) have been reported, which is still far from our results. Although all
the sensors represented in the histogram are working perfectly, the non-uniformity for the RxA is rather
large. But sensors with large resistances (large output power and low power consumption) require large
RxA values, and large RxA always requires a much tighter control in the AlOx deposition than low RxA.
4.1.2.3 IV Characteristics
One characteristic that differentiate an MTJ sensor from a spin valve or other GMR sensor is their non-
linear response of the output voltage of the sensor for an applied current. This non-linear response can
51
be fitted with the Simmons model as presented in Section 2.2.1.2 to extract valuable information about
the barrier thickness t and height φ.
Fig. 4.17 presents a fit of the experimental data to Eq. 2.9 for single MTJ sensors with junction
areas of 2 × 40µm2 and tAl = 14A, both before and after annealing. A slight increase of the barrier
height φ upon annealing, from φ = 2.28 ± 0.06eV to φ = 2.80 ± 0.53eV (23% increase) is observed
due to the diffusion of oxygen that reduces the metallic remains within the Al, promoting a more stable
and homogeneous oxide. The barrier thickness decreased upon annealing from t = 10.33 ± 0.08A
to t = 9.72 ± 0.52A (6% decrease) which indicates that there was some diffusion of oxygen into the
electrodes decreasing the thickness of the barrier. These tendencies have already been reported in
[64], where a 39% increase for the effective barrier height and 11% decrease for the effective barrier
thickness was observed for AlOx MTJs after annealing. Fig. 4.18 shows separately the fit for the
positive and negative branch of the IV curve both before and after annealing in order to analyse barrier
symmetry. It is important to note that after annealing the symmetry of the barrier interfaces improved,
as can be seen by the closeness of the fit results for both positive and negative branches.
(a) Fit of the IV curve for as deposited single sen-sors. Results: t = 10.33±0.08A, φ = 2.28±0.06eV
(b) Fit of the IV curve for annealed single sensors.Results: t = 9.72± 0.52A, φ = 2.80± 0.53eV
Figure 4.17: IV curves for comparison between as deposited and annealed AlOx MTJ sensors.
4.1.2.4 TMR dependence on Voltage
As depicted in Fig. 4.19, the sensors show a dependence of TMR on Vbias . The apparatus used to do the
measurements only allowed to set a current bias, but since current and voltage are directly correlated,
the behaviour is exactly the same in both cases. As can be seen, TMR decreases with increasing Vbias
and a slight asymmetry exists between the positive and negative branches. This difference is due to
asymmetries on the interfaces of the barrier. A V +12
= (0.46±0.04)V was obtained for the positive branch
in annealing free sensors and a slightly higher value for annealed sensors V +12annealed
= (0.52± 0.04)V .
These values represent the voltage at which the TMR drops to half of its original value and can be
obtained from the data in Fig. 4.19. Note that these values are well within the predicted values in
Section 2.2.3 for a good quality AlOx MTJ.
52
(a) Fit of the positive IV curve for asdeposited single sensors. Results: t =10.58± 0.05A, φ = 2.18± 0.03eV
(b) Fit of the negative IV curve for asdeposited single sensors. Results: t =10.07± 0.06A, φ = 2.39± 0.05eV
(c) Fit of the positive IV curve for an-nealed single sensors. Results: t =10.29± 0.03A, φ = 2.57± 0.02eV
(d) Fit of the IV curve for annealed sin-gle sensors. Results: t = 10.11 ±0.02A, φ = 2.57± 0.02eV
Figure 4.18: IV curves of separate branches for comparison between as deposited and annealed AlOx
MTJ sensors.
Figure 4.19: Normalized TMR ratio (TMR/TMRmax) versus voltage for single sensors with and withoutannealing.
4.1.2.5 Voltage Breakdown
To study the voltage breakdown, Vb, the devices were biased with current until their dielectric breakdown,
which occurs when the barrier breaks. The value Vb translates the robustness of the barrier against
electrical breakdown. For an as-deposited single sensor, the voltage breakdown is Vb = 1.71 ± 0.04V
whereas for annealed single sensors it is Vb.anneal = 2.01± 0.03V as depicted in Fig. 4.20. In the case
of series sensors, the expected value of voltage breakdown is 136 times (number of junctions in series)
the value for single sensors. As that value is way above 100V (voltage limit of available measurement
equipment at CTN) we couldn’t actually reach the regime where dielectric breakdown occurs in order
53
to obtain the experimental value. Vb increases after annealing which is coherent with the increase of φ
calculated before, since they are related as mentioned in [67]. Vb also increases with device resistance,
making smaller junction areas more robust against breakdown from strong biasing. However, MTJs with
higher resistance are more prone to destruction by electrostatic discharges.
(a) Single sensors. (b) 136 sensors in series.
Figure 4.20: IV curves until breakdown of AlOx MTJs with and without annealing. Junction area of2× 40µm2 measured without applied magnetic field.
Although non annealed devices present a smaller breakdown voltage than annealed ones, the value
is still much larger than V 12, and therefore does not compromise the usage of the sensors.
4.1.2.6 Noise
The final characterisation of the fabricated sensors are the noise measurements. The aim was to mea-
sure the noise in single and serie sensors, before and after annealing. However, as mentioned before,
high resistance single sensors are very sensitive to electrostatic discharges. Several single sensors
were wirebonded, but the associated electrostatic discharges broke all the barriers. In Fig. 4.21 the
results for the noise measurements in non annealed sensors with 136 junctions in serie are presented
and Fig. 4.22 shows the noise level for one annealed sensor. Some sensors were measured with more
than one voltage bias and the Hooge parameter was obtained by fitting the data to equation 2.17 in the
range of low frequencies (around 10− 100Hz) where the 1/f noise is predominant.
Figure 4.22: Noise level and Detectivity for annealed AlOx sensors with 136 junctions in series of2× 40µm2 each, measured at µ0H = 0T for two different Vbias.
54
(a) Noise level and Detectivity for annealing freesensors for three different resistance and similarVbias = 6.6V .
(b) Noise level and Detectivity for annealing free se-ries sensors for two different Vbias.
Figure 4.21: Noise level and Detectivity for annealing free AlOx sensors with 136 junctions in series of2× 40µm2 each,measured at µ0H = 0T
At high frequencies, the noise of the sensor is dominated by thermal noise as predicted by theory,
since the other components like RTN and 1/f are characteristic of low frequencies. The Detectivity plot is
presented in each image and the values are resumed in Table 4.2. Since sensitivity changes for different
currents applied to the sensor, for each current bias used in the noise setup, sensitivity values were
obtained at the 140Oe setup in order to calculate the detectivity. But some reference values measured
at Ibias = 1× 10−6A are presented in Tab. 4.1.
Annealing Rdut Vbias Detectivity (nT/Hz1/2) Hooge Parameter αtreatment (kΩ) (V) @30Hz @1kHz (10−7µm2)
No
128 6.4 14.5 1.71 1.35 ± 0.014.72 14.7 2.29 1.06 ± 0.01
184 6.55 16.8 2.65 1.58 ± 0.01233 6.61 11.8 2.32 1.77 ± 0.01282 6.68 28.0 4.03 10.8 ± 0.02
Yes 603 6.78 61.6 6.67 16.30 ± 0.022.48 67.9 8.89 2.72 ± 0.04
Table 4.2: Summary of noise measurements for different sensor’s resistance and Vbias, before and afterannealing at 0 Tesla field
As we have an array of sensors, the noise level is expected to be 100 times higher than for a single
sensor according to (αseries/αsingle ×N)1/2. We know from literature and previous works developed at
INESC-MN with similar AlOx stacks that the Hooge parameter is around 2 × 10−9 [23]. From the data
collected we can see that the Hooge parameter diminishes with bias voltage and is bigger for higher
resistance sensors that show also higher noise levels. After annealing, the resistance increases hugely,
contributing for the rise of the α value consistent with the increase of RxA. It is important to note that
for non annealed devices a strong Random Telegraph noise appears that is no longer noticeable after
annealing due to better crystallized magnetic layers. This results in a decline of magnetic fluctuations,
as predicted in Section 2.2.4.3. Detectivity levels vary from 14.5nT/Hz1/2 to 28nT/Hz1/2 at 30Hz for
annealing free sensors and go as high as 67.9nT/Hz1/2 for annealed devices, which indicates that
annealed MTJ sensor with AlOx do not have better performances than the annealing free sensors,
55
which is an improvement in the fabrication process, since this reduces the time needed to have finished
working devices and saves on costs. It is to note that the annealing setup available at INESC-MN can
only hold up to three 1′′ dyes at a given time and the process is usually left over night, which means a
total saving time of about 10 days in the fabrication process.
The noise measurements allow us to determine the minimum detectable field by the sensor. Using
Eq. 2.28 and the maximum detectivity value found for annealed series sensors (as the worst case
scenario), we can calculate the minimum variation of current that can be perceived by the sensor for
various distances between the current line and the sensor, as shown in Fig. 4.23.
Figure 4.23: Minimum detectable variation in the wire current in function of the separation between thecurrent line and the sensor, for a detectivity of 67.9nT/Hz1/2. Example: for a distance of 2.5 cm, theminimum detectable current is 8.5 mA or it can be attributed to noise
56
Chapter 5
Current Sensor
5.1 Sensors integration and Device layout
In this chapter, from the sensors fabricated and characterised previously, we chose 4 arrays of 136
AlOx annealed MTJ sensors with very similar resistance values to integrate in a Wheatstone Bridge
configuration. The reason for choosing annealed sensors instead of annealing free sensors is due to
their larger linear range and sensitivity, which is desirable when measuring high currents that can easily
saturate the sensors. The bridge connections were imprinted in a custom printed circuit board (PCB)
where the sensors can be mounted posteriorly. The layout of the PCB is based on the circuit 2.8 and is
presented in Fig. 5.1.
Figure 5.1: Schematics of the PCB design in Eable. This inicial PCB was developed by Luguang andPedro Ribeiro in 2017.
The sensors are then mounted on top of the PCB using glue and wirebonding the pads of the sensors
with the pads of the PCB as shown by the schematics in Fig. 5.2
The circuit is biased through the channels V + /V− and the output voltage of the bridge (Vout) is
acquire at the OUT + /OUT− terminals. To do so, the PCB was mounted on a Breadboard connected
to a current/voltage source and to a voltmeter to measure Vout.
57
(a) Picture of the PCB with sensors alreadymounted on top.
(b) Schematics of the wirebonding connections tothe PCB. The black arrows correspond to the sensi-tive direction orientation of each sensor
Figure 5.2: Sensors mounted on the PCB and respective connections
A copper wire with 4mm diameter is placed above our PCB so that the induced magnetic field can
be sensed by the bridge. Through the wire passes a current ranging from 0A to 50A.
5.2 Wheatstone Bridge Results
5.2.1 MR characterisation
From the numerous MTJs we fabricated before, 4 array sensors were chosen according to the maximum
proximity of the resistance values. Also, devices with low resistance are preferable because of their
better detectivity levels.
The four sensors chosen were characterised through the magnetoresistance response curve mea-
sured in the 140Oe Setup and the results are presented in Fig. 5.3 and summarised in Tab. 5.1.
Sensor Ibias (µA) R0 (kΩ) MR (%) Sensitivity (mV/Oe) Coercivity (Oe)1
1
388 31.30 1.21 0.242 391 35.44 1.31 0.113 388 38.84 1.52 0.364 391 39.62 1.63 0.47
Table 5.1: Results of the characterisation of the array sensors chosen for the Wheatstone Bridge
The sensors present a very similar behaviour with values of resistance varying only 0.7% between
them and MR, Sensitivity and Coercivity also very close.
With the individual resistance of each sensor, we can now calculate the equivalent resistance of the
Wheatstone bridge by:
58
Figure 5.3: R-H curves of the 4 array sensors chosen to mount on the Wheatstone Bridge.
1
Rbridge=
1
R1 +R2+
1
R3 +R4(5.1)
Rbridge =(R1 +R2)(R3 +R4)
R1 +R2 +R3 +R4(5.2)
Using the values obtained with the characterisation and substituting in the equation above we can
calculate the bridge resistance:
Rbridge =(387662 + 390513)(388491 + 390582)
387662 + 390513 + 388491 + 390582= 389312Ω (5.3)
If the bridge was completely balanced, all the elements would have an equal R resistance value and
the resistance of the bridge would be simply R. In reality the resistances used do not have equal values
of resistance, implying an offset voltage at the output of the bridge.
5.2.2 Wheatstone Bridge transfer curves
With the sensors mounted on the bridge, using again the 140 Oe Setup and biasing the circuit with a
current Ibias = 1µA, we sweeped the magnetic field in a −140/140Oe range and measured the voltage
output of the bridge Vout, thus getting the magnetoresistance curve of the Wheatstone bridge and values
for the voltage offset Voffset = 9.57 ± 0.02mV , sensitivity S = 0.71 ± 0.01mV/Oe and coercivity Hc =
0.21Oe, Fig.5.4. The bridge output has a linear range from −50Oe to 50Oe and a power consumption of
0.4µW . In order to avoid the unwanted saturation of the sensors, a linear range from −30Oe to 30Oe is
to be assumed.
59
Figure 5.4: Wheatstone Bridge voltage output as a function of the Magnetic field.
From the calculated bridge equivalent resistance in 5.3, we can translate the Ibias into a Vbias =
0.39V , so we can calculate the bridge offset as a percentage Voffset
Vbias= 2.5%, and the sensitivity as
SVbias
= 1.82mV/V/Oe. This offset voltage comes from the asymmetry between sensors, either in their
shapes or the presence of some shorted junctions.
From Fig. 5.5 we can see the dependence of the biasing voltage on the bridge output, as a higher
voltage bias gives higher sensitivity values. However, by normalising the sensitivities with the bias
voltage gives a value of around S = 2.1mV/V/Oe, which was expected.
Figure 5.5: Bridge voltage output as a function of Magnetic field, for various bias voltage.
Now placing a copper wire d = 7 ± 0.5mm above the bridge and sweeping the current between 0A
to 50A (see Fig. 5.6), the voltage output was again measured and the results are presented in Fig. 5.7.
60
Figure 5.6: Schematics of the setup for characterising the current sensor.
From the linear fit of the plot we get a sensitivity of S = 0.159 ± 0.002mV/A and an offset voltage of
Voffset = 9.64± 0.01mV = 2.3%.
Figure 5.7: Wheatstone Bridge voltage output as a function of the current passing the copper wire atd = 7± 0.5mm.
Furthermore, we studied the behaviour of the bridge when varying the distance between the wire
and the sensors, ranging from 5± 0.5mm to 40± 0.5mm with wire currents of 10A and 30A. The results
are shown in Fig. 5.8.
To fit the plot, we assume a linear response of the sensor with the generated magnetic field (Vout(mV ) =
S(mV/Oe) × H(Oe) + Voffset) and the dependence of H with distance d given by Ampere’s Law de-
scribed in Eq. 2.28. Joining both gives the expression:
Vout =2× S(mV/Oe)× I
d+ c+ Voffset (5.4)
where c is a constant added to the expression to which we call displacement, to encode the errors of
setting the distance manually.
The results of the fit gave a Voffset = 9.48 ± 0.08mV and Voffset = 9.58 ± 0.03mV for wire currents
of 30A and 10A respectively, sensitivity of S = 0.99 ± 0.06mV/Oe for both cases and a constant c =
61
Figure 5.8: Wheatstone Bridge voltage output as a function of the distance to the copper wire, withI = 10, 30A.
5.3± 0.5mm, c = 5.4± 0.5mm for I = 30, 10A respectively.
5.2.3 Induced Magnetic Field calibration
Since the local magnetic field created by the electrical current gives an output voltage that directly
corresponds to a fixed value of external magnetic field, by having the voltage dependence of the Bridge
on the magnetic field and on the current in the wire, we can easily determine the magnetic field created
by unit of current.
Using the same equation 2.28 as before we can calculate the expected magnetic field generated by
the current line for a d = 7mm distance and for I = 1A:
Hexpected =4π × 10−3 × 1
2π × 7× 10−3≈ 0.286Oe/A (5.5)
To determine this value of induced magnetic field experimentally, we can compare the bridge output
voltage results of measurements using the transport measurement setup, that applies an already cali-
brated external magnetic field in a range of ±140Oe thus obtaining a V-H transfer curve, and the same
bridge output voltage results obtained with the generated magnetic field created by the electrical DC
current driven in the Cu wire ranging from 0 to 50A for a chosen distance, which gives the Iwire − V
characteristic curve. This measurements were already presented above in Fig. 5.4 and Fig. 5.7, for a
7mm distance and a 1µA biasing current.
From the values of sensitivity obtained with the fits of V-H and V-I plots in the linear range it gives:
62
∆V
∆H= 0.71mV/Oe
∆V
∆I= 0.16mV/A (5.6)
Hcreated =∆V∆I∆V∆H
= 0.225Oe/A (5.7)
5.3 Final Sensor and Encapsulation
In order to encapsulate the current sensor in a compact device, a new PCB layout was designed and
ordered with new dimensions of 15mm×15mm as shown in Fig. 5.9(a). When trying to transfer the MTJ
sensors from the previous PCB to the new one, one of the sensors was damaged in the process. This
implies finding a new set of four very similar MTJ sensors, characterise them (Fig. 5.10), wirebond the
bridge connections and characterise the new current sensor (Fig. 5.11, 5.12). The results are resumed
in Tab. 5.2.
(a) Schematics of the new PCB design in Eable (b) Picture of the new PCB with sensors mounted and wire-bonded.
Figure 5.9: Final current sensor.
Sensor R0 (Ω) MR (%) S (mV/Oe) Coercivity (Oe)1 627 35.71 2.5 0.772 609 37.64 1.9 0.583 573 36.97 2.3 0.564 569 37.40 2.3 1.18
Bridge 594 202 1.78 0.5
Table 5.2: Results of the characterisation of the array sensors chosen for the Wheatstone Bridge, with aIbias = 1µA
The new Wheatstone Bridge has a Voffset = 9.8mV when biased with Ibias = 1µA, which corre-
sponds to 1.7%, sensitivity of S = 1.78mV/Oe (S = 3.02mV/V/Oe) and coercivity Hc = 0.5Oe. The new
63
Figure 5.10: R-H curves of the 4 array sensors chosen to mount on the new Wheatstone Bridge.
Figure 5.11: New Wheatstone Bridge voltage output as a function of the Magnetic field measured at the140Oe Setup.
bridge has a linear range of ±40 Oe and a power consumption of only 0.6µW for Vbias = 0.59 V .
5.3.1 Encapsulation
The design and projection of the encapsulation was made using AutoCad, see Fig.5.13(a). Two steps
where included so that the current sensor can be placed in one or the other, depending on the current
values one wishes to measure. The first step is distanced 1 cm from the centre of the current wire,
allowing measurements up to ±200 A, and the second step distances more than 2.5 cm from the wire
64
Figure 5.12: New Bridge voltage output as a function of Magnetic field, for various bias voltage, mea-sured at the 140Oe Setup.
centre, which allows measurements in a range of ±500 A while still operating in the linear range of the
sensor.
Using a 3D printer available at INESC-MN, the encapsulation was finalised and presented in Fig.
5.13(b). In order to achieve better accuracy, the current sensor needs a calibration in situ, as the wire
diameter can change.
(a) AutoCad 3D design of the final encapsulation ofthe current sensor.
(b) Picture of the final encapsulation printed using a 3D printeravailable at INESC-MN and the sensor mounted
Figure 5.13: Final encapsulation of the current sensor
65
With the sensor at the first step of the encapsulation, which corresponds to being distanced 1 cm
from the wire centre, the results for the output of the bridge as the wire current is being sweeped from
0− 50 A are shown in Fig. 5.14. The sensitivity is 0.36 mV/A.
Figure 5.14: New Bridge voltage output as a function of the current passing through the copper wire.
5.3.2 Accuracy
Accuracy is usually defined as a percentage and translates the maximum difference to the actual primary
current being measured is the output of the sensor. Uncertainty, in the case of a current sensor, is
expressed in Amps. The subject of instrumentation accuracy is typically poorly defined and there is little
information available on the methods used in the industry to measure the accuracy of a current sensor.
There are several different sources of accuracy in a current sensor, such as the initial offset at room
temperature, gain error (εg), linearity error (εg), offset drift over temperature and all these combined
drifting over time. Drift over time is usually negligible as MTJs can last more than 10 years [68], although
MgO barriers are more reliable over time than AlOx. The initial offset is the non-zero output of the
sensor when the primary current is zero, that we defined as voltage offset before. This error will not be
accounted for due to the simple method of its elimination with a calibration or with a subtraction of the
voltage offset value. In a plot of Vout versus Iwire, the gain error is equivalent to the error in slope of the
response curve and contrary to the voltage offset, which is constant over all current ranges, gain error is
proportional to the wire current. Linearity error is the straightness of the response curve that corresponds
to the deviation of the measured curve of the sensor from the ideal, and is usually associated with the
saturation of the magnetic core in closed-loop current sensors or the approximation of the saturation
field of the MTJ sensors. Temperature drift was not accounted for, as the process required an oven that
allows measurements in real time which was not available at the current time and place of this thesis.
Besides, in this particular current sensor, it is not critical, since the sensors and the current wire are
separated by 1 cm or more, which allows the heat to dissipate through the air. The overall accuracy of
66
the current sensor is calculated by:
Accuracy =√ε2g + ε2
l (5.8)
In order to measure the accuracy in this work, 11 different values of current were input to the wire in
a random order as in Fig. 5.15(a) and the deviation from the expected value of the bridge output versus
the current applied were compiled in the plot of Fig. 5.15(b).
(a) Iterations done for 110 values, 10 for each cur-rent value I=1,5,10,15,20,25,30,35,40,45,50 A
(b) Bridge output error for a cycle of 110 current values.
Figure 5.15: Accuracy measurements.
The linearity error is calculated by dividing the maximum input deviation by the full scale input, as
a percentage. The gain error is given by the slope of the linear fit in Fig. 5.15(b) (slope = 0.01 mV/A)
divided by the sensitivity of the sensor (Fig. 5.14). Therefore, the linearity error is εl = 3.4% and the
gain error εg = 2.8%. Using the Eq. 5.8, the accuracy of the current sensor is 4.4% and the uncertainty
is 2.2 A, in a range of 50 A. This is a value hard to compare with other current sensors available in the
market, as it is unknown the method used by each company.
67
68
Chapter 6
Conclusions
The main objectives of this thesis was to microfabricate a fully functional 6 inch wafer of AlOx MTJs with
a good TMR and RxA uniformity and develop a current sensor based on a Wheatstone Bridge with the
fabricated sensors as resistive elements.
The wafer was successfully fabricated and fully characterised resulting in a non-uniformity of 8.9% for
the TMR (average value of 36.02%) and 36.6% for the RxA product (average value of 169 kΩµ2) across
more than 7000 sensors with 136 junctions connected in series each. The sensors are fully operational
without any annealing thermal treatment, giving sensitivities of S = 0.94 %/Oe. Some of the sensors
were annealed, which resulted in a substantial increase of the RxA product to 291 kΩµ2 and also on
sensitivity to S = 3.22 %/Oe and TMR to 43.54%. Using the Simmons model on the VI plots, an effective
barrier thickness of Φ = 2.28 ± 0.06 eV was obtained and a 23% increase for annealed sensors, which
is due to the diffusion of oxygen that promotes a more stable and homogeneous oxide. A reduction of
6% was observed for the effective barrier thickness upon annealing, which is explained by the diffusion
of oxygen into the electrodes. When analysing the positive and negative branches of the IV curve
separately, before and after annealing, an improvement of the barrier symmetry is also observed by the
proximity of the values. By biasing some test single structures with increasing current until their dielectric
breakdown, the values for V1/2 = 0.46± 0.04 V and Vb = 1.71± 0.04 V were obtained for annealing free
sensors and V1/2 = 0.52 ± 0.04 V and Vb = 2.01 ± 0.03 V for annealed ones. This translates in a slight
improvement of the robustness of the barrier for annealed sensors, although the usage of the annealing
free devices is not compromised since Vb is much larger than V1/2. Since the expected value of voltage
breakdown for the series sensors is 136 times higher than of the single structures, these devices can
operate at biasing voltages higher than 100 V .
The final characterisation of the sensors was the noise. As expected, the noise is dominated by
thermal noise at high frequencies and by 1/f and RTN at low frequencies. After annealing, the RTN
noise present in annealing free sensors, is no longer noticeable due to better crystallised magnetic
layers. Detectivity levels vary from 14.5 − 28 nT/Hz1/2 at 30 Hz for annealing free sensors and go as
high as 67.9 nT/Hz1/2 for annealed ones. This indicates that annealed devices do not always have
better performances than annealing free devices, which is an improvement in the fabrication process
69
since it allows to save time and in costs. With these noise measurements, we calculated the minimum
detectable current in function of the distance between the current line and the sensor. For a distance of
2.5 cm, the minimum current variation detectable is 8.5 mA.
For the current sensor, four similar MTJ series sensors were chosen and fully characterised. A PCB
with a Wheatstone Bridge circuit was designed and the MTJs were wirebonded to the pads of the PCB.
Under a controlled magnetic field (140 Oe setup) the current sensor was tested and characterised, giving
a Voffset of 1.7%, coercivity Hc = 0.5 Oe and a sensitivity of 1.78 mV/Oe, which with a bias current of
1µA corresponds to S = 3.02 mV/V/Oe. An encapsulation for the current sensor was designed in
AutoCad and printed in a 3D printer so that it can be used under any conditions. This encapsulation
comprises two different steps depending on the range of currents intended to measure. The second step
can measure up to 500 A without saturating the sensor. The sensitivity achieved was S = 0.36 mV/A for
Ibias = 1 µA. Accuracy of the current sensor was tested to be 4.4%.
70
Bibliography
[1] K. P. Gokhale. Wheel speed sensor, May 30 1989. US Patent 4,835,467.
[2] J. Valadeiro. Magnetoresistive sensors with pico-tesla sensitivities. Master’s thesis, Instituto Supe-
rior Tecnico, 2014.
[3] D. Graham, H. Ferreira, P. Freitas, and J. Cabral. High sensitivity detection of molecular recognition
using magnetically labelled biomolecules and magnetoresistive sensors. Biosensors and Bioelec-
tronics, 18(4):483–488, 2003.
[4] S. J. Han, L. Xu, H. Yu, R. J. Wilson, R. L. White, N. Pourmand, and S. X. Wang. CMOS in-
tegrated DNA microarray based on GMR sensors. In Electron Devices Meeting, 2006. IEDM’06.
International, pages 1–4. IEEE, 2006.
[5] cmos.
[6] E. G. Vidal, D. R. Munoz, S. I. R. Arias, J. S. Moreno, S. Cardoso, R. Ferreira, and P. Freitas.
Electronic energy meter based on a tunnel magnetoresistive effect (TMR) current sensor. Materials,
10(10):1134, 2017.
[7] N. Mohan and T. M. Undeland. Power electronics: converters, applications, and design. John Wiley
& Sons, 2007.
[8] S. Ziegler, R. C. Woodward, H. H. C. Lu, and L. J. Borle. Current sensing techniques: A review.
IEEE Sensors Journal, 9(4):354–376, 2009.
[9] J. Quasdorf. A case study: MR vs. hall effect for position sensing. 22:17–26, 11 2005.
[10] P. Ripka. Electric current sensors: a review. Measurement Science and Technology, 21(11):
112001, 2010.
[11] R. Ferreira, E. Paz, P. P. Freitas, J. Ribeiro, J. Germano, and L. Sousa. 2-axis magnetometers
based on full wheatstone bridges incorporating magnetic tunnel junctions connected in series. IEEE
Transactions on Magnetics, 48(11):4107–4110, 2012.
[12] Honeywell. https://sensing.honeywell.com/sensors/current-sensors, 2018. [Online; ac-
cessed 25-May-2018].
71
[13] Tamura. http://tamuracorp.com/products/current-sensor-products/, 2018. [Online; ac-
cessed 25-May-2018].
[14] Melexis. https://melexis.com/en/products/sense/current-sensors, 2018. [Online; accessed
25-May-2018].
[15] DowayTech TMR sensor. http://www.dowaytech.com/en/1392.html, 2018. [Online; accessed
25-May-2018].
[16] T. McGuire and R. Potter. Anisotropic magnetoresistance in ferromagnetic 3d alloys. IEEE Trans-
actions on Magnetics, 11(4):1018–1038, 1975.
[17] R. A. M. A. Ferreira. Ion Beam Deposited Magnetic Spin Tunnel Junctions targeting HDD Read
Heads, Non-volatile Memories and Magnetic Field Sensor Applications. PhD thesis, 2008.
[18] M. N. Baibich, J. M. Broto, A. Fert, F. N. Van Dau, F. Petroff, P. Etienne, G. Creuzet, A. Friederich,
and J. Chazelas. Giant magnetoresistance of (001) Fe/(001) Cr magnetic superlattices. Physical
review letters, 61(21):2472, 1988.
[19] J. M. G. de Almeida. Tunnel Junction Sensors. PhD thesis, Instituto Superior Tecnico, 2009.
[20] B. Dieny, V. S. Speriosu, S. S. P. Parkin, B. A. Gurney, D. R. Wilhoit, and D. Mauri. Giant magne-
toresistive in soft ferromagnetic multilayers. Phys. Rev. B, 43:1297–1300, Jan 1991.
[21] M. D. Cubells-Beltran, C. Reig, J. Madrenas, A. De Marcellis, J. Santos, S. Cardoso, and P. P.
Freitas. Integration of GMR sensors with different technologies. Sensors, 16(6):939, 2016.
[22] M. Julliere. Tunneling between ferromagnetic films. Physics letters A, 54(3):225–226, 1975.
[23] S. Knudde, D. Leitao, S. Cardoso, and P. Freitas. Annealing free magnetic tunnel junction sensors.
Journal of Physics D: Applied Physics, 50(16):165001, 2017.
[24] J. C. Slonczewski. Conductance and exchange coupling of two ferromagnets separated by a tun-
neling barrier. Physical Review B, 39(10):6995, 1989.
[25] J. S. Moodera and G. Mathon. Spin polarized tunneling in ferromagnetic junctions. Journal of
magnetism and magnetic materials, 200(1):248–273, 1999.
[26] J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey. Large magnetoresistance at room
temperature in ferromagnetic thin film tunnel junctions. Physical review letters, 74(16):3273, 1995.
[27] H. Wei, Q. Qin, M. Ma, R. Sharif, and X. Han. 80% tunneling magnetoresistance at room tempera-
ture for thin Al–O barrier magnetic tunnel junction with CoFeB as free and reference layers. Journal
of applied physics, 101(9):09B501, 2007.
[28] K. S. Yoon and J. P. Hong. Observation of giant magnetoresistance in CoFeN/AlOx/CoFeN mag-
netic tunneling junctions employing a nitrogen-doped amorphous CoFeN free layer electrode. Ap-
plied Physics Letters, 110(1):012402, 2017.
72
[29] J. Mathon and A. Umerski. Theory of tunneling magnetoresistance of an epitaxial Fe/MgO/Fe (001)
junction. Physical Review B, 63(22):220403, 2001.
[30] S. Yuasa, T. Nagahama, A. Fukushima, Y. Suzuki, and K. Ando. Giant room-temperature mag-
netoresistance in single-crystal Fe/MgO/Fe magnetic tunnel junctions. Nature materials, 3(12):
868–871, 2004.
[31] S. S. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Hughes, M. Samant, and S.-H. Yang. Giant tun-
nelling magnetoresistance at room temperature with MgO (100) tunnel barriers. Nature materials,
3(12):862–867, 2004.
[32] J. Hayakawa, S. Ikeda, Y. M. Lee, F. Matsukura, and H. Ohno. Effect of high annealing temperature
on giant tunnel magnetoresistance ratio of CoFeB/MgO/CoFeB magnetic tunnel junctions. Applied
Physics Letters, 89(23):232510, 2006.
[33] S. Ikeda, J. Hayakawa, Y. Ashizawa, Y. Lee, K. Miura, H. Hasegawa, M. Tsunoda, F. Matsukura,
and H. Ohno. Tunnel magnetoresistance of 604% at 300 k by suppression of ta diffusion in
CoFeB/MgO/CoFeB pseudo-spin-valves annealed at high temperature. Applied Physics Letters,
93(8):082508, 2008.
[34] R. A. F. C. e Sousa. Magnetic Random Access Memory (MRAM) based on Spin Dependent Tunnel
Junctions. PhD thesis, Instituto Superior Tecnico, 2003.
[35] S. Huang, T. Chen, and C. Chien. Spin polarization of amorphous CoFeB determined by point-
contact Andreev reflection. Applied Physics Letters, 92(24):242509, 2008.
[36] D. Wang, C. Nordman, J. M. Daughton, Z. Qian, and J. Fink. 70% TMR at room temperature for
SDT sandwich junctions with CoFeB as free and reference layers. IEEE Transactions on Magnetics,
40(4):2269–2271, 2004.
[37] S. Cardoso, V. Gehanno, R. Ferreira, and P. P. Freitas. Ion beam deposition and oxidation of spin-
dependent tunnel junctions. IEEE transactions on magnetics, 35(5):2952–2954, 1999.
[38] T. Holmqvist, M. Meschke, and J. P. Pekola. Double oxidation scheme for tunnel junction fabri-
cation. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures
Processing, Measurement, and Phenomena, 26(1):28–31, 2008.
[39] J. G. Simmons. Electric tunnel effect between dissimilar electrodes separated by a thin insulating
film. Journal of applied physics, 34(9):2581–2590, 1963.
[40] J. G. Simmons. Generalized formula for the electric tunnel effect between similar electrodes sepa-
rated by a thin insulating film. Journal of applied physics, 34(6):1793–1803, 1963.
[41] Z. Zeng, P. Khalili Amiri, J. Katine, J. Langer, K. Wang, and H. Jiang. Nanoscale magnetic tunnel
junction sensors with perpendicular anisotropy sensing layer. Applied Physics Letters, 101(6):
062412, 2012.
73
[42] B. Negulescu, D. Lacour, F. Montaigne, A. Gerken, J. Paul, V. Spetter, J. Marien, C. Duret, and
M. Hehn. Wide range and tunable linear magnetic tunnel junction sensor using two exchange
pinned electrodes. Applied Physics Letters, 95(11):112502, 2009.
[43] R. Ferreira, E. Paz, P. Freitas, J. Wang, and S. Xue. Large area and low aspect ratio linear magnetic
tunnel junctions with a soft-pinned sensing layer. IEEE Transactions on Magnetics, 48(11):3719–
3722, 2012.
[44] R. Chaves, S. Cardoso, R. Ferreira, and P. Freitas. Low aspect ratio micron size tunnel magne-
toresistance sensors with permanent magnet biasing integrated in the top lead. Journal of Applied
Physics, 109(7):07E506, 2011.
[45] P. Freitas, R. Ferreira, S. Cardoso, and F. Cardoso. Magnetoresistive sensors. Journal of Physics:
Condensed Matter, 19(16):165221, 2007.
[46] A. V. Silva, D. C. Leitao, J. Valadeiro, J. Amaral, P. P. Freitas, and S. Cardoso. Linearization strate-
gies for high sensitivity magnetoresistive sensors. The European Physical Journal Applied Physics,
72(1):10601, 2015.
[47] S. H. Liao, S. Yuan, and H. Bertram. Biasing characteristics of unshielded, boundary exchange-
coupled SAL/MR sensors. IEEE transactions on magnetics, 29(6):3873–3875, 1993.
[48] J. G. J. Zhu and C. Park. Magnetic tunnel junctions. Materials today, 9(11):36–45, 2006.
[49] J. Zhang and R. White. Voltage dependence of magnetoresistance in spin dependent tunneling
junctions. Journal of applied physics, 83(11):6512–6514, 1998.
[50] S. Sze. Vlsi technology, 2nd, 1988.
[51] B. Oliver, Q. He, X. Tang, and J. Nowak. Tunneling criteria and breakdown for low resistive magnetic
tunnel junctions. Journal of applied physics, 94(3):1783–1786, 2003.
[52] J. B. Johnson. Thermal agitation of electricity in conductors. Physical review, 32(1):97, 1928.
[53] Z. Lei, G. Li, W. F. Egelhoff, P. Lai, and P. W. Pong. Review of noise sources in magnetic tunnel
junction sensors. IEEE Transactions on Magnetics, 47(3):602–612, 2011.
[54] K. Kandiah and F. Whiting. Low frequency noise in junction field effect transistors. Solid-State
Electronics, 21(8):1079–1088, 1978.
[55] J. Scola, H. Polovy, C. Fermon, M. Pannetier-Lecoeur, G. Feng, K. Fahy, and J. Coey. Noise in
MgO barrier magnetic tunnel junctions with CoFeB electrodes: Influence of annealing temperature.
Applied physics letters, 90(25):252501, 2007.
[56] A. M. Nor, T. Daibou, M. Oogane, Y. Ando, and T. Miyazaki. Boron effects on noise in magnetic
tunnel junctions. Journal of Magnetism and Magnetic Materials, 310(2):1917–1919, 2007.
74
[57] C. Ren, X. Liu, B. Schrag, and G. Xiao. Low-frequency magnetic noise in magnetic tunnel junctions.
Physical Review B, 69(10):104405, 2004.
[58] F. Hooge. 1/f noise. Physica B+ C, 83(1):14–23, 1976.
[59] INESC-MN. https://sites.google.com/site/inescmn/home/facilities, 2018. [Online; ac-
cessed 25-May-2018].
[60] R. Varela. Optimization of mr sensor geometry towards picotesla detection. Master’s thesis, Instituto
Superior Tecnico, 2015.
[61] V. Gehanno, P. P. Freitas, A. Veloso, J. Ferrira, B. Almeida, J. Soasa, A. Kling, J. Soares, and
M. Da Silva. Ion beam deposition of Mn-Ir spin valves. IEEE Transactions on magnetics, 35(5):
4361–4367, 1999.
[62] G. A. N. Farinha. Compact size 3D magnetometer based on magnetoresistive sensors. Master’s
thesis, Instituto Superior Tecnico, 2015.
[63] R. W. Dave, G. Steiner, J. Slaughter, J. Sun, B. Craigo, S. Pietambaram, K. Smith, G. Grynkewich,
M. DeHerrera, J. Akerman, et al. MgO-based tunnel junction material for high-speed toggle mag-
netic random access memory. IEEE transactions on magnetics, 42(8):1935–1939, 2006.
[64] R. Sousa, J. Sun, V. Soares, P. Freitas, A. Kling, M. Da Silva, and J. Soares. Large tunneling
magnetoresistance enhancement by thermal anneal. Applied physics letters, 73(22):3288–3290,
1998.
[65] S. Parkin, K. Roche, M. Samant, P. Rice, R. Beyers, R. Scheuerlein, E. O’sullivan, S. Brown,
J. Bucchigano, D. Abraham, et al. Exchange-biased magnetic tunnel junctions and application
to nonvolatile magnetic random access memory. Journal of Applied Physics, 85(8):5828–5833,
1999.
[66] P. P. Freitas, R. Ferreira, and S. Cardoso. Spintronic sensors. Proceedings of the IEEE, 104(10):
1894–1918, 2016.
[67] K. Shimazawa, N. Kasahara, J. Sun, S. Araki, H. Morita, and M. Matsuzaki. Electrical breakdown of
the magnetic tunneling junction with an AlOx barrier formed by radical oxidation. Journal of Applied
Physics, 87(9):5194–5196, 2000.
[68] C. Yoshida, H. Noshiro, Y. Yamazaki, T. Iizuka, Y. Stoh, M. Aoki, S. Umehara, M. Satoh, and
K. Kobayashi. Reliability of MgO tunneling barrier for MRAM device. pages 697–698, 2006.
[69] R. Guerrero, M. Pannetier-Lecoeur, C. Fermon, S. Cardoso, R. Ferreira, and P. Freitas. Low fre-
quency noise in arrays of magnetic tunnel junctions connected in series and parallel. Journal of
Applied Physics, 105(11):113922, 2009.
75
[70] M. Tondra, J. M. Daughton, D. Wang, R. S. Beech, A. Fink, and J. A. Taylor. Picotesla field sensor
design using spin-dependent tunneling devices. Journal of Applied Physics, 83(11):6688–6690,
1998.
[71] X. Han, S. S. Ali, and S. Liang. MgO (001) barrier based magnetic tunnel junctions and their device
applications. Science China Physics, Mechanics and Astronomy, 56(1):29–60, 2013.
[72] S. Parkin and D. Mauri. Spin engineering: Direct determination of the Ruderman-Kittel-Kasuya-
Yosida far-field range function in ruthenium. Physical Review B, 44(13):7131, 1991.
[73] DowayTech. http://www.dowaytech.com/en/sensor/magnetic_field_sensors.html, 2018.
[Online; accessed 25-May-2018].
[74] R. Flores. MgO magnetic tunnel junction sensors in full wheatstone bridge configuration for in-chip
current field detection. Master’s thesis, Instituto Superior Tecnico, 2010.
[75] A. Lopes. MgO magnetic tunnel junction sensors in full wheatstone bridge configuration for electri-
cal current detection. Master’s thesis, Instituto Superior Tecnico, 2012.
76
Appendix A
Runsheet
77
Magnetic Tunnel Junction 1 / 9 INESC-MN
Step 01 Junction Stack Deposition Date Responsible
Step 04 1st DWL Exposure for B.E definition Date
Responsible
Machine DWL
Map LGJTEST Layer cs_v1_L1
Focus Energy Power mW
1st Cross (250,250) 2nd Cross (250,750) Distance (0,500)
(0, 0) (3700, 3700) Xoffset Yoffset
do -1 Ali - Command -
Sample ID
Machine Nordiko 3000
Stack Substrate//Ta(50)/[Ru(150)/Ta(50)]x3/NiFe(30)/CoFeB(30)/Al(7)x2
/CoFeB(30)/Ru(6)/NiFe(30)/MnIr(180)/Ru(150)/Ta(50)
BP 5.8*10-8 Torr
Deposition
date
• Put samples in vapour prime (~30min) (or 1min@110Cº) YES
• Coat samples (Coating Line):
Prog 6 + Prog 2
• Develop samples (Develop Line):
Prog 2 + Prog 6
Bake (60s@ 85deg)
Spin (1min@ 0.8krpm): 1.5µm PR.
Bake (60s@ 110deg) + Cooling (30s)
Spray(3s@ 0.5krpm) + Dissolve(60s)
+ Spin ([email protected] +30s@2krmp)
Magnetic Tunnel Junction 2 / 9 INESC-MN
Optical Inspection Comments
Step 3 1st Etch by Ion Milling Date Responsible
Machine: N3600
Total to etch 680 Å
Total etching
Time
175s x 4 (700s) Total etched Thickness
Batch recipe Etch_junction_first
Wafer Recipe
Process steps
#1 Load wafer #2 [etch pan 45 degrees + cooling down 200s at 45 degrees]3 + etch pan 45 degrees #3 end etch junction
X,Ye.a.
Magnetic Tunnel Junction 3 / 9 INESC-MN
l 45 Degs
Process Timer Assist Gun
Time secs RF FWD Power W
RF REF Power W
Chamber Grid 1 Volts V
Pressure Torr Grid 1 Currents mA
Grid 2 Volts V
Assist Neutraliser Grid 2 Currents mA
Current mA Argon Flow sccm
Voltage V
Argon Flow sccm
Step 4 Photoresist Remove Date Responsible
l Microstrip 3001
Day Start @ End @ Total Type of US
Done Profilometer measurement of etched thickness
-
Magnetic Tunnel Junction 4 / 9 INESC-MN
Step 5 2nd DWL Exposure for Junctions definition Date
Responsible
Machine DWL
Map LGJTEST Layer cs_v1_L2
Focus Energy Power mW
1st Cross (250,250) 2nd Cross (250,750) Distance (0,500)
(0, 0) (3700, 3700) Xoffset -250 Yoffset -250
do -1 Ali -1 Command -
Optical Inspection
Comments
• Put samples in vapour prime (~30min) (or 1min@110Cº) YES
• Coat samples (Coating Line):
Prog 6 + Prog 2
• Develop samples (Develop Line):
Prog 2 + Prog 6
Bake (60s@ 85deg)
Bake (60s@ 110deg) + Cooling (30s)
Spray(3s@ 0.5krpm) + Dissolve(60s)
+ Spin ([email protected] +30s@2krmp)
Spin (1min@ 0.8krpm): 1.5µm PR.
Magnetic Tunnel Junction 5 / 9 INESC-MN
Step 6 2nd Etch by Ion Milling Date Responsible
Machine: N3600 B.P.=_________________
Sample 60deg etch (1st) 137s x 2
30deg etch (2nd) 150s x 2 (last step +100s overetch)
Batch recipe: junction_etch
Wafer Recipe:
#3 etch junction top electrode
Process steps:
#01 Load wafer #02 etch pan 60 deg (137s) #03 cool_down_200s at 60 degrees #04 etch pan 30 deg (150s) #05 cool_down_200s at 30 degrees #06 etch pan 60 deg (137s) #07 cool_down_200s at 60 degrees #08 etch pan 30 deg (250s) #09 end etch junction
l 60 Degs
Process Timer Assist Gun
Time secs RF FWD Power W
RF REF Power W
Chamber Grid 1 Volts V
Pressure Torr Grid 1 Currents mA
Grid 2 Volts V
Assist Neutraliser Grid 2 Currents mA
Current mA Argon Flow sccm
Voltage V
Argon Flow sccm
l 30 Degs
Process Timer Assist Gun
Time secs RF FWD Power W
RF REF Power W
Chamber Grid 1 Volts V
Pressure Torr Grid 1 Currents mA
Grid 2 Volts V
Assist Neutraliser Grid 2 Currents mA
Current mA Argon Flow sccm
Voltage V
Argon Flow sccm
Magnetic Tunnel Junction 6 / 9 INESC-MN
Step 7 Oxide Deposition (Al2O3) Date Responsible
Done Use of calibration sample (Si substrate w/ pen line)
-
Machine: UHVII Total to deposit 1300 Å Rate A/min Total Time Start time End time Total Time
Read v.
Base Pressure (T)
P before plasma (mT)
P working (mT)
Turbo Pump freq. (Hz)
Ar flux (sccm)
Power - Fwd/Ref (W)
Comments
Done Profilometer measurement of oxide thickness (calibration sample)
-
Ellipsometer measurement of oxide thickness (calibration sample) -
Ready for the next Step: ________
Step 8 Oxide Lift-Off Date Responsible
l Microstripe 3001
Day Start @ End @ Total Type of US
- - DWL room
- - DWL room
Done Profilometer measurement of background (oxide) vs stack
-
Ready for the next Step: ________
Magnetic Tunnel Junction 7 / 9 INESC-MN
Step 9 3st DWL Exposure for Top Electrode Definition Date
Responsible
Machine: DWL
Machine DWL
Map LGJTEST Mask cs_v1_L3
Focus Energy The normal value + 10%
because of the pre-development
Power
1st Cross L1 Cross 2nd Cross L1 Cross Distance (17500,0)
(0, 0) Auto do -1 Ali -1
Optical Inspection Comments pre-develop: 20secs
develop:60secs
Step 10 Metallization Date
Responsible
Machine Nordiko 7000
Sequence Metalization
Functions Function Name Conditions Done
Mod2-f9 1min etch 40/60W RF1=40W, RF2=60W, 3.0 mTorr, 50 sccm Ar, time = 60s Mod4-f 19 3000A Aluminim 2 Kw 2 kW, 3.0 mTorr, 50 sccm Ar, time = 80s Mod3-f19 150A TiW 0.5 kW, 3.0 mTorr, 50 sccm Ar +10 sccm N2, time = 27s
• Put samples in vapour prime (~30min) (or 1min@110Cº) √
• Coat samples (Coating Line):
Prog 6 + Prog 2
PRE-DEVELOP 20s (no heating at 110ºC)
• Develop samples (Develop Line):
Prog 2 + Prog 6
Bake (60s@ 85deg)
Bake (60s@ 110deg) + Cooling (30s)
Spray(3s@ 0.5krpm) + Dissolve(60s)
+ Spin ([email protected] +30s@2krmp)
Spin (1min@ 0.8krpm): 1.5µm PR.
Magnetic Tunnel Junction 8 / 9 INESC-MN
Step 11 Metal Lift-Off Date
Responsible
l Microstrip 3001
Day Start @: End @: Total Type of US
Step 12 4th DWL Exposure for Via Definition Date
Responsible
Machine: DWL
Machine DWL
Map LGJTEST Mask cs_v1_L4d
Focus Energy The normal value + 10% because
of the pre-development
Power
1st Cross L1 Cross 2nd Cross L1 Cross Distance (17500,0)
(0, 0) Auto do -1 Ali -1
Optical Inspection Comments pre-develop: 20secs
develop:60secs
• Put samples in vapour prime (~30min) (or 1min@110Cº) √
• Coat samples (Coating Line):
Prog 6 + Prog 2
PRE-DEVELOP 20s (no heating at 110ºC)
• Develop samples (Develop Line):
Prog 2 + Prog 6
Bake (60s@ 85deg)
Bake (60s@ 110deg) + Cooling (30s)
Spray(3s@ 0.5krpm) + Dissolve(60s)
+ Spin ([email protected] +30s@2krmp)
Spin (1min@ 0.8krpm): 1.5µm PR.
Magnetic Tunnel Junction 9 / 9 INESC-MN
Step 13 Oxide Deposition (Al2O3) Date Responsible
Done Use of calibration sample (Si substrate w/ pen line)
-
Machine: UHVII Total to deposit 4000 Å Rate A/min Total Time Start time End time Total Time
Read v.
Base Pressure (T)
P before plasma (mT)
P working (mT)
Turbo Pump freq. (Hz)
Ar flux (sccm)
Power - Fwd/Ref (W)
Comments
Done Profilometer measurement of oxide thickness (calibration sample)
-
Ellipsometer measurement of oxide thickness (calibration sample) -
Ready for the next Step: ________
Step 14 Oxide Lift-Off Date Responsible
l Microstripe 3001
Day Start @ End @ Total Type of US
- - DWL room
- - DWL room
Done Profilometer measurement of background (oxide) vs stack
-
Ready for the next Step: ________