Compact size 3D magnetometer based on magnetoresistive sensors. Engineering Physics

Transcription

1 Compact size 3D magnetometer based on magnetoresistive sensors Gabriel António Nunes Farinha Thesis to obtain the Master of Science Degree in Engineering Physics Supervisor: Prof. Susana Isabel Pinheiro Cardoso de Freitas Examination Committee Chairperson: Supervisor: Member of the Committee: Prof. Pedro Miguel Félix Brogueira Prof. Susana Isabel Pinheiro Cardoso de Freitas Prof. Diana Cristina Pinto Leitão Prof. João Oliveira Ventura September 2015

2 ii

3 Acknowledgments First of all, I want to thank my parents, who gave me the opportunity to be here today writing this thesis, not only for paying all the expenses related with my academic life, but also for the days that they were just there, listening and advising me on life. Also to my little brother, whose curiosity to know everything and always coming to me for answers make me very happy! Like in the day when we went to laboratory and I showed him "my workplace". To professor Susana Freitas, who embraced me as her student and made me a more confident person trusting me with so important and expensive machines by myself, specially on weekends with no engineers nearby. Also for all the guiding, time and effort spending on me and on my work. Not forgetting all those mandatory presentations... I am now a more prepared person to life, it is the true! To Simon Knudde! Well, you were my mentor during this 9 months! Analyzing results and taking conclusions, saying what I should do next and teaching me how to work with so many machines! It was really good to work with you, and now I can look at the results and say that we were always improving it, even when we tried copper! (shhh, don t tell Susana!) To Ricardo Varela, my cleanroom partner, working during weeks and weekends, for all the time we worked side by side facing problems with the machines! To Miguel Neto for all the conversations that we had and for what I learned from you. To Raju, the guy that could notice my humor just by looking at me. To Andreia, Anastacia, Bernardo, João Valadeiro, Marilia (all that noise measurements!), Sara (for reading this thesis) and Rita Soares. Thanks also to the engineers of INESC-MN that helped me and gave me support working with the machines. And by last, to all the people who were daily by my side. Joana Catarino for listening me everyday and for giving me that big hug, which helped me facing every each day; to my friends Catarina Sequeira, Sara Condeixa, João Barata, Alexandre Gomes, André Boné and André Martins. Thank you! iii

4 iv

5 Resumo Magnetómetros são dispositivos usados para medir campos magnéticos com praticamente duzentos anos de história. Apesar de vastamente usados no dia a dia, chegando a ter aplicações espaciais, só recentemente começaram a ser melhorados em aspectos como dimensões e sensibilidade. Ao longo desta tese realiza-se a optimização de sensores magnetoresistivos por efeito de túnel, também classificados como state of the art na detecção magnética e com a vantagem de poderem ser produzidos com dimensões micrométricas. Depositaram-se stacks para produção deste tipo de sensores, com barreiras amorfas de AlOx, com a linearização dos mesmos realizada recorrendo a estratégias como a anisotropia de forma, para configurações top pinned alcançando-se um TMR de 35% e para configurações bottom pinned alcançando-se um TMR de 30%, com camadas livres e fixas compostas por (CoFe)B. Seguidamente incorporou-se NiFe na camada livre permitindo obter coercividades de 0.05 mt e TMR de 29%. Os valores de detectividade obtidos foram de 120 nt/ Hz a baixas frequências e 20 nt/ Hz a altas frequências, com um ruído caracterizado por um parâmetro de Hooge de µm 2. O dispositivo final é um magnetómetro capaz de medir campos magnéticos nas três dimensões espaciais, com sensibilidade de 5.6 mv/mt e uma gama de operação entre -10 mt e +10 mt, constituído por 4 sensores individuais montados em ponte de Wheatstone em cada uma das dimensões (x,y,z). Palavras-chave: Junções de efeito de túnel, magnetómetro 3D, linearização por anisotropia de forma, barreira de AlOx, pontes de Wheatstone. v

6 vi

7 Abstract Magnetometers are devices used to measure magnetic fields with almost 200 years of history. Although vastly used in our daily life, having for example spatial applications, they just recently started to be improved in their sensitivity and dimensions. Along this thesis, optimization of magnetoresistive sensors working by tunnel effect will be presented, classified as state of the art in magnetic detection and with the advantage of being produced with small dimensions. Stacks were deposited to produce this type of sensors, with amorphous barriers of AlOx. Strategies as shape anisotropy were used to allow the linearization of the sensors, achieving for top pinned configurations a TMR of 35% and for bottom pinned configurations a TMR of 29%, with both free and pinned layers composed by (CoFe)B. The incorporation of NiFe on the free layer allowed to have sensors with 0.05 mt of coercivity and TMR of 29%. The detectivity values obtained were 120 nt/ Hz at low frequencies and 20 nt/ Hz at high frequencies, with a noise density characterized by an Hooge parameter of µm 2. The final device is a magnetometer which is capable of measuring magnetic fields in three spatial dimensions, with a sensitivity of 5.6 mv/mt and a detection field in the range of -10 mt and +10 mt, composed by 4 single sensors assembled on Wheatstone bridge configuration for each dimension (x,y,z). Keywords: Magnetic tunnel junctions, 3D magnetometer, shape anisotropy linearization, AlOx barrier, Wheatstone bridge. vii

8 viii

9 Contents Acknowledgments iii Resumo v Abstract vii List of Tables xii List of Figures xv Nomenclature xvii 1 Introduction Motivation State-of-the-art Theoretical Background Magnetism and Magnetic Materials Tunneling Magnetoresistance (TMR) I-V characteristic Bias voltage dependence and breakdown voltage Linearization of magnetic tunnel junctions Magnetic Tunnel Junctions - Bridges Noise in Magnetic Tunnel Junctions Thermal Noise Shot Noise /f Noise Random Telegraph Noise Detectivity Device Micro-Fabrication and Characterization Techniques Deposition of thin films - Nordiko Definition of bottom electrode Lithography Ion milling - Nordiko Resist Strip Definition of the pillar ix

10 nd Lithography Second Etch - Nordiko Oxide Deposition and Liftoff Oxide Deposition - UHV II Oxide liftoff Definition of top electrode Metallization - Nordiko Liftoff Annealing treatment Sensor Characterization Magnetic measurements - VSM Magnetotransport Curve Noise Measurements Optimization of Alumina Junctions Bottom Pinned MTJ Samples Optimization of the barrier parameters Thicker free layer in different barriers Free-layer optimization Top Pinned MTJ samples Deposition of test samples Thicker free layer in different barriers Free layer optimization with different etch conditions Incorporation of NiFe on free layer Study of Redeposition in Top-Pinned Samples Study of IV Curves and TMR bias dependence Noise Study Magnetometer Characterization and Results Bridge assembly and characterization Magnetometer output for 3 axis Conclusions and Outlook 59 Bibliography 64 A Process Runsheet 65 x

11 List of Tables 2.1 Output of a full Wheatstone bridge assembly Deposition conditions used in Nordiko Base Pressure (B. Pressure), Working Pressure (W. Pressure), V + and V are the potentials of each grid respectively Parameters used on the barrier oxidation Color code for the illustrations during the process presented in images as in figure 3.2. The thickness of the layers are not on scale Resume of the parameters used in coating and development of the sample before and after exposition Read values during an etch process. Definition of bottom electrode Deposition conditions of UHV II machine Resume of operating conditions related with the metallization process on Nordiko Samples deposited with different barrier thickness and oxidation time Resume of the results obtained for barrier optimization. Values corresponding to one pillar per sample Summary of the stacks from RP66 and RP67 samples Resume of the results of samples RP66 and RP67 with different free layers Sample s stacks for free layer optimization Summary of the results for free layer optimization. Test of different free layer s thickness Deposition of top pinned test samples Summary of the results for test top pinned samples, RP31 and RP32. Change of the spacer s thickness Stack of top pinned samples RP64 and RP65. Test of different barriers Deposited stacks for top pinned samples RP79 and RP Free layer optimization with NiFe. The values indicated with * correspond to the deposited value, not to the real value after the complete deposition since there is an etch process in between Etch conditions used on Nordiko Resume of the results for free layer optimization with NiFe xi

12 4.14 Resume of noise measurements for both types of free layer before and after annealing, measured at 0 Tesla field xii

13 List of Figures 1.1 The basic fluxgate principle [3] Scheme of a magnetic tunnel junction with SAF, bottom pinned on the left and top pinned on the right Transport in the antiparallel state of a magnetic tunnel junction considering the Fermi golden rule [10] Energy barrier diagram of a tunnel junction with different electrodes Different magnetic responses for a MTJ. On the right a square response with coercivity (Hc), on the left a total linear response with almost no coercitivity. Ms - Saturation Magnetization On the left: scheme of orientation of the demagnetizing field depending of the orientation of magnetization. On the right: scheme of the fields acting on the free layer, parallel anisotropy case Summary of the fields acting on the free layer. Linear transfer curve. Taken from [27] Full Wheatstone bridge configuration in different power supplies Figures comparing the influence of a magnetic field in the noise spectra of a MTJ [30] Nordiko 3000 views and scheme Full stack of a MTJ deposited over a glass substrate. Full legend of the colors can be found on table Resume of lithography process for 1 st etch Lithography equipment available at INESC-MN class 10 cleanroom Machine used for etch process (Nordiko 3600) and scheme of the process Schematic of angles on etch process in Nordiko Scheme of the sample after the resist strip step Image of the structures after first etch and resist strip, taken on optical microscope Resume of the 2 nd lithography process Scheme of the sample after second etch Non-vertical profile of the pillar on 2 nd etch Scheme of oxide deposition over the sample (oxide can be seen in gray color) UHV II machine. Deposition of Al 2 O xiii

14 3.14 Scheme of the sample after oxide liftoff Example of a pillar before and after oxide liftoff process Resume of the 3 rd lithography process Photo of two structures after third lithography. Resist remains outside the pattern Resume of metallization process and Nordiko 7000 design Final device after complete process Annealing equipment used to thermal treatment VSM Setup, model DSM Magnetotransport Equipment, both setups are supplied by Kepco current sources Noise Setup Equipment Correspondent circuit of noise setup VSM measurement performed to the stack:ru 150 Å/CoFeB 30 Å/AlOx 9 Å/CoFeB 30 Å/Ru 9 Å/CoFe 22 Å/IrMn 180 Å/Ru 200 Å/Mg 4 Å/Ru 200 Å. M Sat = 1390 emu/cm 3 of the three magnetic layers summed (maximum point). FL - Free Layer, PL - Pinned Layer, RL - Reference Layer Results of the barrier optimization process with annealing Magnetoresistance results from sample RP66 and RP67 with annealing Result of different free layer s thickness for pillars with 1 50 µm 2, with annealing Results of top pinned samples RP31 and RP VSM measurement to top pinned sample RP64. M Sat = 1400 emu/cm Magnetoresistance results for top pinned samples RP64 and RP Magnetoresistance results for top pinned sample RP79 and RP Results of top pinned samples for NiFe and (CoFe)B free layers. Pillar Size: µm Analysis of coercivity and sensitivity for all samples. The sensitivity values are normalized by the maximum TMR. Sensitivity values correspond only to linear sensors: RP127, RP128, RP129, RP131. Pillar size: µm Redeposition study of samples RP31 and RP TMR dependence of the applied bias current. Sensor s R A = kω.µm Resistance and TMR of an MTJ sensor for different applied currents. Sensor s R A = kω.µm IV curve for an MTJ sensor in parallel and anti-parallel states. Sensor s R A = kω.µm Magnetoresistance response of the sensor before annealing Noise and detectivity results for CoF e free layer sensor before annealing Magnetoresistance response of the sensor after annealing Noise and detectivity results for CoF e free layer sensor after annealing Magnetoresistance response of the sensor before annealing Noise and detectivity results for CoF e + NiF e free layer sensor before annealing Magnetoresistance response of the sensor after annealing xiv

15 4.22 Noise and detectivity results for CoFe + NiFe free layer sensor after annealing Hooge parameter for MTJ sensors as a function of R A. Comparison between MgO and AlOx barrier structures. Adapted from [38]. AlOx-MTJ data obtained with devices at saturation from [39 42]. MgO-MTJ data obtained at linear operation range. Data marked with & from [38]; * from [43]; α from [44]; β from [45]; δ is unpublished data from INESC- MN; σ from [46]; γ from [47]; i from [39][48][49] and ii is unpublished data from INESC- MN. The result achieved in this thesis is presented as a blue square Sensor s stack chosen to bridge magnetometer. Barrier of AlOx made by two step process One inch die with 225 sensors. Dashed lines representing the dicing path Full bridge configuration. Sensors S1 and S2 orientated with opposite sensitive directions to sensors S3 and S Sensor with 2 pads wire bonded to a PCB Overview of the PCB and correspondent schematic Overview of the process and sensors composing the bridge. Magnetoresistance curves characterizing each component Schematic of the correspondent position for each bridge Electrical design of the magnetometer. 3 bridges assembled in series Magnetometer final device. Side and front views with Autocad design Bridge being tested measuring the magnetic field of a permanent magnet Final result of the bridge output measuring a permanent magnet with off-set correction.. 58 xv

16 xvi

17 Nomenclature AFM: AntiFerroMagnetic SAF: Synthetic AntiFerromagnetic AMR: Anisotropic MagnetoResistance TMR: Tunnel MagnetoResistance CPP: Current Perpendicular to Plane UHV: Ultra High Vacuum DWL: Direct Write Laser VSM: Vibrating sample magnetometer EA: Easy Axis α H : Modified Hooge Constant FL: Free Layer H: Magnetic Field H ex : Exchange coupling field GMR: Giant MagnetoResistance MR: Magnetoresistance M Sat : Saturation Magnetization MTJ: Magnetic Tunnel Junction PL: Pinned Layer R A: Resistance Area product RL: Reference Layer xvii

18 xviii

19 Chapter 1 Introduction 1.1 Motivation Magnetic sensors are extremely important nowadays. They are widely used, being integrated in many devices, such as mobile phones, medical devices and navigation tools. This thesis explores further applications, aiming developing/characterizing magnetic sensors to be implemented in for example spatial aircrafts. The first magnetometer sensors used in space applications occurred in 1958, with a magnetometer on board of Sputnik 3 [1]. After this moment, many missions followed, as the Lunik 1 and 2 (1959) that tried to measure the magnetic field of the Moon. The most common instruments to measure magnetic fields in space are based on fluxgate sensors. These were the first magnetic sensors to be developed and, since then, the principle of operation as not changed. Fluxgates are still used nowadays, as they are robust, reliable and have good resolution in magnetic field. However, their mass and volume are not optimized for new generations of satellites, like nano or picosats. It is clear now that factors such as dimensions and weight of each component of a satellite are extremely important when blueprinting a spacecraft. There are already satellites with magnetoresistive sensors, based on AMR (anisotropic magnetoresistance) and GMR (giant magnetoresistance) technologies. But with the recent discoveries and developments in TMR (tunneling magnetoresistance) sensors, they can be the solution for most of the drawbacks found in previous generations of magnetic space sensors. Focusing on this goal, the main idea is the development of a ultralight, low cost and irradiation resistive, ultrasentitive 3D magnetometer based on the magnetoresistive sensing TMR principle, to be integrated at a small footprint device (below 10 mm 3 ) produced at medium-to-large scale. 1

20 1.2 State-of-the-art The first magnetometer was created by a German scientist, Carl Friedrich Gauss in 1832, which assembled a primitive device consisting on a permanent magnet suspend in midair by a fiber. Since that moment, different types of magnetometers were produced, being one of the most outstanding the fluxgate, the so called "King of Space". The first patent on the fluxgate sensor was credited to H. P. Tomas, in As previously stated, Sputnik 3 was the first satellite to carry a magnetometer, in In the early years satellites were larger and parameters such as mass and dimensions were not considered the most important until 1981, when UoSAT-1, an experimental satellite from the University of Surrey, in the UK, was launched from the Vandenberg Air Force Base. This satellite weighted 52 kg and it was the first microsatellite, according to the modern concept of small satellite, and this was the moment when factors as mass, volume and power savings emerged as extremely important requirements. A magnetometer was included in this satellite, consisting in a three-axis fluxgate that provided vector measurements of the Earth magnetic field. The sensor had a resolution of 2 nt and each orthogonal component had a range around 33 µt [1]. Other satellites were launched and many universities followed this example, leading to smaller satellites, for example OSCAR-18 with kg and mm 3, developed by Weber Sate University in Utah in 1990 [2]. Fluxgate sensors are solid-state devices without any moving parts and they work in a wide temperature range, being the best trade-off between resolution, stability and power consumption [1]. They also provide better resolution than magnetoresistive (MR) sensors, as can reach 10-pT resolution and 1 nt long-term stability; 100-pT resolution and 10-nT absolut precision is standard in commercially produced devices [3]. However, fluxgate sensors have cut-off frequencies of a few Hz to khz and a noise density of several pt/ Hz at 1Hz. Figure 1.1: The basic fluxgate principle [3]. The basic sensor principle is illustrated in figure 1.1. The soft magnetic material of the sensor core is periodically saturated in both polarities by the ac excitation field, produced by an excitation current I exc inducing changes in the core permeability, and the dc flux associated with the measured dc magnetic field B 0 is modulated. The voltage V I induced into the sensing coil is proportional to the measured field. Although fluxgates are suitable for space applications, they present some disadvantages such as power consumption. These drawbacks can be overpassed using the new generations of sensors, based on GMR/TMR principles. In 1857, William Thomson (Lord Kelvin) discovered that the resistance of a magnetic material changes when an external magnetic field is applied under the material (in order 2

21 of 3%), an effect know as ordinary magnetoresistance [4 7]. Based on this effect, AMR (anisotropic magnetoresistance) sensors started to appear, being the most used and mature after the fluxgates. Commercial AMR sensors typically have dynamic ranges of hundreds of µt and resolutions in the order of 1 nt, detectivities in the order of nt/ Hz for frequencies higher than 10 Hz and sensitivities in the order of 10 mv/(mt V brigde ). Since the nineties, this area has been widely studied, including progresses in fabrication and optimization of systems with reduced dimensions based on quantum phenomena. Sensors based on GMR were discovered for the first time in 1988 by the groups of A. Fert [8] and P. Grunberg [9]. They not only observed a change of resistance when a magnetic field is applied, but also noticed that was the changing in the direction of the magnetic field that causes the variation, this effect occurs in materials composed by two ferromagnetic layers (typically with a few nanometers) and a non-magnetic material between them. The GMR effect is due to the spin polarization of the electrons, with the ferromagnetic layers working as polarizers/filters. If both magnetizations of the ferromagnetic materials are aligned in the same direction, the electrons polarized in the same way will see a lower resistance than in the case of the layers with opposite magnetization. The discovery of spin filtering effect opened new ways and fields of research, as spin electronics. The measured magnetoresistance is much larger than in the case of AMR and, therefore it s called giant magnetoresistance. For example, in NiFeCo/Cu/Co/Cu multilayers, 16 % of variation in resistance was observed between 0 Oe and 50 Oe at 300 K [10]. In contrast to AMR, GMR devices have higher hysteresis, up to 10% [1], and were used for reading heads in magnetic recording systems. Although with some biasing mechanisms, good sensing properties and repeatability can be achieved. This will, however, result in a higher power consumption. More recently, structures with an insulating layer separating both magnetic layers have been studied. In this case, electrical conduction between the two magnetic layers, supposing that current goes perpendicular to their plane, is allowed by quantum tunneling through the insulator. This phenomenon of tunnel magnetoresistance was first discovered by Meservey and Tedrow [11 13] while doing experiments on spin dependent tunneling. Then, Julliere [14] reported this effect at low temperatures with Co and F e layers separated by Ge. Again, changing the relative magnetic configuration of the adjacent layers, from parallel to anti-parallel the resistance increased about 14 %. Over the years, the TMR ration has been continuously enhanced. In 2004, the groups of S.S. Parkin et al in IBM Almaden and S. Yuasa et al in Japan reported simultaneously a 220% for CoFe/MgO magnetic tunnel junctions (MTJs) [15, 16]. In 2008, Ikeda et al reported a magnetoresistance of 604 % at 300 K with layers of CoFeB/MgO/CoFeB by suppression of Ta diffusion [17]. Although the high values of these parameter, it is also important to guarantee a good linearity and a well defined range of operation of the sensor to obtain good results in output, and, if needed, give up of some parameters ensuring others, as a relation between linearity, coercivity and sensitivity of the sensor. As GMR was later discovered and AMR technology was already mature in the nineties, in 2003 several missions were launched with this kind of sensors, such as the American Ionospheric Observation Nanosatellite Formation (ION-F) [18] and the Canadian CanX-1 [19]. Three other satellites, of the ION-F, all of them weighting between 10 and 15 kg used three-axis AMR magnetic sensors for the measurement 3

22 of the Earth magnetic field up to a 2 accuracy together with CCD cameras. In 2005, an AMR-based sensor was launched on board NANOSAT-01 by the Spanish group of Space Magnetism of the Spanish National Institute of Aerospace Technology (INTA) [10]. In that year, three spanish missions started almost simultaneously: an INTA picosatellite with less than 3 kg and a GMR experiment, this satellite has another AMR sensor for the attitude and orbit control, so the whole experiment consists on comparing both vector sensors, a second one in the frame o NANOSAT with a GMR and a magnetoimpedance vector sensor, and the third one the MAGNETITA, also a GMR vector sensor. This sensors are supposed to measure the geomagnetic field of ± 60 µt with accuracy in the order of 10 nt. Another AMR magnetometer was launched on the TRIO-CINEMA mission in 2012, with a noise floor of less than 50 pt Hz 1/2, above 1 Hz on a 5 V bridge bias [20]. There are other types of sensors as magneto-electric or the search coil. The last one is one of the oldest magnetometers and can detect a magnetic field as weak as 20 ft and can operate from 1 Hz to 1 MHz. However, when applied to navigation, the coil will have a signal either if it feels an AC field or if it moves along a DC field. Considering this, the TMR magnetometers are the best candidate so far among all the other sensors. Already in 2014/2015, the research group INESC-MN obtained 147%/mT of GMR sensitivity with a detection level of 1.0 nt/hz 1/2 at 30 Hz, and for higher frequencies of the order 10 khz a detection level of 71 pt/hz 1/2. To reach this values, strategies as the use of magnetic flux guides to enhance the sensitivity were used in MgO magnetic tunnel junctions [21]. 4

23 Chapter 2 Theoretical Background 2.1 Magnetism and Magnetic Materials Tunneling Magnetoresistance (TMR) Magnetic tunnel junctions (MTJs) emerged as result of the evolution of previous devices and principles, as mentioned before in chapter 1. First reported at low temperatures by Julliere [14] this device is, in the most basic model composed by two ferromagnetic layers with a non-magnetic spacer between them. Is this spacer that distinguishes a MTJ from a Spin-Valve (SV). While the second operates with a conducting layer, a MTJ uses an insulating layer, which force the electrons to tunnel between the two ferromagnets. These kind of devices work based on the principle of magnetoresistante (MR), where a change in electrical resistance (R) occurs when a magnetic field (H) is applied, being this change much higher in MTJs than SVs. MR is defined in expression (2.1): MR = R max R min R min 100 [%], (2.1) being R max the maximum resistance and R min the minimum resistance. The difference of resistances depends on the direction of the magnetic field that is applied. Therefore, the maximum resistance will correspond to the anti-paralell state while the minimum resistance derives from the parallel state. The paralell and anti-parallel alignment is related with the angle between the magnetization of the two ferromagnetic layers. Some strategies were studied to fix the magnetization s direction of one of the layers, namely by using an antiferromagnet material that pins the adjacent ferromagnetic layer, forcing the magnetization of the pinned layer to remain fixed for certain range of fields, while the other correspondent layer s magnetization is free to rotate. Another methods require the deposition of more layers, making use of magnetic principles such as RKKY interaction (Ruderman-Kittel-Kasuya-Yosida) between two ferromagnets with a spacer in between or the exchange bias interaction regarding the antiferromagnet and ferromagnet. These two methods were applied in the structures studied in this thesis. 5

24 A scheme of two magnetic tunnel junctions is presented in figure 2.1. Figure 2.1: Scheme of a magnetic tunnel junction with SAF, bottom pinned on the left and top pinned on the right. Both structures were optimized during this work, with more emphasis on top pinned junctions. As pictured, they are very similar in composition, but diverge in the order of deposition of the layers. In bottom pinned MTJs the free layer comes after the oxide barrier, while in top pinned the free layer appears under the barrier. A buffer with good electrical conductors as tantalum, aluminum or ruthenium is, in general, used to lower the resistance of the device. The buffer will also play an important role on the process, allowing the etch of the ferromagnetic layers and keeping the electrical contact between the structures. As for bottom pinned MTJs, an antiferromagnetic layer is firstly deposited, allowing the pinning of the ferromagnet that is subsequently deposited. Both structures use synthetic synthetic antiferromagnetic structure (SAF), which is based on the principles mentioned before. The spacer has, commonly, a few angstrom (this thesis considered approximately 6Å), while the ferromagnets could achieve values higher than ten angstrom. In the bottom pinned MTJs, the layer just below the oxide barrier plays the role of reference layer, and the layer above corresponds to the free layer. This insulator layer between the two ferromagnets needs to be thin enough to allow the tunneling of the electrons (2-20 Å), which will cross the barrier perpendicularly to it. For this reason a MTJ is characterized as a current perpendicular to the plane (CPP) device. The last layer, capping layer, protects the ferromagnets and should also have a lower resistance. Ferromagnetic materials, typically characterized by an overlaping of incomplete 3d orbitals, have a strong spin imbalance at their Fermi level, resulting in a different number of available states for each spin. Assuming that the electron spin is conserved during the tunneling [14], electrons at the Fermi level of one ferromagnet will tunnel into free equivalent states of the other ferromagnet, so spin up electons will tunnel to spin up empty states in the other ferromagnet. This principle applies to spin down electrons [22], working like two independent channels for each spin, as explained in image 2.2 for a case of anti-parallel ferromagnets. The conductance seen by the electrons depends on their spin, more intrinsically in the density of 6

25 Figure 2.2: Transport in the antiparallel state of a magnetic tunnel junction considering the Fermi golden rule [10]. states of the material, as expressed in 2.2 and 2.3. G p ν 1 ν 2 + ν 1 ν 2, (2.2) G ap ν 1 ν 2 + ν 1 ν 2, (2.3) where G p and G ap extends for parallel and antiparallel conductance, respectively, and ν i meaning the density of states on each ferromagnet (1 and 2) of spin down and spin up electrons. In the case of parallel alignment between the two ferromagnets, the electrons of one spin are the majority and can tunnel to the other electrode, where the majority of the density of states corresponds also to the same spin. On the other hand and as shown in figure 2.2, in case of antiparallel alignment, the spin up electrons are in majority in the first ferromagnet, while the other ferromagnet holds larger density of states available to receive them belongs to spin down electrons, leading to a decrease in conductance. This formulation rephrases the expression for MR, or particularly for MTJs, TMR: T MR = G p G ap Gap 100 [%]. (2.4) The same equation can also be written taking into account the polarization (P i ) of spins in each ferromagnetic layer. The TMR expression can again be rewritten, leading to the Landée Formula: P i = ν i ν i ν i + ν i T MR = 2 P 1 P 2 1 P 1 P 2. (2.5) Materials with larger polarizations, or larger spin imbalance, like Ni, Co and Fe alloys, can achieve higher values of TMR. As seen until this point, none of the expressions presented is related with the characteristics of the barrier itself, that is a current perpendicular to the plane mechanism. It is expected a decrease in sensor resistance with both increasing area of the pillar or decreasing its thickness, once the tunneling probability depends also on the thickness of the barrier, meaning the distance that an electron needs to tunnel, since its wave like behavior leads to a non-null probability of the electrons pass through an insulator barrier. In the case of a too thin barrier, pinholes can appear and the resistance will decrease. 7

26 A parameter that characterizes a barrier is the R A product, that is constant for different areas of pillars and depends on the insulator thickness, going from Ω.µm 2 to MΩ.µm 2. Later on in 1989, Slonczewski [23] finally considered the tunneling between two identical ferromagnetic electrodes, separated by a rectangular potential barrier, assuming that the ferromagnets can be described by two parabolic bands and calculating the vanishing wavefunction inside the tunnel barrier. Reviews on spin-dependent tunneling by Levy and Zhang [24] appeared in 1999 and summarized experimental and theoretical results publish up to this date I-V characteristic The electrical current and voltage response of a MTJ is not linear as in a single resistor. The potential difference created across the two ferromagnetic electrodes through the insulator, depends of the barrier and the height of each ferromagnetic material, φ 1 and φ 2. This height is defined as the work required to remove one electron from the metal conduction band, and it is represented in figure 2.3. (a) No voltage applied across the junction. (b) Voltage applied across the junction. Figure 2.3: Energy barrier diagram of a tunnel junction with different electrodes. If the ferromagnetic materials adjacent to the barrier are different, their heights will also be different. The tunneling current I created by an applied voltage can be given by Simmons model [25], which states that the current has a exponential dependence of the thickness t of the barrier and the heights of each ferromagnetic material, as shown in equation 2.6. [ I exp k t ] φ, (2.6) where k is a constant of the material. Since R=V/I, the resistance of a MTJ increases exponentially with the thickness of the barrier, being function of only two variable parameters, the area of the barrier and its thickness Bias voltage dependence and breakdown voltage The TMR of a magnetic tunnel junction is not constant with the applied voltage. For small values of voltage, around 30 mv, the TMR is almost constant, but for higher values it starts to decrease. This 8

27 TMR decrease depends on the interfaces, barrier type and also on the ferromagnetic materials. The value for which the TMR is reduced for half of its value is known as V 1/2. If the bias voltage across the barrier is increased above a certain value, the insulator barrier will be destroyed, with the consequent formation of pinholes, leading to an abrupt decrease in resistance. This effect is related with the dielectric strength (electric field that results in breakdown) of the insulator, and can be calculated with the expression 2.7. V break = E t, (2.7) where E is the dielectric strength and t the barrier thickness. For an insulator Al 2 O 3, with a dielectric strength of 10 9 V/m and a barrier of 14 Å, V break =1.4 V Linearization of magnetic tunnel junctions The magnetic tunnel junctions have several applications. Depending oh each one, a linear or a square response of the sensor can be produced. For memories (MRAM), for example, a square response is necessary, while for a sensor to detect/read magnetic fields, a linear response is required and preferably with coercivity lower than 0.1 mt. Ms Magnetization Hc Magnetization Magnetic Field Magnetic Field Figure 2.4: Different magnetic responses for a MTJ. On the right a square response with coercivity (Hc), on the left a total linear response with almost no coercitivity. Ms - Saturation Magnetization. A thermodynamic model can be used to describe the magnetic properties of the free layer. The energy of this layer (E f ), in a case of parallel crystalline anisotropy, is given by expression, which includes 5 independent terms: E f = µ 0 H.M f s µ 0H k M f s sin 2 θ µ 0 H p d.m µ 0H f d.m µ 0H N.M. (2.8) Zeeman term: µ 0 H.Ms f. This term relates the energy of interaction between an applied field H and the magnetization of the free ferromagnetic layer Ms f. It can be explained using the atomic 9

28 dipole model, which assumes that the atoms or molecules that constitute the ferromagnetic layer have a magnetic moment µ as if they are permanent dipoles, and so when an external magnetic field is applied, they will align its spin vector with the direction of the field, being the Zeeman energy, the energy required to rotate his spin vector. Intrinsic anisotropic term: 1 2 µ 0H k M f s sin 2 θ. The anisotropic term arises from the fact that the dipoles tend to align in a preferential direction in materials, being stronger along particular crystallographic orientations. The internal field (H k ), aligns the net magnetization in a direction called easy axis. Magnetocristaline anisotropy is defined as the work required to rotate the sample magnetization out of the easy direction. This thesis deals always with magneto crystalline intrinsic anisotropy of the free layer parallel to the pinned layer, since the deposition is always done at the same angle with a magnetic field applied, allowing the crystal to grow always with the same direction. The angle θ is the angle between the magnetization and H k. Demagnetizing terms: µ 0 H p d.m µ 0H f d.m. First term corresponds to demagnetizing energy from the pinned layer which acts in the free layer, and the second is the self-demagnetizing energy of the free layer. These terms are mostly related with the shape of the structure. If a layer is magnetized it creates a magnetic field, that as said before can be imagined as oriented dipoles, and so, these dipoles create at the surfaces of the material an opposite field, called demagnetizing field (H d ). For larger distances and dimensions, this field is almost negligible. However, when shorter distances are used, it starts to have impact in the energy of the free layer, overcoming the crystalline anisotropy term, and so allowing to tune of the sensitivity of the sensor. Additionally, shorter volumes lead to stronger fields. Considering the demagnetizing field on x direction, it can be given by: H f d = N xxm f s cos θ. Considering a pillar with infinite length, with width h and thickness t, the demagnetizing field of the free layer f is defined as: H f d = 4πM sat tf h. Figure 2.5: On the left: scheme of orientation of the demagnetizing field depending of the orientation of magnetization. On the right: scheme of the fields acting on the free layer, parallel anisotropy case. So, reducing the width of the pillar, meaning the free layer and reference layer, the contribution of the demagnetizing energy in this direction will increase and the MR response will be linear without losing TMR. Consequently, an option must to be taken, because the higher the aspect ratio between width and length of the pillar, the larger is the linear response but leading to a lower sensitivity. Another alternative to get a linear response consists in reducing the thickness of the free layer sufficiently, making it superparamagnetic. However, the MR can be considerably reduced as well, although a non-hysteric response is achieved. The thinner the free layer is, the larger the demagnetizing field will also be [26]. 10

29 Néel term: µ 0 H N.M. This term is related with the roughness of the layers, giving rise to Néel Field (H N ). It s also known as Néel coupling (or orange peel) and it is present in almost all ferromagnetic multilayers, inevitable, since it arises during the deposition. Summary On this work, sensors with parallel anisotropy and shape anisotropy were used. There are more techniques to make a linear sensor, such as external biasing and exchange biasing [26]. The conditions that assure the linearity of the sensor can be deduced from the equation 2.8. The minima of this equation leads to three possible solutions. sin θ = 0 θ = 0 θ = π H k N xx M f s = 0 H H p d + H N = 0 cos θ = H Hp d + H N N xx M f s H k (2.9) In the case of H k <N xx M f s, meaning that the demagnetizing field is higher than the H k of the material, a linear magnetic response is obtained, as presented in figure 2.6. A more detailed explanation can be found at [27]. Figure 2.6: Summary of the fields acting on the free layer. Linear transfer curve. Taken from [27]. The demagnetizing field allows the tuning of sensitivity (slope of the curve given by 1/(N M f s H k )), and the demagnetizing field of the pinned layer together with the Néel field (H p d H N ) promote a shift of the whole curve for right or left. 2.2 Magnetic Tunnel Junctions - Bridges A Wheatstone bridge is a measuring device invented by Samuel Christie in 1833, and improved later on by Sir Charles Wheatstone in Composed by two voltage dividers, which are both connected to the same input, and a differential output being the difference between the two dividers. This type of bridge can be composed by different values of variable resistances, depending of the output that is expected to 11

30 obtain. For this thesis, a full bridge configuration was used, with four MTJs playing the role of variable resistors as presented in figure 5.6. (a) Full Wheatstone bridge configuration with 4 MTJs - Voltage Biased. (b) Full Wheatstone bridge configuration with 4 MTJs - Current Biased. Figure 2.7: Full Wheatstone bridge configuration in different power supplies. In this assembly, the resistances are given by R MT J1 = R MT J4 = R + R and R MT J2 = R MT J3 = R R, and the output is presented in table 2.1. Table 2.1: Output of a full Wheatstone bridge assembly. Voltage Biased V = V in R R Current Biased V = I R Once the bridge has a differential output, the noise of the input source is suppressed and its output is an offset-free signal. In the absence of an applied field, the bridge has a null output, since all MTJs have the same resistance. Furthermore, this configuration provides a linear output and the one with larger variation, when compared with other bridge configurations. The bridges mounted during this work are composed by four individual sensors that were processed in the same sample, but diced before the assembly on the PCB, since they need to have different orientations for the magnetization of the pinned layer. 2.3 Noise in Magnetic Tunnel Junctions One of the main goals of this thesis is to achieve a detection level as lower as possible, meaning that the amplitude signal should be higher than the noise level of the sensor. To ensure this condition, it is essential to characterize the devices in terms of noise, or in the end, signal to noise ratio (SNR). Only analyzing the SNR is possible to compare different devices in order to choose the one with the best performance. In the following subsections, the different sources of noise characterizing MTJs will be presented. 12

31 2.3.1 Thermal Noise Johnson-Nyquist noise, also known as thermal noise [28] [29], is a type of noise present in any electrical device. Observed and published in 1927 by J. B. Johnson, and then demonstrated by H. Nyquist, this noise results from the random thermal motion of electrons, being directly proportional to temperature as it is given by the following expression: V th = 4k B T R [V/ Hz], (2.10) where k B is the Boltzmann constant, T is the absolute temperature and R the device resistance. As shown in equation (2.10), thermal noise is independent of the current that flows through the sensor/resistor, due to the fact that the drift velocities of electrons in conductors are small compared with the electrons thermal velocities. It is also noticeable that the thermal noise is independent from the frequency of the signal (white noise type), therefore, it can work as a calibration source to check the confidence of the obtained results, since its a limit to high frequencies Shot Noise Introduced in 1918 by Walter Schottky, shot noise is related with the current that flows through the device, more precisely from the discontinuities in a circuit. The expression that characterizes the shot noise is given by: V shot = 2 q I R 2 [V/ Hz], (2.11) where q is the electron charge, I is the sensor DC bias current and R is the device resistance. Again, it can be seen that shot noise is also independent of frequency but, in this case, is related with electrical charges. Due to this fact, it is in general lower than thermal noise. The noise is characterized by a Poisson distribution since electrical charges have a discrete nature. If the current is high enough, the number of electrons passing through a MTJ will be larger and the expression (2.11) is considered, where the noise assumes a normal distribution. MTJs are an example of a device where this noise can be easily understood. The electrons need to tunnel through a thin insulator barrier, giving rise to small fluctuations in the current, made up by elemental contributions at random time instants /f Noise Also called Flicker Noise or Pink Noise, 1/f noise is a type of noise which decreases (or it s power spectral density) when the frequency is increased. This noise can be found in almost everything in nature, being more well known for its role in electronics, usually related with charge trapping in crystals defects, or electron trapping, which probability favors the energy concentration at low frequencies. The expression that characterizes 1/f noise is given by: 13

32 αh V 1/f = f A R I [V/ Hz], (2.12) where α H is the Hooge parameter, f the frequency, A the area of the sensor, R and I the same parameters of expressions (2.10) and (2.11). The factor area appears here since the number of carriers in a MTJ is proportional to the area of the device, as it is a current perpendicular to the plane (CIP) device. The Hooge parameter characterizes 1/f noise and its intensity, since an higher parameter means higher intrinsic noise, if area and the other parameters remain constant for both sensors. This type of noise is also often characterized by the 1/f knee, meaning the frequency for which this noise is lower than the others types of noise. In the case presented on this thesis, MTJs have magnetic layers, implying that the Hooge parameter accounts for two different components, the electrical one and the magnetic, resulting in: α H = α electric + α magnetic. The magnetic component depends of the field that is applied to the sensor. (a) Voltage noise spectrum vs frequency spectrum at different easy-axis fields for a µm 2 junction without hardaxis bias fields. (b) The voltage-normalized noise spectra. Figure 2.8: Figures comparing the influence of a magnetic field in the noise spectra of a MTJ [30]. As pictured in figure (2.8), it is at zero magnetic field that the magnetic component of the sensor has a higher contribution to the overall measure, and so, all measurements in this work were done at 0 field (magnetic), being the region of higher sensitivity of the MTJ. This analysis allows characterizing the sensor in the operating point, in order to never underestimate the noise level Random Telegraph Noise Telegraph noise, sometimes called Burst Noise, is a second type of noise that is present at low frequencies. It is characterized by a step profile, being similar to a bias current noise with a step change in the noise level. In case of magnetic tunnel junctions, telegraph noise has origin in defects in the tunnel barrier, creating trapping centers for electrons. In these cases, the resistance of the sensor changes suddenly due to the difficulty of tunneling by the electrons. This noise is also associated with random magnetization 14

33 fluctuations in the sensing layer, as can be seen in this work, where the size of the pillar in the order of micrometers, can give origin to more than one magnetic domain, increasing this noise. In the presence of a strong magnetic field, the domains should behave as one, and so, the noise will be reduced. Summary: These are the main sources of noise present in a magnetic tunnel junction. Additionally, there is also, for example, High Frequency Ferromagnetic Resonance Noise which appears at high frequency, although it is not characterized in this thesis. So the Noise Spectral Density (S) of a MTJ is given by the expression S V 2 = S 1/f V 2 + S Shot V 2 hermal + ST V = α H 2 f A R2 I q I R 2 + 4k B T R [V 2 /Hz], (2.13) 2.4 Detectivity The detectivity of a sensor characterizes the lowest variation of field that can be detected, relating the noise level with the sensitivity of the sensor, as presented in equation 2.14: D = S V 2 V H [T/ Hz]. (2.14) It is a value that directly depends on the local applied voltage and magnetic field felt by the sensor, since the sensitivity ( V/ H) of the sensor changes along the magnetic field range, where there will be a window of operation. As expressed in 2.14, a larger sensitivity of the sensor reflects a lower detectivity. To improve the detectivity, bridges are considered, since their output is larger when compared with a single MTJ. 15

34 16

35 Chapter 3 Device Micro-Fabrication and Characterization Techniques In this chapter, all processes related with the fabrication of the device are presented, as well as the machines involved in which step. A brief summary of the run-sheet is shown below: 1. Stack deposition 2. Definition of bottom electrode (a) 1 st Lithography (b) 1 st Ion milling etching (c) Resist strip 3. Definition of the pillar (a) 2 nd Lithography (b) 2 nd Ion milling etching 4. Oxide Deposition and Liftoff 5. Definition of top electrode (a) 3 rd Lithography (b) Metallization (c) Liftoff 6. Annealing 7. Sensor characterization (a) Magnetotransport Curve (b) IV Curve (c) Noise Measurements 3.1 Deposition of thin films - Nordiko 3000 Nordiko 3000 is an ion milling system made to deposit materials for tunnel junctions, spin-valves, passivation layers and others. It has six available targets and it is ready to work with six inch wafers. There is also the possibility to do etching with this machine. A scheme of machine s inside and the main components is presented in figure 3.1. This is a fully automated machine with two broad beam RF ions sources. It s installed in the class 100 clean room. The load lock is pumped by a turbo pump and the main chamber by two pumps, a turbo and a cryogenic one. 17

36 (a) Front view of Nordiko Computer that controls the machine and load lock. (b) Side view of Nordiko The main chamber and one of the RF sources can be seen. (c) Scheme of machine s inside [31]. Figure 3.1: Nordiko 3000 views and scheme. The wafers are mounted in the load lock facing down and after it reaching the right pressure, a robot arm takes one single wafer to the chamber s inside. The wafer is then attached to a substrate table by clamps, around this table is a permanent magnet with a magnetic field of 4 mt. Different conditions can be used during deposition/etching. The table is allowed to rotate from 0-30 rpm, usually using a value of 15 rpm for depositions. The angle between the substrate and the horizontal plane can also be adjusted from 0 to 90 degrees. The RF coil is placed outside the chamber and the grids inside. The RF coils supply a power signal of MHz to the gas inside the chamber generating a plasma. The three grids system accelerate the plasma and allows a uniform collimated ion beam. A shutter protects the substrate while the ion beam 18

37 parameters are adjusted by the machine software. Once they are achieved, the shutter moves and an ion beam hits the target. The material sputtered from the target is then deposited into a nearby substrate to create a thin film. To avoid surface charging, filamentless neutralizers are assembled beside each of the ion sources. The gun below is used for deposition, since the ion beam hits one target, allowing its particles to be ejected onto the substrate. The assist gun is usually used for etch or oxidation. The parameters used during depositions are presented in the table 3.1. Table 3.1: Deposition conditions used in Nordiko Base Pressure (B. Pressure), Working Pressure (W. Pressure), V + and V are the potentials of each grid respectively. Beam Current V + V RF Power Gas Flow Table Angle B. Pressure W. Pressure 24 ma 1022 V -300 V 110 W 2 sccm Xe Torr Torr The deposition of an MTJ stack is composed by an intermediate step of oxidation. After deposited, the barrier layer (Al in this work), needs to be oxidized. This process occurs by the Remote Plasma method, that consists in creating a mixed Ar-O 2 plasma that is not accelerated by the grid, although the plasma itself creates a 12-14V potential difference between the grids. The O 2 ions that fill in the entire chamber, reach the sample with their thermal energy. The RF source is applying power to the gas originating the plasma. Then, the duration of this process can be controlled. The thickness of the barrier will depend of the previous thickness of Al deposited. The parameters used are presented in table 3.2. After the oxidation, the deposition process will continue as previously, as soon as the right conditions have been achieved. Table 3.2: Parameters used on the barrier oxidation. Figure 3.2: Full stack of a MTJ deposited over a glass substrate. Full legend of the colors can be found on table 3.3. Oxygen Flow Argon Flow Grids Voltage RF Power W.Pressure Table Angle ( ) 40 sccm 4 sccm 0V 110 W Torr 70 Nordiko 3000 is the machine used to perform the first step of an MTJ. The thickness of each layer needs to be tuned according with what is to obtain in the end of the process. As it is possible to change almost every parameters related with the ion beam and the deposition, like the grids voltage, the gas flow, the pressure, it allows to attempt different things. Deposition conditions and parameters for barrier oxidation followed, respectively, tables 3.1 and 3.2, except when mentioned otherwise. Table 3.3: Color code for the illustrations during the process presented in images as in figure 3.2. The thickness of the layers are not on scale. Glass Substrate All layers until the Oxide Barrier Oxide Barrier Layers on top of Oxide Barrier Photoresist 19

38 Definition of bottom electrode Lithography After depositing the complete stack of an MTJ, the next step is to define the bottom electrode. For that, the sample needs to be etched, but to define a structure on it, an intermediate step, known as Lithography is performed. Lithography consists in coating the sample with a layer of a photoresist, a light sensitive polymer, illustrated in figure 3.3(a), using for this work a thickness of 1.5 µm. This layer is exposed with a laser according with designs previously projected in a computer program, such as Autocad. The bottom electrode of the sensor is designed in 2D, as presented in figure 3.3(b). The exposed photoresist will be chemically changed, making it soluble in a development solution, resulting in a structure with the thickness of the photoresist remaining on top of the stack, figure 3.3(c). (a) Scheme of the sample after coating with 1.5µm of photoresist. (b) Scheme of the sample after lithography and development. Only the nonexposed photoresist remain on the sample. (c) Scheme of the sample after lithography and development. Only the nonexposed photoresist remain on the sample. Figure 3.3: Resume of lithography process for 1st etch. The equipment needed to perform all the steps previously described is presented in figure 3.4. The room has a filtered light (yellow light) to not interfere with photoresist properties. The coating track is (a) SVG resist coater and developer tracks. (b) 2.0 Heidelberg direct write laser lithography system. HeCd laser, λ = 405nm, critical dimensions down to 0.8 µm. Figure 3.4: Lithography equipment available at INESC-MN class 10 cleanroom. composed by a spinning stage that reaches the maximum of 2500 rpm during the first step. After that, the sample is heated until 87 C in the next stage. Once exposed in the laser, a Heidelberg 2.0 DWLii 20

39 direct write laser, with a maximum resolution of 0.8 µm, using a 405 nm diode laser, it is heated until 110 C and then developed. The summary of this process is presented in table 3.4. Table 3.4: Resume of the parameters used in coating and development of the sample before and after exposition. Coating Parameters Development Parameters First Step Dispense photoresist on the sample and spinning at 800 rpm for 5 sec. Bake at 110 C for 60 seconds. Second Step Spin at 2500 rpm for 30 sec. to obtain 1.45µm thickness. Cool for 30 seconds. Third Step Soft bake at 87 C for 60 seconds. Developer for 60 seconds. The masks exposed in the laser can be draw using AutoCAD and then converted to lic files. These files divide the masks in stripes of 200 µm of width, the pattern followed by the laser. It is important to highlight that the critical structures, as pillars, should be placed in the center of which lic, allowing the laser to draw them without breaking it. Furthermore, the laser as a Gaussian profile, leading to losses in energy in the border of the lics. Once the sample had been exposed and developed, it should be analyzed in the optical microscope, the structures with photoresist should be clean and well patterned. The use of a filter in microscope is essential in case of more development be needed Ion milling - Nordiko 3600 A magnetic tunnel junction is composed by many different materials, resulting in a variety of chemical properties. To avoid completely different etching rates or different selectivities, when etching a layer, a physical process is used. This dry etch method, made by an ion beam, is known as Ion Milling. The machine used to perform this step is Nordiko 3600, and is presented in figures 3.5(a) and 3.5(b). (a) Front view of Nordiko 3600 machine. (b) Side view of Nordiko 3600 machine. (c) Scheme of the sample after first etch. Definition of both electrode, all stack remais on the structure defined. Figure 3.5: Machine used for etch process (Nordiko 3600) and scheme of the process. Identical to Nordiko 3000, Nordiko 3600 is a full-automated machine and is ready to deposit and 21

40 etch. This machine is also composed by 6 targets available for depositions, and two RF ion sources, being similar to the scheme presented with Nordiko 3000 in figure 3.1(c), with the advantage of being ready to larger wafers of 8 inches, and ensuring a better uniformity in the process. After the robot arm takes the sample from the loadlock to the substrate table, it is also allowed to define an angle between the ion beam and the sample. It is important to highlight that the angle values inserted into the program correspond to a angle of +10. For example if the value set is 15, the real value between the sample and the ion beam will be 25. Figure 3.6: Schematic of angles on etch process in Nordiko A table with the parameters used for all etching processes is presented in figure 3.5. These parameters correspond to the read values of the assist gun. As this machine was only used for ion beam milling process of etch, only the assist gun was used. More details about the machine can be found in [31, 32]. Power (W) Table 3.5: Read values during an etch process. Definition of bottom electrode. Reflected Power (W) Voltage + (V) Current + (ma) Voltage (V) Current (ma) Ar Flow (sccm) Xe Flow (sccm) The base pressure of this machine is around 10 8 Torr, while the working pressure during an etch process decreases to 10 4 Torr. For the first etch, which is supposed to define just the structure of the bottom electrode, a real angle of 70 is used, with an etch rate of 1.05 Å/s. The etch process is always performed in steps lower than 220 seconds, meaning for example, if the entire stack has 1000 Å, the process will be executed in 5 steps of 200 seconds of etch with the conditions presented in table 3.5, adding a step of 200 seconds of cooldown between of each etch step. This strategy avoids the burning of the photoresist due to sample heating. As this step pretends to define the complete structure of bottom electrode, in case of over etch this effect will not bring problems to the final device. If the sample had been deposited on glass substrate, after this step should be possible to see through the glass, with only the structures remaining. 22

41 3.2.3 Resist Strip Once the bottom electrode is defined, the photoresist needs to be removed to prepare the sample to a new exposition. For this purpose, a solution, M icrostrip 3001, dissolves the photoresist and has no interactions with the composition of the sample. To completely remove the resist, the sensors are Figure 3.7: Scheme of immersed at a 60 C bath, with ultrasounds. This process takes about half the sample after the resist strip step. an hour. After this process is completed, only structures with the complete stack will remain on the sample, as can be seen in figure 3.8(b). (a) Photo of two structures after first etch and resist strip. (b) Olympus BH3-MJL optical microscope available in yellow room of INESC-MN. Figure 3.8: Image of the structures after first etch and resist strip, taken on optical microscope. 3.3 Definition of the pillar nd Lithography The conditions for second lithography are exactly the same as mentioned before for coating and development. The energy used by the laser can be changed, as it loses power with the utilization. Assuming that this energy remains constant and having the first exposure performed with a value of 45%, the second exposure is executed with a value of 50% to ensure that the photoresist will have a straight profile on the pillar, which is the critical element of our junction. In this exposure, the pads of bottom electrode will be protected with photoresist, as well as the pillar. A good alignment of the sample should be performed through the marks defined in the first lithography, with angle lower than 0.02 mrad. This alignment is done in two steps, first with a macro camera of the laser achieving an angle lower than 1 mrad, and then the use of the micro camera is required to achieve the 0.02 mrad. After development, the sample should be analyzed to ensure the pillars are well defined with the desired geometry, with borders as straight as possible. If the pillars are not well patterned, this step will compromise the final result of the sensors. 23

42 (a) 2D mask used in DWL exposure. (b) Scheme of the sample after second lithography. Definition of the pillar and protection of the contacts in bottom electrode. Figure 3.9: Resume of the 2 nd lithography process Second Etch - Nordiko 3600 The second etch is performed in Nordiko 3600, used also for previous etch step. processing. However, this step requires different conditions during the sample For bottom pinned samples the strategy did not change, remaining the same for all the processed samples. For these, a real angle of 70 was used for etch, reaching the point just after the barrier, and then an angle of 40, going through 1/3 of the anti-ferromagnetic layer. Regarding top-pinned samples, the first approach consisted on applying Figure 3.10: Scheme of the sample after second etch. an angle of 70 until achieve the end of the free layer, and then a lower angle of 40 during 200 seconds. Both structures have the first angle almost perpendicular to the sample, aiming to define the pillar as straight as possible to produce the desired geometry. As presented in figure 3.11, the shape of the pillar depends of the angle used, although larger angles can not be used, since another undesired effects will appear, as for example ion implantation. As this method is based on sputtering of ions by impact on the sample, it is inevitable that some material is redeposited in pillar s sides, which will (truly) compromise the overall performance of the device. In order to reduce this deposited material, the second step, uses a lower angle to clean the walls of the pillar and to avoid conducting channels in parallel. For bottom pinned sample this strategy produced reliable results, however for top pinned samples, the thickness needed to be etched until the end of the free layer was quite larger, which gave rise to even more problems of redeposition. This Figure 3.11: Non-vertical profile of the pillar on 2 nd etch. study is presented further on this document, as the modifications that were set in the process (an angle of 25 was adopted after detecting the problem). 24

43 3.4 Oxide Deposition and Liftoff Oxide Deposition - UHV II The next step consists on insulating our magnetic tunnel junction with an oxide layer. As the photoresist remains on top of the pillar and the contacts, the insulating layer will be deposited over it, as illustrated in figure UHV II is a manual RF sputtering system used to deposit the Al 2 O 3 oxide that ensures the electrical isolation between two metal layers, the top and the bottom electrode of the MTJ. The machine built in INESC-MN is installed in a class clean room and consists in a single chamber able to deposit 6 inch wafers. Once it has been opened to place the samples, it can reach a pressure in the order of 10 7 Torr after 10 hours. Main conditions used to deposit in this machine are presented in table 3.6. Figure 3.12: Scheme of oxide deposition over the sample (oxide can be seen in gray color). The thickness of oxide deposited was the same for all the processed samples, a nominal value of 1000 Å. RF Power Pressure Gas Flow (Ar) Deposition Rate 200 W < (Torr) 45 sccm 12 Å/s Table 3.6: Deposition conditions of UHV II machine. Figure 3.13: UHV II machine. Deposition of Al 2 O 3. The value, of 1000 Å, chosen to the deposited thickness is related with the height of the pillar, since the oxide should cover all the borders of this structure and ensure a complete isolation between the top and bottom electrodes. If not isolated, the sensors will have low resistance, since the current will go through the electrodes directly, avoiding the pillar Oxide liftoff After depositing the oxide, it is not possible to make an electrical contact with the sensor. The photoresist is on top of the pillar, and the contacts are covered by oxide, once again the sample is placed in a ultrasound bath with microstrip. If the thickness of oxide is the expected one and the pillar is well defined, the photoresist will be detached from the sample, opening contacts for both structures. This process could take 12 hours until finished. Figure 3.14: Scheme of the sample after oxide liftoff. 25

44 An optical inspection can be performed in the microscope, with the pillars with photoresist and oxide on top, with darker borders when compared with pillars already open, figure (a) Pillar with oxide reamaining on top. Dark borders indicating oxide presence. (b) Pillar after oxide liftoff. Figure 3.15: Example of a pillar before and after oxide liftoff process. This step is also crucial to the device. If the pillars are not empty of photoresist and oxide, the sensor will work as open circuit. Since the dimensions of photoresist are in the order of micrometer, this liftoff can also be checked on the Dektak profilometer, available at INESC-MN cleanroom. 3.5 Definition of top electrode The last stage of the process is the definition of the top electrode, which allows the connection to the bottom electrode through the pillar. Lithography This lithography step differs from the first two in one detail. After coating, the sample is pre-developed during 20 seconds, without baking it. This step makes the next liftoff process easier to occur. (a) 2D mask used in DWL exposure. (b) Sample scheme after development of the 3 rd lithography. Figure 3.16: Resume of the 3 rd lithography process. The photoresist will be over all the sample, leaving only the pads of bottom electrode and the top electrode open, allowing the deposition of the metal inside. In this step the alignment should also achieve an angle lower than 0.02 mrad, and should be guided by the alignment marks defined in the second lithography. 26

45 Figure 3.17: Photo of two structures after third lithography. Resist remains outside the pattern Metallization - Nordiko 7000 Once the patterning is complete, the sample is ready to deposit the top contact. In this case, the designed structure has cavities which will be empty of photoresist, so the metal can be deposited inside. Nordiko 7000 is the machine used to perform this step. Installed in a class 100 clean room, it s composed by four different modules with different available options. The first module is ready to do flash annealings, so it is not used in this process. The second module does a sputter etch with Ar ions, and it is used to do a soft etch to clean the sample. This etch is executed with two different RF power supplies. The third module is used to deposit TiW, a passivation layer. As the deposition is done with a flow of N 2, the deposited films incorporate also this nitrogen, an so the final composition is also nitrogen dependent (TiW(N)). The fourth module will deposit the material corresponding to the top contact, made of an aluminum alloy (AlSiCu). The deposition occurs by sputtering using Ar ions. (a) Front view of Nordiko 7000, load lock acess. (b) Scheme of Nordiko 7000 modules. (c) Sample scheme after metallization. Figure 3.18: Resume of metallization process and Nordiko 7000 design. As stated before, the process of the MTJ sample only involves modules 2, 4 and 3, by order of process. The sample starts with a soft etch on module 2 to remove the natural oxide formed on top of it, followed by the deposition of 3000 Å of Al 98.5 Si 1.0 Cu 0.5. To finish the process, the sample enters in module 4 to deposit the passivation layer of TiW(N) with 150 Å, working as a protection layer for oxidation and damage. The summary of the conditions is presented in table

46 Table 3.7: Resume of operating conditions related with the metallization process on Nordiko Module Process Power Gas Flow Pressure Deposition Rate 2 Soft etch 40 W/60 W 50 sccm (Ar) 3.0 mtorr - 4 AlSiCu deposition 2 kw 50 sccm (Ar) 3.0 mtorr 37.5 Å/s 3 TiW deposition 0.5 kw 50 sccm (Ar) + 10 sccm (N2) 3.0 mtorr 5.5 Å/s Liftoff To finish the overall process, the photoresist needs to be removed leaving only the contacts and the top electrode. The sample is again placed inside an ultrasound bath with microstrip at 60 C until all the resist is removed. Pictures regarding the final sensor, once the process in completed, are presented in figure (a) Scheme of a magnetic tunnel junction after being processed. (b) Photo of one single sensor after metallization. (c) Photo of 1 inch die after complete process. Figure 3.19: Final device after complete process. 3.6 Annealing treatment The majority of developed work in this thesis was in top-pinned MTJ samples. This approach leads us to a final sensor with no need for annealing, once the orientation of magnetization of the layers is easily defined in the moment of deposition due to the growing properties of the layers in these samples. However, for bottom-pinned samples, an annealing treatment is usually needed to obtain a good exchange bias field between the antiferromagnetic and the adjacent ferromagnetic layer. The blocking temperature (T B ) of the antiferromagnetic layer (Mn 74 Ir 26 ) is about 250 C, so this is the value of maximum temperature. The setup used for this process is presented in figure 3.20(a). The sample is heated until the desired temperature inside an oven is achieved, and remains at that temperature during 30 minutes. All this process occurs at pressures between Torr. As presented further on, annealing can improve the magnetic properties of the films near the interface and promote the diffusion of the oxygen from the electrodes interfaces to the barrier, making it more uniform. At this temperature, T B, the exchange bias vanishes, causing the surface atoms of the ferromagnet to align with the surface atoms of the 28

47 antiferromagnet. Then the sample is cooled down inside a permanent magnet with 1 Tesla until it reaches the room temperature, since TR < TB. (a) Annealing Setup: Turbo Pump, Oven, Permanent Magnet (1 T/104 Oe). (b) Scheme of annealing treatment. Figure 3.20: Annealing equipment used to thermal treatment Sensor Characterization Magnetic measurements - VSM Magnetic tunnel junctions measurements on VSM (vibrating sample magnetometer) are usually done before the sample being processed. This equipment allows a magnetic characterization of the thin film deposited. As will also be presented in some results, a maximum field of 1.3 Tesla can be generated with two separated coils, having a sensitivity of 10 5 emu/cm3. A vibration unit makes the sample vibrates, and as the thin films also Figure 3.21: VSM Setup, model DSM 880. have magnetic properties they will create a magnetic field induced by the vibration, the field is then detected with another set of coils that generate the signal voltage due to the changing flux emanating from the vibrating sample [33]. This gives us the ability to see the inversion of the layers with magnetic behaviors depending of the applied field. All measurements were done at room temperature Magnetotransport Curve The magnetoresistance measurement characterizes the sensor electrically. A current goes through two coils generating a magnetic field. In both setups used the field created goes from 14 mt to +14 mt, with all the measurements done at room temperature. In the automatic setup, figure 3.22(c), the stage moves automatically, allowing the measurement of 225 pillars in less than 3 hours. The data from this setup needs to be treated afterwards. 29

48 In the manual setup, figure 3.22(a), each pillar has to be measured one by one, with someone adjusting the probes for each sensor. The advantage of this setup is related to the fact that a computer program treats all the data, allowing us to choose the right conditions for that measure in particular. (a) Manual Magnetotransport Curve Equipment. Coils (field generated 14 mt, 4/2 probes measurement, optical microscope. (b) Top: Current Source Keithley 220, Bottom Voltmeter Keithley 182 (c) Automatic Magnetotransport Curve Equipment, moving stage, coils (field generated 14 mt, 4 probes measurement, optical microscope. Figure 3.22: Magnetotransport Equipment, both setups are supplied by Kepco current sources Noise Measurements The noise setup is the last stage of the characterization. Before starting the measurements, the sample needs to be diced and wire-bonded to a chip carrier. As previously explained, there is no magnetic field applied in this measurement, being performed at 0 Tesla field. A primary shielded box, pictured in figure 3.23(a), will protect the chip carrier from external noise. The correspondent box s circuit is also presented in figure The circuit is composed by two potentiometers, where it regulates the current that crosses the sensor, with the option to add one more resistance in series for sensors with higher values of resistance, and one battery typically with 9 V. As noticed, the box is not connected to any source or to the power grid. The output coming from the primary box will be connected to an amplifier (SIM910 JFET). The amplifier is supplied by 20 batteries with 1.5 V each and a resistance of 100 MΩ is connected in parallel with the input channel. The gain can be adjusted from 1 to 500 times, being the value used in every measurements 100 times (40 db). With this gain, the amplifier has a constant noise level of 4 nv/ Hz. To perform a complete noise measurement, two sets of two data acquisitions need to be executed. The first two are made in 0-1 KHz range, and the other two from KHz, with the first one being a more detailed measurement. For each range, the sensor is measured with no current and then current is applied. With the first acquisition, it is pretended to eliminate from the data all noise related with other components on the circuit, and for this, the data needs to be treated afterwords. An analysis of the correspondent circuit, which is presented in figure 3.24, will be given. The battery supply (V1) of 9 V can be connected and disconnected from the circuit by switch (SW7). The two potentiometers (Rp1 + Rp2 = Rp) account as a voltage noise source (Vpot). The same logic is used for the sensor with V sens and for the amplifier with V amp. So, the spectrum analyzer (Vout ) is 30

49 (a) Shield box to place the sensor. (b) SIM910 JFET Preamplifier and 30V (c) Tektronix RSA 3308A Spectrum Anabattery box lyzer Figure 3.23: Noise Setup Equipment Figure 3.24: Correspondent circuit of noise setup. located as a voltage divider: Vout = RspIN V amp RampOut + RspIN V amp = G V inamp (3.1) With Vout being the value acquired by the spectrum analyzer, RspIN the input resistance of the analyzer, RampOut the output resistance of the amplifier, V amp and V inamp the output and input of the amplifier respectively, and G the selected gain. Considering that the sensor is being supplied, the noise contribution of each resistive element for V inamp, assuming a Thevenin analysis where each voltage source is replaced by a short circuit is: V inamp 2 = Ramp k Rp Rsen + Ramp k Rp 2 V sens2 + Rsen Rsen k Rp + Ramp 2 V amp2 + Rsen Rp k Ramp + Rp 2 (3.2) Taking in account that there is no current through the sensor, its noise level is only the thermal noise (Vth ), V sens = Vth = 4 kb T Rsens, so V ini=0 amp = V inamp (V sens = Vth ). These two expressions relate the two acquired measurements for each case mentioned before. Therefore, to have the output 31 V pot2.

50 coming from the sensor with different applied currents, the two measurements are joined together by the expression (3.3). ( V sens 2 Ramp Rp R sen + Ramp Rp ( V in 2 amp = (V in I=0 amp) 2 (Vth 2 V sens 2 ) ) 2 ( = V in 2 amp (V in I=0 amp) 2 ) 2 Ramp Rp, R sen + Ramp Rp ) 2 V 2 th. (3.3) Ramp Rp R sen + Ramp Rp It is assumed that the noise from the other components is constant with the current applied. 32

51 Chapter 4 Optimization of Alumina Junctions The final goal of this thesis is to have a magnetometer working with linear sensors and good TMR response, followed by a good detectivity and a low noise level. To achieve these requirements, it was necessary to continue a process of optimization, which was already started by the PhD student Simon Knudde at INESC-MN. The results presented in this chapter correspond to the optimization process performed with both bottom pinned and top pinned magnetic tunnel junctions with Alumina barrier (AlOx), an amorphous barrier. The composition of the sensing layer and reference layer is (Co 70 Fe 30 ) 10. The complete process corresponding to bottom pinned samples is presented in the following sequence: 1. Optimization of the barrier thicknesses and time of oxidation 2. Test of different free layer thickness for different barriers 3. Free layer optimization The first samples deposited with a bottom pinned stack have a free layer of 30 Å. To check the magnetic stability of the deposited stack, measurements on VSM were performed, after an annealing treatment, as presented in section 3.6, with 30 minutes at 250 C and cooling down inside the 1 T permanent magnet. The result is presented in figure 4.1. As mentioned before, this measurement can be performed on unpatterned samples, which will only give information about the magnetic behavior of the deposited stack. Applying a magnetic field in the range of -600 to 600 mt is possible to see the inversion of all magnetic layers. For high values of field, positive or negative, all layers will be aligned along the direction of the applied field. When the field starts to decrease, the first layer to invert its magnetization can be the pinned layer (PL) or the reference layer (RL), depending on which side the sample is orientated. The pinned layer is coupled with the antiferromagnet (180Å IrMn), while the reference layer is coupled with the pinned layer by RKKY coupling, and so it is not expected a symmetry for positive and negative fields in this results. For smaller values of applied field, the only sensitive layer is the free layer. The inversion of magnetization corresponding to this layer, can be seen in the inset of figure 4.1. It is centered in zero, but with a coercivity value of 1 mt. 33

52 VSM Measurements 1 Easy Axis Hard Axis Normalized Magnetization M/Ms FL: CoFeB 30A PL RL FL PL RL FL Normalized Magnetization M/Ms PL RL FL PL RL FL Magnetic Field (mt) Magnetic Field (mt) Figure 4.1: VSM measurement performed to the stack:ru 150 Å/CoFeB 30 Å/AlOx 9 Å/CoFeB 30 Å/Ru 9 Å/CoFe 22 Å/IrMn 180 Å/Ru 200 Å/Mg 4 Å/Ru 200 Å. M Sat = 1390 emu/cm 3 of the three magnetic layers summed (maximum point). FL - Free Layer, PL - Pinned Layer, RL - Reference Layer. For higher values of field, the effect of SAF will be present in on direction, and the effect of both SAF and exchange bias coupling with the antiferromagnet in the other direction. The larger plateau, between -140 mt to -1 mt, will correspond to the sum of these two effects, with the pinned layer inverting its magnetization from higher negative fields until -140 mt. On the other side, from 1 mt to 76 mt, a smaller plateau is obtained corresponding just to the RKKY coupling created by the SAF, with the inversion of magnetization of the reference layer. In summary from VSM analysis, we conclude that the stack is behaving as expected magnetically, with good couplings between the desired layers. The response is also centered in zero but with some coercivity. 4.1 Bottom Pinned MTJ Samples Optimization of the barrier parameters In the first set of samples, an optimization of the barrier thickness and oxidation time was performed. The barrier was deposited and oxidized with the parameters presented before in section 3.1 The complete set of samples deposited is presented in table

53 Table 4.1: Samples deposited with different barrier thickness and oxidation time. Sample ID RP24 RP25 RP26 RP27 RP28 RP29 RP30 Layer Composition & Layer Thickness (Å) Ru CoFeB AlOx 7Å + 10s 7Å + 20s 7Å + 30s 9Å + 10s 9Å + 20s 9Å + 30s 9Å + 40s CoFeB Ru CoFe IrMn Ru Mg Ru The thickness of the Al layer varies from 7 Å to 9 Å and the time of oxidation from 10 to 40 seconds. The sample RP27 had peeling problems and so the corresponding results are not presented in figure 4.2. TMR (%) Sample RP24 Pilar Size: 2x50 µm 2 Sample RP25 Pilar Size: 1x50 µm 2 Sample RP26 Pilar Size: 2x50 µm 2 Sample RP28 Pilar Size: 2x50 µm 2 Sample RP29 Pilar Size: 1.5x50 µm 2 Sample RP30 Pilar Size: 1.5x50 µm 2 Sample ID TMR R A H c H f (%) (kω.µm 2 ) (mt) mt RP RP RP RP RP RP Magnetic Field (mt) Figure 4.2: Results of the barrier optimization process with annealing. Table 4.2: Resume of the results obtained for barrier optimization. Values corresponding to one pillar per sample. Change the oxidation time of a barrier can lead to three different results: under oxidation, over oxidation, or complete oxidation. In a case of under oxidation the resistance will decrease, since the Al is a conductor material and there will be a loss in TMR due to the non-polarizing effect of Al. Over oxidation can contribute to damage the layer after the barrier, in this case the reference layer, that will lose again the polarizing effect and contribute to loss in TMR. As follows, the point that is pretended to achieve is complete oxidation, the compromise between the previous two. For a 7 Å thick barrier, the best result was achieved for sample RP25 with a time of oxidation of 20 seconds. Sample RP24 has a low TMR value and low resistance, which affects the RA product, so it is under oxidized, and sample RP26 has a higher RA product but with loss in TMR, meaning it is over oxidized. 35

54 For a 9 Å barrier, both samples RP28 (20s) and RP29 (30s) gave good results, when compared with the previous case for 7 Åbarrier. As they have a thicker barrier, it takes more time to achieve a point of over oxidation as in the case of sample RP30 (40s). It can also be noticed that from the 7 Å samples to 9 Å there is a general decrease in H f, representing the shift of the curve for one side as explained before, due to the decrease on the fringe field originated in the layers below the barrier, since there is also a RKKY and static field coupling between these layers, with the spacer being the barrier Thicker free layer in different barriers The previous results show a non-linear response of all MTJs deposited. As explained before in section 2.1.4, one of the parameters that directly influence the strength of the demagnetizing field of the free layer is its thickness, as well as its magnetization, since it affects the crystalline anisotropy. Based on this fact, the free layer thickness was increased to 120 Å to assure the condition of H k < N xx Ms f. The thickness of the barrier when increased, leads to an exponential increase of the junction resistance, since the tunneling probability also decreases exponentially [25], and so, sensors with higher values of resistance were processed. It s important to notice that the oxidation time was already optimized for 7 Å thickness. The deposition of these barriers occurs in two different steps, first a 7 Å layer is deposited and oxidized, and then another layer of 7 Å (RP66) or 5 Å (RP67) is deposited and again oxidized. In the case of 5 Å, the time of oxidation will not damage the layer below since it s precisely the layer that was already oxidized. Relatively to SAF, from the VSM results, it is possible to conclude that there was a good coupling between the desired layers, and so the thickness of Ru was also decreased for 6Å with an increase of the pinned layer (CoFe) to 30 Å and to 40 Å on the reference layer (CoFe)B. Table 4.3: Summary of the stacks from RP66 and RP67 samples. RP66 Sample ID RP67 Layer Composition & Layer Thickness (Å) Ru CoFeB AlOx 7Å + 20s + 7Å + 20s 7Å + 20s + 5Å + 20s CoFeB Ru 6 6 CoFe IrMn Ru The results are presented in figure 4.3 and resume on table 4.4. Table 4.4: Resume of the results of samples RP66 and RP67 with different free layers. Resistance (Ω) R A (kω.µm 2 ) H f (mt) H c (mt) TMR (%) Pillar Size/Sample RP66 RP67 RP66 RP67 RP66 RP67 RP66 RP67 RP66 RP µm µm µm

55 30 Sample RP66 Pilar Size: 1x50 µm 2 Rmin:1.19E+3 Ω Hf: 1.43 mt Hc: 0.11 mt 30 Sample RP67 Pilar Size: 1x50 µm 2 Rmin:8.22E+2 Ω Hf: 1.27 mt Hc: 0.18 mt 25 Pilar Size: 1.5x50 µm 2 Rmin:8.33E+2 Ω Hf: 1.18 mt Hc: 0.23 mt 25 Pilar Size: 1.5x50 µm 2 Rmin:5.11E+2 Ω Hf: 0.93 mt Hc: 0.15 mt 20 Pilar Size: 2x50 µm 2 Rmin:6.63E+2 Ω Hf: 1.13 mt Hc: 0.40 mt 20 Pilar Size: 2x50 µm 2 Rmin:3.61E+2 Ω Hf: 0.92 mt Hc: 0.23 mt TMR (%) 15 TMR (%) 15 Pilar Size: 3x50 µm 2 Rmin:2.44E+2 Ω Hf: 0.52 mt Hc: 0.70 mt Magnetic Field (mt) Magnetic Field (mt) Figure 4.3: Magnetoresistance results from sample RP66 and RP67 with annealing. The TMR values are in accordance with the values obtained before in figure 4.2. For larger pillar dimensions the results show an increase in TMR, nevertheless the value of R A is kept constant, table 4.4, and so the variations of TMR only appear due to the fact that the sensors do not achieve saturation for the range of applied field. As expected, for the thicker barrier (RP66), higher values of resistance were achieved. The values of H f are also in accordance with the results obtained before, and we can also notice a decrease for larger pillar s areas, since the demagnetizing field created by the pinned layer and which affects the free layer is also decreasing. The values of coercivity are also increasing for larger widths, since this leads to a decrease in the relative strength of the demagnetizing field of the free layer, and so contributing to a square response Free-layer optimization In the last series of samples presented in table 4.3, it was studied the behavior of the sensor response for different free layers thickness. The time of oxidation was also reduced to 15 seconds based on results from other samples processed apart this work. The results are presented in figure 4.4. The values of TMR are normalized to the maximum value of each pillar, since the sample RP85 had a low resistance and so, lower TMR, probably due redeposition problems, keeping the magnetic behavior. On the inset of the figure a zoom of the saturation points is presented. For a free layer of 70 Å, a square response still appears, converging to the results obtained for a 30 Å thicker layer, in the beginning of the overall process of optimization. For a free layer of Å, the result is a linear response, as expected by the results of figure 4.3, which confirms the increasing of linearity with the increasing thickness of the FL. 37

56 Table 4.5: Sample s stacks for free layer optimization. Sample ID RP82 RP83 RP84 RP85 RP86 Layer Composition & Layer Thickness (Å) Ru CoFeB AlOx 2 (7Å + 15s) CoFeB Ru CoFe IrMn Ru TMR/TMR max TMR/TMR max Magnetic Field (mt) Magnetic Field (mt) Sample RP82 FL= 70 Å Sample RP83 FL= 90 Å Sample RP84 FL= 110 Å Sample RP85 FL= 120 Å Sample RP86 FL= 130 Å Sample TMR H c H f R A (%) (mt) (mt) (kω.µm 2 ) RP RP RP RP RP Figure 4.4: Result of different free layer s thickness for pillars with 1 50 µm 2, with annealing. Table 4.6: Summary of the results for free layer optimization. Test of different free layer s thickness. As mentioned in section 2.1.4, the demagnetizing field can be given by: H f d = 4πM sat t f /h. If this field is larger enough, meaning that its contribution is larger than difference between H p d and H n as presented in figure 2.6, it can be assumed that H sat H d, with H sat the saturation field (this assumption is valid for an infinite shape approach, meaning that the width of the pillar is much bigger than the length). This is the effect observed in figure 4.4, the increase of the free layer thickness t, leads to an increase in saturation field, tuning the sensitivity of the sensors. 38

57 4.2 Top Pinned MTJ samples Top pinned MTJs have the advantage of usually not requiring annealing treatment, since the direction of magnetization of the pinned layer may be set during the growth of the crystal in the deposition, thanks to the different order of deposition relatively to bottom pinned samples. This only happens in the case of a barrier of AlOx, since in the case of MgO this is not possible, and also for IrMn. The best results were also achieved with this kind of structures. The corresponding optimization process consisted: 1. Deposition of test samples 2. Thicker free layer for different barriers 3. Free layer optimization 4. Incorporation of NiFe in the free layer Deposition of test samples The first set of samples deposited relied on the structure of bottom pinned MTJs, with the corresponding order for the layers of a top pinned sensor. The purpose of this deposition was to have a initial start point to work, so only the thickness of Ru, the spacer related with RKKY coupling, was changed. Table 4.7: Deposition of top pinned test samples. Sample ID Layer RP31 RP32 Layer Composition & Layer Thickness (Å) Ru IrMn CoFe Ru 9 8 CoFeB AlOx 9Å + 20s 9Å + 20s CoFeB Ru Mg 3 3 Ru Mg 3 3 Ru From the deposition conditions, it is not expected a big difference in the magnetoresistance results presented in figure 4.5. Although the results of these samples have good value of TMR, the overall yield of sensors working with a good signal in each sample was quite low. The resistance values obtained are not in accordance with the values obtained for the same barrier in bottom pinned samples, indicating possible problems of redeposition in the majority of the pillars (a redeposition study is presented in section 4.3, as well as the TMR vs R A graphic), since this effect is characterized by lower values of resistance and a decrease in TMR. 39

58 To achieve linear responses, an optimization of the free layer thickness was done, with an increase of the barrier thickness to have higher values of resistance and reducing the H f parameter. 35 TMR (%) Sample R31 Pilar Size: 1.5x50 µm 2 Sample RP31 Pilar Size: 3x50 µm 2 Sample RP32 Pilar Size: 1.5x50 µm 2 Sample RP32 Pilar Size: 3x50 µm Sample Pilar Size TMR R A H c H f ID (µm 2 ) (%) (Ω.µm 2 ) (mt ) (mt ) RP31 RP32 1.5x x x x Magnetic Field (mt) Figure 4.5: Results of top pinned samples RP31 and RP32. Table 4.8: Summary of the results for test top pinned samples, RP31 and RP32. Change of the spacer s thickness Thicker free layer in different barriers Following the strategy used for bottom pinned samples, a thicker free layer was tested in two different barriers, with new parameters also for pinned and reference layer. Table 4.9: Stack of top pinned samples RP64 and RP65. Test of different barriers. RP64 Sample ID RP65 Layer Composition & Layer Thickness (Å) Ru IrMn CoFe Ru 6 6 CoFeB AlOx 2 (7Å + 20s) 7Å + 20s + 5Å + 20s CoFeB Ru Mg 3 3 Ru Mg 3 3 Ru VSM measurements were performed to the sample RP64 to verify the magnetic behavior of the magnetic layers, presented in figure

59 Normalized Magnetization M/Ms Easy Axis Hard Axis FL: CoFeB 120A PL RL FL VSM Measurements PL RL FL Magnetic Field (mt) Magnetic Field (mt) Normalized Magnetization M/Ms PL RL FL PL RL FL Figure 4.6: VSM measurement to top pinned sample RP64. M Sat = 1400 emu/cm 3. The result of figure 4.6 can be compared with the previous results for bottom pinned MTJs with annealing, although in this case a thicker free layer was deposited, which directly influences the magnitude of its magnetization. While before the pinned layer remained fixed for fields below -140 mt, in a top pinned configuration without annealing this field was reduced to -74 mt. The RKKY interaction between the pinned layer and the reference layer, also suffered a reduction from 76 mt to 48 mt. The coercivity value in this case is lower, following the results obtained before for thicker free layers, resulting from a lower magneto crystalline anisotropy [26]. In this case, a shift of the curve is observed of about 1.5 mt, since the texture of the layers changed, caused by the order of deposition, the deposited materials will have different roughness, giving rise to a different Neél coupling. From the results obtained before, a linear behavior was expected for these samples. Again, this set of samples had problems of redeposition, with a very low yield, and so the results presented in figure 4.7 do not belong to the same size of pillar in each sample. From the results obtained, we can see that two regimes are present in the magnetoresistance behavior. In both samples the sensor is almost linear for fields far from the zero, but near zero there is a suddenly change/jump. This can be explained from the fact that, in this case there was no annealing treatment and so, we have an amorphous free layer of CoFeB. After annealing, the Boron will migrate and the free layer crystallizes. As we have a thicker free layer it will break in magnetic domains, and so the layer will invert its magnetization in two different phases, giving rise to the result obtained. On the other hand, we should also consider that the pillar is defined on the second etch, which is a physical method by ion beam sputtering, and in this case the free layer is one of the last layers to be etched, which can originate shadow effects from the photoresist, leading to a non-vertical profile of the pillar. Later on, in chapter 4.5, it is presented the result of a sensor totally linear after annealing. 41

60 Sample R64 Pilar Size: 2x30 µm 2 Hf: 0.44 mt Hc: 0.19 mt Sample RP65 Pilar Size: 3x30 µm 2 Hf: 0.99 mt Hc: 0.43 mt TMR (%) Magnetic Field (mt) Figure 4.7: Magnetoresistance results for top pinned samples RP64 and RP Free layer optimization with different etch conditions Due to the redeposition problems that affected the top pinned samples processed before, a new strategy for second etch was used. As before, the process starts by an etch at 70 (real angle) until 20Å after the end of the free layer, assuring that the increase of the barrier thickness after oxidation is covered and that the free layer is totally etched with the same profile until its end. The next step consists in a etch with a low angle of 25 during 500 seconds. With this second etch, it is pretended to clean the pillar s sides only, since the etch rate on the vertical in this case is almost negligible. Table 4.10: Deposited stacks for top pinned samples RP79 and RP80 Sample ID RP79 RP80 Layer Composition & Layer Thickness (Å) Ru IrMn CoFe Ru 6 6 CoFeB AlOx 2 (7Å + 15s) 2 (7Å + 15s) CoFeB Ru Mg 3 3 Ru Mg 3 3 Ru With the conditions presented before, the amount of pillars with TMR larger than 29%, increased from 32% to 74% and the values of resistance are now according with the expected values. The result is 42

61 presented in figure 4.8. As seen before, thicker free layers (sample RP80) lead to higher demagnetizing 35 TMR (%) Sample RP79 Pilar Size: 1x50 µm 2 Rmin:2.78E+2 Ω Hf: 1.71 mt Hc: 0.29 mt Sample RP80 Pilar Size: 1x50 µm 2 Rmin:2.95E+2 Ω Hf: 1.35 mt Hc: 0.05 mt Magnetic Field (mt) Figure 4.8: Magnetoresistance results for top pinned sample RP79 and RP80. fields and so larger saturation fields with a loss on sensitivity, with sample RP80 with a sensitivity of 1.94%/mT, while sample RP79 has a sensitivity of 2.71%/mT Incorporation of NiFe on free layer. The next step of this optimization processes was the incorporation of NiFe in the free layer. This material is classified as a soft magnetic material, meaning that exhibits lower coercivities when compared with (CoFe)B. The material in the interfaces of the barrier remains the same, and so a decrease in TMR is not expected, although these two materials have different polarizations [34], the difference is negligible. The barrier used remains the same, a two-step oxidation barrier with 14 Å of deposited Al. The different stacks deposited are presented in table Due to the available targets in Nordiko 3000, the deposition of the complete stack was done in two machines. All layers until the N if e layer, were deposited in Nordiko Over the thickness indicated with *, a 50 Å layer of Tantalum was deposited to protect the NiFe layer from oxidation. After this deposition on Nordiko 3600, the sample was moved to Nordiko 3000 where was exposed to an etch process to remove this last layer, first a vertical etch with an angle of 70 during 60 seconds and then 20 more seconds with different conditions as presented in table The different etch conditions will influence the roughness on top of the NiFe layer, which can result or not, on a favorable coupling between the NiFe and (CoFe)B layer. The results are presented in figure 4.9. From the first two samples, RP125 and RP126, the behavior of the magnetoresistance curve represents the same response as seen before for thinner (CoFe)B free layers, meaning that the coupling 43

62 Table 4.11: Free layer optimization with NiFe. The values indicated with * correspond to the deposited value, not to the real value after the complete deposition since there is an etch process in between. Sample ID RP125 RP126 RP127 RP128 RP129 RP130 RP131 Layer Composition & Layer Thickness (Å) Ru IrMn CoFe Ru CoFeB AlOx 2 (7Å + 15s) CoFeB NiFe 40* 40* 60* 80* 100* 120* 120* Ru Ta Ru Ta Table 4.12: Etch conditions used on Nordiko Sample ID Etch Current (ma) Grid V + (V) Grid V (V) Etch Angle ( ) RP RP RP RP RP RP RP between the two ferromagnetic layers was not achieved. In the other samples, it is possible to notice a good improvement on the results. Lower values of coercivity were obtained, with the sample RP127 achieving a good TMR value and sensitivity, as can be seen in figure 4.9. While the samples composed by free layers with only (CoFe)B could easily achieve values of TMR in the order of 35%, in this case the maximum TMR was about 29%. The roughness induced by the incorporation of NiFe could lead to this decrease in TMR, since the difference in polarizations is not strong enough to explain this change, although in this case the thickness of CoFeB was reduced. Samples RP130 and RP131 are composed by the thicker free layers, showing an increase in coercivity. This result is also expected since there is an optimal point that relates the thickness of the layer and its coercivity, figure 4.10, for larger thickness, larger magnetic domains start to appear, the magnetization can have components out of the plane, and so increasing the coercivity value. It can also be noticed that for etch conditions with an angle of 70, the parameter H f is always smaller. Still, the samples etched with these conditions always present worse results than the samples etched with 50. Sample RP128 had low TMR, and sample RP130 although the results presented in table 4.13 are good, from the magnetoresistance response in figure 4.9 it is easily seen that do not have the desired behavior for a sensor. From figure 4.10, it is also possible to see the variation of sensitivity, decreasing with the increase of thickness. In resume, the use of a soft etch with a lower angle is more adequate in this case. 44

63 Sample RP125 Sample RP126 Sample RP127 Sample RP128 Sample RP129 Sample RP130 Sample RP TMR/TMR max Magnetic Field (mt) Sample TMR (%) H c (mt) H f (mt) R min (Ω) RP RP RP RP RP RP RP Figure 4.9: Results of top pinned samples for NiFe and (CoFe)B free layers. Pillar Size: µm 2. Table 4.13: Resume of the results for free layer optimization with NiFe. 0.3 Analysis of Coercivity and Sensitivity Sensitivity Coercivity 3 Sensitivity (%/mt) Hc (mt) 0 RP125 RP126 RP127 RP128 RP129 RP130 0 RP131 Sample Figure 4.10: Analysis of coercivity and sensitivity for all samples. The sensitivity values are normalized by the maximum TMR. Sensitivity values correspond only to linear sensors: RP127, RP128, RP129, RP131. Pillar size: µm Study of Redeposition in Top-Pinned Samples During the optimization process for top pinned samples, problems related with redeposition of material were detected. This effect consists in material that gets attached to the pillar s side, which will give raise to a low resistance sensor, since the material deposited in the sides will work as conducting channels in parallel with the MTJ pillar. The redeposition effect is more accentuated in top pinned MTJs, due to the fact that the second etch 45

64 is longer when compared with bottom pinned samples, and so more material will be projected. To solve this problem, the strategy already presented in section was adopted, consisting in cleaning the sides of the pillar during 500 seconds at an angle of 25. The advantage of this strategy is presented in figure Redeposition Study % of Sensors with TMR >29% -RP79: 74% -RP31: 32% Sample: RP79 Sample: RP31 30 TMR (%) Maximum RxA: RP79: 2.07x10 5 Ω.µm 2 RP31: 9.27x10 2 Ω.µm RxA (normalized) Figure 4.11: Redeposition study of samples RP31 and RP79. The result of the sample RP31 is characteristic of redeposition, with a lower value of TMR followed by a decrease in resistance. With this new strategy, the amount of pillars with TMR larger than 29%, was improved from 32% to 74%. 4.4 Study of IV Curves and TMR bias dependence One of the characteristics that makes an MTJ sensor different from a spin valve or other GMR sensors, it is their non-linear response between the current applied to the sensor and its output voltage. Also the TMR will depend of the bias used. These measurements were performed on a sensor with the stack of sample RP129 (CoFeB and NiFe free layer) with a deposited thickness of Al of 14 Å, and with R A = kω.µm 2 From the figure 4.12, it can be seen that the variation of TMR is not symmetric for both positive and negative currents. This effect can be explained from the fact that the electrodes of the sensor are also asymmetric, caused by defects on the oxidation step. If the barrier is under oxidized, there will be a thin layer of Al that will contribute for a different polarization and so the result is not symmetric as expected. The TMR is reduced to half of its nominal value at a bias voltage of 0.37 V. The TMR dependence is not completely understood, but it can be explained by two different ways. Using a Slonczewski parabolic model, a decrease in TMR is obtained simply because the voltage across 46

65 25 Magnetoresistance vs Current 20 Magnetoresistance (%) Pillar size: 2x30 µm Current Applied (10-6 A) Figure 4.12: TMR dependence of the applied bias current. Sensor s R A = kω.µm 2. Figure 4.13: Resistance and TMR of an MTJ sensor for different applied currents. Sensor s R A = kω.µm IV Curve Antiparallel State Parallel State Current Applied (A) Sensor Output (V) Figure 4.14: IV curve for an MTJ sensor in parallel and anti-parallel states. Sensor s R A = kω.µm 2. the barrier shifts the bands of the electrode into which the electrons tunnel downward towards higher energy states. The second purposed cause, relies on spin flip events induced by inelastic scattering of magnons. The tunneling electrons that arrive at the second ferromagnet, as hot electrons, with higher energy than the Fermi level of this electrode, may lose their energy by magnon emission, and then flipping the electron spin. When a bias current/voltage increase, this effect will also increase. In figure 4.13, it is possible to see that with the increase of bias current, the difference between the two states anti-parallel and parallel is decreasing (TMR decrease), and the resistance of the sensor also depends of the bias used, for the same reasons explained before. The assymetry of this effect can also be noticed by the difference between the resistance values of negative (lines) and positive values (dots). The last figure shows the non-linearity between the current applied and the voltage, for both states 47

66 parallel and anti-parallel. To perform this measurement, the sensor is inside a magnetic field of +10 mt for one state, and -10 mt for another, ensuring that both states parallel and anti-parallel are reached. 48

67 4.5 Noise Study The last characterization of the sensors is related with noise. One stack with just CoFeB in the free layer and another with CoFeB + NiFe were tested before and after annealing. These two stacks were chosen taking in account the previous results, with a free layer of 115 Å for the first case, and the stack of sample RP129 in the second case. The first result is presented in figure All sensors were measured with more than one current value, to check the confidence of results. The Hooge parameter is determined by fitting the data to the equation Noise Measurements Resistance (Ω) MR: % R min : Ω Pilar Size: 2 50 µm 2 R A= 250 kω.µm 2 Hf: 0.41 mt Hc: 0.48 mt 0 mt: I bias =15.27 µa Sens: V/mT I bias =8.97 µa Sens: V/mT Ru 150 A IrMn 180 A CoFe 30 A Ru 6 A CoFeB 40A AlOx 14 A CoFeB 115 A Ru 200 A Mg 3 A Ru 200 A Mg 3 A Noise Level (V/ Hz) I= A I= A Thermal Noise Detectivity (mt/ Hz) Detectivity Frequency(Hz) Magnetic Field (mt) Figure 4.15: Magnetoresistance response of the sensor before annealing Frequency (Hz) Figure 4.16: Noise and detectivity results for CoF e free layer sensor before annealing. For this sensor, the Hooge parameter, α H, was not determined, by the reason that this sensor shows a evident response of Random Telegraph Noise, with a step at low frequencies, originated by different magnetic domains on the free layer a 0 field. At high frequencies, the sensor tends to thermal noise as expected, since the other components (RTN, 1/f) are characteristic from low frequencies. The detectivity determined is: 65 nt/ Hz at 30 Hz, and 9.18 nt/ Hz at 10 KHz. After annealing and as mentioned before, the crystal structure of the free layer will be improved, working as a single domain magnetic layer. In this case, the sensor has a complete linear response, although losing sensitivity. The R A parameter increased due to the uniformity of the barrier achieved, after the migration of oxygen at higher temperatures. For each sensor, there is only one point where it has the maximum value of detectivity, in this case, although a higher value of current gives rise to a higher noise level, the detectivity will be also higher, thanks to the increase of sensitivity of the sensor. The detectivity determined is: 120 nt/ Hz at 30Hz, and 20 nt/ Hz at 10 khz. The Hooge parameter is usually used to compare different sensors in terms of noise level, more precisely the 1/f component. The noise level is expected to change with the type of barrier and the ferromagnetic materials of the MTJ. As the barrier is always the same, the results almost do not change. With the incorporation of NiFe on the free layer, corresponding to the sensor presented on figure 4.20, it has a stable and well coupled layer. Consequently, the effect that appeared before in figure 4.16, related 49

68 Noise Measurements 7200 I= A αh = µm2 I= A αh = µm2 Thermal Noise MR: % Pilar Size: 2 40 µm 2 Resistance (Ω) R A= 451 kω.µm Hf: 1.88 mt Hc: 0.03 mt mt: I =0.85 µ A 6200 Sens: bias -5 V/mT I bias =8.19 µ A 6000 Sens: V/mT Ru 150 A IrMn 180 A CoFe 30 A Ru 6 A CoFeB 40A AlOx 14 A CoFeB 115 A Ru 200 A Mg 3 A Ru 200 A Mg 3 A Detectivity (mt/ Hz) 6800 Rmin : Ω Noise Level (V/ Hz) Magnetic Field (mt) Noise Measurements Hc: 0.02 mt 0 mt: I bias =15.27 µa Sens: V/mT I bias =8.08 µ A Sens: V/mT Magnetic Field (mt) Ru 150 A IrMn 180 A CoFe 30 A Ru 6 A CoFeB 40A AlOx 14 A CoFeB 30 A NiFe 100 A* Ru 150 A Ta 50 A Ru 150 A Ta 50 A 10 Detectivity Frequency(Hz) Figure 4.19: Magnetoresistance response of the sensor before annealing. Detectivity (mt/ Hz) Hf: 3.71 mt 10-7 R A= 216 kω.µm 2 Noise Level (V/ Hz) Rmin : Ω Pilar Size: 2 40 µm 2 Resistance (Ω) 10 3 Frequency (Hz) I= A αh = µm2 I= A αh = µm2 Thermal Noise MR: % Figure 4.18: Noise and detectivity results for CoF e free layer sensor after annealing Frequency(Hz) 10 Figure 4.17: Magnetoresistance response of the sensor after annealing Detectivity Frequency (Hz) Figure 4.20: Noise and detectivity results for CoF e + N if e free layer sensor before annealing. with RTN, is now negligible. The annealing treatment does not change the result obtained for the Hooge parameter, although the noise level increases due to the higher resistance value obtained after annealing. Same results were previous obtained in literature, reporting that Hooge parameter remains constant for annealing temperatures and bias voltage. It is also reported that, at low frequencies annealing can improve the stability of the magnetic layers, reducing the noise level, when comparing the result of figure 4.16 with results after annealing [35 37]. A summary of the results obtained is presented in table 4.14, as a graphic with data from different sources and with different barriers in figure The value obtained for these sensors are in agreement with previous results, showing an increase when compared to other AlOx barriers measured in saturation 50

69 Noise Measurements Resistance (Ω) MR: % R min : Ω Pilar Size: 2 40 µm 2 R A= 472 kω.µm 2 Hf: 5.08 mt Hc: 0.05 mt 0 mt: I bias =15.2 µa Sens: V/mT I bias =8.06 µa Sens: V/mT Ru 150 A IrMn 180 A CoFe 30 A Ru 6 A CoFeB 40A AlOx 14 A CoFeB 30 A NiFe 100 A* Ru 150 A Ta 50 A Ru 150 A Ta 50 A Noise Level (V/ Hz) I= A α H = µm 2 I= A α H = µm 2 Thermal Noise Detectivity (mt/ Hz 10-3 Detectivity * Frequency(Hz) Magnetic Field (mt) Frequency (Hz) Figure 4.21: Magnetoresistance response of the sensor after annealing. Figure 4.22: Noise and detectivity results for CoFe + NiFe free layer sensor after annealing. points, where the magnetic contribution to noise is smaller. The result of this work is presented as a blue point in the figure. Table 4.14: Resume of noise measurements for both types of free layer before and after annealing, measured at 0 Tesla field. Free Layer Annealing Detectivity (nt/ Hz) Hooge (10 9 µm 2 ) CoFe No CoFe Yes CoFe+NiFe No CoFe+NiFe Yes

70 Figure 4.23: Hooge parameter for MTJ sensors as a function of R A. Comparison between MgO and AlOx barrier structures. Adapted from [38]. AlOx-MTJ data obtained with devices at saturation from [39 42]. MgO-MTJ data obtained at linear operation range. Data marked with & from [38]; * from [43]; α from [44]; β from [45]; δ is unpublished data from INESC-MN; σ from [46]; γ from [47]; i from [39][48][49] and ii is unpublished data from INESC-MN. The result achieved in this thesis is presented as a blue square. 52

71 Chapter 5 Magnetometer Characterization and Results 5.1 Bridge assembly and characterization The purpose of this work was to build a magnetometer capable of measuring magnetic fields in three dimensions. With optimized sensors, characterized by a large linear range, it is now possible to assemble them in bridge configuration, leading to a device free of offset and with a differential output, avoiding noise contributions from common sources to all sensors. To use sensors assembled in bridge configuration, it is mandatory that they have the same values of resistance. To achieve this goal, 4 inch wafers with 16 dies were processed, with the stack indicated in figure 5.1, with each died being a square of 1 inch side, as presented in figure 5.2. With this strategy, it is possible to have more uniform sensors, characterized by the same values of R A and TMR (28%). Ru 150Å IrMn 180 Å CoFe 30 Å Ru 6 Å CoFeB 40 Å AlOx 14 Å CoFeB 115 Å Ru 200 Å Mg 3 Å Ru 200 Å Mg 3 Å Ru 200 Å Figure 5.1: Sensor s stack chosen to bridge magnetometer. Barrier of AlOx made by two step process. Figure 5.2: One inch die with 225 sensors. Dashed lines representing the dicing path. These dies were diced in the dicing machine Disco DAD-321, following the paths represented as dashed lines in figure 5.2, resulting in groups of six available sensors. From each group of sensors, only one sensor is used in the following processes. 53

72 Figure 5.3: Full bridge configuration. Sensors S1 and S2 orientated with opposite sensitive directions to sensors S3 and S4. To achieve a bridge response, the sensors need to be assembled with different directions, as presented in figure 5.3, with sensors indicated as S1 and S2 changing their resistance in the opposite way of sensors S3 and S4. The sensors are connected to a PCB through a wire bonding process, figure 5.4, that connects two of the four pads of the sensor to the external device. The PCB used is presented in figure 5.5(a), with the correspondent schematic of connections in figure 5.5(b). This PCB has four connections, with V+ and V- being the output of the bridge, and OUT+ and OUT- the input. It was fabricated to INESC-MN, being present in previous projects as a test device. The final device of this thesis is mainly limited by the dimensions of this structure. Figure 5.4: Sensor with 2 pads wire bonded to a PCB. OUT+ V+ V- OUT- V+ S1 S3 S2 S4 = S1 S3 OUT+ OUT- S4 S2 OUT+ V- V+ OUT- V- (a) PCB used to assemble sensors in bridge configuration. (b) Schematic of the PCB and equivalent bridge circuit. S1, S2, S3 and S4 represent the four sensors connected to the PCB. Figure 5.5: Overview of the PCB and correspondent schematic. Four sensors were selected and mounted with the required configuration. The individual magnetoresistance response of each sensor is presented in figure 5.6(a), for a current bias of 6 µa. All sensors have the same pillar size of 2 50 µm 2. Since the magnetic tunnel junctions do not have a linear IV response, the bridge was measured at 54

73 Resistance (Ω) Sensor 1: H c = 0.14 mt H f = 1.99 mt R min = 2.75 KΩ Sensor 2: H c = 0.09 mt H f = 2.16 mt R min = 2.75 KΩ Sensor 3: H c = 0.14 mt H f = 2.24 mt R min = 2.79 KΩ Sensor 4: H c = 0.11 mt H f = 2.10 mt R min = 2.77 KΩ Pillar Size: 2 50 µm 2 I Bias = 6 µa Sensor 1 Sensor 2 Sensor 3 Sensor Magnetic Field (mt) Voltage (mv) I=10 µ I=100 µ I=200 µ I=300 µ I=400 µ I=500 µ I=800 µ V Bridge (mv) Applied Current (µa) Magnetic Field (mt) (a) Output of the four sensors that compose the bridge after annealing. (b) Bridge output for different current values. Determination of the optimal point of operation. Voltage (mv) Bridge Output (mv) I bias = 0.4 ma V = mv min V max = 55.4 mv mt = -6.3 mv P = 2.02 mw Power Sens = 5.60 mv/mt H c = 0.16 mt H f = 0.09 mt Pillar Size: 2 50 µm 2 R: R min - R max R 1 : KΩ R 2 : KΩ R 3 : KΩ R 4 : KΩ Magnetic Field (mt) (c) Bridge output for an applied current of 0.4 ma and overall characterization. (d) Bridge measurement with 4 probes. Figure 5.6: Overview of the process and sensors composing the bridge. Magnetoresistance curves characterizing each component. different applied currents to find the point where its variation of output was maximum, as presented in figure 5.6(b). The maximum output corresponds to an applied current of 400 µa, as can be seen on the inset of figure 5.6(b), with a total variation of 122 mv from -14 mt to +14 mt. This result meets the results showed before for single sensors, since from the determined IV curves, a single sensor has its TMR decreased for half of its maximum value at 37 mv, in this case assuming that the current is equally distributed for each branch of the bridge, the current that crosses one sensor will be 200 µa for a total bias of 400 µa, and 150 µa for a total bias of 300 µa, leading to a potential of mv on one sensor, where the variation of TMR starts to be too small and so the overall output. The bridge has an offset of -6.3 mv, corresponding to 5% of the total variation, originated by the difference in resistance between the four sensors. The power consumption is 2.02 mw for a bias of 400 µa. The sensitivity increased from 10 4 V/mT for single sensors, to 10 3 V/mT. 55

74 5.2 Magnetometer output for 3 axis To measure the three dimensions of space, three different PCBs in bridge configuration were combined. The schematic related with the position of the sensors is presented in figure 5.7. One bridge was placed perpendicular to the other two, measuring fields in the vertical direction. The other two sensors are placed with a difference of 90 in the same plane, measuring the X and Y directions. Once connected, the device has two inputs of current, with all bridges being supplied with the same value since they are arranged in series, and 6 outputs corresponding to each of V+ and V- of the 3 bridges, as presented in figure 5.8. Figure 5.7: Schematic of the correspondent position for each bridge. Figure 5.8: Electrical design of the magnetometer. 3 bridges assembled in series. The device was incorporated inside a box, designed in AutoCAD and created/printed in the 3D printer of INESC-MN, a Witbox printer from BQ. In figure 5.9(a) is presented the AutoCAD design, as well as the dimensions of the device, in figure 5.9(b) the device s inside and connections as presented in figure 5.7, with the 2 inputs and 6 outputs mentioned before, and in figure 5.9(c) the front view of the magnetometer. (a) Autocad design of the magnetometer. Dimensions: 3.2x3.8x3.8cm. (b) Side view of the final magnetometer. (c) Front view of the final magnetometer. Figure 5.9: Magnetometer final device. Side and front views with Autocad design. When compared with another developed magnetometers, as the case of 2D-axis Magnetometer, from R. Ferreira et. al [50], this magnetometer has higher values of power consumption and less sen- 56

75 sitivity, but with the advantage of do not have coercivity and have a higher linear range, allowing the measurement of larger fields without having problems in case of saturation, since it s an hysteresis-free device. The coercivity of the device is about 0.16 mt, less than half of the value from R. Ferreira et. al with 0.42 mt. The power consumption of device when biased with the same value of current, 0.3 ma, is about 1.51 mw, relatively larger when compared to the other case, with only 0.3 mw. The offset parameter of this bridge is another advantage, being just 5% when compared with the 10% of R. Ferreira et. al. Figure 5.10: Bridge being tested measuring the magnetic field of a permanent magnet. The magnetometer was tested measuring the magnetic field created by a permanent magnet, that was placed at a constant distance of 5 centimeters from the device, figure At this distance, the field created by the permanent magnet will never cause the saturation of the sensors, ensuring that the measurement is performed in the linear range. Then, the angle between the magnet and the magnetometer was changed, following a circumference. The first measurement was performed in just one axis, with a sinusoidal behavior. This result is presented in figure After the data being collected, it needs to be treated, correcting the offset value of the bridge. Once corrected, the maximum and minimum value of the sinusoidal curve should have the same module, a result that was not achieved, probably due to bad placement of the sensor. When measured in two axis, the module of the field should be constant, with a circumference as result. This result was achieved, measuring a magnetic field of 1.05 mt with an error of 0.04 mt. 57