MnNi-based spin valve sensors combining high thermal stability, small footprint and pTesla detectivities

Magnetoresistive sensors (MR) are already widely used in automotive applications, nondestructive testing, bacteria detection or biomagnetic signal mapping.¹ These sensors work at room temperature, have a tunable micrometric spatial resolution and demonstrate adequate noise levels. Many of these systems use spin-valve (SV) structures with low noise and field sensitivities around nT.² A critical part of the SV is the antiferromagnetic (AFM) layer to provide a magnetic reference (pinned layer) through exchange biasing of and adjacent ferromagnet (FM). The AFM/FM has to yield large exchange bias field, be thermally stable and have a low resistance to ensure high MR ratio.³ MnNi, although not so widely used as MnIr or MnPt, gives high exchange bias field combined with high robustness at larger temperatures, with reported blocking temperature for a thickness of 300 Å of 400°C.⁴ As deposited MnNi exhibits a (111) FCC structure. After annealing, crystalizes into a (111) FCT structure, becoming ordered into an AFM state, and consequently leading to the onset of exchange bias.⁵ Buffer layers, as Ta or NiFeCr, are used to improve the properties of the AFM layer and full SV stack by promoting an increase in grain size and reduced interface roughness.^6,7 In this case, both buffer layers are used due to their capability to induce a strong FCC (111) crystalline structure in the MnNi, correlated with an increase in MR.^8,9

For example, in specific biomedical experiments or industrial applications one always requires a robust sensor, that is able to maintain its properties, either during consecutive hours of experiments or in environments where elevated temperatures are present. Nozières et al. demonstrated that SV sensors enclosing a MnNi antiferromagnetic (AFM) layer, required a longer time to degrade when compared to other AFM materials.³ Although MnNi-based SV exhibit smaller change in voltage (signal) with magnetic field, they display improved reliability and are more affordable. In this work, we optimize the multilayer stack and magnetic annealing conditions for MnNi-based SVs, taking into account as performance indicators the magnetoresistance, noise level, and enhanced thermal stability. Furthermore, to improve the signal-to-noise ratio (SNR) of the sensor, 4 MnNi based SVs are deposited vertically profiting from a reduced device footprint.¹⁰ 

In this work, the top-pinned SV stacks were deposited by ion beam deposition on a N3600.⁸ The optimum structure was as follows: buffer 50 / Ni₈₀Fe₂₀ 28 / Co₉₀Fe₁₀ 22 / Cu 22 / Co₉₀Fe₁₀ 30 / Ru 8 / Co₉₀Fe₁₀ 26 / Ni₈₀Fe₂₀ 7 / Mn₅₀Ni₅₀ 300 / Ta 90 (thicknesses in Å). Magnetic annealing is then performed in vacuum, with an applied magnetic field (0.5 T) during heating and cooling steps to set the exchange-bias. The sensors were patterned with length l=50 μm and height h from 2 μm to 5 μm using photolithography and ion beam milling. Electrical contacts (2 contacts architecture) were defined by liftoff of Al_98.5Si_1.0Cu_0.5 3000 Å/ 150 Å TiWN. Sensors and electrical contacts were passivated with 1000 Å of Al₂O₃. For the 0vertically packed structures, the insulator spacer was also grown by ion beam deposition without vacuum break, where the assist gun is used for oxidation.

All devices were characterized by measuring the transfer curve [MR(H)] and noise. Transfer curve was measured within a field range of ±14 mT and ±200 mT and the noise was measured at H=0 at the most sensitive point of the sensor. In the noise measurements, the sensor was placed inside a mu-metal box and was biased using a circuit powered by battery. The sensor signal was amplified by a SRS SIM910 low noise amplifier and its power density was acquired by a TEKTRONIX RSA3308A real time spectrum analyzer. The measurement was performed in a range from dc to 100 kHz with a resolution bandwidth of 2 Hz (dc-1 kHz range) and 200 Hz (dc-100 kHz range).

Two different buffers (Ta, NiFeCr) and a thin NiFe layer were used as underlayers adjacent to the MnNi, and their impact on H_f was studied. Figure 1a) and 1b) show that regardless the buffer thickness, H_f is always lower for NiFeCr indicating a lower interfacial roughness.¹⁰ In addition, figure 1c) shows that the optimum annealing temperature for such a stack is 375°C. Taking into considerations all optimized parameters, the final SV shows MR=6% compared with previous work,³ a coercive field (H_c) less than 4.8 A/m and H_f =0.6 kA/m.

Figure 2a) displays a schematic drawing and a microscopic image of the microfabricated sensor. Figure 2b) shows the main parameters obtained from the transfer curves. The slight increase of the MR for higher h can be explained by a larger overlap area between the SV and the electrical contact which reduces the contact resistance. An increase in the saturation field and a consequent sensitivity reduction is observed for a lower h due to free layer self-demagnetizing field.¹¹ H_c around 50 A/m (0.6 Oe) is obtained, showing a negligible change with h. H_f depends on the field created at the sensing layer by the demagnetizing field of the pinned and the Neél coupling.¹² Taking this into account, the curve shifts towards higher field values with increasing h.

The detectivity (D) [Tesla/Hz^1/2] is obtained by the ratio between the noise and the sensitivity (S) [V/T]. Noise [V/ Hz^1/2] in the SV sensors² is described by the thermal and 1/f noise contributions. Figure 2c) shows the noise level S_V(V/Hz^1/2) of the single SV for h = 2, 3 and 4 μm measured at operation point (H=0). Figure 2d) shows the corresponding field detection level. In the thermal noise in SVs, the only parameter that can be tuned is the resistance, while in the 1/f noise term, the resistance and current bias can both be changed. For higher h, the resistance is lower and therefore the noise level is also lower. At 100 Hz the obtained noise level is 19 nV/Hz^1/2 and 12 nV/Hz^1/2 for h=2 μm and h=4 μm, respectively; the thermal noise level is 2.3 nV/Hz^1/2 and 1.5 nV/Hz^1/2 for h=2 μm and h=4 μm, respectively, being consistent with previous results for top pinned SV.¹³ Since the sensitivity increases with h, the detectivity is improved for larger h dimensions. Therefore, and at thermal level, one obtains a detection level of 1 nT/Hz^1/2 for h=2 μm, and 571 pT/Hz^1/2 for h=4 μm. These results are promising for biomagnetic mapping where pT or fT detectivities are required.¹⁴ There are some strategies that already improve the detectivity by enhancing the sensitivity combining magnetic flux guides with SV sensors¹⁵ or by decreasing the noise reducing the sensor resistance (see section III C).

In addition, the thermal stability depends directly on the AFM material used.³ Figure 3 compares the synthetic antiferromagnetic (SAF) MnNi SV and top pinned MnIr SV (Ta 10/ NiFe 28/ CoFe 23/ Cu 26/CoFe 25/ MnIr 70 / Ta 30; thickness in Å) in terms of its thermal stability. The samples were heated for 15 min at each temperature without magnetic field, cooled down and measured.

For the MnIr-based SV, no changes in MR are visible up to 160°C. Afterwards, MR value drops vanishing for 200 °C. In contrast, for the MnNi-based SV with SAF, MR decreases only by 10% of the initial value until 275°C. MnNi has a higher reported blocking temperature,⁴ which is further strengthen by using a SAF where the interlayer coupling remains after at such temperatures.¹⁶ Also, the increase of the minimum resistance and consequent decrease of MR can also be explained by Mn interdiffusion,¹⁷ which occurs at lower temperature in MnIr SV, while for MnNi SV was observed only at 225°C.¹⁸ Finally, the sensitivity is highly affected with these behaviors, although for MnNi SV the sensitivity decreases by 30% of the initial value until 275 °C.

Four SVs were deposited on top of each other separated by an insulator layer (450 Å NiFeCrOx). This vertical packaging allows one to achieve low resistance without comprising the spatial resolution, when compared to a side-by-side parallel configuration. To improve further the sensitivity of the final sensing device a configuration of series of SV was also employed. Consequently, an array with X sensors connected in series, Y in parallel and Z are packed vertically were implemented. Thus, (i) with the same device footprint the number of sensors can be multiplied by the number of SV stacked vertically yielding lower levels of noise, and (ii) with fixed number of sensors the total device footprint can be decreased while maintaining the detectivity values.

Figure 4a) shows the normalized transfer curve for 2 configurations with Z=1 and Z=4 maintaining the device footprint (same XY). For ZXY=1/1/1 the MR is 6%, the saturation field (μ₀H_sat) = 7 mT and the μ₀H_f = 0.2 mT. For ZXY=4/1/1 the MR is 4.8 %, μ₀H_sat = 19 mT and μ₀H_f = 1 mT. MR decreases for Z=4 due to the higher lateral contact resistance in the vertical packaging. The increase on H_f is mainly due to an increase on the Néel coupling, since the top SVs were deposited over SV/spacer and the roughness increases compared to the substrate. The saturation field (μ₀H_sat) for Z=4 increases to 12 mT when compared with Z=1 leading to a sensitivity decrease. Assuming that the anisotropy field and the free layer demagnetizing field are the same in both cases (similar stack and h), this behavior suggests the presence of a magnetostatic coupling between the packed SVs. Consequently, this effect can be minimized by increasing the spacer thickness, however its roughness cannot compromise the growth of the following layers.

Figure 4b) shows the noise level for 4 different configurations: two mentioned before and two arrays composed of ZXY=1/20/15 and ZXY=4/15/5. Comparing the first two with the same footprint, Z=4 has lower resistance and thus lower noise. At thermal level, the noise is 1.7 nV/Hz^1/2 and 1.1 nV/Hz^1/2 for Z=1 (R=182 Ω) and Z=4 (R= 69 Ω), respectively. In terms of sensitivity, sensor with Z=1 shows 1.2 V/T while Z=4 shows 0.4 V/T, which leads to detectivities of 1.4 nT/Hz^1/2 and 2.8 nT/Hz^1/2, respectively.

The resistance for arrays of 300 SVs (ZXY=1/20/15 and ZXY=4/15/5) are R_Z=1 = 243 Ω and R_Z=4 = 201 Ω which are the expected values taking into account the single sensor resistance. This means that the current is flowing equally for all the SVs. At thermal level, the noise is 2.1 nV/Hz^1/2 and 1.8 nV/Hz^1/2 for planar (Z=1) and vertically packed (Z=4) arrays geometries, respectively. The sensitivity is 21.9 V/T and 5.2 V/T which leads to detectivities of 95 pT/Hz^1/2 and 346 pT/Hz^1/2 for planar and vertical array configuration, respectively. These arrays show a spatial resolution of 0.4 mm² for Z=1 and 0.1 mm² for Z=4. Even with an increase in the saturation field for Z=4 configurations and consequent decrease in sensitivity, the detectivity values are still promising compared with other strategies which resort to large area arrays or flux guides coupled to a single sensor¹⁹ occupying areas higher than 1 mm².

This paper highlights the use of MnNi as AFM material on top pinned SVs. This material combined with NiFeCr as buffer and optimum annealing conditions, results in unpatterned SV with MR=6% and H_f = 0.6 kA/m. Moreover, MnNi SVs showed improved thermal stability with MR decreasing only 10 % until 275 °C. The detectivity of a single SV was improved by vertical packaging without compromising the spatial resolution. Arrays of 300 SVs achieved detectivity of 95 pT/Hz^1/2 and 346 pT/Hz^1/2 for planar (1 SV) and vertical (4 packed SVs) configuration, respectively, competing with sensors coupled with flux guides while offering the advantage of higher spatial resolution. Consequently, MnNi based and vertically packed SV are a promising sensing tool for biomagnetic field measurements, given that high thermal and magnetic stability for long experiments are obtained, together with a high spatial resolution and improved detectivity in the pT range.

M. Silva acknowledges FCT for scholarship grant PD/BD/128206/2016 within the Doctoral Programme AIM-Advanced Integrated Microsystems and support through POPH. D C Leitao acknowledges financial support through FSE/POPH. INESC-MN acknowledges FCT funding through the Associated Laboratory IN.