Magnetic, magnetoresistive and low-frequency noise properties of tunnel magnetoresistance sensor devices with amorphous CoFeBTa soft magnetic layers

Figures 2(a) and (b) show the bright-field (BF) STEM image and EDS map of the spin-valves with t_CFBT = 20 nm annealed at 350 °C for 1 h. The CoFeB electrodes in the FL and RL have been crystallized with a [001] out-of-plane texture, whereas the CFBT layer remains amorphous as shown by the halo of the nano-beam electron diffraction pattern shown in the inset of figure 2(a). Relatively flat interfaces in the spin-valve structure are seen in figures 2(a) and (b). The spin-valve with t_CFBT = 80 nm for the FL is shown in figures 2(c) and (d). The film interfaces are much rougher than those of the spin-valve with t_CFBT = 20 nm, indicating that the surface roughness of the CFBT amorphous layer increases with increasing layer thickness.

Zoom In Zoom Out Reset image size

The interface roughness between the CoFeB electrodes and the MgO barrier influences the TMR properties. The surface roughness of the test sample structure of substrate/Ta (2)/Ru (10)/Ta (2)/Ru (10)/Ta (5)/CFBT (t_CFBT = 20–80)/Ta (0.3)/CoFeB (3)/MgO (2)/Ta (2) was measured by an atomic force microscope (AFM). Figures 3(a) and (b) show the AFM images for t_CFBT = 20 nm for the as-deposited state and after annealing at 350 °C, respectively. The arithmetic average roughness (R_a) and the peak-to-valley (P–V) values were 0.17 nm and 1.9 nm for the as-deposited film and 0.25 nm and 2.7 nm for the annealed film, respectively. In addition to the increase in the roughness amplitude, the grain-like surface morphology became finer by annealing, which may be associated with the crystallization of CoFeB. As shown in figure 3(c), R_a increases monotonically as t_CFBT increases, consistent with the STEM images (figure 2).

Zoom In Zoom Out Reset image size

Figure 4(a) shows the magnetization–magnetic field (M–H) curves of the spin-valve film with t_CFBT = 20 nm after the first annealing at T_1st = 350 °C under a magnetic field in the y-direction and after the second annealing at T_2nd = 200 °C under a magnetic field in the x-direction. The magnetization is expressed by the product of saturation magnetization and thickness (4πM_s t). The kinks of the M–H curves indicated by the arrows in figure 4(a) correspond to the effective pinning fields (H_pin) for the RL. By the first annealing, a pinning of the RL (μ₀ H_pin = 35 mT) and magnetic anisotropy in the FL (μ₀ H_k = 1.8 mT) were induced in the y-direction. By the second annealing, however, the initial pinning field was erased and reset in the x-direction with μ₀ H_pin = 19 mT, while the anisotropy of the FL remains in the y-direction with a reduced strength of μ₀ H_k = 1.2 mT. Thus, H_k of the FL is orthogonal to H_pin of the RL, which is often referred to as 'crossed-anisotropy configuration'.

Zoom In Zoom Out Reset image size

Figure 4(b) shows the dependence of μ₀ H_pin on t_CFBT. For both first and second annealing, μ₀ H_pin is lowered with increasing t_CFBT, which may be attributed to the larger film roughness as t_CFBT increases, as shown in figures 2 and 3. For each t_CFBT, μ₀ H_pin is reduced by the second annealing at 200 °C compared to the values after the first annealing at 350 °C. This is because 200 °C is not the optimal annealing temperature for the highest exchange bias by the IrMn layers. Figure 4(c) shows the dependence of μ₀ H_k on t_CFBT. There is a trend that μ₀ H_k decreases as t_CFBT increases. For each t_CFBT, μ₀ H_k is reduced by the second annealing compared to the values after the first annealing.

Figure 5(a) shows the R–H transfer curves of the spin-valve devices with t_CFBT = 20 nm annealed at various conditions, where R is expressed as the TMR ratio (=ΔR/R_P). The resistance-area product in the parallel magnetization state (RA_P) was ∼40 kΩ μm². After the first annealing at 350 °C, the device shows a square-shaped R–H_y curve with a coercivity (μ₀ H_c) of 0.5 mT and a small interlayer coupling field between the RL and the FL (μ₀ H_int) of 0.3 mT. By annealing the film again at T_2nd = 200 °C under a magnetic field in the x-direction, the crossed-anisotropy between the RL and the FL results in a linear R–H_x curve with a negligible μ₀ H_c, which is suitable for magnetic field sensors. As T_2nd increases to 250 °C and 280 °C, the R–H_x curves become hysteretic with larger μ₀ H_c.

Zoom In Zoom Out Reset image size

The change of the shape of the R–H_x curve for T_2nd = 200 °C–280 °C shown in figure 5(a) is explained by assuming two magnetic anisotropy axes in the FL induced by the two-step annealing process. In the ideal two-step annealing process, the magnetic anisotropy of FL is induced only in the y-direction with an energy K_y by the first annealing, and the second annealing induces only an exchange bias pinning to PL and RL in the x-direction. However, the second annealing may induce an additional magnetic anisotropy in the FL in the x-direction (K_x ), as shown in figure 5(b). As the magnetization of the RL is strongly fixed within the external field range for the R–H measurements (μ₀ H_ext = ± 5 mT), we consider only the behavior of the magnetization of the FL by the Stoner–Wohlfarth model in a single-domain regime [16]. The total magnetic energy of the FL is expressed as

$$\begin{equation}{E_{{\text{tot}}}} = - M{H_{{\text{ext}}}}{\text{cos}}\theta \, - {\left( {\sqrt {{K_x}} {\text{cos}}\theta + \sqrt {{K_y}} {\text{sin}}\theta } \right)^2},\end{equation} \tag{ 2 }$$

where M is the magnetization of the FL, θ is the angle between H_ext and M. The anisotropy field H_k has a relationship of H_k = 2 K M⁻¹ . Solving $\frac{{\partial {E_{{\text{tot}}}}}}{{\partial \theta }} = 0$ under $\frac{{{\partial ^2}{E_{{\text{tot}}}}}}{{\partial {\theta ^2}}} > 0$ gives θ that minimizes E_tot for given H_kx and H_ky . As shown in figure 4(c), H_ky induced by the first annealing decreased after the second annealing at T_2nd = 200 °C; thus it is expected that H_ky decreases further and H_kx increases for higher T_2nd. For simplicity, we assume μ₀(H_kx + H_ky ) = 1.5 mT regardless of T_2nd. Figure 5(c) shows calculated R–H_x curves for variations of H_kx using the angular dependence of TMR [17]: $R\left( \theta \right) = \frac{{{R_ \bot }}}{{1 + {P^2}\cos \theta }}$, where ${R_ \bot }$ and P are the resistance at θ = 90° and the tunneling spin polarization obtained by the Julliere's equation [18] (P² ∼0.44 for a full TMR ratio of 160%), respectively. The FL offset field (H_int) is not included in these calculations. The R–H_x curve for H_kx = 0 shows a loop with H_c = 0 similar to the experimental result for T_2nd = 200 °C. As H_kx increases, the R–H_x curves show hysteresis loops with nonzero H_c, which qualitatively agrees with the experiments for T_2nd = 250 °C and 280 °C. Thus, the additional magnetic anisotropy induced in the FL in the x-direction explains the hysteretic R–H_x loops for T_2nd = 250 °C and 280 °C. Therefore, we conclude that T_2nd = 200 °C is the optimal annealing temperature to obtain an appropriate crossed anisotropy in the TMR devices using the CFBT-based FL.

Figure 6 shows the R–H_x curves with t_CFBT = 20–80 nm after the two-step annealing at T_1st = 350 °C and T_2nd = 200 °C. All devices show a gradual change of R as H_x changes due to the crossed anisotropy. The anisotropy fields were similar for all t_CFBT (2μ₀ H_k ∼2.3 mT), while the TMR ratios were largely dependent on t_CFBT. The lower TMR ratio for larger t_CFBT may be explained by the larger film roughness as seen in figures 2 and 3. As shown in the inset of figure 6, the sensitivity defined by the TMR ratio/2μ₀ H_k was ∼65%/mT for t_CFBT = 20 and 40 nm and decreased for larger t_CFBT. Hence, we focus on the TMR device with t_CFBT = 20 nm for the following studies.

Zoom In Zoom Out Reset image size

We measured the noise spectra of a TMR sensor device with t_CFBT = 20 nm patterned into a 50 μm diameter circle. V_b was fixed at 30 mV, where the positive sign of V_b corresponds to the flow of electrons from the FL to the RL. Figure 7(a) shows the R-H_x curve and the numerically obtained magnetic field sensitivity defined by 1/R_min · (dR/dH_x), where R_min is the resistance in the P-state. The R–H_x curve shows a gradual change of R as H_x changes between μ₀ H_x = 0 mT and +3 mT, due to the magnetization rotation of the FL. A maximum sensitivity of ∼70%/mT is obtained at μ₀ H_x between +1.0 mT and +1.7 mT, where the R-H_x loop is almost nonhysteretic.

Zoom In Zoom Out Reset image size

Figure 7(b) shows the noise voltage density $\sqrt {{S_{\text{v}}}}$ at μ₀ H_x = –5 (P-state), +1, +2 (intermediate magnetization states), and +5 mT (AP-state) for the same device biased at V_b = 30 mV. For these noise measurements, we first applied a magnetic field of μ₀ H_x = +5 mT to saturate the device in the AP-state, and then set it to the required magnetic field for the noise measurements. It was important to wait for a few minutes to equilibrate the magnetization state of the device before starting the noise measurements [19]. The noise of the pre-amplifier is also shown in figure 7(b), indicating that the electronics noise of the measurement setup is significantly smaller than the noise of the TMR device. The white noise level observed in the TMR device in f > a few kHz is larger in the AP-state than in the P-state because of the larger resistance in the AP-state. The value of $\sqrt {{S_{\text{v}}}}$ in the low-f regime (<1 kHz) is similar between the P-state and the AP-state, whereas the low-f noise in the intermediate magnetization states was much larger than those in the P- and AP-states, suggesting magnetic fluctuation dominates the 1/f noise in the intermediate magnetization states [20]. Figure 7(c) shows the magnetic field dependence of $\sqrt {{S_{\text{v}}}}$ at f = 10 Hz and also the Hooge's noise parameters (α_H) for μ₀ H_x = −2.0 mT, +1.0 mT, +2.0 mT and +4.0 mT was obtained by the relationship of ${\alpha _{\text{H}}} = fA{S_{\text{V}}}/V_{\text{b}}^2$ [21], where A is junction area (1962 μm²) and the thermal and pre-amplifer noises were subtracted from the measured $\sqrt {{S_{\text{v}}}}$. The high noise peaks observed at μ₀ H_x = –0.25 mT and +2.0 mT may be attributed to the fluctuation of magnetic domains nucleated in the MTJ when the FL starts to rotate from the P-state and the AP-state, respectively. The magnetic field dependence of detectivity D is shown in figure 7(d). The minimum D of ∼2 nT/$\sqrt {{\text{Hz}}}$ at f = 10 Hz is obtained at μ₀ H = +1.0 mT, where the sensivity and $\sqrt {{S_{\text{v}}}}$ are ∼70%/mT and ∼40 nV/$\sqrt {{\text{Hz}}}$, respectively.

Figure 8(a) shows the noise voltage spectra of the TMR device with t_CFBT = 20 nm at various V_b. The magnetization state was set to the intermediate state by applying μ₀ H_x = +1 mT, where the minimum D was obtained at V_b = 30 mV (figure 7(d)). A typical 1/f noise characteristic was observed at V_b = 20 mV, 40 mV, 80 mV and 100 mV, where the Hooge's noise parameter was approximately constant α_H ∼4 × 10⁻⁸ μm². At V_b = 60 mV, however, a Lorentzian noise spectrum was observed, indicating random telegraph noise (RTN) with a center frequency of ∼700 Hz. The V_b-dependences of the noise spectra of the same device in the P- and AP-magnetization states were also measured (data not shown), where $\sqrt {{S_{\text{v}}}}$ increased linearly with increasing V_b between 20 mV and 100 mV with an approximately contant α_H ∼1 × 10⁻⁸ μm² in both magnetization states. No RTN was observed in either P- and AP-magnetization states. It is noteworthly that similar RTN was observed in other devices with the identical shape and size; however, the V_b where the RTN appeared and the center frequency of the RTN differed from device to device. In addition, the appearance of RTN was intermittent and did not necessarily reproduce in repeated measurements under nominaly idential conditions (see supplemental material (available online at stacks.iop.org/54/095002/mmedia)). Such characteristics of RTN in TMR devices have been reported previously [20–24] and explained by the thermally activated hopping of small magnetic domain(s) whose volume fraction is much smaller than 1% of the total volume of the FL [20, 24]. The strong V_b-dependence of the RTN may indicate that spin transfer torque plays a role in the activation process of the magnetic domain hopping, as discussed by Arakawa et al [24].

Zoom In Zoom Out Reset image size

The value of $\sqrt {{S_{\text{v}}}}$ at f = 10 Hz is shown in the inset of figure 8(a). $\sqrt {{S_{\text{v}}}}$ increases approximately in proportion to V_b between 20 mV and 80 mV, except at 60 mV where the RTN appears. Figure 8(b) shows the f-dependence of D of the same device. The depepdence of the TMR ratio on V_b shown in the inset of figure 8(b) was used to calculate D. As $\sqrt {{S_{\text{v}}}}$ scales linearly with V_b for between 20 mV and 80 mV except for 60 mV, almost constant D of 2.2 nT/$\sqrt {{\text{Hz}}}$ at 10 Hz was obtained regardless of V_b.

We also investigated the dependence of $\sqrt {{S_{\text{v}}}}$ on the junction area A for the TMR device with t_CFBT = 20 nm. Note that the device was patterned into circular shapes with diameters ranging from 5 μm to 50 μm and the Ar ion milling was performed down to the MgO tunnel barrier, thus A defines the area of PL, RL and MgO barrier. The FL was patterned to a much larger area (100 μm × 400 μm). As shown in figure 9, $\sqrt {{S_{\text{v}}}}$ shows a relationship of $\sqrt {{S_{\text{v}}}}$ ∝ 1/$\sqrt A$, indicating an approximately constant α_H of 4 × 10⁻⁸ μm² from the relationship of ${\alpha _{\text{H}}} = fA{S_{\text{V}}}/V_{\text{b}}^2$, also as reported previously in a NiFeCo/Al₂O₃/CoFe MTJ [25]. Since the TMR ratio and its V_b-dependence are independence of A, a constant sensitivity is obtained regardless of A. Thus, smaller D is obtained as A increases.

Zoom In Zoom Out Reset image size

We have obtained linear R–H response curves with negligibleH_c using the TMR devices with a CFBT (20 nm)/Ta (0.3 nm)/CoFeB (3 nm)-FL under an optimal two-step annealing condition (T_1st = 350 °C and T_2nd = 200 °C), which is suitable for magnetic field sensor applications. The properties of the CFBT-based TMR sensor are summarized in table 1 in comparison with reported TMR sensors with NiFe and CoFeSiB soft magnetic layers. The noise voltage density $\sqrt {{S_{\text{v}}}}$ at 10 Hz is normalized by dividing by V_b because $\sqrt {{S_{\text{v}}}} \propto {V_{\text{b}}}.$ Note that these data are for single devices without array structure or integration with magnetic flux concentrator. Fujiwara et al reported a very high sensitivity of up to 253%/mT using a NiFe (70 nm)/Ru (0.9 nm)/CoFeB (3 nm)-FL with a small μ₀ H_k of 0.3 mT by a two-step annealing [10, 26]. For these sensors, only the noise level in the P-state has been reported to be $\sqrt {{S_{\text{v}}}} /{V_{\text{b}}}$ ∼3.3 nV/$\sqrt {{\text{Hz}}}$/mV, which is much larger than that of our device of ∼0.7 nV/$\sqrt {{\text{Hz}}}$/mV in P-state. In addition, our device shows a smaller D than those reported for the TMR sensors with NiFe-based FLs [13, 27], which is explained by a higher sensitivity and a lower noise in our device with the CFBT-based FL.

Table 1. Properties of TMR single devices with CFBT (this work), NiFe and CoFeSiB soft magnetic layers. T_an: annealing temperature (the numbers in parentheses are T_an of the second annealing for two-step annealing process), A: junction area, μ₀ H_k: magnetic anisotropy field of FL, S: sensitivity, $\sqrt {{S_{\text{V}}}} /{V_{\text{b}}}$: noise voltage density normalized by bias voltage, α_H: Hooge's noise parameter, D: detectivity. 'H_bias' and 'soft pin' denote external bias field and softly-pinned exchange bias for H_k of FL.

Soft magnetic layer	Tan (°C)	A (μm2)	μ0 Hk (mT)	TMR ratio (%)	S (%/mT)	$\sqrt {{S_{\text{V}}}} /{V_{\text{b}}}$ at 10 Hz (nV/$\sqrt {{\text{Hz}}}$/mV)	αH (μm2)	D (nT/$\sqrt {{\text{Hz}}}$) at 10 Hz	Reference
CoFeBTa	350 (200)	1962	1.2	160	70	1.3 0.7 (P-state)	4 × 10−8	2.2	This work
Ni80Fe20	325 (300)	3200	0.3	150	253	3.3 (P-state)	3.3 × 10−7 (P-state)		[10, 12]
	325	300	1.0 (Hbias)	170				12.8	[13]
	330	2463	4.2 (soft pin)	191	22.7	2.9	2.1 × 10−7	16.1	[27]
CoFeSiB	375 (300)		0.28	228	400				[11]
	325	300	1.0 (Hbias)	200			4 × 10−9	4.5	[13]

The TMR sensors with a CoFeSiB-based FL have been reported to show a very high sensitivity of 400%/mT due to a small μ₀ H_k of 0.28 mT although the noise properties have not been discussed in the paper by Kato et al [11]. Huang et al reported a smaller α_H, but a larger D using sensors with a CoFeSiB-based FL, where the FL is biased by external magnetic field [13], than those of our sensors. However, a direct comparison between the performances of our sensors with the CFBT-based FL and those with the CoFeSiB-based FL is not possible because of the different techniques for H_k of FL; i.e. by two-step annealing vs. by external magnetic field. Thus, we need a side-by-side comparison for benchmarking these soft magnetic layers under the same sample structures and measurement conditions.