Sub-200 ps spin transfer torque switching in in-plane magnetic tunnel junctions with interface perpendicular anisotropy

Ultrafast spin transfer torque (STT) switching in an in-plane MgO magnetic tunnel junction with 50 nm × 150 nm elliptical shape was demonstrated in this paper. Switching speeds as short as 165 ps and 190 ps at 50% and 98% switching probabilities, respectively, were observed without external field assistance in a thermally stable junction with a 101% tunnelling magnetoresistance ratio. The minimum writing energy of P-AP switching for 50% and 98% switching probability are 0.16 pJ and 0.21 pJ, respectively. The observed ultrafast switching is believed to occur because of partially cancelled out-of-plane demagnetizing field in the free layer from interface perpendicular anisotropy between the MgO layer and the Co20Fe60B20 layer. High J/Jc0 ratio and magnetization nucleation at the edge of free layer, which result from the reduced perpendicular demagnetizing field, are possibly two major factors that contribute to the ultrafast STT switching. (Some figures may appear in colour only in the online journal) Ultrafast spin transfer torque (STT) switching in the subns regime is one of the key issues for spin transfer torque random access memory (STT-RAM) development. One of the crucial limitations for ultrafast switching is the incubation delay induced by pre-switching oscillation [1]. Several approaches have been proposed to minimize pre-switching oscillations in order to improve the switching speed in spin valves (SVs), such as developing all perpendicular structures [2], applying hard axis field to set the free layer equilibrium away from the easy axis [3], and adding an extra perpendicular polarizer [4–6]. As of now, limited work has been done on sub-nanosecond STT switching in magnetic tunnel junctions (MTJs). Minimum switching times of 400–580 ps at 50% switching probability has been reported in conventional inplane MTJs [7, 8]. By adding perpendicular polarizer, Liu et al showed 100% switching at 500 ps with external field assistance in their MTJ+SV device [9]. Rowlands et al achieved 50% switching probability at 120 ps under zero bias field in the full orthogonal MTJ [10]. In this paper, we report ultrafast switching (165 ps–10 ns) in CoFeB–MgO MTJs with good tunnelling magnetoresistance (TMR) ratio around 100% and large coercivity (100 Oe) under zero bias field. With a basic conventional stack structure, the sample exhibits ultrafast switching in the sub-200 ps 0022-3727/12/025001+04$33.00 1 © 2012 IOP Publishing Ltd Printed in the UK & the USA J. Phys. D: Appl. Phys. 45 (2012) 025001 H Zhao et al Figure 1. (a) MTJ resistance versus magnetic field loop at room temperature. The red curve is tested before switching probability measurement and the blue curve is obtained after switching probability measurement. (b) Switching probability dependence on pulse width with various pulse amplitudes on P-AP side. Each curve corresponds to the same setting voltage on pulse generator. The inset figure shows the change in pulse shape from 100 to 400 ps with the same setting amplitude. Because of the pulse peak attenuation, the labelled voltage in figure 1(b) is the peak voltage at the pulse duration corresponding to 50% switching probability8. For example, the first curve (purple, triangle-to-left) has the nominal pulse amplitude at 2.4 V for long pulses. The labelled value is 1.89 V, which means the peak value at 165 ps pulse width with 50% switching probability. (c) Pulse voltage as a function of pulse width at 50% switching probability for AP-P and P-AP switching. And the dashed line is the breakdown voltage at different pulse widths. regime while maintaining all the requirements for STT-RAM application. The MTJ samples’ stacking structure is as follows: (bottom electrode)/PtMn (15 nm)/Co70Fe30 (2.3 nm)/Ru (0.85 nm)/Co40Fe40B20 (2.4 nm)/MgO (0.83 nm)/Co20Fe60B20 (1.7–2.0 nm)/(top electrode). Here we used Fe-rich free layer from 1.7 to 2.0 nm, which has a strong perpendicular interface anisotropy [11, 12], but still retains the easy axis in plane. The sample was post-annealed at 300 ◦C under 1 T magnetic field for 2 h. MTJ devices in this paper were patterned into 50 nm × 150 nm elliptical nanopillars. The samples were characterized at room temperature for resistance versus applied field (R–H ) loop and switching probability. The switching probability measurement was performed by a sequence similar to our previous work [8] under zero bias field. Each probability value was calculated by 200 switching trials with the free layer magnetization preset by a 1μs reset pulse. The 100 ps–10 ns switching pulse was generated by the picosecond bipolar voltage pulse generator 10070A, which has a rise and fall time of 65 ps and 85 ps, respectively. Figure 1(a) shows the R–H loop of a MTJ sample with 2.0 nm free layer. The sample has the TMR ratio of 101% and the coercivity of 100 Oe. By averaging from a group of similar devices, the thermal stability is estimated to be above 65 kBT according to the hard-axis magnetoresistance curve fitting method [13]. The loop is centred at−45 Oe due to the coupling with pinned layer which is compensated by an external field during all the following switching probability measurements. The nearly overlapping blue and redR–H loops were obtained before and after the switching probability measurement, respectively, and showed that no partial breakdown of the barrier or change of magnetic properties had occurred. The switching probability as a function of pulse width is plotted in figure 1(b), where each curve represents the setting same pulse amplitude. The labelled voltage is the pulse peak voltage on the device at the pulse duration corresponding to 50% switching probability8. Please find more description in the figure caption of figure 1(b) and footnote 8. We observed 50% switching probability at 165 ps and 98% switching probability at 190 ps. Moreover, the switching probability curves were very steep and did not display a switching probability plateau because of the half precession period jitter as observed in some metallic SVs [1]. Furthermore, the observed sub-200 ps switching implies that incubation delay did not occur as a result of pre-switching oscillation. To calculate the writing energy, we did an integration based on the pulse shape for each pulse width by Ew = ∫ V 2(t)/R dt . The minimum writing energy of P-AP switching for 50% and 98% switching probabilities are 0.16 pJ and 0.21 pJ, respectively. During the measurement, the samples can generally survive 103–104 writing circles for the sub-200 ps switching. And we also calculated the endurance in our sample according to [14], with 1.107 V, 500 ps pulse width, the failure rate is 3.25×10−4. The same switching probability measurement was also done for AP-P switching. We plot the pulse amplitude versus pulse width at 50% probability in figure 1(c) together with the breakdown voltage, which was measured from 20 MTJs’ breakdown point with identical barrier thickness at various pulse widths. The figure shows that the achievable minimum switching time is limited by the breakdown voltage of the device. With the same applied voltage, the current through the device in the P state is about twice the value in the AP state due to the resistance difference in each state. Therefore, for P-AP switching higher voltages can be reached, thus allowing shorter switching times, as shown in figure 1(c). Two other MTJs of the same size but with thinner free layers (1.90 nm, 1.73 nm) were also measured for ultrafast switching probabilities in the sub-ns regime. The results were summarized in figure 2. The red line indicates the current density at which the oxide barrier breaks down at different pulse widths. Again, we see that the minimum measured switching time is limited by the breakdown voltage, especially 8 For pulses below 300 ps, the pulse shape turns to triangle as shown in the inset of figure 1(b) from equipment constraints and the peak voltage starts to decrease with pulse width. Therefore, with the same setting voltage, real pulse voltage varies in the ultrafast switching end. For example, at 165 ps pulse width, the 2.4 V nominal pulse amplitude is reduced to 1.89 V. All the pulse voltage values used in this paper are the peak voltage measured by Tektronix DPO72004BO scilloscope (20 GHz bandwidth and 50 GHz sampling rate) multiplied by the reflection coefficient at the MTJ end ( = 2RMTJ/(RMTJ + Z0)).