Electric-Field-Controlled MRAM: Physics and Applications
163
(long τ ), Kanai et al. have shown through simulations that longer pulse widths are
more susceptible to thermal effects. Thermally-induced random stray fields from
the pinned layer can cause phase shifts to the switching probability, again causing
transition probabilities to tend towards 0.5. This is shown by the contour that becomes
lower in intensities and more diffused with large pulse widths. This decoherence due
to thermal effects also affects the higher-order peaks (more precessional cycles) in
these contours more pronouncedly—the switching probability decreases with the
second, third peaks and so on in the contour. This behavior is also evident in Fig. 8b.
While precessional switching is fast (<1 ns), there are several drawbacks. Firstly, as
seen from Fig. 8b, the switching probability has a sinusoidal dependence with pulse
width which implies an accurate pulse width is required to switch with high accuracy.
Secondly, an external in-plane field is required. Thirdly, the probability distributions
may not have the same starting offset (Fig. 8b). Experimentally, it can be seen that
different devices may have different offsets. This contributes to a further increase in
errors for the fully precessional scheme. We will address some of these concerns and
EF-MRAM’s application in actual circuits in the next section.
3.2 Interplay of Electric-Field and Spin-Transfer Torque
Proposals that use only E-field is limited by the unidirectionality of the VCMA effect,
i.e. typically only when a positive voltage is applied to the free layer does the surface
anisotropy decrease. There are 2 documented ways to get this to work with while
using a unidirectional current. The first scheme, proposed by Alzate et al. [73] relies
on the fact that the spin-torque effect can still be relevant when the current is large.
Alzate et al. reported that the introduction of a small STT leakage current can
act as a small effective magnetic field. The STT current, in this case, is not able
to induce switching on its own. Since this effect is opposite of what is expected of
the antidamping torque, this was attributed to some non-trivial effect of the fieldlike torque in STT acting in conjunction with the EF. This scheme should be highly
dependent on the exact spin dynamics or the shape of the voltage pulse— there is no
reason why the P state does not revert to the AP state as the voltage is reduced after
writing to the P state. This may explain the high write error rates (WER). Lastly, the
slight shift in the bias field required also implies an imbalance in the stability of the
P and AP states. This might be a bottleneck of this scheme at elevated temperatures
where the coercivity of the MTJ and hence its thermal stability is expected to shrink
[74].
The second scheme was proposed by Kanai et al. [21] and is depicted in Fig. 10.
As shown in Fig. 10a, two pulses are applied to the MTJ, the first one, with higher
amplitude, reduces the anisotropy and causes precession while the second one with
lower amplitude stabilizes the final magnetization with STT. Since STT is directional,
the 2nd pulse is opposite in direction to first for AP → P switching as shown by
Fig. 10b. Lower WER was reported in this scheme compared to pure precessional
switching alone since the STT pulse flattens the oscillatory switching probability of
163
(long τ ), Kanai et al. have shown through simulations that longer pulse widths are
more susceptible to thermal effects. Thermally-induced random stray fields from
the pinned layer can cause phase shifts to the switching probability, again causing
transition probabilities to tend towards 0.5. This is shown by the contour that becomes
lower in intensities and more diffused with large pulse widths. This decoherence due
to thermal effects also affects the higher-order peaks (more precessional cycles) in
these contours more pronouncedly—the switching probability decreases with the
second, third peaks and so on in the contour. This behavior is also evident in Fig. 8b.
While precessional switching is fast (<1 ns), there are several drawbacks. Firstly, as
seen from Fig. 8b, the switching probability has a sinusoidal dependence with pulse
width which implies an accurate pulse width is required to switch with high accuracy.
Secondly, an external in-plane field is required. Thirdly, the probability distributions
may not have the same starting offset (Fig. 8b). Experimentally, it can be seen that
different devices may have different offsets. This contributes to a further increase in
errors for the fully precessional scheme. We will address some of these concerns and
EF-MRAM’s application in actual circuits in the next section.
3.2 Interplay of Electric-Field and Spin-Transfer Torque
Proposals that use only E-field is limited by the unidirectionality of the VCMA effect,
i.e. typically only when a positive voltage is applied to the free layer does the surface
anisotropy decrease. There are 2 documented ways to get this to work with while
using a unidirectional current. The first scheme, proposed by Alzate et al. [73] relies
on the fact that the spin-torque effect can still be relevant when the current is large.
Alzate et al. reported that the introduction of a small STT leakage current can
act as a small effective magnetic field. The STT current, in this case, is not able
to induce switching on its own. Since this effect is opposite of what is expected of
the antidamping torque, this was attributed to some non-trivial effect of the fieldlike torque in STT acting in conjunction with the EF. This scheme should be highly
dependent on the exact spin dynamics or the shape of the voltage pulse— there is no
reason why the P state does not revert to the AP state as the voltage is reduced after
writing to the P state. This may explain the high write error rates (WER). Lastly, the
slight shift in the bias field required also implies an imbalance in the stability of the
P and AP states. This might be a bottleneck of this scheme at elevated temperatures
where the coercivity of the MTJ and hence its thermal stability is expected to shrink
[74].
The second scheme was proposed by Kanai et al. [21] and is depicted in Fig. 10.
As shown in Fig. 10a, two pulses are applied to the MTJ, the first one, with higher
amplitude, reduces the anisotropy and causes precession while the second one with
lower amplitude stabilizes the final magnetization with STT. Since STT is directional,
the 2nd pulse is opposite in direction to first for AP → P switching as shown by
Fig. 10b. Lower WER was reported in this scheme compared to pure precessional
switching alone since the STT pulse flattens the oscillatory switching probability of