This section shows the results of the thickness dependence of the various layers. Multiple wedges are grown to vary the IrMn, Co and Pt thicknesses. The remanence, coercivity and exchange biases are compared. For some wedges, only OOP field-cooled results are shown, as the OOP properties are easier to measure via polar MOKE. While they can’t be compared directly to the IP case, they still give an indication of the thickness dependent behavior.
5.3.1 IrMn layer
The IrMn thickness tIrMndependence is shown in figure 5.7, using a Ta(3)/Pt(4)/Co(1.5)/IrMn(0-15)/Pt(2) (thicknesses in nm) OOP field-cooled sample. As this layer is optimized first, the Pt and Co thickness do not yet correspond to the final values that are used. The remanence is 100%
for all thicknesses and therefore not shown in the graph. The values for HEBand HCare found by fitting the different MOKE loops that are taken at points along the wedge. Refer to section 2.3 for details about coercivity, exchange bias and remanence.
The first observation to make is the coercivity peak at around 3 nm of IrMn. This can be ascribed to IrMn spins that rotate freely with the Co spins, hence the disappearance of exchange bias for tIrMn < 3 nm. The behavior corresponds to the theoretical predictions of the Spin-Glass model introduced in section 2.4, which assigns this behavior to a low anisotropy region in the antiferromagnet. At thicknesses > 4 nm, the coercivity is still increased compared to tIrMn=0 which indicates that a low anisotropy region of the IrMn layer rotates partially with the Co, even for thick layers.
The thickness dependence of the exchange bias somewhat corresponds to the prediction of the SG-model. For imperfect F/AF interfaces, the exchange bias shows a peak just after the critical thickness. Below this thickness, no exchange bias is visible as all AF spins rotate together with the Co spins. The decrease in exchange bias for tIrMn > 6 nm can be explained by the Mauri model that predicts that in thicker layers, the AF spins parallel to the interface form a gradual domain wall (see section 2.4.1). A thickness of tIrMn = 6 nm is considered best for the final sample design, as it corresponds to the largest measured exchange bias.
OOP MOKE, OOP field cooled
H
ext,
Figure 5.7: Measured exchange bias and coercivity for IrMn thicknesses between 0 and 15 nm by analyzing polar MOKE loops. The peak in the coercivity is the result of the rotation of the IrMn spins together with to Co. Hence, no exchange bias is seen at a thickness corresponding to the coercivity peak.
5.3.2 Co layer
The Co thickness tCo dependence are shown for both the case without dusting layer (sample:
Ta(3)/Pt(3)/ Co(0-3)/IrMn(6)/Ta(1.5), figure 5.8a) and with dusting layer (sample: Ta(3)/Pt(3)/
Co(0-3)/Pt(0.3)/ IrMn(6)/Ta(1.5), figure 5.8b) as the Co thickness dependence is significantly different between both cases. A standard IP field-cooling procedure is applied to both samples.
For both cases, the exchange bias is completely frozen into the IP configuration after the field-cooling procedure, hence no exchange bias is seen in the OOP MOKE loops.
From the graph, a critical tCocan be noticed below which no coercivity is present: 0.6 nm without Pt dusting and 0.4 nm with Pt dusting. Moreover, there appears to be almost no magnetic signal at all below this critical thickness. This indicates that the Co may be intermixing with the adjacent layers, resulting in a partially dead magnetic layer [50]. Intermixing can also be present at the Co/IrMn interface, resulting in the observed low anisotropy region for thin IrMn, as discussed in the previous section.
Above the critical thickness, the coercivity and remanence rapidly increase to form a square easy-axis loop. A competition between the surface anisotropy of the Pt/Co interface and the shape anisotropy (which scales with the volume of the Co layer) determines if the Co is OOP. For thicknesses below 1.6 nm, the surface anisotropy is strong enough to force the Co in the OOP configuration. At higher thicknesses the shape anisotropy becomes dominant and the Co is in the IP configuration. At exactly 1.6 nm, Keff≈0 and the surface anisotropy is equal to the shape anisotropy.
The insertion of the Pt dusting layer reduces intermixing with the IrMn. This results in a reduced dead layer thickness causing emergence of OOP anisotropy at thinner Co thickness. The critical thickness at which the Co goes IP remains at 1.6 nm, indicating that the surface anisotropy is not increased by much. However, a coercivity increase along the whole range is observed, which
OOP MOKE, IP field cooled
Without dusting layer
With dusting layer
a.
b. H
ext, ,
Figure 5.8: a: Co thickness dependence for an IP cooled sample. The field-cooling procedure removed all the exchange bias in the OOP direction. For Co thick-ness < 0.6 nm, intermixing with adjacent layers prevents the emergence of OOP Co.
Between 0.6nm and 1.6 nm the surface anisotropy induced by the Pt/Co interface overcomes the shape anisotropy which results in OOP anisotropy. b: The same be-havior is seen for the sample with dusting layer. The Co is OOP for lower thickness due to the increased surface anisotropy induced by the extra Pt layer.
indicates that the extra Pt has an effect on the OOP anisotropy. The exact influence of the Pt dusting layer on the surface anisotropy or saturation magnetization is therefore not entirely clear and leaves room for further investigation.
Co thicknesses between 0.9 nm-1.35 nm (without dusting layer) and 0.7 nm-1.2 nm (with dust-ing layer) seem to be most ideal for further experiments due to their high remanence and high coercivity.
5.3.3 Pt layer
A Ta(0.5)/Pt(0-6)/Co(1.35)/IrMn(6)/Ta(1.5) wedge sample is used to study the Pt thickness (tPt) dependence. The result for an OOP field-cooled wedge is shown in figure 5.9.
OOP MOKE, OOP field cooled
H
ext3
1 5
,
Figure 5.9: Pt thickness dependence of the coercivity, exchange bias and reme-nance. Below a critical thickness of 1.3 nm, no clear values could be extracted from the MOKE loops and the values are set to zero. Beyond the critical thickness of 1.3 nm, HEB,HC and MR steadily increase for thicker Pt. This is most likely due to smoother growth which increase the surface interactions betwen Pt/Co and even be-tween Co/Irmn, hence the increase in exchange bias. The wiggle at tPt=3.8 nm is most likely due to contamination at the surface, which distorted the measurement.
Below a critical thickness of 1.3 nm, the Pt layer is insufficiently thick to induce enough OOP anisotropy (PMA) in the Co layer and does not result in a clear coercivity or exchange bias. Above this critical thickness, the coercivity, exchange bias and remanence steadily rise with increasing Pt thickness. This might seem surprising as the the induced anisotropy at the Pt/Co interface is only dependent on the surface area, not on volume. It might be the result of better Pt interface quality. The increase in exchange bias is even more surprising as the Pt layer does not interact with the IrMn layer. A possible explanation could be that the quality of the Co/IrMn surface is also enhanced, resulting in a reduced amount of defects that decrease the exchange bias magnitude.
According to the simulations in chapter 3 (figure 3.3), the optimal Pt thickness for spin Hall injection is 3 nm of Pt. At this thickness, sufficient coercivity, remanence and exchange bias are
visible in figure 5.9 (dashed line). Therefore, it is concluded that 3 nm of Pt is the optimal value for the switching experiments in chapter 6.
5.3.4 Ta seed & capping layer
The seed layer thickness dependence is investigated by growing multiple sheets with various Ta thicknesses. No wedge measurement is conducted as the critical properties of the stack already changed at a very thin layer of Ta, which made it hard to measure it in a wedged sample. The results of the OOP field-cooled sheets are shown in figure 5.10. As can be seen, even a very thin seed layer of 0.5 nm dramatically enhances the remanence and coercivity of the sample. Ta is known to grow extremely flat and with few defects when sputtered, improving the growth of any materials deposited on top of it [51][52]. An increase in Ta thickness above 0.5 nm does not change the coercivity or exchange bias.
OOP MOKE, OOP field cooled
H
extFigure 5.10: Easy-axis hysteresis loops for different Ta seed layer thickness. The samples are OOP field-cooled. A clear reduction of remanence and coercivity is seen if no seed layer is used. A very thin seed layer enhances the properties of the layer dramatically.
The capping layer for optical study of the samples is chosen to be exactly 1.5 nm of Ta. Previous studies have shown that this is the critical thickness at which the Ta fully oxidizes, and there-fore becomes completely transparent [53]. No notable difference in coercivity, exchange bias or remanence is found between 1.5 nm of Ta or a 2 nm Pt capping layer which is used for the non-optical and non-current related experiments.