Timescales and contribution of heating and helicity effect in helicity-dependent all-optical switching

The sample used in this study was a Pt (2 nm)/Co (0.6 nm)/Pt (2 nm) triple-layer grown on a 5 nm Ta buffer layer. The film was deposited on a Corning glass substrate with a thickness of 0.13 mm at room temperature by direct-current (DC) magnetron sputtering. The base pressure of the sputtering system was better than 4 × 10^–5 Pa. The Ar pressure during growth was 0.5 Pa. The sputtering rate with a DC of 40 mA was 0.041, 0.084 and 0.048 nm·s⁻¹ for Ta, Pt and Co, respectively. The Ta buffer layer was employed to improve the Pt/Co interface smoothness and the (111) orientation and enhance the Co layer’s perpendicular magnetic anisotropy. This is confirmed in the MOKE hysteresis loop shown in Fig. 2. Here Fig. 2a is measured by the MOKE image system used in the AOS experiments and Fig. 2b is measured by the vibrating-sample magnetometer (VSM). Both loops prove that the sample has a well-defined perpendicular magnetic anisotropy.

Figure 3 is a schematic diagram of the double pump experimental setup, a combination of time-resolved pump-probe and MOKE image systems. For the MOKE imaging system part, the light-emitting diode provides a 465 nm light to the system. The objective lens chosen in this experiment is HALO, and the numerical aperture (NA) is 0.38. Since such an objective and blue light were used for illumination, domains as narrow as 612 nm can be resolved. A Ti:sapphire laser amplifier system with 150 fs pulse duration, 800 nm central wavelength, and a 1000 Hz repetition rate was used. For the dual-pumping measurements, the pulse was split into two pulses. The first pump pulse was LP and used to heat the sample’s electron/spin systems. The second pump pulse was circularly polarized (CP), delayed with respect to the first LP pulse, and used to switch the sample’s magnetic state, as illustrated in Fig. 1a. The power of each pump beam was individually adjusted for the desired power combination. The two pump beams were combined at a beam splitter before focusing onto the Pt/Co/Pt triple-layer from the substrate side. The spot size was measured as 38 \(\mathrm{\upmu m}\) in diameter using a charge-coupled device (CCD) beam profiler which gave a laser fluence of \(8.83\times {10}^{-2}\) mJ·cm⁻² at an average laser power of \(1\mathrm{ \upmu W}\). The sample was mounted on a motorized 3-axis nanomax flexure stage. The magnetization of the sample was initially saturated along the perpendicular direction of the sample plane, defined as \({M}^{\uparrow }\) state. When the sample was exposed to the dual-pump beams, the stage was scanned over a 300 \(\mathrm{\upmu m}\) distance at a velocity of 10 μm·s⁻¹. After laser excitation, the magnetic domain state was recorded as a MOKE image via wide-field MOKE microscopy. The sample was then re-magnetized to the \({M}^{\uparrow }\) state, and a reference MOKE image was taken. The MOKE images presented are the subtractions of each pair of these images, where most of the effects from the surface morphology are eliminated.

HD-AOS induced by LP and CP pulse pairs with different power combinations at a fixed delay time, 1.6 ps, between them were investigated first. The energy transfer for the electron-spin system to the lattice system is mediated by the phonons, which takes about 1.6 ps to reach the spin-electron-lattice thermal equilibrium state [30‐32] with the system in a largely demagnetized state when the second CP pulse arrives. Figure 4a shows each subtracted MOKE image centered along the scanning path of the dual-pump beam with a field of view of \(60 \, \mathrm{ \upmu m}\times 380 \, \mathrm{ \upmu m}\). The horizontal axis shows the total power of the LP and CP pulse pairs increasing from 100 to 180 \(\mathrm{\upmu W}\), while the vertical axis gives the power of the CP pulse increasing from 10 \(\mathrm{\upmu W}\) to the total power of each column. The power of the LP pulse used for each image is the difference between the total power and the CP pulse power. The MOKE images, framed in blue dashed lines, exhibit AOS with only CP pulses incident. The switching ratio is calculated by dividing the total number of switched pixels by the total number of pixels within a 20 μm × 200 μm window in each image of Fig. 4a, resulting in switching ratio mapping, as shown in Fig. 4b. In Fig. 4a, the images of the first-row show that a random domain state is generated by 10 \(\mathrm{\upmu W}\) beam power combined with various power preheat LP. The switching ratio range of each image in the first row is 43% to 64% (as shown in Fig. 4b), which indicates a multidomain state caused by the low power of CP.

Once the CP beam power is increased to 20 \(\mathrm{\upmu W}\) and beyond, HD-AOS is observed with a total power window of 120 \(\mathrm{\upmu W}\)(\(10.6\) mJ·cm⁻²) to 160 \(\mathrm{\upmu W}\) (\(14.1\) mJ·cm⁻²). The switching ratio is above 80% across this whole power window. It indicates that the power window for pairs of LP and CP pulses is the same as for a single CP pulse when the second CP arrives at a 1.6 ps time delay. In the column of the 160 \(\mathrm{\upmu W}\) total power, multidomain patterns can be seen to emerge in the center of the laser beam path. When the total power is increased to 180 \(\mathrm{\upmu W}\), the multidomain pattern is created regardless of beam power combinations. Even though the switching ratio is over 60% under the 180 \(\mathrm{\upmu W}\) total beam power induced, the laser causes irreversible changes in the sample’s magnetic properties leading to increased coercivity of the exposed area. Therefore, 180 \(\mathrm{\upmu W}\) is excluded from the quoted power window of HD-AOS. The occurrence of the multidomain states under high power laser pumping is due to the laser heating the sample to sufficiently high levels to demagnetize the sample again after HD-AOS. As shown in the images framed in dashed red lines in Fig. 4a, the laser-swept area remains a uniformly switched magnetic domain even when the power of the CP beam is reduced to 20 \(\mathrm{\upmu W}\) with the samples preheated by the LP pulse. The corresponding switching ratios are all above 80%, as shown in Fig. 4b.

Therefore, the minimum power of the CP pulse required to achieve HD-AOS with LP preheated is 20 μW for our sample. When the sample is sufficiently demagnetized, circularly polarized illumination with a power threshold as low as 1.77 mJ·cm⁻², reduced by 80% compared to that without-preheating, is sufficient to achieve HD-AOS as demonstrated by the images framed in red dashed lines in Fig. 4a. This proves that laser heating plays an essential role in HD-AOS of the Pt/Co/Pt triple-layer, where only a single magnetic lattice exists, compared to HID-AOS in RE-FM alloy/multilayers [5, 6]. This discovery reveals that most of the required pulse energy is used to heat the spin system in a single-pump-induced HD-AOS event in the Pt/Co/Pt structure [21]. The helicity effect requires only a small portion of the power threshold.

As indicated in Fig. 1b, the delay time between the LP and CP on HD-AOS is a critical factor. To study this effect in detail, the delay time was set from 0 to 10 ps, with a step size of 0.2 ps for the first 2 ps and then 0.5 ps afterward. The CP beam power was increased from 20 to 100 \(\mathrm{\upmu W}\) with a step size of 10 \(\mathrm{\upmu W}\), while the LP was decreased from 100 to 20 \(\mathrm{\upmu W}\), so that the total power was fixed at 120 \(\mathrm{\upmu W}\), which was the minimum total laser power needed for HD-AOS in the sample. The switching ratio of HD-AOS was extracted for each MOKE image captured at every delay time, quantified via image processing using ImageJ [33], and plotted as a function of time delay in Fig. S5 (see Supporting Information for details). The MOKE images and the extracted switching ratio of two representative power combinations are displayed in Fig. 5. The interference of the two pump pulses at the zero-delay point induced a multidomain state, which leads to an approximate 50% switching ratio for every curve in Fig. 5c, d at the zero-delay point. With the delay time increasing, the switching ratio increases first and reaches its highest point, approximately 90%, when the time delay is about 1 ps for all the power combinations. However, after the initial rise, the switching ratio significantly differs in its dependence on the LP and CP time delay between these two power combinations. For the case of LP power 40 \(\mathrm{\upmu W }\,(3.53\)mJ·cm⁻²) and CP power 80 \(\mathrm{\upmu W}\), the switching ratio drops sharply when the time delay between the two pulses is longer than 2 ps. It decreases to less than 20% when the time delay is longer than 3 ps, as shown in Fig. 5c. On the other hand, for the case of LP power 80 \(\mathrm{\upmu W}\,(7.06\)mJ·cm⁻²) and CP power 40 \(\mathrm{\upmu W}\), the switching ratio stays at its highest value (~ 90%) for the time delay from 1 to 4 ps. When the time delay is longer than 4 ps, the switching ratio drops very slowly and gradually to around 60% up to 300 ps. When the time delay is longer than 300 ps, the switching rate drops dramatically below 50%, as shown in Fig. 5d. The gradual decrease in the switching rate between 4 and 300 ps delay is large due to the shrinking of the switching area. As shown in Fig. 5b, the width of the central black trace decreased with the delay time increase while the switching rate was calculated within a window of fixed size of 20 μm × 200 μm. These two different processes are also evidenced in their MOKE images at different time delays, as presented in Fig. 5a and b. In Fig. 5a, for the case of LP power of 40 μW, there is no sign of switching at 6 ps, while Fig. 5b shows a clear switching at the same delay time, but with a larger LP power of 80 μW.

Atomistic spin dynamics modelling combined with a two-temperature model has been employed to simulate the demagnetization rate and magnetization recovery after the LP pulse for both cases, i.e., of 40 and 80 \(\mathrm{\upmu W}\), and the results are superimposed on their switching rate curves in Fig. 5c and d, respectively. To simplify the analysis, we consider the spin system is coupled to the temperature profile of the electrons in the two-temperature model. It shows that the magnetization recovery (spin cooling) time after laser excitation increases from a couple of picoseconds to a few hundreds of picoseconds as the LP power increases from 40 to 80 \(\mathrm{\upmu W}\), which was also observed in the time-resolved MOKE measurements of the sample shown in Fig. S6. A red dotted horizontal line is drawn at the 50% switching ratio (left-hand y-axis) in both Fig. 5c and d. In Fig. 5c, HD-AOS occurs when the 80 \({\upmu \text{W}}\) CP pulse arrives within a time delay between 0.2 and 2 ps, shaded in grey. In Fig. 5d, HD-AOS occurs when the 40 \(\mathrm{\mu W}\) CP pulse arrives within a time delay from 0.2 to over 300 ps, also shaded in grey. Comparing the sample magnetic states within these two grey areas where HD-AOS is achieved, one can see a common minimum of the sample magnetization of around 60% of the saturation magnetization. This indicates that the demagnetized state upon which the CP pulse impinges is a key factor in achieving HD-AOS.

The relationship between HD-AOS switching ratio and laser ellipticity was further investigated by using a single pump, and the results are included in Fig. 6. The light ellipticity is tuned by the quarter-wave plate and the number on the left of Fig. 6a indicates the quarter wave plate’s angle. Here \(45^\circ\) gives the right-handed circularly polarized (RCP) light, and \(0^\circ\) gives the LP light. The laser power is fixed at \(120 \, \mathrm{ \upmu W}\) in this experiment. The elliptically polarized light can be regarded as a combination of LP and CP. The switching ratio was found to decrease as the laser polarization changed from circular to linear. This is consistent with a previous finding on laser-induced domain wall motion, where wall displacement decreases as laser polarization changes from circular to linear [16]. For these single pulse cases, the LP and CP photons arrived simultaneously, and the heating effect from the LP photons lags; thus, the CP photons fail to achieve HD-AOS. This can account for the fact that HD-AOS has not been observed in a wider range of material systems because, generally, ultrafast laser heating effects lag behind its helicity effect. This also explains a previous observation that a longer laser pulse duration gives a higher switching ratio under the same laser power AOS [34]. As shown in Fig. S6, the higher laser fluence takes longer to reach the highest demagnetized state as pointed out previously [35]. With the dual-pump laser pulses, we expect that HD-AOS would occur in many other magnetic materials where the transient magnetization states needed for the CP lighted driven HD-AOS can be achieved by controlling the strength of the LP pulse and the delay time.

The essential role of heating in HD-AOS alone cannot explain the dramatically different time-delay dependence between the above two cases since the second CP pulse also heats the sample, which reduces the magnetization. As shown in Fig. 7a, the first LP pulse increases both \({M}^{\uparrow }\) and \({M}^{\downarrow }\) domains’ temperature but not close to the Curie temperature (\({T}_{\mathrm{c}}\)). While due to the MCD effect, the second CP pulse heats the \({M}^{\uparrow }\) domain’s temperature closer to \({T}_{\mathrm{c}}\). Once it switched to \({M}^{\downarrow }\) domain, it absorbs fewer CP photons and makes its temperature away from \({T}_{\mathrm{c}}\). However, this phenomenological mechanism should not be sensitive to the time sequence of the CP and LP pulses. As shown in Fig. 7b, if the CP pulse arrived first, it still gives a temperature difference of \({T}_{1}^{\uparrow }\) and\({T}_{1}^{\downarrow }\). And then, the second LP pulse increases both domains’ temperature but keeps the temperature difference. Then the HD-AOS still can happen. Also, in Fig. 7c, the elliptically polarized pulse can be regarded as a mixture of CP and LP pulses. Based on the MCD effect, it still induces a temperature difference between \({M}^{\uparrow }\) and \({M}^{\downarrow }\) domains. However, the \(60 \, \mathrm{ \upmu W \, LP}+60 \mathrm{ \,\upmu w \,CP}\) (LP arrived first) gives the same AOS ratio as the 120 \(\mathrm{\upmu W}\) CP pulse, but if they arrived simultaneously (elliptically polarized pulse), the switching ratio is lower than the formers. In contrast, the AOS is not observed if the CP pulse arrived earlier than the LP pulse. Moreover, in our double pump experiments (especially the \(100 \, \mathrm{ \upmu W}\) LP + \(20 \, \mathrm{ \upmu W}\) CP), the temperature difference between these two domains is much smaller than using a single \(120\,\mathrm{ \upmu W}\) CP pump, while these two experiments give a nearly same switching ratio. Here, the spin heat capacity (\({C}_{\mathrm{s}}\)) change with the spin system’s temperature (\({T}_{\mathrm{s}}\)) was not considered. MCD is introduced to explain the laser-induced domain wall motion [16]. We would like to emphasize that several more effects should be considered to fully explain our results based on the MCD mechanism.

To clarify this picture, atomistic spin dynamics modelling combined with a two-temperature model has been applied again to simulate the demagnetization rate and magnetization recovering excited by both the LP and CP pulses. Figure 8a and b shows the case when two pulses are 5 ps apart. A red dashed horizontal line is drawn at 60% of the saturation magnetization value in Fig. 8, which is the maximum magnetization observed in the two grey areas in Fig. 5. The spin-flip energy barriers related to the spin temperatures are also added and represented by \({E}_{\mathrm{fa}}^{t}\) and \({E}_{\mathrm{fb}}^{t}\), respectively, where t represents the delay time. To identify the lag between the helicity and heating effects, we show the corresponding \({E}_{\mathrm{f}}^{t}\) at four-time points, namely when: the first pulse arrival time (\(t=0\)), the first demagnetization peak occurs (\(t=1\,\mathrm{ ps}\)), the second pulse arrival time (\(t=5\,\mathrm{ ps}\)), and when the second demagnetization peak occurs (\(t=6\,\mathrm{ ps}\)). It can be seen that in the case of Fig. 8a with the LP 80 \({\upmu \rm W}\) + CP 40 \({\upmu \rm W}\) pulse combination, at 5 ps delay time, the sample’s magnetization has only recovered 20%. This corresponds to a lower energy barrier \({E}_{\mathrm{fb}}^{5}\) between \({M}^{\uparrow }\) and \({M}^{\downarrow }\) states upon the arrival of the 40 \({\upmu \rm W}\) CP pulse, and HD-AOS takes place in this case. In the case of Fig. 8b with LP 40 \({\upmu \rm W}\)+ CP 80 \(\upmu \text{W}\) pulse pair, at 5 ps delay time, the sample magnetization recovers to around 80% of its saturation value. This corresponds to a high energy barrier \({E}_{\mathrm{fa}}^{5}\) between \({M}^{\uparrow }\) and \({M}^{\downarrow }\) states upon the arrival of the 80 \({\upmu \rm W}\) CP pulse, and HD-AOS doesn’t take place. This is because the CP pulse’s heating effect takes more than 0.3 ps to demagnetize the sample’s magnetization to its second peak. During the CP pulse duration, the sample’s magnetization is only reduced to 70% of its saturation value, as marked by two short vertical dash lines in Fig. 8b. The only explanation for this observation is that the onset time of the helicity effect from the CP pulse is instant, and the duration of the helicity effect is less than 200 fs which is close to the laser pulse width of 150 fs. Even though the energy barrier \({E}_{\mathrm{fb}}^{6}\) is reduced further by the heating effect of the CP pulse itself, the helicity effect has already disappeared at this point, and HD-AOS cannot be triggered anymore.