1: \documentclass[preprint,preprintnumbers,showpacs,amsmath,amssymb]{revtex4}
2: \usepackage{epsfig}
3: \begin{document}
4:
5: \title{Magnetization damping in polycrystalline Co ultra-thin films:
6: Evidence for non-local effects}
7: \author{J-M. L. Beaujour, J. H. Lee, A. D. Kent} \affiliation{Department of Physics, New York University, 4
8: Washington Place, New York, NY 10003, USA}
9: \author{K. Krycka and C-C. Kao} \affiliation{Brookhaven National Laboratory, Upton, New York 11973, USA}
10: \date{\today}
11:
12: \begin{abstract}
13: The magnetic properties and magnetization dynamics of
14: polycrystalline ultra-thin Co layers were investigated using a
15: broadband ferromagnetic resonance (FMR) technique at room
16: temperature. A variable thickness (1 nm $\leq t \leq$ 10 nm) Co
17: layer is sandwiched between 10 nm thick Cu layers (10 nm Cu$|t$
18: Co$|$10 nm Cu), while materials in contact with the Cu outer
19: interfaces are varied to determine their influence on the
20: magnetization damping. The resonance field and the linewidth were
21: studied for in-plane magnetic fields in field swept experiments at
22: a fixed frequency, from 4 to 25 GHz. The Co layers have a lower
23: magnetization density than the bulk, and an interface contribution
24: to the magnetic anisotropy normal to the film plane. The Gilbert
25: damping, as determined from the frequency dependence of the
26: linewidth, increases with decreasing Co layer thickness for films
27: with outer Pt layers. This enhancement is not observed in
28: structures without Pt layers. The result can be understood in
29: terms of a non-local contribution to the damping due to spin
30: pumping from Co through the Cu layer and spin relaxation in Pt
31: layers. Pt layers just 1.5 nm thick are found to be sufficient to
32: enhance the damping and thus act as efficient ``spin-sinks.'' In
33: structures with Pt outer layers, this non-local contribution to
34: the damping becomes predominant when the Co layer is thinner than
35: 4 nm.
36: \end{abstract}
37:
38: \maketitle
39:
40: \section{introduction}
41: The magnetization dynamics of ultra-thin magnetic layers ($<$10
42: nm) is of great scientific and technological interest, as such
43: layers are widely used in spin-injection and transport studies as
44: well as in magnetic devices. There has been particular interest in
45: non-local effects whereby layers separated from a magnetic layer
46: influence its magnetization dynamics through non-magnetic (NM)
47: metallic contact layers, i.e., via conduction electrons. Such
48: effects were modeled and studied in the early 1970's by Silsbee
49: \emph{et al.} \cite{silsbee}. More recently, a scattering theory
50: approach has been employed to describe the enhancement of the
51: damping \cite{brataas1,brataas2}. There have also been experiments
52: \cite{mizukami,heinrich,foros} which indicate quantitative
53: agreement with this theory based on interface parameters that can
54: be determined from \emph{ab-initio} calculations \cite{xia} as
55: well as transport experiments \cite{bauer}.
56:
57: The current interest in this mechanism is at least threefold.
58: First, it is a fundamental mechanism of damping that can provide important
59: information on interface and bulk spin diffusion. Second, the
60: effect is known to play an important role in current-induced
61: magnetization excitations in spin-transfer devices \cite{sun}. In
62: such devices, a few nanometer thick magnetic layer is
63: embedded between NM layers which separate it from a thick
64: ferromagnetic layer that sets the spin-polarization of the current
65: \cite{nanopillar}. The threshold current density for magnetic
66: excitations is proportional to the damping \cite{sunPRB}. In order to understand
67: the physics of spin transfer, it is therefore important to
68: investigate the effect of adjacent NM layers on the magnetic
69: relaxation of ultra-thin films. Finally, from a technological
70: point of view, this process provides a way to engineer the
71: damping, which is important for high speed magneto-electronic
72: devices.
73:
74: Mizukami \textit{et al.} studied the Gilbert damping of sputtered
75: NM$|t$ Py$|$NM films as a function of the FM layer thickness
76: ($2\leq t \leq10$ nm) and for different adjacent non-magnetic
77: metals NM=Cu, Pd and Pt using a X-band FMR technique
78: \cite{mizukami}. The damping was found to be consistent with the
79: spin pumping picture: increasing with decreasing Py thickness for
80: the films with NM=Pt and Pd only. Further, the magnetization
81: damping of Cu$|$Py$|L$ Cu$|$Pt structures as a function of Cu
82: layer thickness $L$ and with fixed Py thickness, showed evidence
83: for a non-local effect. However, the non-local damping has been
84: studied mainly in NM$|t$ FM$|$NM structures as a function of the
85: FM layer thickness and varying the material directly in contact
86: with the FM layer \cite{mizukami, foros}. There have also been no
87: experimental studies of polycrystalline Co layers, which are
88: widely used in spin-transfer devices.
89:
90: In this paper, we report systematic studies of the thickness
91: dependence of the linewidth and Gilbert damping of
92: ultra-thin Co layers in $||y_1$ Pt$|$Cu$|t$ Co$|$Cu$|y_2$ Pt$||$
93: structures. The thickness of the Cu layers in contact with Co is
94: kept fixed at 10 nm, which is chosen to be less than the spin-diffusion
95: length in Cu. The Pt layers have no direct interface with
96: the FM layer. The structure is modified by removing one or both of
97: the Pt layers ($y_1$ or $y_2=0$). The observation of changes in the
98: Gilbert damping confirm the non-local nature of this damping
99: mechanism and the data allow for quantitative analysis of the
100: interface spin-mixing conductances in the scattering theory
101: approach.
102:
103: The paper is organized as follows. In section II, the film
104: fabrication and the FMR setup are described. Section III explains
105: the method of analysis of the resonance field and linewidth. In
106: section IV, the resonance field and the effective demagnetization
107: field are studied as a function of Co layer thickness. The FMR
108: linewidth and the Gilbert damping data are presented. This is
109: followed by a discussion of the dependence of the Gilbert damping
110: on the Co layer thickness for films with and without Pt and a
111: quantitative analysis of the data.
112:
113: \section{Experimental technique}
114: Four series of films were fabricated with the layer structure
115: $||y_1$ Pt$|$10 nm Cu $|t$ Co$|$10 nm Cu$|y_2$ Pt$||$ on GaAs
116: substrates. First, symmetric structures with a variable thickness
117: of Co between Pt and Cu layers were grown with $y_1=y_2=1.5$ nm
118: and $1 \leq t \leq 10$ nm. The second set of samples were
119: asymmetric, without the top Pt layer ($y_1=1.5$ nm and $y_2=0$),
120: and with Co of 1.5, 2 and 2.5 nm thickness. In addition, films
121: without Pt layers ($y_1=y_2=0$), with Co of 2 and 3 nm thickness
122: were fabricated. In the fourth and final set of films, the
123: thickness of the Pt layers was varied from 0 to 5 nm in a
124: symmetric way with $y_1=y_2=0$, 1.5, 3 and 5 nm with a fixed Co
125: layer thickness of $t=2$ nm. Note that the studies in which Pt
126: layer thicknesses were varied focused on thin Co layers because
127: the non-local effect on the damping were found to be
128: significant only in Co layers thinner than 4 nm.
129:
130: Films were prepared by a combination of electron-beam (Co, Pt) and
131: thermal (Cu) evaporation in an UHV system at a base pressure of $5
132: \times 10^{-8}$ Torr on polished semi-insulating $4 \times 6$
133: mm$^2$ GaAs wafers. Note that the chamber is equipped of an
134: in-situ wedge growth mechanism that enables the fabrication of a
135: number of samples with different Co or Pt thickness in the same
136: deposition run, without breaking vacuum. The evaporation rate for
137: Co and Cu were 0.5 and 0.8 \AA /sec respectively.
138:
139: \begin{figure}
140: \begin{center}\includegraphics[width=8cm]{fig0.eps}
141: \vspace{0 mm} \caption{X-ray data of $||$Pt$|$Cu$|t$
142: Co$|$Cu$|$Pt$||$ films with $t=$1.5, 3 and 6 nm. The curves for
143: the films with 3 and 6 nm Co layer were shifted up for clarity.
144: The vertical solid, dashed and dotted lines show the Q for nominal
145: fcc Pt, Cu and Co respectively.} \vspace{-5 mm}
146: \end{center}
147: \end{figure}
148:
149: X-ray measurements were conducted on $||$Pt$|$Cu$|t$
150: Co$|$Cu$|$Pt$||$ with $t=1.5$, $3$ and $6$ nm (Fig. 1). The
151: reflectivity scans were taken at 19 keV using a Bicron point
152: detector. The intensity was first normalized by a beam-monitoring
153: ion chamber, and then each scan was rescaled to reach a maximum of
154: one. The films are polycrystalline and Pt, Cu and Co have an fcc
155: structure with (111) as the crystallographic growth direction.
156: Note that the peak at Q$\sim 3$ \AA$^{-1}$ shifts slightly between
157: the (111) nominal values for Co and Cu depending on how much Co is
158: present. An undistorted bcc (the most intense peak of 110 for Co
159: would be at 3.15 \AA$^{-1}$) is clearly not present, but a
160: distorted bcc with a lattice of 2.8 \AA \ cannot be ruled out.
161: Note that the peak of intensity at $Q\approx2.79$ \AA \ from the
162: Pt layer is strong evidence against interdiffusion at the Pt$|$Cu
163: interface.
164:
165: The surface topography of the films $||$Pt$|$Cu$|$Co$|$Cu$||$ and
166: $||$Cu$|$Co$|$Cu$||$ were studied using non-contact Atomic Force
167: Microscopy (AFM) (Fig. 2). From the analysis of the AFM images, a
168: rms roughness of $\sigma=1.4\pm0.1$ nm was found for films with Pt
169: underlayer and $\sigma=1.3\pm0.2$ nm for films without Pt. The
170: lateral correlation length were $\xi=19.6\pm1.7$ nm and
171: $\xi=18.0\pm0.4$ nm respectively. There is no clear difference in
172: the rms roughness and the correlation length of the film with and
173: without Pt, suggesting that 1.5 nm Pt does not alter the growth of
174: the subsequent Cu$|$Co$|$Cu trilayers. The rms roughness is about
175: $6\%$ of the film thickness, which is typical of films prepared by
176: evaporation.
177:
178: \begin{figure}
179: \begin{center}\includegraphics[width=8cm]{fig1B.eps}
180: \vspace{0 mm} \caption{Grey scale images of AFM scans performed on
181: 1 $\mu m ^2$ areas of a magnetic film (a) $||$Cu$|$Co$|$Cu$||$ and
182: (b) $||$Pt$|$Cu$|$Co$|$Cu$||$ with a Co layer of 2 nm thickness.
183: The rms values of surface roughness is $\sigma=1.3\pm0.2$ nm for
184: (a) and $\sigma=1.4\pm0.1$ nm for (b), and the lateral correlation
185: length is $\xi=18.0\pm0.4$ nm and $\xi=19.6\pm1.7$ nm
186: respectively.} \vspace{-5 mm}
187: \end{center}
188: \end{figure}
189:
190: FMR measurements were conducted in an in-plane field geometry at
191: room temperature employing a coplanar waveguide (CPW) as an ac
192: magnetic field generator and inductive sensor \cite{barry}. The
193: CPW was fabricated on a 350 $\mu$m thick semi-insulating and
194: polished GaAs wafer from a 200 nm thick Au film patterned using
195: bi-layer photolithographic process. It is characterized by a
196: transmission line of 50 $\mu$m width, a gap to the ground plate of
197: 32 $\mu$m and a length of 4 mm, which is designed to have 50
198: $\Omega$ impedance above 4 GHz. The CPW was placed into a brass
199: cavity, with its ends connected directly to the ports of a Network
200: Analyzer. Care was taken to avoid magnetic components in the
201: cavity and in all contacts to the CPW. FMR spectra were measured
202: by placing the magnetic sample metal face down on the CPW and
203: sweeping the external magnetic field at fixed microwave frequency
204: while measuring the S-parameters of the transmission line. Our
205: setup enables measurement of the FMR response of Co layers as thin
206: as 1 nm. Fig. 3a shows the geometry of the measurements. The
207: applied field produced by an electromagnet is directed along the
208: axis of the transmission line and perpendicular to the ac magnetic
209: field generated by the CPW. The applied field was in the film
210: plane, and was monitored with a Hall probe sensor that was
211: calibrated using electron paramagnetic resonance (EPR) on
212: 2,2-diphenyl-1-picrylhydrazyl (dpph), a spin 1/2 system. The
213: measured absorption of dpph is shown in Fig. 3b. The resonance
214: fields were always in agreement with the readings from the
215: Gaussmeter within 10-15 Gauss.
216:
217: The FMR response was recorded at different frequencies in the
218: range 4-25 GHz. The spectra is measured as the relative change in
219: the transmitted power versus applied field. At 13 GHz for example,
220: the absorption from 5 nm thick Co film at resonance leads to a
221: 0.66$\%$ decrease in the transmission. Thus, the susceptibility of
222: the magnetic films only causes a small change in the impedance of
223: the CPW and the absorption line can be analyzed as a small
224: perturbation to the CPW transmission.
225: \begin{figure}
226: \begin{center}\includegraphics[width=7cm]{fig2.eps}
227: \vspace{-2 mm} \caption{a) The FMR setup and field geometry. b)
228: Absorption of dpph at different frequencies. The resonance field
229: $H_{\text{res}}$ depends linearly on frequency: $H_{\text{res}}=(2
230: \pi \hbar / g \mu _B)f$, and is used to verify the calibration of
231: a Hall sensor used in this study.} \vspace{-5 mm}
232: \end{center}
233: \end{figure}
234:
235: \section{method of analysis of the resonance field and linewidth}
236: The dynamics of the magnetization can be described in the
237: classical limit by the Landau-Lifshitz equation of motion:
238: \begin{equation}
239: \label{eq.1} {{1}\over{\gamma}}{{\partial \vec{M}}\over{\partial
240: t}}=\vec{M}\times \vec{H}_{\text{eff}} + {{\text{G}}\over{\gamma
241: ^2 M_{\text{s}}^2}} \vec{M} \times {{\partial
242: \vec{M}}\over{\partial t}} \;,
243: \end{equation}
244: where $\vec{H}_{\text{eff}}$ is the effective field, $\vec{M}$ is
245: the magnetization vector and $\text{G}$ is the Gilbert damping
246: constant. The gyromagnetic ratio, $\gamma = g \mu _B / \hbar$, is
247: proportional to $g$, the Land\'{e} gyroscopic factor. For a film
248: magnetized in the film plane in an ac field, the resonance
249: condition is \cite{kittel}:
250: \begin{equation}
251: \label{eq.2} \left({{2 \pi f}\over{\gamma}} \right)^2 =
252: H_{\text{res}} \left(H_{\text{res}}+4 \pi M_{\text{eff}}
253: \right)\;,
254: \end{equation}
255: where for a continuous film, the effective demagnetization field
256: is given by:
257: \begin{equation}
258: \label{eq.3} 4 \pi M_{\text{eff}} = 4 \pi M_{\text{s}} + {{2
259: K_{\text{s}}}\over{M_{\text{s}}\ t}}\;.
260: \end{equation}
261: $M_{\text{s}}$ is the saturation magnetization. The uniaxial
262: anisotropy field $H_{\text{s}}=2K_{\text{s}} / M_{\text{s}}t$ is
263: characterized by a 1/t thickness dependence, where the anisotropy
264: originates from interface and/or strain-magnetoelastic
265: interactions. If $K_{\text{s}}$, the uniaxial anisotropy constant,
266: is negative, $H_\text{s}$ is directed out-of the film plane,
267: corresponding to a perpendicular component to the magnetic
268: anisotropy. Note that by assuming $M_{\text{s}}$ is independent of
269: thickness the uniaxial anisotropy constant, extracted from $4 \pi
270: M_{\text{eff}}$ versus $t$, can be overestimated.
271:
272: The Gilbert damping is determined by the frequency dependence of
273: the FMR linewidth $\Delta H$ \cite{inhomogeneity}:
274: \begin{equation}
275: \label{eq.4} \Delta H(f) = \Delta H_0 + {{4 \pi
276: G}\over{\gamma ^2 M_{\text{s}}}}f\;,
277: \end{equation}
278: where the slope of $\Delta H(f)$ is the intrinsic contribution to
279: the linewidth, and is proportional to the Gilbert damping constant
280: $\text{G}$. $\Delta H_0$, the zero-frequency intercept, is usually
281: considered to be an extrinsic contribution to the linewidth.
282: $\Delta H_0$ is sensitive to the film quality: the highest quality
283: films typically exhibit a smallest residual or zero field
284: linewidth \cite{inhomogeneity, celinski}.
285:
286: \section{Resonance field}
287: \begin{figure}
288: \begin{center}\includegraphics[width=7cm]{fig3.eps}
289: \vspace{-2 mm}\caption{Typical absorption line at 13 GHz for
290: $||$Pt$|$Cu$|t$ Co$|$Cu$|$Pt$||$ film with $t=$10, 3 and 1.5 nm. With decreasing Co thickness
291: the absorption line shifts to higher field and broadens.}
292: \end{center}
293: \end{figure}
294: Fig. 4 presents the normalized FMR peak at 13 GHz for a selection
295: of $||$Pt$|$Cu$|t$ Co $|$Cu$|$Pt$||$ films. The absorption lines
296: were normalized by substracting the background signal and dividing
297: by the relative change in transmission at resonance. The lineshape
298: of the FMR curves is typically Lorentzian. Note that we observed
299: asymmetric lineshapes at some frequencies, associated with the
300: measurement method. Distortion of the FMR lineshapes originates
301: from the mixing of the absorptive and dispersive components of the
302: susceptibility (see Ref. \cite{kalarickal}). Fig. 5 shows the
303: thickness dependence of the resonance field $H_{\text{res}}$ at 14
304: GHz for films with and without 1.5 nm Pt. The resonance field is
305: practically constant ($H_{\text{res}} \approx 1.2$ kOe) when the
306: Co layer thickness is larger than 5 nm. For thinner layers,
307: $H_{\text{res}}$ increases with decreasing $t$, reaching a value
308: of 1.6 kOe for the film with 1 nm Co layer. The films with the
309: same thickness of Co exhibit about the same resonance field,
310: independent of the presence of the Pt layers. The run-to-run
311: variations in magnetic properties from small changes in film
312: deposition conditions, for instance, are significantly less than
313: the changes in $H_{\text{res}}$ observed in the very thin film
314: limit.
315:
316: \begin{figure}
317: \begin{center}\includegraphics[width=7cm]{fig4.eps}
318: \vspace{-2 mm}\caption{Thickness dependence of the resonance field
319: at 14 GHz.}
320: \end{center}
321: \end{figure}
322:
323: The resonance field was measured at different frequencies (Fig.
324: 6). The effective demagnetization, $4 \pi M_{\text{eff}}$, and the
325: g-factor were found by fitting $f^2/H_{\text{res}}$ $vs.$
326: $H_{\text{res}}$ to Eq. 2. The slope of $f^2/H_{\text{res}}$ gives
327: the g-factor and the zero frequency intercept provides the
328: effective demagnetization field (see the inset of Fig. 6). Films
329: with equal Co layer thickness have nearly the same $g$ and $4 \pi
330: M_{\text{eff}}$. This suggests that the Pt underlayer and capping
331: layer do not affect the magnetic properties of the films.
332: \begin{figure}
333: \begin{center}\includegraphics[width=7cm]{fig5.eps}
334: \vspace{-2 mm}\caption{The frequency dependence of the resonance
335: field for a $||$Pt$|$Cu$|t$ 3 nm Co $|$Cu$|$Pt$||$ film. The inset
336: shows $f^2/H_{\text{res}}$ $vs.$ $H_{\text{res}}$.}
337: \end{center}
338: \end{figure}
339: The effective field exhibits a clear thickness dependence,
340: decreasing from about 15 kOe to about 8 kOe when the Co thickness
341: varies from 10 nm to 1 nm (Fig. 7). The thickness dependence of
342: the effective field can be understood as originating from the
343: presence of an uniaxial anisotropy field that depends on thickness
344: as 1/$t$. The magnetization density is assumed to be independent
345: of the film thickness, as Co and Cu are immiscible. The best fit
346: to Eq. 3 gives a saturation magnetization density of
347: $M_{\text{s}}=1131$ emu/cm$^3$ and an uniaxial anisotropy constant
348: $K_{\text{s}}=-0.46$ erg/cm$^2$. The value of $M_{\text{s}}$ is
349: smaller than the magnetization density of bulk fcc Co
350: ($M_{\text{s}}$=1400 emu/cm$^3$). The negative sign of
351: $K_{\text{s}}$ reflects a perpendicular component of magnetic
352: anisotropy. As noted earlier, the origin of this perpendicular
353: anisotropy may be a magnetic anisotropy associated with Co$|$Cu
354: interfaces or magnetoelastic interactions associated with strain
355: in the Co layers, which increases with decreasing Co thickness.
356: The results are in good agreement with magnetometry studies of 150
357: nm thick epitaxial Co$|$Cu multilayers grown on GaAs substrates,
358: with Co layer thickness ranging from 0.5 to 4 nm \cite{ohandley}.
359: The authors found an average magnetization density
360: $M_{\text{s}}$=1241 emu/cm$^3$ and an anisotropy constant
361: $K_{\text{s}}=-0.47$ erg/cm$^2$. The weighted average value of the
362: Land\'{e} g-factor is $g=$2.49$\pm$0.14 (inset of Fig. 7). This is
363: larger than the value reported in the literature for fcc phase Co
364: ($g$=2.15) \cite{wiedwald}. Note that in thin layers there is a
365: larger uncertainty in $g$ making it difficult to determine the
366: thickness dependence of the g-factor.
367:
368: \begin{figure}
369: \begin{center}\includegraphics[width=7cm]{fig6.eps}
370: \vspace{-2 mm}\caption{The effective demagnetization field $4 \pi
371: M_{\text{eff}}$ $vs.$ 1/$t$. The solid line is the best fit based
372: on Eq. 3. The inset shows the thickness dependence of the
373: Land\'{e} $g$ factor. The dashed line gives the value of $g$ for
374: fcc bulk (2.15).}
375: \end{center}
376: \end{figure}
377:
378: \section{Linewidth and Gilbert damping}
379: The linewidth was studied as a function of frequency and Co layer
380: thickness. As mentioned earlier (section IV), we observed
381: asymmetry in the FMR lineshape at some frequencies. In those
382: cases, the half power linewidth was extracted using the procedure
383: described in Ref. \cite{kalarickal}. The linewidth at 10 and 14
384: GHz is plotted versus the thickness in Fig. 8 for the films with
385: and without Pt. For thick films ($t\geq 5$ nm) with two Pt outer
386: layers, $\Delta H$ is practically independent of thickness.
387: However, the linewidth of thinner films ($t < 5$ nm) increases
388: strongly with decreasing Co layer thickness. The three series of
389: films show different increases of the linewidth for thin layers.
390: For instance, the film with 2 nm Co layer thickness and no Pt has
391: a linewidth of 120 Oe at 10 GHz. This is smaller than that of the
392: film with a Pt underlayer ($\Delta H$=180 Oe), and that with two
393: Pt layers ($\Delta H$=260 Oe).
394: \begin{figure}
395: \begin{center}\includegraphics[width=7cm]{fig7.eps}
396: \caption{Thickness dependence of the linewidth at 10 and 14 GHz
397: for the films of series without and with Pt. The insets show the
398: corresponding data on a log-log scale.}
399: \end{center}
400: \end{figure}
401: \begin{figure}
402: \begin{center}\includegraphics[width=7cm]{fig8.eps}
403: \caption{Typical frequency dependence of the linewidth for the
404: film $||$Pt$|$Cu$|$Co$|$Cu$|$Pt$||$ with Co layer thickness of 3
405: nm. $\Delta H_0$ and $d \Delta H / df$ are extracted from
406: the linear best fit (solid line).}
407: \end{center}
408: \end{figure}
409: The frequency dependence of $\Delta H$ enables a determination of
410: the Gilbert damping and the inhomogeneous contribution to the
411: linewidth. Fig. 9 shows the frequency dependence of $\Delta H$ for
412: a symmetric film with Pt layers ($y_1=y_2=1.5$ nm) and a 3 nm
413: thick layer of Co. The linewidth depends linearly on frequency,
414: with a zero frequency offset. Below 10 GHz, the data points are
415: more scattered. In this frequency range, the resonance field is of
416: the order of or smaller than the film saturation field, and the
417: absorption line becomes asymmetric. A linear fit to the data is
418: shown in Fig. 9. The thickness dependence of the slope $d \Delta H
419: / df$ and intercept $\Delta H_0$ are shown in Fig. 10. The two
420: parameters exhibit similar thickness dependence: decreasing with
421: increasing Co layer thickness. Therefore, both changes in the
422: Gilbert damping and inhomogeneous broadening contribute to the
423: enhanced linewidth for thin layers. $\Delta H_0$ approaches zero
424: for thick Co reflecting the good quality of these layers. The
425: slope is also practically constant when $t \geq 5$ nm, and it is
426: about three times larger for the thinnest films with two Pt
427: layers.
428: \begin{figure}
429: \begin{center}\includegraphics[width=7cm]{fig9.eps}
430: \caption{Thickness dependence of the two contributions to the
431: linewidth: $\Delta H_0$ and $d \Delta H / df$}
432: \end{center}
433: \end{figure}
434:
435: The Gilbert damping G was estimated from $d \Delta H / df$ (Fig.
436: 10) with Eq. 4. The thickness dependence of the damping is shown
437: in Fig. 11 for the films with and without Pt layers. The damping
438: of $||$Cu$|$Co$|$Cu$||$ films is practically independent of the
439: thickness, and it is about equal to the damping of the films with
440: two Pt layers and thick Co layers ($t>5$ nm). Note, however, that
441: the inhomogeneous contribution to the linewidth increases with
442: decreasing Co thickness (Fig. 10). The films with one or two Pt
443: layers show an increase in the damping when the Co layer is
444: thinner than 4 nm. 1.5 nm thick Pt layers are thus sufficient to
445: lead to the enhancement of the damping.
446:
447: In order to investigate whether the damping is a function of the
448: Pt outer layers, films with 2 nm thick Co layers and variable Pt
449: layer thickness were studied. The film structure was $||y$
450: Pt$|$Cu$|$2 nm Co$|$Cu$|y$ Pt$||$. The results are shown in the
451: inset of Fig. 11. The films without Pt have two times smaller
452: damping than the films with Pt. However, remarkably, the damping
453: does not increase further as the Pt layer thickness is increased
454: beyond 1.5 nm. It saturates at a value of about $6 \times 10^8$
455: s$^{-1}$. This result clearly shows that the main origin of the
456: enhancement of the damping in thin Co layers are the Cu$|$Pt
457: interfaces. Note that the films of this series, including the ones
458: with $y=0$ and $y=1.5$ nm were fabricated a few months after the
459: other series of films. It can be seen that the Gilbert damping of
460: the samples with $y=0$ and $y=1.5$ nm, plotted in the inset, are
461: consistent with the films deposited earlier (shown in the main
462: part of Fig. 11).
463: \begin{figure}
464: \begin{center}\includegraphics[width=7cm]{fig10.eps}
465: \caption{The magnetic damping versus ferromagnetic layer
466: thickness. The solid line is the best fit of Eq. 5 to G($t$) for
467: the films with two Pt layers. The dashed grey line is the
468: calculated thickness dependence of G for the films with a single
469: Pt$|$Cu interface, and using the parameters found in the earlier
470: fitting. Inset: the magnetic damping versus the thickness of the
471: Pt layers for the films $||y$ Pt$|$Cu$|$Co$|$Cu$|y$ Pt$||$ with 2
472: nm Co layer.}
473: \end{center}
474: \end{figure}
475:
476: \section{Discussion and interpretation}
477: There are a number of mechanisms which lead to the broadening of FMR linewidth
478: in very thin magnetic films. In the present study, particularly from the effect of
479: ``remote'' Pt layers on the linewidth, it is clear that non-local interactions are important. We
480: discuss this in detail below and follow this with a brief discussion of two-magnon
481: scattering, which has been extensively discussed in the context of damping in ultra-thin films.
482:
483: The dependence of $\text{G}$ on the thickness of the FM and on the
484: presence of the Pt can be understood in term of spin pumping
485: induced enhancement of the magnetic damping. In this model, the
486: precessing magnetization of the thin Co layer generates a spin
487: current that flows through the Cu layers. In absence of Pt
488: ($||$Cu$|$Co$|$Cu$||$ system), there is not a significant
489: additional damping with decreasing Co layer thickness because the
490: Cu layers are 10 nm thick, which is much thinner than the spin
491: diffusion length in Cu ($\lambda _{\text{sd}} \approx$ 350 nm at
492: room temperature \cite{jedema}). When the Cu layer is adjacent to
493: Pt, a strong spin scatterer, the spin current is absorbed in Pt.
494: There is no back flow of spin current. As a consequence, the
495: Gilbert damping of the FM is enhanced. For a symmetric film with
496: two Pt layers, the effect is more pronounced because the pumping
497: current is absorbed at the two interfaces with Pt. Furthermore,
498: above 1.5 nm Pt, the Gilbert damping does not depend on the
499: thickness of Pt. In assuming that the film is continuous, a Pt
500: layer as thin as 1.5 nm is thus sufficient to act as a perfect
501: spin sink, i.e., the entire spin current is absorbed. This is in
502: agreement with the results in Ref. \cite{bass}, where it was
503: reported that the spin loss parameter at the Pt$|$Cu interface is
504: 0.9 $\pm 0.1$, which implies about 70$\%$ spin randomization at
505: the interface. For very thin Co layers ($t\leq 4$ nm), the
506: non-local damping becomes predominant. It must be mentioned that
507: for the films without Pt capping layer, the top Cu layer, which is
508: in contact with air, might be covered by a native oxide layer. The
509: FMR measurements were conducted within a few days of fabrication
510: of the films. While $a$ $priori$ the native oxide layer may affect
511: the damping, our results indicate that this oxide layer must not
512: play a very important role. Indeed, assuming that the oxide layer
513: acts as a spin scatterer, like Pt, it would then be expected that
514: the damping $d \Delta H/df$ of $||$Pt$|$Cu$|$Co$|$Cu$||$ and
515: $||$Pt$|$Cu$|$Co$|$Cu$|$Pt$||$ with the same Co layer thickness
516: will be equal. Fig. 10 shows that this is not the case.
517:
518: In a symmetric structure $||$NM$_2|$NM$_1|$FM$|$NM$_1|$NM$_2||$,
519: where NM$_2$ is a perfect spin sink and NM$_1$ thickness is much
520: smaller than the spin diffusion length of the material, the
521: Gilbert damping is \cite{brataas2}:
522: \begin{equation}
523: \label{eq.5} \text{G}(t) = \text{G}_0 + \left({{g \mu
524: _B}\over{e}}\right)^2 {{\tilde {\text{G}}_{\text{eff}}^{\uparrow
525: \downarrow} S^{-1}}\over{t}}\;.
526: \end{equation}
527: $\text{G}_0$ is the residual Gilbert damping (bulk damping), $S$
528: is the surface area of the sample and $\tilde
529: {\text{G}}_{\text{eff}} ^{\uparrow \downarrow}$ is the effective
530: spin mixing conductance. $\tilde{\text{G}}_{\text{eff}} ^{\uparrow
531: \downarrow}$ accounts for the spin mixing conductance at the
532: Co$|$Cu and Cu$|$Pt interfaces:
533: \begin{equation}
534: \label{eq.6} {{1}\over{\tilde{\text{G}}_{\text{eff}}^{\uparrow
535: \downarrow}S^{-1}}} =
536: {{1}\over{\tilde{\text{G}}_{\text{Co}|\text{Cu}}^{\uparrow
537: \downarrow}S^{-1}}} + {{1}\over{{\text{G}}_{\text{Cu}}}} +
538: {{1}\over{\tilde{\text{G}}_{\text{Cu}|\text{Pt}}S^{-1}}} \;,
539: \end{equation}
540: where $\text{G}_{\text{Cu}}=1/(2L \rho)$ is the conductance per
541: spin of the Cu layer of thickness $L$ and resistivity $\rho$.
542: Using the literature value of the resistivity for pure Cu, $\rho =
543: 1.7 \times 10^{-8}\ \Omega$m, the conductance per spin of 10 nm
544: Cu layer is $\text{G}_{\text{Cu}}= 2.94 \times 10^{15}$ $\Omega
545: ^{-1}$m$^{-2}$. In the films $||$Pt$|$Cu$|$Co$|$Cu$|$Pt$||$
546: studied here, the Cu thickness (10 nm) is much smaller than the Cu
547: spin diffusion length. In addition, we found that a 1.5 nm Pt
548: layer is sufficient to saturate the additional Gilbert damping,
549: and therefore is a perfect spin sink. The best fit of the
550: thickness dependence of the magnetic damping gives
551: $\tilde{\text{G}}_{\text{eff}} ^{\uparrow \downarrow}S^{-1} =
552: (0.34 \pm 0.05) \times 10^{15}$ $\Omega ^{-1}$m$^{-2}$ and
553: $\text{G}_0=2.09 \pm 0.44$ s$^{-1}$, using the average value of
554: the g-factor reported in section IV. Note that the conductance per
555: spin of a 10 nm Cu layer is about 10 times larger than the
556: estimated effective spin mixing resistance, meaning that the main
557: contributions to the resistance in the layered structures
558: originates from that associated with the spin mixing at the
559: Cu$|$Co and at the Cu$|$Pt interfaces. After correction of the
560: Sharvin conductance \cite{brataas2}, the effective spin mixing
561: conductance is $\text{G}_{\text{eff}} ^{\uparrow
562: \downarrow}S^{-1}=0.26 \pm 0.04 \times 10^{15}$ $\Omega ^{-1}
563: \text{m}^{-2}$. Using the theoretical values of the mixing
564: conductance for Cu$|$Co ($0.55 \times 10^{15}$ $\Omega
565: ^{-1}$m${^{-2}}$) and Cu$|$Pt ($1.36 \times 10^{15}$ $\Omega
566: ^{-1}$m$^{-2}$) given in \cite{brataas2} and the conductance per
567: spin of Cu as calculated above, the effective spin mixing
568: conductance of the films is predicted to be $0.34 \times 10^{15}$
569: $\Omega ^{-1}$m$^{-2}$. The experimental value of the effective
570: spin mixing is about 30$\%$ smaller than the calculated value. A
571: similar FMR study was conducted on sputtered films with the same
572: structure. An effective spin mixing conductance
573: $\text{G}_{\text{eff}} ^{\uparrow \downarrow}S^{-1}=0.41 \pm 0.05
574: \times 10^{15}$ $\Omega ^{-1} \text{m}^{-2}$ was found. The result
575: implies that the interface spin mixing conductance depends on the
576: film deposition method.
577:
578:
579: We now briefly discuss FMR line broadening due to two magnon
580: scattering. The scattering of magnons by defects and imperfections
581: at the surface and interface of a thin ferromagnetic film produces
582: modes that are degenerate with the FMR mode ($k$=0)
583: \cite{twomagnon}, leading to an additional contribution to the
584: linewidth. The mechanism is operative when the magnetization
585: vector lies in the film plane, and it is not operative in the
586: perpendicular geometry. Recently, Arias and Mills developed a
587: theory of the two-magnon scattering contribution to the FMR
588: linewidth in ultra-thin films \cite{mills}. The theory was found
589: to successfully explain the frequency dependence of the linewidth
590: of epitaxially grown thin films, with defects of rectangular shape
591: \cite{lindner, urban}. The AFM images (Fig. 2) show that there is
592: no evidence of anisotropy in the topology of the defects at the
593: surface of the polycrystalline films. In this case, according to
594: \cite{mills}, the two-magnon contribution is proportional to
595: $H_{\text{res}}^2$ and $H_{\text{s}}^2$. As shown in Fig. 9, the
596: linewidth increases linearly with the frequency in the range 4-25
597: GHz. In plotting the data as $\Delta H$ versus $H_{\text{res}}$,
598: the trend is similar to a square root dependence, which does not
599: agree with the expected $H_{\text{res}}^2$ dependence predicted by
600: \cite{mills}. Furthermore, the authors calculated that one of the
601: signatures of the linewidth broadening from two-magnon scattering
602: is $\Delta H \propto H_{\text{s}}^2$, where the coefficient of
603: proportionality contains information related to the roughness of
604: the film. As the uniaxial anisotropy $H_{\text{s}}$ scales as
605: $1/t$ (Eq. 3), it is predicted that the linewidth increases as
606: $1/t^2$ for the very thin films. The insets of Fig. 8a and 8b
607: shows the linewidth versus the Co layer thickness in log-log plot
608: at two frequencies, $f=10$ and $f=14$ GHz respectively. From the
609: linear best fit, we found that $\Delta H$ increases as $1/t^{n}$
610: with $n \approx 0.8$. We can conclude that the Arias and Mills
611: model of two-magnon scattering contribution to the linewidth
612: cannot be used to explain the thickness dependence of $\Delta H$.
613: While, in general a two-magnon contribution to the linewidth may
614: be present, it cannot explain the strong increase of the linewidth
615: for very thin films. This can be seen by comparing the films with
616: two Pt layers and those with only the Pt underlayer. The films
617: grown with a Pt underlayer ($y_1=1.5$ nm) and with or without top
618: Pt ($y_2=1.5, 0$ nm) are expected to have similar microstructure,
619: since the underlayer structure is identical ($||$1.5 nm Pt$|$10 nm
620: Cu). As a consequence, in the picture of two-magnon scattering
621: induces linewidth broadening, the linewidth of those films should
622: be the same. However, the thickness dependence of $\Delta H$ shows
623: that the films with two Pt layers have a larger linewidth than
624: those with a single Pt layer.
625:
626: \section{conclusion}
627: We have conducted a FMR study of polycrystalline ultrathin Co
628: films embedded between 10 nm Cu. The outer interfaces of Cu were
629: placed in contact with different environments, by adding Pt
630: layers. We have found that very thin Co layers magnetization
631: relaxation processes depend on the non-local environment. The Co
632: layers exhibit a lower magnetization density compared to the bulk
633: material, and an uniaxial anisotropy field perpendicular to the
634: film plane. The large difference between the g-factor value of the
635: ultra-thin Co layer compared to the bulk fcc Co remains an open
636: question.
637:
638: The FMR linewidth, studied as a function of the Co layer thickness
639: and of the non-local environment, increases with decreasing Co
640: layer thickness. We have provided evidence that the two-magnon
641: scattering mechanism cannot explain the thickness dependence of
642: the linewidth. The inhomogeneous contribution to $\triangle H$
643: increases with decreasing Co layer thickness, independently of the
644: presence of the Pt layer. However, the Gilbert damping, the
645: intrinsic contribution to the linewidth, was found to be function
646: of the Co layer thickness and depends on the non-local
647: environment. In particular, the Gilbert damping increases
648: significantly with decreasing Co layer thickness only in the
649: presence of a Pt$|$Cu interface. The thickness dependence of G is
650: consistent with the theory of spin pumping. By changing the outer
651: interface environment of $|$Cu$|$Co$|$Cu$|$ by adding or removing
652: Pt in contact with Cu, we gave unambiguous experimental evidence
653: that the damping enhancement is a non-local mechanism. The
654: non-local damping becomes predominant compared with other
655: relaxation mechanism when the Co layer is a few nanometer thick
656: ($t\leq$4 nm).
657:
658: In summary, the adjacent layers material as well as the non-local
659: environment are critical for understanding the magnetization
660: relaxation in ultra-thin FMs. This important result must be taking
661: into account in the study of spin transfer devices.
662:
663: We appreciate very stimulating discussions with C. E. Patton and
664: J. Z. Sun. This research is supported by NSF-DMR-0405620 and by
665: ONR N0014-02-1-0995.
666:
667:
668:
669:
670: \begin{references}
671:
672: \bibitem{silsbee} R. H. Silsbee, A. Janossy, and P. Monod, Phys. Rev. B \textbf{19},
673: 4382 (1979).
674:
675: \bibitem{brataas1} Y. Tserkovnyak, A. Brataas, and G. E. W. Bauer,
676: Phys. Rev. Lett. \textbf{88}, 117601 (2002); \emph{\emph{ibid}}
677: Phys. Rev. B \textbf{66}, 224403 (2002).
678:
679: \bibitem{brataas2} Y. Tserkovnyak, A. Brataas, G. E. W. Bauer, and B. I. Halperin, Rev. Mod. Phys. \textbf{77}, 1375 (2005).
680:
681: \bibitem{mizukami} S. Mizukami, Y. Ando, and T. Miyazaki, Jpn. J. Appl. Phys. \textbf{40} 580 (2001); Phys. Rev. B \textbf{66},
682: 104413 (2002).
683:
684: \bibitem{heinrich} B. Heinrich, Y. Tserkovnyak, G. Woltersdorf, A. Brataas, R. Urban, and G. E. W.
685: Bauer, Phys. Rev. Lett. \textbf{90}, 187601 (2003).
686:
687: \bibitem{foros} J. Foros, G. Woltersdorf, B. Heinrich, and A. Brataas, J. Appl. Phys. \textbf{97}, 10A714
688: (2005).
689:
690: \bibitem{xia} K. Xia, P. J. Kelly, G. E. W. Bauer, A. Brataas, and I. Turek, Phys. Rev. B \textbf{65}, 220401 (2002) and
691: ref. therein.
692:
693: \bibitem{bauer} G. E. W. Bauer, Y. Tserkovnyak, D. Huertas-Hernando, and A. Brataas, Phys. Rev. B \textbf{67}, 094421 (2003) and
694: ref. therein.
695:
696: \bibitem{sun} J. Z. Sun, J. Appl. Phys. \textbf{97}, 10C714
697: (2005).
698:
699: \bibitem{nanopillar} see for example, J. A. Katine, F. J. Albert, R. A. Buhrman, E. B. Myers, and D. C. Ralph, Phys. Rev. Lett.
700: \textbf{84}, 3149 (2000) ; B. Oezyilmaz, A. D. Kent, D. Monsma, J.
701: Z. Sun, M. J. Rooks, and R. H. Koch, \emph{ibid} \textbf{91},
702: 067203 (2003).
703:
704: \bibitem{sunPRB} J. Z. Sun, Phys. Rev. B \textbf{62}, 570 (2000).
705:
706: \bibitem{barry} W. Barry, I.E.E.E Trans. Micr. Theor. Techn. MTT\textbf{34}, 80
707: (1996).
708:
709: \bibitem{kittel} C. Kittel, Introduction to Solid State Physics, 7th Edition,
710: p.505.
711:
712: \bibitem{inhomogeneity} See, for example, D. L. Mills and S. M. Rezende in \emph{Spin Dynamics in Confined Magnetic Structures II}, edited by B. Hillebrands and K. Ounadjela (Springer, Heidelberg, 2002), pp. 27$-$58.
713:
714: \bibitem{celinski} W. Platow, A. N. Anisimov, G. L. Dunifer, M. Farle, and K. Baberschke, Phys. Rev. B \textbf{58}, 5611 (1998) ; Z. Celinski and B. Heinrich, J. Appl. Phys. \textbf{70} (10), 5935 (1991).
715:
716:
717: \bibitem{kalarickal} S. S. Klarickal, P. Krivosik, M. Wu, C. E.
718: Patton, M. L. Schneider, P. Kabos, T. J. Silva, J. P. Nibarger, J.
719: Appl. Phys. \textbf{99}, 093909 (2006).
720:
721: \bibitem{ohandley} C. H. Lee, H. He, F. J. Lamelas, W. Vavra, C. Uher, and R. Clarke, Phys. Rev. B \textbf{42}, R1066 (1990); G. Bochi, O. Song, and R. C. O'Handley, \emph{ibid.} \textbf{50}, 2043
722: (1994).
723:
724: \bibitem{wiedwald} U. Wiedwald, M. Spasova, M. Farle, M. Hilgendorff, and M. Giersig, J. Vac. Sci. Technol. A\textbf{19},
725: 1773 (2001).
726:
727: \bibitem{chikazumi} S. Chikazumi and C. D. Graham, Physics of Ferromagnetism,
728: Oxford University Press (1997), p.73.
729:
730: \bibitem{bland} J. A. C. Bland, B. Heinrich (Eds.), Vols. I and II.
731: Ultrathin Magnetic Structures, (Springer, Berlin 1994).
732:
733: \bibitem{bass} H. Kurt, R. Loloee, K. Eid, W. P. Pratt, and J. Bass, Appl. Phys. Lett. \textbf{81} (25), 4787
734: (2002).
735:
736: \bibitem{jedema} F. J. Jedema, A.T. Filip, and B.J. van Wees, Nature (London) 410, 345 (2001).
737:
738: \bibitem{twomagnon} M. Sparks, R. Loudon, and C. Kittel, Phys. Rev. \textbf{122}, 791 (1961); R. C. Fletcher, R. C. LeCraw, and E. G. Spencer, Phys. Rev. \textbf{117}, 955 (1960) ; C. E. Patton, J. Appl.
739: Phys. \textbf{39}, 3060 (1968) ; M. J. Hurben, and C. E. Patton,
740: J. Appl. Phys. \textbf{83}, 4344 (1998).
741:
742: \bibitem{mills} R. Arias and D. L. Mills, Phys. Rev. B \textbf{60}, 7395
743: (1999).
744:
745: \bibitem{urban} R. Urban, B. Heinrich, G. Woltersdorf, K. Ajdari, K. Myrtle, J. F. Cochran, and E. Rozenberg, Phys. Rev. B \textbf{65}, 020402(R) (2001)
746:
747: \bibitem{lindner} J. Lindner, K. Lenz, E. Kosubek, K. Baberschke,
748: D. Spoddig, R. Meckenstock, J. Pelzl, Z. Frait, and D. L. Mills,
749: Phys. rev. B \textbf{68}, 060102(R) (2003).
750:
751: \bibitem{pelzl} J. Pelzl, R. Meckenstock, D. Spoddig, F. Schreiber, J. Pflaum and
752: Z. Frait, J. Phys.: Condens. Matter \textbf{15}, S451-S463 (2003).
753:
754: \bibitem{yuan} S. J. Yuan, L. Wang, R. Shan, and S. M. Zhou, Appl. Phys. A \textbf{79}, 701
755: (2004).
756:
757: \end{references}
758:
759: \end{document}
760: