Thermal spin injection and interface insensitivity in permalloy/aluminum metallic nonlocal spin valves

We present measurements of thermal and electrical spin injection in nanoscale metallic nonlocal spin valve structures. Informed by measurements of the Seebeck coefficient and thermal conductivity of representative films made using a micromachined Si-N thermal isolation platform, we use simple analytical and finite-element thermal models to determine limits on the thermal gradient driving thermal spin injection and calculate the spin-dependent Seebeck coefficient to be $-0.5\mu \mathrm{V}/\mathrm{K}>S_s>-1.6\mu \mathrm{V}/\mathrm{K}$. This is comparable in terms of the fraction of the absolute Seebeck coefficient to previous results, despite dramatically smaller electrical spin injection signals. Since the small electrical spin signals are likely caused by interfacial effects, we conclude that thermal spin injection is less sensitive to the ferromagnetic/nonmagnetic interface, and possibly benefits from the presence of oxidized ferromagnets, which further stimulates interest in thermal spin injection for applications in sensors and pure spin current sources.

Figure 1

(a) Schematic layout of the NLSV under electrical spin injection, where a large charge current driven through a FM nanowire creates a spin accumulation and pure spin current in a NM channel that is detected with a second FM. (b) The nonlocal resistance $R_{\mathrm{NLE}}=V_{\mathrm{NLE}}/I$ for a $L=900\mathrm{nm}$ device at $78\mathrm{K}$, where the relative alignments of the two FM contacts are indicated with paired arrows. (c) Thermal spin injection is achieved by passing current $I$ only through FM1, creating a thermal gradient at the NM/FM interface that injects spin into the NM. (d) False-color SEM micrograph of the nanoscale circuit defining the NLSV. Sizes given indicate the designed widths of nanowires; measured geometries appear in Table 1.

Figure 2

$V_{\mathrm{channel}}$ vs $I$ characteristics for the NM channel (contacts made as shown schematically in the inset) are highly linear across the entire range of applied $I$ in contrast to both the three-terminal contact resistance [Fig. 6] and nonlocal resistance measurements [Figs. 5 and 6]. Measurements for two NLSVs are shown for two temperatures. Dashed lines show linear fits.

Figure 3

Nonlocal resistance signals $R_{\mathrm{NLE}}=V_{\mathrm{NLE}}/I$ in electrical [(a) and (b)] and $R_{\mathrm{NLT}}=V_{\mathrm{NLT}}/I$ in thermal [(c) and (d)] spin injection for both 500 nm and 1300 nm nominal FM spacing. (e) The electrical spin signal $\mathrm{\Delta }{R}_{\mathrm{NLE}}$ vs $L$ with the fit to the 1d spin diffusion equation. This fit gives ${\lambda }_{\mathrm{NM}}=760\pm 50\mathrm{nm}$.

Figure 4

(a) Comparison of electrically driven spin signal $\mathrm{\Delta }{R}_{\mathrm{NLE}}$ (blue spheres) to various models based on the 1d spin diffusion equation [39]. Matching of the electrical signal is only possible using a strongly reduced value of interfacial spin polarization, regardless of the model employed. (b) Comparison of $\mathrm{\Delta }{R}_{\mathrm{NLE}}$ reported here (blue spheres) to similar Py-based NLSVs with various NM channels [3, 15, 44, 45]. In each case the lines represent a 1d spin diffusion model that explains the signal size. Relevant parameters and spin resistances are also given. Note especially the large signal that matches predictions of the transparent interface model for the Py/Cu device originally used to observe the SDSE [15]. NLSVs used in this study have much reduced electrical signal, but maintain the same thermally driven spin signal.

Figure 5

(a) $IV$ characteristic for the electrical spin injection configuration [Fig. 1] measured separately for parallel and antiparallel states of the FM nanowires for the $L=500\mathrm{nm}$ device at 78 K. (b) The corresponding $IV$ characteristic for the thermal spin injection configuration [Fig. 1]. (c) Subtraction of the parallel and antiparallel curves in (a) gives the highly linear response of electrical spin injection, while the corresponding subtraction for thermal injection yields a spin signal dominated by the $I^2$ term indicating thermal generation of a spin accumulation in the NM. In both (c) and (d), data for both $L=500\mathrm{nm}$ and $L=1300\mathrm{nm}$ are shown. Fitted values of spin signal are also shown.

Figure 6

(a)–(b) Two-body thermal models used to analytically model the $T$ profile in the devices. (c)–(d) Resulting $IV$ curves show significant curvature as a result of heating and thermoelectric effects. Data are shown for the $L=1300\mathrm{nm}$ NLSV at 78 K, but similar curvature is seen at room $T$ and for other devices. Inset: The simplified thermal profile used to estimate a maximum possible $\mathbf{\nabla }T$ of 33 K/micron from our data. (e) The three-terminal contact resistance (shown schematically in upper inset) $IV$ characteristic shows small but clearly measurable nonlinearity (lower inset).

Figure 7

Measured Seebeck coefficients for the constituent thin films vs $T$. Each film was deposited on a thermal isolation platform, and the measured Seebeck coefficient is relative to Cr/Pt leads. The estimated lead contribution has been subtracted here, so that this plot compares estimated absolute Seebeck coefficients. Inset: Scanning electron micrograph of the thermal isolation platform we use for thermal properties measurements.

Figure 8

(a) 2d geometry and mesh used for FEM thermal calculations. (b) Thermal profile resulting from heat dissipated in FM1 chosen to give the correct $\mathrm{\Delta }T$ at FM2. Inset: Dashed red line shows the region of the 2d cross-sectional slice used for the FEM model. (c)–(d) Resulting $T$ and $dT/dx$ profiles for the $L=1300\mathrm{nm}$ NLSV at the height $\approx 50\mathrm{nm}$ above the substrate at the peak of the broad maximum in $dT/dx$.