1: 

2: %\documentclass[aps,prl,preprint,groupedaddress]{revtex4}

3: %\documentclass[aps,prl,preprint,superscriptaddress]{revtex4}

4: \documentclass[aps,prl,twocolumn,groupedaddress,dvips]{revtex4}

5: \usepackage[dvips]{graphicx}

6: % You should use BibTeX and apsrev.bst for references

7: % Choosing a journal automatically selects the correct APS

8: % BibTeX style file (bst file), so only uncomment the line

9: % below if necessary.

10: \bibliographystyle{apsrev}

11: \begin{document}

12: % Use the \preprint command to place your local institutional report

13: % number in the upper righthand corner of the title page in preprint mode.

14: % Multiple \preprint commands are allowed.

15: % Use the 'preprintnumbers' class option to override journal defaults

16: % to display numbers if necessary

17: \preprint{}

18: %Title of paper

19: \title{Saturation of Spin-Polarized Current in Nanometer Scale Aluminum Grains}

20: 

21: 

22: % repeat the \author .. \affiliation etc. as needed

23: % \email, \thanks, \homepage, \altaffiliation all apply to the current

24: % author. Explanatory text should go in the []'s, actual e-mail

25: % address or url should go in the {}'s for \email and \homepage.

26: % Please use the appropriate macro foreach each type of information

27: % \affiliation command applies to all authors since the last

28: % \affiliation command. The \affiliation command should follow the

29: % other information

30: % \affiliation can be followed by \email, \homepage, \thanks as well.

31: %\email[]{dragomir.davidovic@physics.gatech.edu}

32: %\homepage[]{Your web page}

33: %\thanks{}

34: %\altaffiliation{}

35: \author{Y. G. Wei, C. E. Malec, D. Davidovi\'c }

36: \affiliation{School of Physics, Georgia Institute of Technology,

37: Atlanta, GA 30332}

38: %Collaboration name if desired (requires use of superscriptaddress

39: %option in \documentclass). \noaffiliation is required (may also be

40: %used with the \author command).

41: %\collaboration can be followed by \email, \homepage, \thanks as well.

42: %\collaboration{} FG^{as}

43: %\noaffiliation

44: \date{\today}

45: \begin{abstract}

46: We describe measurements of spin-polarized tunnelling via discrete

47: energy levels of single Aluminum grains. In high resistance

48: samples ($\sim G\Omega$), the spin-polarized tunnelling current

49: rapidly saturates as a function of the bias voltage. This

50: indicates that spin-polarized current is carried only via the

51: ground state and the few lowest in energy excited states of the

52: grain. At the saturation voltage, the spin-relaxation rate

53: $T_1^{-1}$ of the highest excited states is comparable to the

54: electron tunnelling rate: $T_1^{-1}\approx 1.5\cdot 10^6 s^{-1}$

55: and $10^7s^{-1}$ in two samples. The ratio of $T_1^{-1}$ to the

56: electron-phonon relaxation rate is in agreement with the

57: Elliot-Yafet scaling, an evidence that spin-relaxation in Al

58: grains is governed by the spin-orbit interaction.

59: 

60: \end{abstract}

61: 

62: % insert suggested PACS numbers in braces on next line

63: \pacs{73.21.La,72.25.Hg,72.25.Rb,73.23.Hk}

64: % insert suggested keywords - APS authors don't need to do this

65: %\keywords{}

66: %\maketitle must follow title, authors, abstract, \pacs, and \keywords

67: \maketitle

68: 

69: \section{introduction}

70: 

71: Electron tunnelling through single nanometer scale metallic grains

72: at low temperatures can display a discrete energy level

73: spectrum.~\cite{ralph} Tunnelling spectroscopy of the energy

74: spectra have led to numerous discoveries, including Fermi-Liquid

75: coupling constants between quasiparticles,~\cite{agam} spin-orbit

76: interactions,~\cite{davidovic,petta} and superconducting

77: correlations in zero-dimensional systems.~\cite{black} Some

78: information regarding the spin of an electron occupying a discrete

79: level can be obtained using spin-unpolarized tunnelling, such as

80: spin-multiplicity and electron g-factors.~\cite{ralph}

81: 

82: In this letter we report on spin-polarized tunnelling via discrete

83: energy levels of single aluminum grains. Spin-polarized electron

84: transport permits studies of spin relaxation and spin

85: dephasing.~\cite{johnson,jedema1} By comparison, spin-unpolarized

86: spectroscopy is suitable for the studies of energy relaxation in

87: the grains.~\cite{agam,ralph} Since spin-relaxation times are

88: generally many orders of magnitude longer than energy relaxation

89: times, spin-unpolarized spectroscopy is not an easy tool to study

90: spin-relaxation in the grains and spin-polarized tunnelling is

91: needed. We find that some electron spin-relaxation times in Al

92: grains are exceptionally long compared to bulk Al with similar

93: disorder, on the order of $\mu s$.

94: 

95: Spin-polarized transport via metallic grains has recently

96: generated a lot of theoretical

97: interest.~\cite{braun,weymann,vandermolen,wetzels,cottet1} In

98: addition, there is a major effort to study nano-spintronics using

99: carbon-nanotubes; see Ref.~\cite{cottet} and references therein.

100: Spin-coherent electron tunnelling via nanometer scale normal

101: metallic grains has been confirmed in arrays~\cite{zhang1,ernult}

102: and in single grains.~\cite{bernard} However, the electron

103: spin-relaxation time $T_1$ in a metallic grain has not been

104: reported yet.

105: 

106: \section{Sample fabrication}

107: 

108: Our samples are prepared by electron beam lithography and shadow

109: evaporation, similar to the technique described

110: previously.~\cite{davidovic} First we define a resist bridge

111: placed 250 nm above the Si wafer; this bridge acts as a mask. Next

112: (Fig.~\ref{fig1}-A), we deposit 11 nm permalloy (Py =

113: Ni$_{0.8}$Fe$_{0.2}$) onto oxidized silicon substrate at $4\cdot

114: 10^{-7}$ Torr base pressure, measured near the gate valve, along

115: the direction indicated by the arrow. Then we rotate the sample by

116: 36 degrees without breaking the vacuum and deposit 1.2 nm of

117: Al$_2$O$_3$ by reactive evaporation of Al,~\cite{davidovic} at a

118: rate of 0.35 nm/s, at an oxygen pressure of $2.5\cdot 10^{-5}$

119: Torr. Now, oxygen flow is shut down. When pressure decreases to

120: the $10^{-7}$ Torr range, we deposit a 0.6 nm thick film of Al, as

121: shown in Fig.~\ref{fig1}-B. Al forms isolated grains with a

122: typical diameter of 5 nm. The grains are displayed by the scanning

123: electron microscope (SEM) image in Fig.~\ref{fig1}-D. Finally we

124: deposit another 1.2 nm layer of Al$_2$O$_3$ by the reactive

125: evaporation and top it of by an 11 nm thick film of Py

126: (Fig.~\ref{fig1}-C).

127: \begin{figure}%[p]

128: \includegraphics[width=0.45\textwidth]{fig1.eps}

129: \caption{A, B, C: Sample fabrication steps. D. Image of Al grains.

130: E, F: Image of a typical sample. G, H: I-V curves at the base

131: temperature.~\label{fig1}}

132: \end{figure}

133: We make many samples on the same silicon wafer, and vary the

134: overlap from 0 to 50 nm and select the devices with the highest

135: resistance, as they have the smallest overlap. Figs.~\ref{fig1}-E

136: and F show SEM images of a typical device.

137: 

138: \section{Discrete energy levels}

139: 

140: Transport properties of the samples at low temperatures were

141: measured using an Ithaco current amplifier. The samples were

142: cooled down to $\approx 0.035K$ base temperature. The sample leads

143: were cryogenically filtered to reduce the electron temperature

144: down to $\approx 0.1K$.

145: 

146: The majority of samples ($>$80\%) exhibit Coulomb Blockade at low

147: temperature. About 150 samples were measured at 4.2K and 16

148: samples at $0.035 K$. In this paper we describe two samples. The

149: I-V curve of two samples are shown in Fig.~\ref{fig1}-G and H. The

150: tunnelling current increases in discrete steps as a function of

151: bias voltage, corresponding to discrete electron-in-a-box energy

152: levels of the grain.

153: 

154: 

155: In sample 1, the average electron-in-a-box level spacing caused by

156: electron geometric confinement is $\delta\approx 0.8meV$, which

157: corresponds to diameter of $D\approx 6nm$ assuming a spherical Al

158: grain. The average current step is ${\overline I}\approx 0.47pA$.

159: We make a connection with the tunnelling rates from the leads to

160: the grain and the measured current response.  The tunnel junctions

161: are highly asymmetric, and therefore one of the tunnelling rates

162: is much smaller than the other, and thus rate limiting. Throughout

163: this paper, we choose the rate limiting step to be across the left

164: junction, corresponding to the tunnelling rate $\Gamma_L$.

165: Therefore, our measured current corresponds to the average

166: tunnelling-in rate of $\overline{\Gamma}_L={\overline

167: I}/2|e|\approx 1.5\cdot 10^6 s^{-1}$. Similarly, in sample 2,

168: $\delta\approx 2.7 meV$, $D\approx 4nm$, and

169: ${\overline\Gamma_L}\approx 9.6\cdot 10^6 s^{-1}$.

170: 

171: 

172:  The

173: spin-conserving energy-relaxation in Al grains takes place by

174: phonon emission with the relaxation rate~\cite{agam}

175: \begin{equation}

176: {\tau_{e-ph}^{-1}(\omega)}=\left(\frac{2}{3}E_F\right)^2\frac{\omega^3\tau_e\delta}{2\rho\hbar^5

177: v_S^5}, \label{eph}

178: \end{equation}

179: where $E_F=11.7 eV$ is the Fermi energy, $\omega$ is the energy

180: difference between the initial and the final state, $\rho=2.7

181: g/cm^3$ is the ion-mass density, and $v_s=6420 m/s$ is the sound

182: velocity. We obtain $\tau_{e-ph}^{-1}(\delta)\approx 1.6\cdot 10^9

183: s^{-1}$ and $4.1\cdot 10^{10} s^{-1}$ in samples 1 and 2,

184: respectively. Sample 2 has significantly larger relaxation rate

185: because of the larger level spacing. Since the tunnelling rates in

186: our samples are $\sim 10^6s^{-1}$, if the grain is excited by

187: electron tunnelling in and out, it will instantly relax to the

188: lowest energy state accessible by spin-conserving transitions.

189: 

190: As shown by Fig.~\ref{fig2}, the energy levels exhibit Zeeman

191: splitting as a function of an applied magnetic field. In sample 1,

192: the I-V curve probes the same energy spectrum at  negative and

193: positive bias voltage. This is evident from the equivalence of the

194: magnetic field dependencies at negative and positive bias. The

195: lowest tunnelling threshold is two fold degenerate at zero

196: magnetic field, showing that $N_0$, the number of electrons on the

197: grain before tunnelling in, is even. The conductance peaks are

198: similar in magnitude at negative bias, because the first

199: tunnelling step, in which an electron tunnels in to the grain

200: through the higher resistance junction, is rate limiting. At

201: positive bias, the first conductance peak is much larger than the

202: subsequent conductance peaks, because the first tunnelling step

203: takes place via the lower resistance junction, and the rates are

204: limited by the electron discharge process across the high

205: resistance junction.

206: 

207: In sample 1, the first two peaks split corresponding to g-factors:

208: $g=1.83\pm 0.05$ and $1.95\pm 0.05$. Slight reduction of the

209: g-factors from $2$ indicates spin-orbit interaction in

210: Al.~\cite{ralph} The avoided level crossings  clearly are resolved

211: in Fig.~\ref{fig2}, near points $(-11.5mV,5T)$ and

212: $(-13mV,11.5T)$. The corresponding avoided crossings at positive

213: bias are located near $(13.5mV,5T)$ and $(15.5mV,11.5T)$,

214: respectively. In the regime where g factors are slightly reduced,

215: the spin-orbit scattering rate ($\tau_{SO}^{-1}$) can be obtained

216: from the avoided crossing energies $\Delta_{SO}\approx

217: 0.1meV$.~\cite{adam} Theory predicts that

218: $\tau_{SO}\approx\hbar\delta/\pi\Delta_{SO}^2$,~\cite{adam} within

219: a factor of two. Thus, we obtain $\tau_{SO}^{-1}\approx 5.5\cdot

220: 10^{10} s^{-1}$. By the Elliot-Yafet relation,~\cite{yafet}

221: $\tau_{SO}^{-1}$ is related to the elastic scattering rate

222: $\tau_e^{-1}$: $\tau_{SO}^{-1}=\alpha \tau_e^{-1}$. Assuming

223: ballistic grain, $\tau_{e}^{-1}\approx v_F/D=3.4\cdot 10^{14}

224: sec^{-1}$. We obtain $\alpha\approx 1.6\cdot 10^{-4}$, in

225: excellent agreement with $\alpha\approx 10^{-4}$ in Al thin

226: films.~\cite{jedema3}

227: 

228: 

229: \begin{figure}%[p]

230: \includegraphics[width=0.45\textwidth]{fig2.eps} \caption{A, B: Differential conductrance (gray) versus bias

231: voltage and the applied magnetic field in sample 1 at the base

232: temperature.~\label{fig2}}

233: \end{figure}

234: 

235: \section{Spin-polarized tunnelling}

236: 

237: Now we discuss magnetoresistance from the spin-polarized

238: tunnelling. In the magnetic field range of $\pm 50 mT$,

239: approximately 90\% of the samples do not display any of the

240: tunnelling magnetoresistance effect (TMR) . By contrast, we tested

241: about 10 tunnelling junctions without the embedded grains and with

242: similar resistance (empty junctions) at 4.2 K. All of the empty

243: junctions exhibit a significant TMR in this field range,

244: comparable to $10\%$. Approximately one half of the empty

245: junctions display a simple spin-valve effect. So, the absence of

246: TMR for electron tunnelling via grains shows that the

247: spin-dephasing rate $T_2^{-1}$ in 90\% of the samples must be much

248: larger than the tunnelling rate.

249: 

250: Nevertheless, approximately 10\% of the samples with embedded

251: grains display significant TMR, so the dephasing must be weak,

252: e.g. $T_2^{-1}$ must be smaller than or comparable to the

253: tunnelling rate in these samples. $T_2$ variation among different

254: samples could be explained by magnetic defects, such as

255: paramagnetic impurities from the Py layer. Paramagnetic impurities

256: are common sources of dephasing.~\cite{bergman} The defects would

257: be located on the grain surface, since bulk Al does not support

258: paramagnetism. Since the number of atoms on the surface is

259: relatively small ($\sim$1000), we could occasionally obtain a

260: sample free of impurities.  More insight into the nature of $T_2$

261: in this device will require a more in depth theoretical study.

262: 

263: A majority of the samples with nonzero TMR show positive TMR near

264: the Coulomb-Blockade conduction threshold; only about 30\% of the

265: samples show negative TMR. The sign of TMR in quantum dots is

266: determined by the interplay between charging effects and

267: spin-accumulation.~\cite{barnas,ernult} For any given sample, the

268: data in this paper correspond to the voltage range within the

269: first step of the Coulomb staircase. In this range the sign of TMR

270: is found to be constant as expected.

271: 

272: TMR in our devices usually does not display a simple spin-valve

273: effect. We believe this is because there are spin-dependent

274: interactions inside the grain that induce a complicated TMR even

275: when the magnetic transitions in the drain and source leads are

276: sharp and as expected. For example, a rotation of stray magnetic

277: field acting on the grain will alter the direction of the

278: spin-quantization axis in the grain, thereby changing the

279: conductance~\cite{braun}. A rotation or a switch of a remote

280: domain can change the tunnelling current through the grain via the

281: magnetic field generated by the domain. Similarly, the orientation

282: of the nuclear spin in the grain can change the quantization axes

283: via the hyperfine interaction.

284: \begin{figure}%[p]

285: \includegraphics[width=0.45\textwidth]{fig3.eps} \caption{A-F: Spin-valve effect in current versus

286: applied magnetic field in two samples at the base temperature. The

287: current magnitude is reduced in the antiparallel

288: state.~\label{fig3}}

289: \end{figure}

290: 

291: We select only those samples that display a simple spin-valve TMR

292: effect, which is shown in Figs.~\ref{fig3}. Fig.~\ref{fig3}-A is

293: the TMR of sample 1 at a bias voltage corresponding to the second

294: current plateau. TMR is barely resolved in this case, since the

295: current changes by only about 40fA. We do not have good data to

296: display TMR at the first current plateau. By comparison,

297: Figs.~\ref{fig3}-B and C display TMR at bias voltage where the

298: number of electron-in-a-box levels energetically available for

299: tunnelling-in are approximately 19 and 48, respectively. To

300: facilitate comparisons, the current intervals on the vertical axes

301: in Figs.~\ref{fig3} A-C and D-F have equal lengths.

302: 

303: The main observation in this letter is that $\Delta

304: I=I_{\uparrow\uparrow}-I_{\uparrow\downarrow}$ is nearly constant

305: with current above a certain current. There is hardly any increase

306: in $\Delta I$ between Figs.~\ref{fig3} B and C and between

307: Figs.~\ref{fig3} E and F. This behavior is shown in more detail

308: Fig.~\ref{fig4}-A and B, which displays $\Delta I$ versus bias

309: voltage. $\Delta I$ versus negative bias voltage in sample 1 is

310: fully saturated at the third current plateau; at the second

311: current plateau, $\Delta I$ is already at one half of the

312: saturation value. Similarly, in sample 2 $\Delta I$ reaches

313: saturation at the second current plateau. Our samples should be

314: contrasted with ordinary ferromagnetic tunnelling junctions, where

315: $\Delta I$ is  proportional to the current over a significantly

316: wider range of bias voltage~\cite{moodera1,zhang2}.

317: 

318: \section{Interpretation of the results}

319: 

320: In coulomb-blockade samples containing magnetic leads, the

321: electrochemical potential difference between the island and leads

322: can jump when the magnetization in one of the leads changes

323: direction~\cite{vandermolen}. This can lead to a sudden shift in

324: energy levels, producing a jump in current that is constant as a

325: function of bias voltage. The shift in energy levels is seen as a

326: discontinuity near zero magnetic field in Fig. 2, and is $\sim 0.1

327: mV$.

328: 

329: To show that the electrocemical shift is not responsible for the

330: saturation of the spin-polarized current with voltage in our

331: sample, we performed other measurements by sweeping the magnetic

332: field both on and between the current plateaus, coming up with

333: similar values for the electrochemical shift.  The shift is lower

334: than the average level spacing of 0.8 meV and 2.7 meV for sample 1

335: and sample 2 respectively.  Therefore, since we measured

336: magnetoresistance in the middle of the current plateau, the

337: threshold voltage shift should not effect our measurements of the

338: saturation in $\Delta I$.

339: 

340: %This sets up the long argument

341: To explain $I_{\uparrow\uparrow}-I_{\uparrow\downarrow}=const$, we

342: must discuss the relative magnitudes of three rates:

343: $\tau_{e-ph}^{-1}$, the rate of energy relaxation from excited to

344: lower energy states by spin-conserving phonon emission;

345: $\Gamma_L$, the rate electrons tunnel into the grain; and

346: $T_1^{-1}$, the rate of transitions between levels that result in

347: an electron flipping its spin orientation. $\tau_{e-ph}^{-1}$ is

348: obtained theoretically, the measured I-V spectrum fixes the

349: tunnelling rate, and $T_1^{-1}$ is obtained from the saturation in

350: $I_{\uparrow\uparrow}-I_{\uparrow\downarrow}$ with bias voltage.

351: 

352: Finally we must deduce the relative magnitude of $T_1^{-1}$. The

353: rate of spin-flip transitions is expected to be significantly

354: smaller than $\tau_{e-ph}^{-1}$~\cite{yafet}. In this case the

355: ground state would not necessarily be accessible by energy

356: relaxation. The grain could remain in an excited, spin-polarized

357: state, as sketched in Fig.~\ref{fig4}-C. These spin-polarized

358: excited states are responsible for spin accumulation in the

359: antiparallel magnetic configuration of the leads. If the

360: relaxation rates for the spin-flip transitions are much smaller

361: than the tunnelling rate, then various spin-polarized states would

362: have similar probabilities, which are determined by the tunnelling

363: rates. In the antiparallel configuration of the leads, the

364: probabilities of the excitations with spin up would be enhanced by

365: $1+ P$ and probabilities of the excitations with spin down would

366: be suppressed by $1-P$, where $P$ is the spin-polarization in the

367: leads. In the parallel configurations, the probabilities of the

368: excitations with spin up and spin down are the same. In this

369: regime, $I_{\uparrow\uparrow}-I_{\uparrow\downarrow}$ is

370: proportional to the current, similar to the usual ferromagnetic

371: tunnelling junctions.

372: 

373: It is reasonable to expect that the spin-flip rate

374: $T_1^{-1}(\omega)$ increases rapidly with energy difference

375: $\omega$ between the initial and the final state~\footnote{In bulk

376: metals the spin-orbit scattering rate increases rapidly with

377: electron excitation energy}. If $T_1^{-1}(\omega)$ exceeds the

378: tunnelling rate above some $\omega$, then the excitations with

379: energy $>\omega$ will occur with a reduced probability in the

380: ensemble of states generated by tunnelling in and out. Thus

381: $\Delta I$ is limited by tunnelling via the ground state and those

382: low lying spin-polarized states where $T_1^{-1}(\omega)<\Gamma_L$.

383: $\Delta I$ versus bias voltage approaches saturation approximately

384: when $T_1^{-1}(\omega)=\Gamma_L$, where $\omega$ is the highest

385: excitation energy in the ensemble of spin-polarized states

386: generated by tunnelling in and out: $\omega\approx\delta

387: \frac{I}{|e|\Gamma_L}$. This is how we determine the

388: spin-relaxation time $T_1(\omega)$ at an energy $\omega$ in a

389: given sample.

390: 

391: \begin{figure}%[p]

392: \includegraphics[width=0.45\textwidth]{fig4.eps}

393: \caption{A and B: $\Delta

394: I=|I_{\uparrow\uparrow}-I_{\uparrow\downarrow}|$ versus bias

395: voltage in samples 1 and 2, respectively, at the base temperature.

396: The numbers near the circles indicate how many doubly degenerate

397: electron-in-a-box levels are available for tunnelling in. C:

398: Possible spin-polarized electron configurations caused by electron

399: tunnelling in and out, before an electron tunnels in, at the

400: second current plateau, for $N_0$ even.~\label{fig4}}

401: \end{figure}

402: 

403: In sample 1, $\Delta I$ is at 50\% of the saturation value at the

404: second current plateau, and $\Delta I$ is saturated at the third

405: current plateau. At the second current plateau, the

406: spin-relaxation rate of the highest energy excited state generated

407: by tunnelling must be close to the tunnelling rate.  Since the

408: spin relaxation is very rapid in configurations more than

409: $3\delta$ above the ground state, and $N_0$ is even as noted

410: above, the grain spends most of the time among the five

411: configurations shown in Fig.~\ref{fig4}-C: $N_0$, $N_0^+$,

412: $N_0^-$, $N_0^{++}$, and $N_0^{--}$. The highest energy

413: spin-polarized states are $N_0^{++}$ and $N_0^{--}$. Thus,

414: $T_1^{-1}(3\delta)\approx \Gamma_L=1.5\cdot 10^6 s^{-1}$. In

415: sample 2, this analysis leads to $T_1^{-1}(2\delta)\approx 10^7

416: s^{-1}$.

417: 

418: Now we discuss the origin of spin relaxation and its rapid

419: enhancement with the energy difference. Note that the rate of

420: spin-conserving transitions in Eq.~\ref{eph} increases as

421: $\omega^3$. We suggest that the electron-phonon transition rates

422: without and with spin flip scale by the Elliot-Yafet relation:

423: $T_1^{-1}(\omega)=\alpha'\tau_{e-ph}^{-1}(\omega)$. This scaling

424: would certainly explain the rapid increase in spin-relaxation rate

425: with excitation energy. In metallic films, it is well established

426: that the Elliot-Yafet scaling applies for both elastic and

427: inelastic scattering processes, with $\alpha\approx

428: \alpha'$.~\cite{jedema3}

429: 

430: 

431: In sample 1, Eq.~\ref{eph} leads to

432: $\tau_{e-ph}^{-1}(3\delta)\approx 4\cdot 10^{10} s^{-1}$. Since

433: $T_1^{-1}(3\delta)\approx 1.5\cdot 10^6 s^{-1}$, we obtain

434: $\alpha'\approx 0.4\cdot 10^{-4}$. Similarly, in sample 2,

435: $\tau_{e-ph}^{-1}(2\delta)\approx 3.3\cdot 10^{11} s^{-1}$ and we

436: obtain $\alpha'\approx 0.3\cdot 10^{-4}$. $\alpha'$ agrees with

437: $\alpha\approx 1.5\cdot 10^{-4}$ obtained earlier, within an order

438: of magnitude. So the ratio of $\tau_{e-ph}$ and $T_1$ is in

439: agreement with the Elliot-Yafet scaling. This is an evidence that

440: the spin-flip transitions in Al grains are driven by the

441: spin-orbit interaction. By this relaxation mechanism, the spin of

442: an electron on the grain is coupled to the phonon continuum via

443: the spin-orbit interaction. An electron in an excited

444: spin-polarized state relaxes by an emission of a phonon, which has

445: an angular momentum equal to the difference between the initial

446: and final electron spin.

447: 

448: \section{Conclusion}

449: 

450: In summary, we have observed spin-coherent electron transport via

451: discrete energy levels of single Al grains. Spin polarized current

452: saturates quickly as a function of bias voltage, which

453: demonstrates that the ground state and the lowest excited states

454: carry spin polarized current. Higher excited states have a

455: relaxation time shorter than the tunnelling time and they do not

456: carry spin-polarized current. The spin-relaxation time of the

457: low-lying excited states is $T_1\approx 0.7\mu s$ and $0.1\mu s $

458: in two samples. Finally, the ratio of the spin-flip transition

459: rate and the electron-phonon relaxation rate is in quantitative

460: agreement with the Elliot-Yafet scaling ratio, an evidence that

461: the spin-relaxation transitions are driven by the spin-orbit

462: interaction.

463: 

464: This work was performed in part at the Georgia-Tech electron

465: microscopy facility. We thank Matthias Braun and Markus Kindermann

466: for consultation. This research is supported by the DOE grant

467: DE-FG02-06ER46281 and David and Lucile Packard Foundation grant

468: 2000-13874.

469: 

470: \bibliography{career1}

471: 

472: \end{document}

473: