1: \documentclass[twocolumn,showpacs,prb,preprintnumbers,amsmath,amssymb]{revtex4}

2: 

3: \usepackage{graphicx}

4: \usepackage{dcolumn}

5: \usepackage{bm}

6: 

7: \begin{document}

8: 

9: \preprint{Rapid Communications}

10: 

11: \title{Evidence for multiple impurity bands in sodium-doped silicon MOSFETs}

12: 

13: \author{T.~Ferrus}

14: \email{taf25@cam.ac.uk}

15: \author{R.~George}

16: \author{C.~H.~W.~Barnes}

17: \author{N.~Lumpkin}

18: \author{D.~J.~Paul}

19: \author{M.~Pepper}

20: 

21: 

22: \affiliation {Cavendish Laboratory, University of Cambridge, Madingley Road, CB3 0HE, Cambridge, United Kingdom}

23: 

24: 

25: \date{\today}

26:              

27: 

28: \begin{abstract}

29: 

30: We report measurements of the temperature dependent conductivity in a silicon MOSFET that

31: contains sodium impurities in the oxide layer. We explain the variation of conductivity in

32: terms of Coulomb interactions that are partially screened by the proximity of the metal gate. 

33: The study of the conductivity exponential prefactor and the localisation length as a function of gate 

34: voltage have allowed us to determine the electronic density of states and has provided arguments

35: for the presence of two distinct bands and a soft gap at low temperature.

36: 

37: \end{abstract}

38: 

39: \pacs{71.23.Cq, 71.55.Gs, 71.55.Jv, 72.15.Rn, 72.20.Ee, 72.80.Ng, 73.20.At, 73.40.Qv}

40: 

41: \maketitle

42: 

43: Since the invention of the silicon MOSFET, understanding the influence of impurities, especially

44: sodium contamination, on device performance has been a priority and continues to provide a rich 

45: system for investigation by experimental and theoretical physicists alike. The electronic

46: properties of sodium doped MOSFETs were first studied by Fowler and Hartstein \cite {Fowler1, Fowler2}

47: in the 1970s. They reported a single, broad peak in the subthreshold drain current against gate 

48: voltage and attributed it to the formation of an impurity band induced by the presence of sodium ions 

49: near the Si-SiO$_2$ interface. Further studies of narrow channel devices ($\sim$\,100\,nm) demonstrated

50: a series of reproducible sharp peaks \cite{Fowler3}, while later experiments found evidence for resonant

51: tunneling between localised states in the channel. \cite{Fowler4, Kopley, Popovic} For sufficiently

52: low impurity concentrations, the overlap between neighbouring localized electron wavefunctions

53: and consequently the hybridisation of their excited states is predicted to be reduced \cite{Erginsoy},

54: splitting the single impurity band observed at high concentrations into the ground and excited bands as

55: modeled by Ghazali. \cite{Ghazali} Increasing the resistivity of the silicon substrate reduces the scattering

56: from acceptors at the Si-SiO$_2$ interface, allowing the possibility for such a band splitting to be 

57: experimentally observed in the transport. In this paper, we will present evidence for the observation 

58: of two separate impurity bands with a soft gap, based on analysis of the temperature dependent

59: conductivity below 20\,K.

60: \newline\indent

61: The device we used is a MOSFET fabricated on a (100) oriented p-silicon wafer and was subsequently 

62: patterned in the circular Corbino geometry to eliminate Hall voltages and possible leakage paths. The 

63: effective gate channel length and interior width were respectively 1\,$\mu$m and 346\,$\mu$m. A high 

64: resistivity wafer (10$^4$\,$\Omega$.cm) provided a background concentration of less than 

65: $10^{12}$\,cm$^{-3}$ of boron corresponding to a mean distance between impurities of 1\,$\mu$m. A 

66: 35\,nm gate oxide was grown at 950\,$^{\circ}$C in a dry, chlorine-free oxygen atmosphere. The 

67: phosphorous implanted and aluminium sputtered contacts were highly metallic and Ohmic at all 

68: temperatures investigated. Sodium ions were introduced onto the oxide surface by immersing the

69: device in a $10^{-7}$\,N solution of high purity sodium chloride (99.999\,$\%$) in de-ionised water. The

70: surface of the chip was dried with nitrogen gas and an aluminium gate subsequently evaporated.

71: To observe or remove the low temperature conductivity structures, the mobile ions are drifted through 

72: the oxide to the Si-SiO$_2$ interface, or returned to the Al-SiO$_2$ interface by applying either a $+4$\,V 

73: or a $-4$\,V DC gate-substrate bias for 10\,min at $65^\circ$C before the device is cooled down to helium

74: temperature at which sodium looses its diffusivity in the oxide. All measurements were performed using 

75: standard low-noise lockin techniques with an amplifier of 10$^8$V/A. The AC excitation was maintained 

76: at 15\,$\mu$V with a frequency of 11\,Hz. Suitable RC filters were employed to eliminate any DC offset from the

77: amplifier. The gate voltage was controlled by a high resolution digital to analog converter. All experiments

78: were performed in an $^3$He cryostat and the temperature was measured by a calibrated germanium

79: thermometer.

80: \newline\indent

81: Fig.\,1 shows the conductivity $\sigma$ of our device at 300\,mK versus gate voltage $V_\textup{g}$ for the

82: case where the sodium ions had been drifted to the Si-SiO$_2$ interface. Two groups of peaks appear clustered

83: around $V_g\,=\,-2$\,V and $-0.5$\,V and separated by a region of low conductivity and limited by noise. 

84: The origin of the peaks themselves will be discussed later. Following a $-4$\,V drift, no structure was detectable

85: over the full range of gate voltages investigated but a difference in threshold voltage of 0.2\,V was found 

86: at 77\,K between the characteristics of the device following $+\,4$\,V and $-4$\,V drifts. This was attributed

87: to the presence of mobile charges close to the Si-SiO$_2$ interface at a concentration of 

88: 3.7$\times 10^{11}$\,cm$^{-2}$ corresponding to a mean impurity separation of 16\,$\pm$\,1\,nm. In a reference

89: device where no sodium was introduced, no sub-threshold conductivity peaks and no shift of the threshold

90: voltage appeared for any drift conditions investigated. Tunneling through the oxide and other leakage currents

91: were discounted as the gate leakage current was below 50\,fA at 4.2\,K and was approximately constant over 

92: the range of gate voltages used. We remark also that no hysteresis and thus no charging effects that have

93: been reported in similar devices \cite {Travlos} were observed here.

94: \newline\indent

95: The presence of two distinct ranges of $V_\textup{g}$ where peaks appear suggests the possibility

96: of a split impurity band (Fig.\,1). Such a splitting into a ground and an excited state is expected to 

97: happen for low-doping concentration \cite{Ghazali, Serre} if one takes into account the overlaps 

98: between impurity wavefunctions and uses a multi-band formalism \cite{Klauder}. The 

99: conductivity of a Si-MOSFET is not directly related to the density of states and the fact that we see

100: two regions of high conductivity separated by a region of low conductivity is only an indirect indication

101: that the device density of states consists of two bands separated by a gap. In fact, through

102: the Kubo formalism, conductivity tends to be related to local paths through a disordered device but

103: density of states is a global property.  In order to show that  the density of states splits into two bands,

104: we have looked at the temperature dependence of the conductivity at a series of different gate voltages.

105: For all gate voltages studied, the conductivity decreases non-monotonically as temperature is lowered

106: (Fig.\,2). In the range 1\,K to 20\,K, we observe the characteristics of hopping conduction so that the 

107: conductivity $\sigma(T)$ is fitted to the generalised equation :

108: 

109: \begin{figure}

110: \centering

111: \resizebox{!}{6cm}{\includegraphics{Figure1}}

112: \caption{The source drain conductivity versus gate voltage at 300\,mK following a +4\,V drift and a

113: -4\,V drift (dotted line).}

114: \end{figure}

115: 

116: \begin{eqnarray}\label{eqn:equation1}

117: \sigma = \sigma_0\,T^{-\gamma s}\,\textup{e}^{-\left(\frac{T_0}{T}\right)^s}

118: \end{eqnarray}

119: \noindent

120: where $T_0$ and $\sigma_0$ depend on gate voltage. 

121: 

122: \begin{figure}

123: \centering

124: \resizebox{!}{5.7cm}{\includegraphics{Figure2}}

125: \caption{a) Temperature dependence of the conductivity for $V_\textup{g}$\,=\,0.1

126: ($\circ$), $-0.24$ ($\bullet$),$ -1.48$\,V ($\star$) and $-2.26$\, ($\triangleleft$), b) Variation of the 

127: reduced $\chi^2$ with $s$ for $\gamma\,=\,2$  for the gate voltages listed showing minima at 

128: $s$\,=\,0.412, 0.406, 0.394 and 0.385 respectively and c) Variation of the reduced $\chi^2$ with 

129: $\gamma$ for $s$ equal to the optimum value found in 2(b), for the appropriate gate voltages, 

130: displaying consistency with a minimum at $\gamma\,=\,2$ (dotted line). Lines in b) represent the 

131: exponent for the Mott hopping regime (left) and Efros and Schklovskii regime (right).}

132: \end{figure}

133: 

134: \begin{figure}

135: \centering

136: \resizebox{!}{6cm}{\includegraphics{Figure3}}

137: \caption{Temperature dependence of the conductivity for $V_\textup{g}$\,=\,0.1 ($\circ$),

138: $-0.24$ ($\bullet$), $-1.48$\,V ($\star$) and $-2.26$\, ($\triangleleft$) for the optimum values of $\gamma$

139: and $s$ defined in Eq.\,1. For clarity, curves are shifted downward respectively by 0, 1, 2 and 3 from their 

140: original values.}

141: \end{figure}

142: 

143: The best values for $\gamma$ and $s$ are found by minimizing the value of the reduced chi-square 

144: deviation $\chi^2$ with standard procedures. Figs\,2(a)$-$(b) show the optimum values determined for 

145: four  gate voltages, one point in the threshold region ($V_g$\,=\,0.1\,V), the upper band ($V_g\,=\,-0.24$\,V), 

146: the gap ($V_g\,=\,-1.48$\,V) and the lower band ($V_g\,=\,-2.26$\,V). The resulting fits for $\sigma(T)$

147: are valid over three orders of magnitude in $\sigma$ (Fig.\,3). Studies in gate voltage show that 

148: $\gamma\,=\,1.98\,\pm\,0.04$ and $s\,=\,0.39\,\pm\,0.02$ for V$_{\textup{g}}$ below 0.25\,V. Above

149: this point, the value of $s$ decreases rapidly towards $s\,=\,1/3$ for $T\,\geq\,4$\,K and the range is 

150: better described by Mott hopping conduction \cite{Mott}. It is worth noticing that the smallest values of 

151: $s$ are obtained for band centre regions $-0.8\,V\,<\,V_{\textup{g}}\,<\,-0.5\,V$ and 

152: $-2.2\,V\,<\,V_{\textup{g}}\,<\,-2.1\,V$ for which the hopping lengths are relatively smaller and the Coulomb

153: interactions weaker.

154: \newline\indent

155: We emphasise that the use of temperature dependent exponential prefactors allows the fine distinction 

156: between Mott ($s\,=\,1/3$), Efros-Schklovskii ($s\,=\,1/2$) \cite {Efros} and~the regime under study 

157: ($s\,\sim\,0.39$). The finding that $\gamma\, \sim\,2$ is consistent with the formulation for $\sigma(T)$ 

158: given by Allen and Adkins \cite{Allen, Mansfield} if it is rederived for the 2D case. This gives 

159: the conductivity in terms of both the localization length $\xi$ and the density of states at the Fermi level

160: $n(E_F)$. Accounting for the fact that Coulomb interactions between electrons in different localized states

161: modifies the density of states close to the Fermi level so that $n(E)\,=\,N_0\,\left|\,E-E_F\,\right|^p$, we

162: obtain :

163: 

164: \begin{eqnarray}\label{eqn:equation2}

165: \sigma = \sigma_0\,T^{-\gamma\left(\frac{p+1}{p+3}\right)}\,\textup{e}^{-\left(\frac{T_0}{T}\right)^{\frac{p+1}{p+3}}}

166: \mbox{ with } \gamma\,=\,2

167: \end{eqnarray}

168: 

169: \begin{eqnarray}\label{eqn:equation3}

170: \sigma_0 = \frac{A_0}{\xi^2 \left(p+3\right)^2} {T_0}^{2\left(\frac{p+1}{p+3}\right)}

171: \end{eqnarray}

172: 

173: \begin{eqnarray}\label{eqn:equation4}

174: k_BT_0 ={ \left(\frac{p+3}{p+1} \right) } ^ {\frac{p+3}{p+1}}  { \left[ \frac{ \left( {p+1} \right)^3 } 

175: { \pi N_0 \xi^2 } \right] }^ {\frac{1}{p+1}}

176: \end{eqnarray}

177: 

178: \noindent

179: where $A_0$ is a constant depending on the electronic properties of bulk silicon and $k_B$ the Boltzmann

180: constant.

181: \newline\indent

182: The conditions given by Allen \cite{Allen} for the use of the equations Eqs.\,2-4 are satisfied in our device. 

183: Recent calculations showed that sodium ions in the oxide may trap either one or two valence electrons 

184: against the Si-SiO$_2$ interface and that the wavefunctions of the localized states remain hydrogen-like.

185: \cite{Barnes}  Also, hopping conduction is present in our device but Coulomb interactions are such as the 

186: only mobile electrons are found in an energy band of few $k_BT$ around the Fermi level. The resulting Coulomb 

187: gap for $s\,=\,0.39$ (i.e. $n(E)\,\sim\,|\,E-E_F\,|^{0.30}$) is much sharper than for the Efros regime where

188: $n(E)\,\sim\,|\,E-E_F\,|$. This behaviour has been predicted by Blanter and Raikh \cite{Blanter} while studying

189: 2D systems localized by disorder. They have shown that a metallic gate close to the interface

190: provides image potentials that modify the density of states to the form $n(E)\,=\,n(E_F)+N_0\,|\,E-E_F\,|^{1/3}$ 

191: at $T$\,=\,0\,K and thus gives an exponential dependence of $T^{-0.4}$ for the conductivity.  This behaviour is

192: explained by the fact that the oxide thickness plays the role of the screening length and that initial and final states

193: become electrostatically independent at low temperature when the hopping length $R$ becomes greater than

194: twice the oxide thickness $d$. This then produces a crossover from the Efros to the Mott regime. The localization

195: length in our device was approximately estimated by fitting the conductivity with $p\,=\,0$, using the Mott formula

196: for $k_B T_0$ and taking the 2D value for the density of states. For $V_g\,=\,-0.4$\,V, this gives 

197: $\xi\,\sim\,$22\,nm and a hopping length $R\,\sim\,$63\,nm at 1\,K. Our device has a gate oxide of $d\,=\,35$\,nm 

198: and the Coulomb interactions may be screened by the electrostatic gate as $R\sim\,2d$. This demonstrates the 

199: device may well be in the regime described by Blanter and Raikh. Fits of the conductivity using $\gamma\,=\,2$

200: and $s\,=\,0.4$ are still valid for $V_g$ smaller than 0.25\,V and from 1\,K to 18\,K typically, so we will proceed

201: with these values. In this regime, the density of states at $E_F$ is given by :

202: 

203: \begin{eqnarray}\label{eqn:equation5}

204: n(E_F)\,=\,n_0(E_F)-\frac{2\pi{n_0(E_F)}^2de^2}{4 \pi \epsilon_0 \epsilon_r} \nu

205: \end{eqnarray}

206: \noindent

207: \begin{eqnarray}\label{eqn:equation6}

208: n_0(E_F) = {\left( \frac{N_0}{2\pi} \right) }^{\frac{1}{2}}  {\left( \frac{4 \pi \epsilon_0 \epsilon_r}{2 e^2 d^2} \right) }^{\frac{1}{3}}

209: \end{eqnarray}

210: 

211: \noindent

212: where $\epsilon_r$ is the permittivity of silicon.

213: \newline\indent

214: These equations have been derived at $T\,=\,0$K. At higher temperature but for $k_BT\,\ll\,V(d)$, where $V$ is the coulomb 

215: potential energy, the gap is partly filled. The Coulomb interactions then become negligible for energies below $V(d)$ and 

216: the density of states saturates such as $n_0(E_F)\,=\,n_0(E_F+V(d))$.\cite{Aleiner} This condition decreases the value 

217: of $\nu$ from 1 down to 0.05. Combining (3) and (4), the parameters $N_0$ and $\xi$ were expressed in terms of $T_0$ 

218: and $\sigma_0$, values easily accessed experimentally. The density of states was extracted using equation (5) and (6).

219: To find the value of $A_0$, we used $\Xi$\,=\,8.91\,eV for the effective deformation potential of acoustic phonons 

220: \cite{Fischetti} as well as 3800\,m.s$^{-1}$ for the speed of surface acoustic phonons in silicon \cite{SAW}. This gives 

221: $A_0$\,=\,6.69 $\times$ $10^{-19}\,e^2.h^{-1}.\textup{m}^2$. 

222: 

223: \begin{figure}

224: \centering

225: \resizebox{!}{6cm}{\includegraphics{Figure4}}

226: \caption{Variation of the DOS at $E_F$ with gate voltage. The dashed and dotted lines represent

227: respectively the background density due to the conduction band tail and the upper band tail. The inset is the

228: relative variation of the DOS to the background in the same units.}

229: \end{figure}

230: 

231: The density of states is shown in Fig.\,4. The value we find is more than one order of magnitude lower than the pure 2D

232: case ($\sim\,1.6 \times 10^{14}\,$eV$^{-1}.$cm$^{-2}$ with valley degeneracies). The large background is predominantly 

233: due to the conduction band tail which spreads over the full range of $V_g$ studied. The upper band also has a

234: significant tail. The presence of density of states tails is a common occurance in disordered systems with low

235: impurity concentration and localized wavefunctions \cite{Halperin, Zittartz, Kane} but its linear shape for $V_g\,<\,0\,V$

236: is unusual. It has been attributed to the formation of regions of constant local potential energy at the Si-SiO$_2$

237: interface and containing a random number of charges.\cite{Arnold} Two regions of higher density are superimposed

238: on the background density and correspond to the upper and lower groups of peaks. This confirms that the structure

239: observed in the conductivity is due to the presence of two separate bands. By numerically subtracting the background

240: density and integrating over the appropriate gate voltage and by supposing a linear relation between the gate voltage

241: and the surface potential energy, we estimate the upper band contains approximately 3 times the number of states

242: as the lower band. This value is an upper bound as energetically deeper states could not be accessed experimentally.

243: \newline\indent

244: The variation of the localization length $\xi$ (Fig.\,5) follows that of the density of states, showing

245: the two bands do correspond to the more conductive regions. The value of $\xi$ decreases

246: rapidly when approaching the threshold voltage. This is expected as the region between

247: $V_g\,=\,0\,$V and $V_g\,=\,0.4\,$V is both in the band tails of the conduction band and the

248: upper band, the conduction band edge being well above 0.4\,V. Thus, this region is still

249: a region of strong localization. Nevertheless, the value of $\xi$ is expected to rise once the

250: Fermi energy crosses the conduction band edge. Finally, taking into account fitting errors as well as the 

251: discrepancy in the value of the deformation potential, we estimate the localization length within 5 $\%$ 

252: and the density of states within 6 $\%$. From the same derivation which gave (1), (2) and (3), the 

253: transmission coefficient is obtained :

254: 

255: \begin{figure}

256: \centering

257: \resizebox{!}{5.9cm}{\includegraphics{Figure5}}

258: \caption{Variation of the localisation length ($\bullet$) and the transmission

259: coefficient at 1K ($\circ$) with gate voltage.}

260: \end{figure}

261: \noindent

262: \begin{eqnarray}\label{eqn:equation7}

263: \textup{ln}\left(T_E\right) =\,-\frac{2}{p+3}\left(\frac{T_0}{T}\right)^{\frac{p+1}{p+3}}

264: \end{eqnarray}

265: 

266: The experimental variation of the transmission coefficient in gate voltage (Fig.\,5)

267: shows that the conductivity of the two bands mostly comes from the higher mobility

268: of the states of energies within the upper and lower bands and less from the increase in the

269: density of states.

270: \newline\indent

271: In conclusion, we have observed an unusual hopping regime with an exponent 0.4 which results from the

272: screening of the Coulomb interactions by the metal gate. We have shown that, consequently, both the 

273: localization length and the density of states can be extracted from the temperature dependence

274: of the conductivity. This analysis has given strong evidence for the existence of two separate bands and

275: a soft gap at low temperature. This may result from the splitting of the impurity band into a lower and

276: an upper band in presence of Coulomb interactions. The formation of the two bands results itself from the 

277: presence of a low concentration of sodium impurities close to the Si-SiO$_2$ interface. The conditions

278: for the observation of such a formation may be the creation of deep but well separated impurity potentials

279: at the interface, resulting in electron localization. Finally, the electron screening may be sufficiently weak

280: and the disorder not too strong to allow a Mott-Hubbard transition to take place. The two bands could

281: then possibly be Hubbard bands.

282: \newline\indent

283: We would like to thank Drs T. Bouchet and F. Torregrossa from Ion Beam System-France for

284: the process in the device as well as funding from the U.S. ARDA through U.S. ARO grant number DAAD19-01-1-0552.

285: 

286: \begin{references}

287: 

288: \bibitem{Fowler1} F. F. Fang, A. B. Fowler, Phys. Rev. 169, 619 (1967)

289: 

290: \bibitem{Fowler2} A. Hartstein, A. B. Fowler, Phys. Rev. Lett. 34, 1435 (1975)

291: 

292: \bibitem{Fowler3} A. B. Fowler, A. Hartstein and R. A. Webb, Phys. Rev. Lett. 48, 196 (1982)

293: 

294: \bibitem{Fowler4} A. B. Fowler, G. L. Timp, J. J. Wainer and R. A. Webb, Phys. Rev. Lett. 57, 138 (1986)

295: 

296: \bibitem{Kopley} T. E. Kopley, P. L. McEuen and R. G. Wheeler, Phys. Rev. Lett. 61, 1654 (1988)

297: 

298: \bibitem{Popovic} D. Popovic, A. B. Fowler and S. Washburn, Phys. Rev. Lett. 67, 2870 (1989)

299: 

300: \bibitem{Erginsoy} C. Erginsoy, Phys. Rev. 80, 1104 (1950)

301: 

302: \bibitem{Ghazali} A. Ghazali, A. Gold and J. Serre, Phys. Rev. B 39, 3400 (1989)

303: 

304: \bibitem{Travlos} V. Ioannou-sougleridis, A. G. Nassiopoulou and A. Travlos, Nanotechnology, 14, 11174 (2003)

305: 

306: \bibitem{Serre} J. Serre and A. Ghazali and A. Gold, Phys. Rev. B 39, 8499 (1989)

307: 

308: \bibitem{Klauder} J. R. Klauder, Ann. Phys. (N.Y.) 14, 43 (1961)

309: 

310: \bibitem{Mott} N. H. Mott, J. Non-Cryst. Solids 1, 1 (1969)

311: 

312: \bibitem{Efros} A. L. Efros and B. I. Schklovskii, J. Phys. C 8, L49 (1975)

313: 

314: \bibitem{Allen} F. R. Allen and C. J. Adkins, Phil. Mag. 26, 1027 (1972)

315: 

316: \bibitem{Mansfield} R. Mansfield, S. Abboudy and P. Fozooni, Phil. Mag. B 57, 6, 777 (1988)

317: 

318: \bibitem{Barnes} C. H. W. Barnes and A.V. Moroz, to be published

319: 

320: \bibitem{Blanter} Y. M. Blanter and M. E. Raikh, Phys. Rev. B 63, 075304 (2001)

321: 

322: \bibitem{Aleiner} I. L. Aleiner and B. I. Shklovskii , Phys. Rev. B 49, 13721(1994)

323: 

324: \bibitem{Fischetti} M. V. Fischetti and S. E. Vaux, J. Appl. Phys. 80, 2234 (1996)

325: 

326: \bibitem{SAW} G. T. Andrews, M. J. Clouter and J. Zuk, Semicond. Sci. Technol. 19, 1306 (2004) 

327: 

328: \bibitem{Halperin} B. I. Halperin and M. Lax, Phys. Rev. 148, 722 (1966)

329: 

330: \bibitem{Zittartz} J. Zittartz and J. S. Langer, Phys. Rev. 148, 741 (1966)

331: 

332: \bibitem{Kane} E. O. Kane, Phys. Rev. 131, 79 (1963)

333: 

334: \bibitem{Arnold} E. Arnold, Appl. Phys. Lett., 25, 705 (1974)

335: 

336: 

337: \end{references}

338: 

339: 

340: \end{document}

341: