Tunable Light Emission from Quantum-Confined Excitons in TiSi2

Broadband antireflection on the silicon surface realized by Ag nanoparticle-patterned black silicon. Y. Wang , Y. P. Liu , H. L. Liang , Z. X. Mei , X...
0 downloads 0 Views 191KB Size
NANO LETTERS

Tunable Light Emission from Quantum-Confined Excitons in TiSi2-Catalyzed Silicon Nanowires

2006 Vol. 6, No. 9 2140-2144

Alex R. Guichard,*,† David N. Barsic,† Shashank Sharma,‡ Theodore I. Kamins,‡ and Mark L. Brongersma† Geballe Laboratory for AdVanced Materials, Stanford UniVersity, 476 Lomita Mall, Stanford, California 94305, and Quantum Science Research, Hewlett-Packard Laboratories, 1501 Page Mill Road, Palo Alto, California 94304 Received June 5, 2006; Revised Manuscript Received July 13, 2006

ABSTRACT Visible and near-infrared photoluminescence (PL) at room temperature is reported from Si nanowires (NWs) grown by chemical vapor deposition from TiSi2 catalyst sites. NWs grown with average diameter of 20 nm were etched and oxidized to thin and passivate the wires. The PL emission blue shifted continuously with decreasing nanowire diameter. Slowed oxidation was observed for small nanowire diameters and provides a high degree of control over the emission wavelength. Transmission electron microscopy, PL, and time-resolved PL data are fully consistent with quantum confinement of charge carriers in the Si nanowire core being the source of luminescence. These light emitting nanowires could find application in future CMOS-compatible photonic devices.

Semiconductor nanowires (NWs) have exceptional electronic and optical properties, which make them ideal candidates for fundamental physics studies and next-generation nanoelectronic and photonic devices. For practical applications, chemical vapor deposition (CVD) is their preferred method of growth because of its widespread use in industry. CVD grown NWs are already used extensively as the key elements in nanoscale electronic devices, such as p-n junctions, field effect transistors, and logic circuits.1-3 More recently, they have also proven useful as basic building blocks for photonic components, including optically and electrically pumped lasers, polarization-sensitive detectors, electrooptic modulators, and solar cells.4-7 The majority of such NW-based optical applications have involved the use of direct band gap II-VI and III-V semiconductors, and photonic device applications employing indirect band gap Si NW have not yet been developed. Although high-efficiency electroluminescence (EL) and photoluminescence (PL) have already been observed from Si nanocrystal-doped silica films, devices based on this material exhibit oxide wearout during charge injection, and device reliability is limited.8,9 These issues can potentially be circumvented in the nanowire geometry where charge can be injected directly into the wire at both ends without traversing an oxide. For this reason, CVD-grown Si NWs could provide a promising new route to CMOS* Corresponding author. E-mail: [email protected]. Phone: (650) 723-6466. † Stanford University. ‡ Hewlett-Packard Laboratories. 10.1021/nl061287m CCC: $33.50 Published on Web 08/09/2006

© 2006 American Chemical Society

compatible light-emitting diodes. Unfortunately, the most popular catalyst for CVD growth of Si NWs is Au, a deeplevel trap in Si that causes fast nonradiative decay of excited carriers. Not surprisingly, we have not observed luminescence from Au-catalyzed Si wires, despite numerous attempts. Here, we employ TiSi2-catalyzed nanowires in our optical experiments. Ti impurities do not produce trap levels near the middle of the Si band gap, and TiSi2 is frequently used as a contact material to active regions in Si integratedcircuit technology.10 Other groups have already reported light-emission from Si NW samples grown by a variety of synthesis techniques.11-13 However, additional systematic studies are needed to demonstrate conclusively that light emission can be obtained from quantum-confined excitons in the Si nanowire core. In particular, PL lifetime studies have been lacking to distinguish the potential light emission from quantum-confined excitons14-16 from other well-known luminescence sources in nanoscale Si systems, including radiative recombination from surface states at Si-SiO2 interfaces, impurity-center recombination within the crystal, defects in the passivating oxide, and chemiluminescence from linear Si-chain molecules.17 In this Letter, we present an in-depth study of the optical emission from TiSi2-catalyzed Si NWs and show that the data are fully consistent with the NW PL being due to radiative recombination of quantum-confined excitons inside the Si NWs.15,17,18 Moreover, we demonstrate that the NW

Figure 2. Average nanowire diameter as a function of oxidation time (diamonds, from transmission electron microscopy analysis; solid line, predicted from Massoud model for a planar surface). The error bars indicate the standard deviation from the mean core diameter for each sample.

Figure 1. (a) Scanning electron micrograph of as-grown TiSi2catalyzed Si nanowire (NW) sample used for photoluminescence experiments. (b) High-resolution transmission electron microscopy (TEM) micrograph of an as-grown Si NW with native oxide. (c) Dark field TEM micrograph of a NW oxidized for 30 min. The bright region within the gray NW is the crystalline Si core.

luminescence can be tuned by controlling the nanowire diameter. To this end we performed a combined transmission electron microscopy (TEM), PL, and time-resolved PL study of the size-dependent optical properties of the NWs. The results of this study could enable the realization of a range of Si NW-based photonic devices using CMOS-compatible fabrication methods. TiSi2-catalyzed Si NWs were grown by depositing Ti on an Sb-doped, n-type Si(100) wafer. The samples were then annealed at 900 °C in H2 for 5 min to form TiSi2 catalyst islands. To grow Si NWs, the samples were exposed to a mixture of SiH4, HCl, and H2 at 680 °C for 30 min; the scanning electron micrograph in Figure 1a) shows a representative nanowire sample. Following NW growth, the samples were etched in 2% HF for 1 min to remove the native oxide from the NWs. The samples were then thermally oxidized at 950 °C in dry O2 for times ranging from 10 to 120 min. Each oxidation step was followed by annealing in Ar at 950 °C for 10 min to ensure diffusion of the oxidant species to the Si-SiO2 interface.19 High-resolution TEM and dark-field TEM (HR-TEM, DF-TEM) were performed after each oxidation using a Philips CM20 field-emission TEM. PL measurements were performed using the 488 nm line of an Ar+ ion laser as the excitation source, with a pump intensity of 10 W/cm2. The luminescence spectra were taken with a Si charge-coupled device coupled to a grating spectrometer. Time-resolved PL measurements were performed by acousto-optical modulation of a 6 W/cm2 excitaNano Lett., Vol. 6, No. 9, 2006

tion beam and PL detection with a GaAs photomultiplier tube in conjunction with a multichannel photon-counting system. The modulation frequency of the excitation beam was 400 Hz, and the time resolution of the system was 160 ns. The TEM micrographs in parts b and c of Figure 1 depict a HR-TEM micrograph of an as-grown NW kept in air and a DF-TEM micrograph of a NW after thermal oxidation for 30 min, respectively. The HR-TEM image clearly shows the fringe contrast from the 〈111〉 lattice planes and a 1.8 nm thick native oxide. The DF-TEM image shows contrast between the c-Si core and the amorphous oxide around it. Images taken from as-grown and oxidized samples show continuous Si wire cores with a 3-6 Å standard deviation in the measured diameters along the wires. From both types of images the wire diameters can be obtained with an accuracy that is significantly better than the distribution in NW diameters in every sample. Figure 2 shows the measured diameters of the crystalline Si NW cores as a function of oxidation time. Every data point in the graph shows an average wire diameter based on measurements on 20 different NWs, and the error bar provides a measure of the standard deviation of the size distribution. The TEM analysis of NWs clearly shows a progressive decrease in the NW diameter and in the width of the size distribution. It is important to note that the wires are not fully oxidized even for extended oxidation time, and a self-limited crystalline Si NW core diameter of about 3.3 nm is found. Figure 3 shows normalized PL spectra of Si NWs after thermal oxidation for 10, 15, 30, 40, and 120 min. After 10 min of oxidation, a broad luminescence band is observed, peaking at 1.55 eV and with a full width at half maximum (fwhm) of 330 meV. For longer oxidation times, the luminescence band maintains a constant fwhm and shifts to the blue. For short oxidation times, the peak energy shifts rapidly with increasing oxidation time, and for longer 2141

Figure 3. Photoluminescence spectra taken after 10, 15, 30, 40, and 120 min of oxidation of TiSi2-catalyzed Si nanowires. Inset plots photoluminescence peak energy as a function of oxidation time.

oxidation times the peak energy saturates at around 1.66 eV, consistent with a self-limited NW diameter. The inset shows a plot of the PL peak energy vs average Si nanowire diameter measured by DF-TEM, demonstrating the blue shift of peak luminescence wavelength with decreasing diameter. Figure 4 shows the dependence of the photoluminescence decay lifetime as a function of the detection energy for a NW sample oxidized for 30 min. The inset shows a typical PL decay trace taken at energy of 1.65 eV. The lifetimes at this and other energies in the range from 1.45 to 1.9 eV were obtained by fitting such decay traces with a stretched exponential decay function of the following form

[( ) ]

IPL(t) ) I0 exp -

t τd

β

where I0 is the PL intensity at t ) 0, τd is the decay lifetime, and β is a constant between 0 and 1, where 1 corresponds to single-exponential decay.20 The lifetimes are in the tens of microsecond range and tend to decrease with increasing detection energy. The magnitude of β monotonically decreases from 0.63 to 0.38, indicating that the decay becomes increasingly multiexponential at higher detection energies. The HR-TEM and DF-TEM results in Figure 1 demonstrate that the diameter of CVD grown Si NWs can be reduced from about 16 nm to about 3.5 nm. Figure 2 clearly shows that the oxidation kinetics slow significantly as the oxidation progresses. This slowed oxidation can be exploited to produce NWs of well-controlled diameter, allowing systematic studies of size-dependent physical properties. To understand the decrease of the oxidation rate, we compared the oxidation rate to those predicted by conventional DealGrove and Massoud models for oxidation of planar surfaces.21,22 The solid line in Figure 2 shows the predicted Si core diameter, assuming the oxide shell grows at a rate that is equal to the oxidation rate predicted by the Massoud 2142

Figure 4. Variation of the decay lifetime and β with detection energy in the range from 1.46 to 1.91 eV, appropriate to the nanowire sample oxidized for 30 min. The inset shows a photoluminescence decay trace taken at 1.66 eV from TiSi2-catalyzed Si nanowire sample oxidized for 30 min. By use of a stretched exponential fit, a decay constant of 41 µs and β of 0.63 were found.

oxidation model for a planar (111) interface. Using fits from experimental data, the Massoud model predicts an oxide thickness of 18 nm after 15 min of oxide growth at 950 °C, which would consume the entire crystalline Si core.22 Several two-dimensional oxidation models have been formulated that treat the oxide as an incompressible viscous fluid,23 and others have used experimental data to fit the oxidation growth rates using empirical parameters for stress and oxide viscosity.24,25 Each of these models uses a different method to ascertain a stress (σnn) normal to the Si-oxide interface and an isotropic stress within the oxide, which is treated as a hydrostatic pressure (σo). Liu, et al.23 reported a similar reduction in the oxidation rate for pillars of Si defined by electron-beam lithography and found that oxidation stopped when the diameter of the crystalline core reached a few nanometers. This self-limiting oxidation is due to stress at the Si-oxide interface increasing the energy cost for creating the extra volume required for the forming oxide, and thus decreasing the oxidation reaction rate constant, k, by a factor exp(-σnn*(VSiO2 - VSi)/kBT), where VSiO2 and VSi are the volumes of SiO2 and Si, respectively.23 Furthermore, as noted in ref 23, the equilibrium concentrations of oxidant in the oxide can be decreased by a similar exponential factor containing σo, which reduces the oxidation rate as well. Thermodynamic components of the driving force for the oxidation reaction could be modified as well and slow the oxidation process. For example, in a small diameter nanowire the driving force for the creation of self-interstitial Si atoms during the formation of a SiO2 molecule will be altered. The final wire diameter after 120 min of oxidation is smaller than the 4.9 nm Bohr exciton radius of bulk silicon.16 For this reason, quantum-mechanical effects are expected to noticeably change the electronic band structure and optical properties of the wires. Quantum-confinement theory (QCT) for nanoscale Si structures predicts that the band gap (and Nano Lett., Vol. 6, No. 9, 2006

thus the PL peak energy) increases and the PL decay lifetime decreases with decreasing size.13,26 This behavior has been observed for many nanostructured Si systems, such as porous Si and Si nanocrystals.8,9,18,27 The results presented in Figures 2, 3, and 4 are consistent with the QCT as well. Figure 3 shows relatively broad PL spectra, which would be expected for NW samples with a distribution in sizes and thus band gaps. The reduction in NW diameter during oxidation observed by TEM is accompanied by a blue shift of the PL peak position. After 40 min of oxidation, both the NW diameter and the shift in the PL peak energy saturate, with final values of 3.5 nm and 1.66 eV, respectively. The observed blue shifts as a function of measured core diameter are in reasonable agreement with theoretical predictions made by Delerue et al. on Si NWs, who predicts a band gap energy dependence on diameter of ∼1/d1.39.18 A quantitative difference can be expected because the calculations were made for single H-passivated NWs; our PL experiments probe oxidized wires with a distribution of sizes. In addition, stresses on the order of a few hundred megapascals are potentially present inside the Si core, as discussed above. These stresses are expected to increase during the NW oxidation and to further change upon cooling the wires after the final Ar anneal at 950 °C for 10 min. The latter stress results from the difference in thermal expansion coefficients of Si (2.6 × 10-6 °C-1) and SiO2 (0.4 × 10-6 °C-1).28,29 The observed change in peak PL wavelength is thus a net effect of size- and strain-induced changes in the band gap. Studies of Si thin film growth on Ge and oxidationinduced strain show that roughly 1% tensile or compressive strain can cause significant splitting of the valence band edges and a reduction in the band gap on the order of 100 meV.30-32 Taking the stress within the Si to be equal to that at the oxideSi boundary, we calculated33 that the strain resulting from the oxidation-induced radial stress at the interface is on the order of 0.27%, which would result in an approximately 27 meV change in band gap.30-32 Strain due to the thermal contraction during cooling is of the same magnitude but of opposite sign.33 Although the net strain and the corresponding strain-induced modification of the band gap may differ somewhat from the calculated values, it seems that the large (∼500 meV) blue shift in the PL peak position can only be explained by quantum size effects. The magnitudes of the PL decay lifetimes are in the tens of microsecond range, similar to those observed for other luminescent nanostructured Si systems.15,27 The lifetimes tend to decrease with increasing detection energy. This behavior is also in agreement with QCT, which predicts that smaller wiressemitting higher energy photonssshould exhibit shorter decay lifetimes.13,18 For the NW samples in this study, the magnitude of β differs from unity and depends on the detection energy. Values of β < 1 are commonly observed in many luminescent systems and arise from different physical mechanisms. For example, a distribution in NW cross-sectional sizes and shapes, charge transport along the NW axis, and/or coupling of excitons to different nonradiative decay sites could both give rise to non-single Nano Lett., Vol. 6, No. 9, 2006

exponential decay. However, for well-separated and properly passivated Si nanoparticles β-values close to 1 have been observed.27 In summary, we have demonstrated room-temperature photoluminescence from TiSi2-catalyzed Si NWs that is attributed to the recombination of quantum-confined charge carriers in crystalline Si. HR-TEM and DF-TEM measurements verify that the wire core diameters are on the order of the Bohr exciton radius. PL studies show a blue shift and a reduction in PL lifetime with decreasing diameter that is in agreement with other experimental and theoretical studies on Si nanostructures in the literature. These Si NWs are ideal candidates for CMOS-compatible, electroluminescent, nanostructured-Si devices and could potentially enable low-cost fabrication of on-chip light emitters, large-area displays, and possibly electrically pumped, Si-based lasers.34 Acknowledgment. The authors thank Rohan Kekatpure for valuable discussions. This work was supported by a DoD MURI sponsored by the AFOSR (F49550-04-1-0437), by DARPA, and by Hewlett-Packard. References (1) Zheng, G. F.; Lu, W.; Jin, S.; Lieber, C. M. AdV. Mater. 2004, 16, 1890. (2) Yu, J. Y.; Chung, S. W.; Heath, J. R. J. Phys. Chem. B 2000, 104, 11864. (3) Wang, D.; Wang, Q.; Javey, A.; Tu, R.; Dai, H.; Kim, H.; McIntyre, P. C.; Krishnamohan, T.; Saraswat, K. C. Appl. Phys. Lett. 2003, 83, 2432. (4) Huang, M. H.; Mao, S.; Feick, H.; Yan, H. Q.; Wu, Y. Y.; Kind, H.; Weber, E.; Russo, R.; Yang, P. D. Science 2001, 292, 1897. (5) Wang, J. F.; Gudiksen, M. S.; Duan, X. F.; Cui, Y.; Lieber, C. M. Science 2001, 293, 1455. (6) Greytak, A. B.; Barrelet, C. J.; Li, Y.; Lieber, C. M. Appl. Phys. Lett. 2005, 87, 151103. (7) Huynh, W. U.; Dittmer, J. J.; Alivisatos, A. P. Science 2002, 295, 2425. (8) Kovalev, D.; Heckler, H.; Ben-Chorin, M.; Polisski, G.; Schwartzkopff, M.; Koch, F. Phys. ReV. Lett. 1998, 81, 2803. (9) Castagna, M. E.; Coffa, S.; Monaco, M.; Caristia, L.; Messina, A.; Mangano, R.; Bongiorno, C. Physica E 2003, 16, 547. (10) Kamins, T. I.; Williams, R. S.; Basile, D. P.; Hesjedal, T.; Harris, J. S. J. Appl. Phys. 2001, 89, 1008. (11) Lyons, D. M.; Ryan, K. M.; Morris, M. A.; Holmes, J. D. Nano Lett. 2002, 2, 811. (12) Li, C. P.; Sun, X. H.; Wong, N. B.; Lee, C. S.; Lee, S. T.; Teo, B. K. Chem. Phys. Lett. 2002, 365, 22. (13) Yu, D. P.; Bai, Z. G.; Wang, J. J.; Zou, Y. H.; Qian, W.; Fu, J. S.; Zhang, H. Z.; Ding, Y.; Xiong, G. C.; You, L. P.; Xu, J.; Feng, S. Q. Phys. ReV. B 1999, 59, 2598. (14) John, G. C.; Singh, V. A. Phys. ReV. B 1995, 52, 11125. (15) Delley, B.; Steigmeier, E. F. Appl. Phys. Lett. 1995, 67, 2370. (16) Hybertsen, M. S. Phys. ReV. Lett. 1994, 72, 1514. (17) Cullis, A. G.; Canham, L. T.; Calcott, P. J. J. Appl. Phys. 1997, 82, 909. (18) Delerue, C.; Allan, G.; Lannoo, M. Phys. ReV. B 1993, 48, 11 024. (19) Using diffusion data from ref 22, the diffusion coefficient for O in SiO2 on Si(100) is calculated to be 4.3 × 10-9 cm2/s. After 10 min of an inert anneal, this corresponds to a diffusion length of 15.9 µm, a length much greater than the thickness of the oxide surrounding the NW. Hence, we conclude that all of the oxidant has diffused to the Si/oxide interface and reacted. (20) Pavesi, L.; Ceschini, M. Phys. ReV. B 1993-I, 48, 17625. (21) Deal, B. E.; Grove, A. S. J. Appl. Phys. 1965, 36, 3770. (22) Massoud, H. Z.; Plummer, J. D.; Irene, E. A. J. Electrochem. Soc. 1985, 132, 1745. 2143

(23) Liu, H. I. Fabrication and Properties of Si nanostructures. Ph.D. Thesis, Stanford University, 1995. (24) Kao, D. B.; McVittie, J. P.; Nix, W. D.; Saraswat, K. C. IEEE Trans. Electron DeVices 1988, 35, 25. (25) Sutardja, P.; Oldham, W. G. IEEE Trans. on Electron DeVices 1989, 36, 2415. (26) Sanders, G. D.; Chang, Y. C. Phys. ReV. B 1992, 45, 9202. (27) Brongersma, M. L.; Polman, A.; Min, K. S.; Tambo, T.; Atwater, H. A. Appl. Phys. Lett. 1998, 72, 2577. (28) Yim, W. M.; Paff, R. J. J. Appl. Phys. 1974, 45, 1456. (29) Callister, W. D. Materials Science and Engineering: An Introduction, 5th ed.; John Wiley & Sons: New York, 2000; p 662. (30) Van de Walle, C. G.; Martin, R. G. Phys. ReV. B 1986, 34, 5621. (31) Nayak, D. K.; Chun, S. K. Appl. Phys. Lett. 1994, 64, 2514. (32) Shiraishi, K.; Nagase, M.; Horiguchi, S.; Kageshima, H.; Uematsu, M.; Takahashi, Y.; Murase, K. Physica E 2000, 7, 337.

2144

(33) Using a viscosity-dependent stress model similar to that developed in ref 23, we calculate a radial compressive stress at the Si-SiO2 interface of 0.43 GPa after 120 min of oxidation. Using an elastic modulus of 161 GPa (the mean of the 〈111〉, 〈110〉, and 〈100〉 values reported in ref 28), the corresponding strain is found to be 0.27%. Using the thermal expansion values given in the text, the thermal strain is calculated to be of magnitude 0.2%; this component should be tensile since the Si core decreases in size more than does the oxide sheath as the system cools. Therefore, the thermal strain should reduce the compressive strain created during oxidation. If the two components are of comparable magnitude, the net strain in the wire is small. (34) Dal Negro, L.; Cazzanelli, M.; Pavesi, L.; Ossicini, S.; Pacifici, D.; Franzo, D.; Priolo, F.; Iacona, F. Appl. Phys. Lett. 2003, 82, 4636.

NL061287M

Nano Lett., Vol. 6, No. 9, 2006