Open Access
Add to BibSonomyAbstract
A feasible way to synthesize a surface-enhanced Raman scattering (SERS) substrate has been developed, where Ag nanoparticles (AgNPs) of different size and morphology are assembled on the surface and sidewalls of the aligned carbon nanotube (CNT) arrays via magnetron sputtering and high-temperature annealing. Our results show that the optimized substrate is performed by annealing temperature at 450 °C for 30 min. The state of the obtained AgNPs makes a significant contribution to the high sensitivity of SERS to R6G molecules, and the substrate has an enhancement factor (EF) on the order of ~1010. Meanwhile, the Ag/CNT arrays keep a good reproducibility with the average RSD values being less than 0.01 for all major Raman peaks. The temporal stability of our substrates has been also appeared, which indicates that the Ag/CNT arrays can be used as stable substrates for the production of enhanced SERS signals for up to three months under ambient conditions.
© 2014 Optical Society of America
1. Introduction
Surface-enhanced Raman scattering (SERS) is sufficiently sensitive for single molecule detection, which has been commonly used as a molecular sensing technology [1,2]. It is widely accepted that electromagnetic (major contribution) and chemical enhancement contributes to SERS enhancement. While field enhancement occurs over a nanostructure surface, SERS signals are strongest on “hot spots” between nanoparticles. Most work has focused on different substrates, which is to optimize the density, sensitivity, reproducibility and lifetime [3,4]. We know that optical scattering and Raman light collection occur in a three-dimensional (3D) focal volume [5]. In order to maximize the Raman scattering, the substrates should contain enough hot spots in a 3D volume that is matched to the optical system. Compared to flat or two-dimensional (2D) substrate, 3D substrate is able to form more hot spots [6]. Many 3D substrates were described and found with sensitive detection, such as Au/TiO2 arrays [7], porous silicon [8], nano-cavity arrays [9], Au nanoparticles decorated black silicon [10], silver nanoparticles protected by monolayer graphene [11], Cu nanolines [12], etc. In our previous study, the CNT arrays coated by Au-sol through a simple drop method was investigated, however, the sensitivity of the SERS substrate remains modest owing to a limited number of hot spots on the sidewalls of CNT arrays [6].
Therefore, further efforts are needed to design and assemble highly effective substrates, with high-density metal nanoparticles that have nanogaps to form sufficient hot spots for better SERS enhancement performance. Compared with Au or Cu, the plasmonic spectral window covered by Ag nanostructures is much larger, in the entire range from visible to infrared [11]. Furthermore, the operation of Ag nanostructures can be carried out at a much higher frequency without apparent signal intensity loss, because of the absence of inter-band absorptions [11,13,14].
In this paper, we describe the experiments of a tunable 3D SERS substrate based on vertically aligned CNTs and Ag nanoparticles fabricated by magnetron sputtering and high temperature annealing.
2. Structures and preparation
Our SERS substrate is shown in Fig. 1(a), aligned-CNT arrays decorated by Ag nanoparticles, which provides a paradigm to realize 3D SERS substrates with high-density nanoparticles. Carbon nanotube has huge specific surface areas, which can adsorb more nanoparticles, especially for three-dimensional CNT arrays. The near-field for a silver substrate is stronger than a gold substrate, which could explained as Ag being a much better plasmonic material due to lower loss in the visible and NIR ranges [15]. With SERS, it was reported that Ag nanoparticles are able to enhance the efficiencies of surface optical process by 14-15 orders of magnitude [2]. The advantages of the structure are the strong electromagnetic interaction of Ag nanoparticles on a single CNT wall, Ag nanoparticles on different neighboring CNT walls, and the separation area of CNTs and AgNPs (coupling effect of CNT and AgNPs [16]), expressed in Fig. 1(b).
Fig. 1 Schematic of (a) the AgNPs/CNTs SERS substrate and (b) the enhancement effects of Raman signal of Ag nanoparticles. There are three major effects.
In order to observe the hot spots effects clearly, a simulation method of DDA (discrete dipole approximation) is used. The simulation condition is incident light of 632.8 nm laser, Ag nanoparticles size of 100 nm, 10 nm nanogap between two nanoparticles, surrounding medium refractive index of 1.33, and the results are shown in Fig. 2.
Fig. 2 Electromagnetic field simulation of (a) a single Ag nanoparticle and (b) two Ag nanoparticles. The axis unit is nm.
The incident out-of-plane wave is polarized parallel to the Ag substrates. The electric field enhancement is observed near the sharp edge of the Ag nanoparticles or at the junction nanogap of two neighboring AgNPs. Moreover, these simulations reveal that electromagnetic interaction was enhanced at the hot spot between Ag nanoparticles by almost 9 times. Ideally, with a justified Raman laser wavelength, an optimized nanogap for different morphology Ag nanoparticles, and an extended proportion of metallic-type CNTs, a greater enhancement of the signal could be achieved [17].
Vertically aligned CNTs preparation: The vertically aligned CNTs were synthesized on a Si substrate by chemical vapor deposition (CVD). In order to obtain orderly CNT arrays with good quality [18–20], it is important to choose suitable catalyst layer, and balance the ratio of hydrogen/ethylene, deposition time and temperature. Firstly, a 50 nm Mo, 10 nm Al and 5 nm Fe layer as the catalyst were orderly deposited on a Si wafer, with an E-beam evaporation [6]. Secondly, the prepared wafer was put into a vacuum quartz tube reactor and heated to 720 °C; a buffer gas of high purity H2 (50 sccm) was introduced continuous during the CVD process. Then C2H4 (150 sccm) as the carbon source was flowed through the quartz tube for 10 min of the CNT array growth. Finally, the buffer gas was cooled to room temperature.
Ag nanoparticles formation on CNT arrays: The Ag nanoparticles were formed by magnetron sputtering and high temperature annealing process. Firstly, a silver film (about 20 nm thickness) was deposited on the top of the CNT arrays by magnetron sputtering, as demonstrated in Fig. 3(a) (“before annealing” part). Then, the Ag/CNT arrays were put into a vacuum quartz tube, and high purity H2 gas (40 sccm) was introduced continuously during the whole annealing process. The samples were then heated up to a high temperature of 350 (or 400 or 450) °C. At 350 (or 400 or 450) °C, the samples were annealed for 60 (or 30 or 40) min, to cause the silver to diffuse along the nanotubes walls (“after annealing” part). Finally, the quartz tube was cooled down to room temperature. The process provides a high density of Ag nanoparticles on the surface and sidewalls of the vertically aligned CNTs, with nanogaps between particles (Fig. 3(b)).
Fig. 3 SEM images of highly dense AgNPs decorated on the aligned CNT arrays. SEM images of top morphology and cross section of (a) Ag film after deposition (before annealing), (b) AgNPs on the top and side walls of aligned CNT arrays along the perpendicular axis after annealing.
Ag nanoparticle formation on the CNT surfaces depends on the diffusion barrier of the Ag atoms, nucleation rate, and surface wetting [5]. Each CNT/CNT intersection provides a nucleation site for the formation of Ag nanoparticles [21], which leads to discrete AgNPs on the surface and walls of the aligned CNT arrays. High resolution scanning electron microscopy (SEM) shows that the surface and walls of pre-prepared aligned CNT arrays are coated by various-sized high-density Ag nanoparticles, as shown in Fig. 3(b).
3. Raman experiments
SERS measurements: All measurements were made using a Raman spectrometer and the excitation wavelength was 632.8 nm with power of 10 mW. The collective objective was 20 × of numerical aperture 0.4, at work distance of 5.46 mm. The integration time was 6 s, because the SERS signal was oversaturated at higher power and integration time. SERS measurements were carried out with Rhodamine 6G (R6G) as a SERS probe molecule. Each sample for SERS measurements was prepared by dropping a coating of 5 μL R6G aqueous solutions onto the substrate. There is no other dry process before measurement.
3.1 Different annealing condition effect
There are two important influence factors to AgNPs size and distribution, annealing temperature and time duration. In order to investigate annealing condition effect, we prepared different SERS substrates with four couple of different temperature and time duration, (Sample A: 350 °C, 1 h), (Sample B: 400 °C, 1 h), (Sample C: 450 °C, 1 h) and (Sample D: 450 °C, 30 min). SEM photos of these samples are shown in Fig. 4, and the size of Ag nanoparticles on the surface for sample A, B, C, D is respectively about 120 nm, 45 nm, 70 nm and 100 nm. The size of AgNPs formed on aligned CNT arrays along the perpendicular axis for each sample is a little smaller than that on the surface. The reason could be that the silver film become silver nanoparticles with thermal decomposition when annealing temperature of 200 °C. With the annealing temperature and time increasing, the small silver nanoparticles gradually reunited. But when the temperature reaches a certain level, due to substantial evaporation of silver, the number and size of silver nanoparticles were reduced. In our preparation of the samples, the amount of silver is fixed, and silver nanoparticles diffuse along the carbon nanotubes wall from top to bottom by magnetron sputtering and high temperature annealing process, the AgNPs formed on the CNT arrays were found to be smaller than those on the surface.
Fig. 4 SEM photos of AgNPs surface and sidewalls of (a) sample A, (b) sample B, (c) sample C and (d) sample D.
The SERS spectra of 0.5 × 10−4 M and 0.5 × 10−6 M R6G molecules obtained from the four substrates are shown in Figs. 5(a) and 5(b). The detailed assignment of the Raman spectral features has been reported previously [6]. The peak intensity enhanced with the increase of R6G concentration. We found that most SERS-active particles are concentrated in the 80 - 100 nm fraction [22]. It was reported that a calculated enhancement keeps increasing with the increase in particle size until about 110 nm [23]. In our experiments, sample D showed the best SERS sensitivity among the four substrates. From SEM photos, the AgNPs (Sample D) are distributed on the surface and sidewalls of CNT arrays with ~100 nm size. When the two influencing factors mentioned above (annealing temperature and time duration) were taken into account, the substrate was accomplished by annealing temperature at 450 °C for 30 min for our samples.
Fig. 5 SERS spectra of R6G molecular collected on a set of Ag/CNT arrays at different R6G concentrations of (a) 0.5 × 10−4 M and (b) 0.5 × 10−6 M adsorbed on four different substrates (Sample A, B, C, D)
3.2 Reproducibility of the SERS substrate
To test whether the as-prepared SERS substrates are able to give reproducible SERS signals, we collected SERS spectra of R6G molecular with a concentration of 0.5 × 10−5 M from 8 randomly selected places on the Ag/CNTs sample (Sample A), shown in Fig. 6(a).Qualitatively, the Raman signals maintained their fine features and intensities. The Raman intensity of one major peak at 1502 cm−1 is given in Fig. 6(b), which has been changed a little.
Fig. 6 Raman intensity of R6G (10−5 M) collected (a) on the randomly selected 8 places on the Ag/CNTs (Sample A), (b) at 1502 cm−1 of 8 places.
The relative standard deviation (RSD) curve of 8 SERS spectra is used to estimate the reproducibility of SERS signals, shown in Fig. 7.From the RSD-SERS graph, RSD values of signal intensity can be looked up directly, the average RSD values of all Raman peaks are observed to be below 0.01, revealing a good reproducibility across our sample. The table of RSD values corresponding to the major Raman peaks is shown in Table 1, which means that the Raman peaks exhibit good consistency and reproducibility. The same results are attained for other samples.
Table 1. RSD Values for the Major Peaks of the R6G SERS Spectrum
3.3 Stability of the SERS substrate
To investigate the temporal stability of our substrates, we compared the freshly prepared substrate and the substrate stored in air for three months for SERS detection of R6G molecular, expressed in Fig. 8(a).There is no shift in the major Raman peaks positions. The Raman intensity at 1502 cm−1 versus exposure time is shown in Fig. 8(b). After 14 days’ aerobic exposure, the intensity at 1502 cm−1 dropped to only 77% (1452/1887) of intensity in 7 days. Impressively, the signals reduced to 49% (917/1887) of intensity in 7 days and 88% (917/1037) of intensity in 28 days, even after 90 days’ aerobic exposure. The intensity has an obvious reduction before 28 days’ aerobic exposure, but almost no reduction after 28 days. Similar observations were observed for the other Raman peaks. The signals displayed without noticeable loss of the Raman peak features, but a sizable reduction in intensities. The reasons could be [11]: (1) Ag nanostructures could be oxidized in air. (2) Formation of surface-bound silver oxide leads to morphological changes of the Ag nanoparticles. (3) The active plasmonic material (Ag) is depleted, resulting in the reduction of intensity of the surface plasmon resonance. There is a report [11] that a monolayer graphene was used to protect Ag nanoparticles, in which the signal intensities are 80% of their corresponding original intensity, compared to 47.9% with uncovered substrates. We would make use of monolayer graphene to protect our substrates in the next step.
Fig. 8 Raman intensity of R6G (10−4 M) from the Ag/CNTs (450 °C, 40 min) (a) at different time points and (b) at 1502 cm−1 along with exposure time
3.4 EF calculation
The SERS enhancement factor (EF) may be strongly dependent on the exact SERS conditions: substrate, analyte, excitation wavelength, etc. There is a wide variety of definitions for EF [24], as shown in Table 2.
Table 2. Three Major Definitions of the EF, SMEF for Single Molecular EF, SSEF for Substrate SERS EF, AEF for Analytical EF.
Whereis the SERS intensity of the single molecular (SM) under consideration, is the average Raman intensity per molecule for the same probe. NSurf and NVol is the average number of adsorbed molecules in the scattering volume for the SERS measurement and non-SERS measurement, cRS and cSERS is the concentration of the solution used for non-SERS measurement and SERS measurement, respectively.
In the paper, due to our measurement condition (no more dry process for R6G) and potential analytical applications, the AEF is estimated by measuring the fluorescence of the R6G on SiO2 substrate with the same conditions and measurement systems as the SERS measurement in Fig. 5. By this means [25,26], we calculated the Raman scattering cross section compared with the fluorescence cross section. The 0.5 × 10−4 M R6G substrate was selected here for the EF calculation.
The fluorescence spectra of 0.5 × 10−4 M R6G molecules on SiO2 substrate is 8376 counts at 1354 cm−1, whereas the corresponding SERS spectra of our AgNPs/CNT arrays substrate (Sample D) is 29537 counts, shown in Fig. 9.Hence the SERS signal is about 3.5 times higher than the fluorescence signal.
Fig. 9 Raman intensity of R6G at a concentration of 0.5 × 10−4 M adsorbed on SiO2 and SERS substrate
The fluorescence cross-section of R6G is 2.5 × 10−16 cm2 [2]; the SERS cross-section of R6G is about 3.5 times larger (Fig. 9), about 8.8 × 10−16 cm2. We know that the conventional Raman scattering section is the order of 10−30 cm2 [27]. Owing to resonance, the Raman scattering of R6G is enhanced by a factor of about 104 [25,26], which could be removed in our EF calculation. The EF values for R6G on different SERS substrate samples were calculated and given in Table 3.With these values, it can be estimated that the EF of AgNPs/CNT arrays is about on the order of ~1010. EF values differed for each AgNPs/CNT arrays, which means different morphology and size of Ag nanoparticles has different enhancement effects. From the high EF values of Sample D with Ag nanoparticle size of 100 nm, we can conclude that for Ag nanoparticles, the size of 100 nm could be used for achieving possibly high enhancement. To some extent, it is similar to the report of the influence of particle size on the hot spots strongly coupled Ag nanoparticles [22]. As seen in Table 3, there is a little difference in AEFs with mode energy. This can be explained by the electromagnetic (EM) enhancement theory of the SERS. The underlying plasmon resonance at two wavelengths (one at the incident laser wavelength, the other at stokes wavelength) contributes to the EM EFs in SERS. It is not possible to simultaneously fulfill both resonance conditions to all Stokes shifts, due to the finite spectral width of the plasmon resonance [24].
Table 3. Values of Enhancement Factor for the Major Raman Peaks, Measured with Different Sample
In order to gain higher enhancement factor, the optimization for our substrate could be: (1) varying the size of nanoparticles to tune the resonance frequency matching the laser frequency [3], or tuning excitation source wavelength to different SERS substrates; (2) decreasing the gaps between nanoparticles to further enhance the local E-field inside the gaps; (3) controlling the uniformity of nanoparticles to stable enhancement; (4) and protecting the Ag oxidation to prolong the lifetime of the substrates.
4. Conclusion
In summary, we have developed vertically aligned carbon nanotubes coated by Ag nanoparticles through the magnetron sputtering and annealing process, which provides a way for 3D SERS substrate with a high EF. The results demonstrate that the degree of AgNPs on the walls of CNTs and the annealing conditions are the key factors determining the SERS signal enhancement. Our substrates with good sensitivity, reproducibility and stability would be useful for applications of sensing and SERS detection as an analytical tool.
Acknowledgments
We would like to thank Prof. L.W. Lin at University of California @Berkeley for CNT arrays sample help, Prof. G. Chen, Dr. T.C. Gong and W.J. Ren at Chongqing University for Raman experiments and simulation analysis help. It is funded by National Natural Science Foundation of China (No. 61376121), Natural Science Foundation of Chongqing, China (CSTC 2011BB2076) and the Fundamental research Funds for the central Universities (106112013CDJZR 125502, 20003, 20008).
References and links
1. K. Kneipp, Y. Wang, H. Kneipp, L. T. Perelman, I. Itzkan, R. R. Dasari, and M. S. Feld, “Single molecule detection using surface-enhanced Raman scattering (SERS),” Phys. Rev. Lett. 78(9), 1667–1670 (1997). [CrossRef]
2. S. Nie and S. R. Emory, “Probing single molecules and single nanoparticles by surface-enhanced Raman scattering,” Science 275(5303), 1102–1106 (1997). [CrossRef] [PubMed]
3. Y. H. Sun, K. Liu, J. Miao, Z. Y. Wang, B. Z. Tian, L. N. Zhang, Q. Q. Li, S. S. Fan, and K. L. Jiang, “Highly sensitive surface-enhanced Raman scattering substrate made from superaligned carbon nanotubes,” Nano Lett. 10(5), 1747–1753 (2010). [CrossRef] [PubMed]
4. Z. Q. Tian, B. Ren, J. F. Li, and Z. L. Yang, “Expanding generality of surface-enhanced Raman spectroscopy with borrowing SERS activity strategy,” Chem. Commun. (Camb.) 43(34), 3514–3534 (2007). [CrossRef] [PubMed]
5. S. Lee, M. G. Hahm, R. Vajtai, D. P. Hashim, T. Thurakitseree, A. C. Chipara, P. M. Ajayan, and J. H. Hafner, “Utilizing 3D SERS active volumes in aligned carbon nanotube scaffold substrates,” Adv. Mater. 24(38), 5261–5266 (2012). [CrossRef] [PubMed]
6. J. Zhang, T. Fan, X. L. Zhang, C. H. Lai, and Y. Zhu, “Three-dimensional multi-walled carbon nanotube arrays coated by gold-sol as a surface-enhanced Raman scattering substrate,” Appl. Opt. 53(6), 1159–1165 (2014). [CrossRef] [PubMed]
7. X. Li, G. Chen, L. Yang, Z. Jin, and J. Liu, “Multifunctional Au-coated TiO2 nanotube arrays as recyclable SERS substrates for multifold organic pollutants detection,” J. Adv. Fun. Mater. 20(17), 2815–2824 (2010). [CrossRef]
8. P. A. Yu, S. N. Terekhov, and I. A. Khodasevich, “Silver-coated nanoporous silicon as SERS-active substrate for investigation of tetrapyrrolic molecules, ” Proc of SPIE, The International Conference on Coherent and Nonlinear Optics Novel Photonics Materials; Optics and Optical Diagnostics of Nanostructures, Minsk, Belarus, 6728, 672828, 2007, http://proceedings.spiedigitallibrary.org.
9. W. D. Li, F. Ding, J. Hu, and S. Y. Chou, “Three-dimensional cavity nanoantenna coupled plasmonic nanodots for ultrahigh and uniform surface-enhanced Raman scattering over large area,” Opt. Express 19(5), 3925–3936 (2011). [CrossRef] [PubMed]
10. Y. L. Deng and Y. J. Juang, “Black silicon SERS substrate: effect of surface morphology on SERS detection and application of single algal cell analysis,” Biosens. Bioelectron. 53, 37–42 (2014). [CrossRef] [PubMed]
11. X. Y. Li, J. Li, X. M. Zhou, Y. Y. Ma, Z. P. Zheng, X. F. Duan, and Y. Q. Qu, “Silver nanoparticles protected by monolayer graphene as a stabilized substrate for surface enhanced Raman spectroscopy,” Carbon 66, 713–719 (2014). [CrossRef]
12. C. Wang, Y. C. Chang, J. Yao, C. Luo, S. Yin, P. Ruffin, C. Brantley, and E. Edwards, “Surface enhanced Raman spectroscopy by interfered femtosecond laser created nanostructures,” Appl. Phys. Lett. 100, 023107 (2012), doi:. [CrossRef]
13. G. Niaura, A. K. Gaigalas, and V. L. Vilker, “Surface-enhanced Raman spectroscopy of phosphate anions: adsorption on silver, gold, and copper electrodes,” J. Phys. Chem. B 101(45), 9250–9262 (1997). [CrossRef]
14. A. Kudelski, “Structures of monolayers formed from differnet HS-(CH2)2-X thiols on gold, silver and copper: comparative studies by surface-enhanced Raman scattering,” J. Raman Spectrosc. 34(11), 853–862 (2003). [CrossRef]
15. P. R. West, S. Ishii, G. V. Naik, N. K. Emani, V. M. Shalaev, and A. Boltasseva, “Searching for better plasmonic materials,” Laser Photonics Rev. 4(6), 795–808 (2010). [CrossRef]
16. T. Makaryan, S. Esconjauregui, M. Gonçalves, J. W. Yang, H. Sugime, D. Nille, P. R. Renganathan, P. Goldberg-Oppenheimer, and J. Robertson, “Hybrids of carbon nanotube forests and gold nanoparticles for improved surface plasmon manipulation,” ACS Appl. Mater. Interfaces 6(8), 5344–5349 (2014). [CrossRef] [PubMed]
17. M. Mahjouri-Samani, Y. S. Zhou, X. N. He, W. Xiong, P. Hilger, and Y. F. Lu, “Plasmonic-Enhanced Carbon Nanotube Infrared Bolometers,” Nanotechnology 24(3), 035502 (2013). [CrossRef] [PubMed]
18. Y. Q. Jiang, P. B. Wang, and L. W. Lin, “Characterizations of contact and sheet resistances of vertically aligned carbon nanotube forests with intrinsic bottom contacts,” Nanotechnology 22(36), 365704 (2011). [CrossRef] [PubMed]
19. Y. Q. Jiang and L. W. Lin, “A two-stage, self-aligned vertical densification process for as-grown CNT forests in supercapacitor applications,” Sens. Actuators A Phys. 188, 261–267 (2012). [CrossRef]
20. W. Cho, M. Schulz, and V. Shanov, “Growth and characterization of vertically aligned centimeter long CNT arrays,” Carbon 72, 264–273 (2014). [CrossRef]
21. H. M. Duong, K. Ishikawa, J. Okawa, K. Ogura, E. Einarsson, J. Shiomi, and S. Maruyama, “Mechanism and optimization of metal deposition onto vertically aligned single walled carbon nanotube arrays,” J. Phys. Chem. C 113(32), 14230–14235 (2009). [CrossRef]
22. S. R. Emory and S. Nie, “Screening and enrichment of metal nanoparticles with novel optical properties,” J. Phys. Chem. B 102(3), 493–497 (1998). [CrossRef]
23. P. P. Fang, J. F. Li, Z. L. Yang, L. M. Li, B. Ren, and Z. Q. Tian, “Optimization of SERS activities of gold nanoparticles and gold-core–palladium-shell nanoparticles by controlling size and shell thickness,” J. Raman Spectrosc. 39(11), 1679–1687 (2008). [CrossRef]
24. E. C. Le Ru, E. Blackie, M. Meyer, and P. G. Etchegoin, “Surface enhanced Raman scattering enhancement factors: a comprehensive study,” J. Phys. Chem. C 111(37), 13794–13803 (2007). [CrossRef]
25. P. Hidebrandt and M. Stockburger, “Surface-enhanced resonance Raman spectroscopy of Rhodamine 6G adsorbed on colloidal silver,” J. Phys. Chem. 88(24), 5935–5944 (1984). [CrossRef]
26. K. Kneipp, Y. Wang, H. Kneipp, I. Itzkan, R. R. Dasari, and M. S. Feld, “Population pumping of excited vibrational states by spontaneous surface-enhanced Raman scattering,” Phys. Rev. Lett. 76(14), 2444–2447 (1996). [CrossRef] [PubMed]
27. Y. C. Li, J. Zhou, K. Zhang, and C. Q. Sun, “Gold nanoparticle multilayer films based on surfactant films as a template: Preparation, characterization, and application,” J. Chem. Phys. 126(9), 094706 (2007). [CrossRef] [PubMed]
