12504

2007, 111, 12504-12507 Published on Web 08/09/2007

Origin of Gate Hysteresis in Carbon Nanotube Field-Effect Transistors Joon Sung Lee,†,‡ Sunmin Ryu,† Kwonjae Yoo,† Insung S. Choi,‡ Wan Soo Yun,*,† and Jinhee Kim*,† Korea Research Institute of Standards and Science, Daejeon 305-600, Korea, and Department of Chemistry and School of Molecular Science (BK21), KAIST, Daejeon 305-701, Korea ReceiVed: June 17, 2007; In Final Form: July 26, 2007

We have studied gate hysteresis of carbon nanotube field-effect transistors (CNFETs) on silicon oxide substrates in an ultrahigh vacuum (UHV) at low temperatures. It is found that the hysteresis is neither reduced by thermal annealing at temperatures over 300 °C under UHV nor significantly affected by independent adsorption of ammonia or water at T ) 56 K. However, the hysteresis decreases greatly upon coadsorption of water and ammonia below condensation temperatures and restores completely with desorption of the adsorbed water layer. On the basis of these results, it is concluded that the main cause of gate hysteresis in CNFETs on silicon oxide substrate is charge transfer between the carbon nanotube and charge traps at the silicon oxide/ ambient interface. We propose a mechanism for gate hysteresis that involves surface silanol groups as the major sources of screening charges. This surface silanol model is supported by results from scanning surface potential microscopy (SSPM).

Because of its superior performance and ultrahigh sensitivity to adsorbed chemicals, carbon nanotube field-effect transistors (CNFETs) have been a subject of numerous studies that are focused mostly on device applications. Yet, several issues still need to be clarified for better utilization of the CNFET. One of the unresolved issues is gate hysteresis, which has been widely observed in CNFETs. Because of its advancing nature in general, gate hysteresis in CNFETs had usually been attributed to injection of screening charges into/from traps in the gate oxide.1-3 Later, yet another view of this phenomenon was suggested; the hysteresis is due to the surface-bound water layer, which works as a charge trap or mediator.4 This explanation has since been widely quoted, but several issues were raised about the role of water. The possibility of water molecules being charge traps has been questioned,5 along with the plausibility of ionic movements at low temperatures.6 It has also been pointed out that ionic charge transfer via a water layer should cause a retarding hysteresis rather than an advancing one.1 In a practical viewpoint, gate hysteresis should be controlled to ensure reproducible and stable operation of CNFETs. Reduction of hysteresis has been realized via various methods such as adoption of thin high-κ dielectric top gates,7 electrolyte gating,8 and so forth. However, for application as sensors for chemicals such as NO2 and NH3,9,10 the device should be constructed in an open geometry in which CNTs are exposed to ambient conditions. This requirement renders the aforementioned measures against gate hysteresis inapplicable. Recently, there has been some progress in these regards; passivation of silicon oxide substrate with a self-assembled monolayer (SAM),2 thermal annealing on a surfactant-covered device,11 and short* Corresponding authors. E-mail: [email protected]; wsyun@kriss. re.kr. † Korea Research Institute of Standards and Science. ‡ KAIST.

10.1021/jp074692q CCC: $37.00

pulsed measurements12 were shown to be effective for hysteresis reduction or elimination. In addition, adoption of thin high-κ dielectrics in back-gated CNFETs,13 which enables FET operation within a smaller gating bias range, has been shown to be effective in circumventing hysteresis-related problems. Notwithstanding these practical successes, the cause of hysteresis still needs to be clarified; more detailed knowledge of the hysteresis may provide hints for improving not only CNFETs but also other low-dimensional nanoelectronic devices such as nanowire devices. To investigate the origin of gate hysteresis in CNFETs with silicon oxide-backgate dielectric, we did electron-transport measurements combined with gas adsorption at low temperatures under a UHV. We show that molecularly adsorbed water cannot fully explain the hysteresis. Also, we show that gate hysteresis in a back-gated CNFET on a silicon oxide substrate can be reduced greatly by ammonia coadsorbed with water at low temperatures. On the basis of these results, the main cause of gate hysteresis in CNFETs on silicon oxide substrate is deduced to be charge transfer between the CNT and charge traps at the silicon oxide/ambient interface. In addition, we propose a mechanism for the hysteresis involving surface silanol groups as the major sources of screening charges, which can also explain the humidity dependence of gate hysteresis at room temperature.4,9,12,14 This surface silanol model of hysteresis is supported by surface potential measurements using scanning surface potential microscopy (SSPM). The CNFET used in this work was fabricated on a degenerately p-doped silicon substrate covered with a 550-nm-thick thermally grown oxide layer. Single-walled carbon nanotubes (SWCNTs) were grown directly on the substrate by chemical vapor deposition.15 Individual CNTs were located using an atomic force microscope (AFM). Source and drain contacts were formed on a CNT as thermally evaporated 10-nm-thick Pd © 2007 American Chemical Society

Letters

J. Phys. Chem. C, Vol. 111, No. 34, 2007 12505

Figure 1. Conductance-gate voltage (G-Vg) graphs of the CNFET device: (a) after storage under UHV, (b) after a thermal annealing over 300 °C for 10 h, (c) before and (d) after the injection of H2O gas at 56 K, (e) after the first injection of NH3 gas, (f) after the temperature rise to 165 K, (g) after the second cooldown to 56 K, (h) after the second injection of NH3 gas, and (i) after the temperature rise to 295 K. The conductance in the graphs is dc conductance measured with Vds ) 10 mV.

defined by e-beam lithography. The channel length and diameter of the CNT were 4 µm and 2 nm, respectively. When measured in air, the CNFET exhibited a p-channel-dominant ambipolar character with a large advancing gate hysteresis. The sample was inserted into a modified Omicron UHV-VT-SPM system equipped with gas-injection attachments and a residual gas analyzer. After storage under UHV (curve a in Figure 1), gate hysteresis of the device decreased to nearly 50% of that measured in ambient air, which can be ascribed to desorption of water layer.4 After thermal annealing at 300 °C for 10 h under UHV (curve b), the device became n-channel-dominant, with a change in the threshold voltages. Such a change in device character can be explained by a modification of Fermi level alignment at the CNT-metal contacts due to desorption of gas adsorbates such as oxygen from the electrode surface.16 However, the hysteresis was not reduced further by this thermal annealing. It is known that water molecules bound on the hydroxylated silica surface are fully desorbed at temperatures up to 200 °C.17 This persistence of hysteresis contradicts the report that mostly surface-bound water molecules cause gate hysteresis in CNFETs.4 After the thermal annealing, the sample was cooled down to 56 K, and its conductance versus gate voltage (G-Vg) was measured (curve c). An advancing hysteresis of ∼6 V was still observed at this low temperature. With the sample temperature held at 56 K, H2O gas was injected onto the sample.18 Curve d was obtained after the H2O injection. The hysteresis almost did not change from curve c to d, except for a slight decrease in the p-branch. Thus, it follows that the existence of molecular water by itself does not cause gate hysteresis at this low temperature. Curve e was obtained after NH3 gas had been additionally injected onto the sample already covered by water ice. The hysteresis near the n-branch threshold almost disappeared in e.

Figure 2. Schematic of the hysteretic surface charging process involving silanol groups: (a) at the start of a gate sweep with a positive Vg, (b) at a negative Vg in the decreasing sweep, (c) near the negative end of the decreasing sweep, and (d) at a positive Vg in the subsequent increasing sweep. The circled plus marks represent protons released from surface silanol groups.

This result seemed quite surprising in that, in a control experiment without water, adsorption of ammonia alone did not reduce the hysteresis at all.19 As the sample temperature was being raised, thermal desorption of NH3 and H2O from the sample was detected by the residual gas analyzer above 80 and 145 K, respectively. Curve f was obtained at 165 K. Here the usual advancing gate hysteresis is fully recovered. Now it becomes clear that the ammonia intermixed in the water-ice layer actually had suppressed the hysteresis. When the sample temperature was lowered again to 56 K, the hysteresis remained the same, as shown in curve g. After NH3 gas was injected onto the sample again, the gate hysteresis decreased a bit (curve h), but not to the extent when the full water multilayer was intact. This limited response to ammonia was caused by cooperation of the additional ammonia with the few remaining water molecules that survived the previous desorption. As the sample temperature was raised afterward,

12506 J. Phys. Chem. C, Vol. 111, No. 34, 2007

Letters

Figure 3. (a) Topography of the CNFET from a noncontact AFM scan. (b-d) Surface potential images obtained at room temperature under UHV (b) after Vg was set to +20 V and back to 0 V, (c) after Vg was set to -20 V and back to 0 V, and (d) after a brief thermal annealing at 200 °C. The imaged area is 1.9 × 4.8 µm2. Surface potential profiles along dotted lines in b-d are given in e, in which the curves are offset to adjust surface potential values over Pd electrodes in each image to 0 V.

the hysteresis fully recovered over T ) 113 K with desorption of ammonia. Curve i was obtained at room temperature after the gas experiment. It shows no substantial difference from curve b except for a slight increase in the n-channel conductance. This full recovery of the hysteresis with the gas desorption dictates that the observed reduction of hysteresis does not involve any irreversible chemical modification of the sample surface. The data presented above strongly suggest that the sources of gate hysteresis in the CNFET are mostly confined to the surface. It is known that the dielectric constant, , of 30% ammonia-water ice at 77 K is ∼4.5,20 compared to  ≈ 3.2 of pure crystalline water ice below 130 K.21 Because any possible modification of the charge injection barrier at the CNT surface by such a secondary dielectric layer cannot explain the reduction of hysteresis, charge injection into the bulk oxide charge traps can be excluded from the main sources of hysteresis. This leads us to conclude that the cause of the reduction in hysteresis was a noncovalent, temporary modification of the silicon oxide surface, by cooperation of ammonia and water. In particular, we speculate that the ammonia molecules deactivated potential charge traps or sources on the oxide surface, possibly with charge transfers or hydrogen bonds between the oxide surface and ammonia, assisted and stabilized by water molecules.22 We propose that silanol groups on the silicon oxide surface (≡SiOH) are the charge traps or sources subject to the deactivation. A schematic of the gate-screening process involving surface silanol groups is given in Figure 2. In this model, silanol groups may lose their protons and electrons at negative Vg. While the released protons are trapped nearby, the electrons are transferred into conductors (CNT or electrodes) via fielddriven hopping. This field-driven lateral charge hopping (or charge diffusion) should be similar to the “wetting of the surface by injected charges” observed in a contact electrification study on Al2O3 surfaces.23 The trapped net positive charges add to the gate potential, causing a negative shift of the G-Vg curve as drawn by the dotted curve in Figure 2. This model not only allows gate hysteresis without surface water at low temperatures but also helps in understanding the role of surface water in gate hysteresis. In this work, we have shown that water ice at low temperatures does not work as a charge trap by itself, probably because of the immobility of protons in ice below 190 K.21 At room temperature, the situation changes. On ionization of silanol groups, the released protons can be subsequently soaked into the water layer, forming H3O+; or the water layer may promote ionization of the silanol groups. The water layer here works as a proton absorbent with an intrinsic charge-screening capability.6 It actually works as a

charge trap, but the corresponding charge transfer is mediated by the intrinsic surface charge sources, the silanol groups. In this way, humidity may influence the magnitude of hysteresis. This model can also resolve the aforementioned paradox of mobile ions causing advancing hysteresis.4,14 According to this model, the gate-screening charges trapped near the CNT should be mainly positive. To check this, we performed SSPM measurements24 on the CNFET under UHV. After Vg had been set to -20 V and back to 0 V, a heavy positive charging was observed near the CNT (Figure 3c). On the contrary, a much-weaker negative charging was detected after Vg had been set to +20 V and back to 0 V (Figure 3b), thus demonstrating a tendency of positive charging25 that supports the surface silanol model. Figure 3d was obtained after a thermal annealing at 200 °C, where trapped charges become thermally dissipated. The major experimental findings from this work can be summarized as follows: (1) gate hysteresis in CNFETs on silicon oxide substrates was not removed by thermal annealing in a UHV at temperatures over 300 °C, (2) transfer characteristics of a CNFET were not significantly changed by independent adsorption of ammonia or water at T ) 56 K, (3) hysteresis was greatly reduced by coadsorption of ammonia and water below condensation temperatures, and (4) the hysteresis was restored with thermal desorption of the adsorbed molecular layer. On the basis of these results, it was concluded that the main cause of gate hysteresis in CNFETs on silicon oxide substrate is charge transfer between CNT and charge traps at the silicon oxide/ambient interface. Additionally, a mechanism for the hysteresis involving surface silanol groups was proposed and supported by SSPM. Because these results may help to develop and interpret schemes for control of gate hysteresis, further efforts are required to verify this idea and apply it to control the properties of low-dimensional electronic devices using CNTs or nanowires. Acknowledgment. We thank Byung-Chill Woo, Woon Song, and Sanghun Kim for their assistance with instrumentation, Jae-Ryoung Kim for help with sample fabrication, and Noejung Park for helpful discussions. This work was supported by MOST through KOSEF. References and Notes (1) Peillard, A. R.; Rotkin, S. V. IEEE Trans. Nanotechnol. 2005, 4, 284-288. (2) McGill, S. A.; Rao, S. G.; Manandhar, P.; Xiong, P.; Hong, S. Appl. Phys. Lett. 2006, 89, 163123. (3) (a) Fuhrer, M. S.; Kim, B. M.; Du¨rkop, T.; Brintlinger, T. Nano Lett. 2002, 2, 755-759. Radosavljevic´, M.; Freitag, M.; Thadani, K. V.;

Letters Johnson, A. T. Nano Lett. 2002, 2, 761-764. (b) Freitag, M.; Chen, J.; Tersoff, J.; Tsang, J. C.; Fu, Q.; Liu, J.; Avouris, P. Phys. ReV. Lett. 2004, 93, 076803. (4) Kim, W.; Javey, A.; Vermish, O.; Wang, Q.; Li, Y.; Dai, H. Nano Lett. 2003, 3, 193-198. (5) Sung, D.; Hong, S.; Kim, Y.-H.; Park, N.; Kim, S.; Maeng, S. L.; Kim, K.-C. Appl. Phys. Lett. 2006, 89, 243110. (6) Vijayaraghavan, A.; Kar, S.; Soldano, C.; Talapatra, S.; Nalamasu, O.; Ajayan, P. M. Appl. Phys. Lett. 2006, 89, 162108. (7) Yang, M. H.; Teo, K. B. K.; Gangloff, L.; Milne, W. I.; Hasko, D. G.; Robert, Y.; Legagneux, P. Appl. Phys. Lett. 2006, 88, 113507. (8) Siddons, G. P.; Merchin, D.; Back, J. H.; Jeong, J. K.; Shim, M. Nano Lett. 2004, 4, 927-931. (9) Bradley, K.; Gabriel, J.-C. P.; Briman, M.; Star, A.; Gru¨ner, G. Phys. ReV. Lett. 2003, 91, 218301. (10) Kong, J.; Franklin, N. R.; Zhou, C.; Chapline, M. G.; Peng, S.; Cho, K.; Dai, H. Science 2000, 287, 622-625. (11) Li, J.; Zhang, Q.; Li, H.; Chan-Park, M. B. Nanotechnology 2006, 17, 668-673. (12) Lin, H.; Tiwari, S. Appl. Phys. Lett. 2006, 89, 073507. (13) (a) Hur, S.-H.; Yoon, M.-H.; Gaur, A.; Shim, M.; Facchetti, A.; Marks, T. J.; Rogers, J. A. J. Am. Chem. Soc. 2005, 127, 13808-13809. (b) Cao, Q.; Xia, M.-G.; Shim, M.; Rogers, J. A. AdV. Funct. Mater. 2006, 16, 2355-2362. (14) Bradley, K.; Cumings, J.; Star, A.; Gabriel, J.-C. P.; Gru¨ner, G. Nano Lett. 2003, 3, 639-641. (15) Kong, J.; Soh, H. T.; Cassell, A. M.; Quate, C. F.; Dai, H. Nature 1998, 395, 878-881. (16) (a) Derycke, V.; Martel, R.; Appenzeller, J.; Avouris, P. Appl. Phys. Lett. 2002, 80, 2773-2775. (b) Cui, X.; Freitag, M.; Martel, R.; Brus, L.; Avouris, P. Nano Lett. 2003, 3, 783-787.

J. Phys. Chem. C, Vol. 111, No. 34, 2007 12507 (17) Zhuravlev, L. T. Colloids Surf. A 2000, 173, 1-38. (18) The actual gas dosage is believed to have exceeded 3 Langmuirs. Under the experimental conditions, the water-ice layer must have been deposited as a thin porous film, thus not providing a good passivation over the oxide surface. About the morphology of water ice, see the following: Stevenson, K. P.; Kimmel, G. A.; Dohna´lek, Z.; Smith, R. S.; Kay, B. D. Science 1999, 283, 1505-1507. (19) In fact, it did not cause any remarkable changes in G-Vg curves. This lack of ammonia response coincides with results from previous studies: charge transfer from ammonia to CNT is negligible, and chemisorption or charge doping of ammonia to CNT only occurs via other preadsorbed species such as oxygen or water. For references, see refs 9 and 10 and the following: Feng, X.; Irle, S.; Witek, H.; Morokuma, K.; Vidic, R.; Borguet, E. J. Am. Chem. Soc. 2005, 127, 10533-10538. Bauschlicher, C. W., Jr.; Ricca, A. Phys. ReV. B 2004, 70, 115409. (20) Lorenz, R. D. Icarus 1998, 136, 344-348. (21) Cowin, J. P.; Tsekouras, A. A.; Iedema, M. J.; Wu, K.; Ellison, G. B. Nature 1999, 398, 405-407. (22) Teunissen, E. H.; van Duijneveldt, F. B.; van Santen, R. A. J. Phys. Chem. 1992, 96, 366-371. (23) Lambert, J.; Saint-Jean, M.; Guthmann, C. J. Appl. Phys. 2004, 96, 7361-7369. (24) Sommerhalter, C.; Matthes, T. W.; Glatzel, T.; Ja¨ger-Waldau, A.; Lux-Steiner, M. C. Appl. Phys. Lett. 1999, 75, 286-288. (25) This tendency agrees with the results given in Figure 3b of ref 2 which imply that the trapped screening charges causing hysteresis in CNFETs without surface passivation are mainly positive. Figure 1a and c of ref 12, in which current-gate voltage (I-Vg) curves obtained with 0 V-based, short-pulsed Vg are located closer to the curves obtained with decreasing Vg sweeps than to the curves with increasing Vg sweeps, also conform to the tendency of positive charging.