NANO LETTERS
Chemical Optimization of Self-Assembled Carbon Nanotube Transistors
2005 Vol. 5, No. 3 451-455
Ste´phane Auvray,† Vincent Derycke,*,† Marcelo Goffman,† Arianna Filoramo,† Oliver Jost,‡ and Jean-Philippe Bourgoin† Laboratoire d’Electronique Mole´ culaire, CEA-DSM, CEA Saclay, 91191 Gif-sur-YVette, France, and Institut fu¨r Werkstoffwissenschaft der TU Dresden, D-01062 Dresden, Germany Received November 26, 2004; Revised Manuscript Received January 10, 2005
ABSTRACT We present the improvement of carbon nanotube field effects transistors (CNTFETs) performances by chemical tuning of the nanotube/ substrate and nanotube/electrode interfaces. Our work is based on a method of selective placement of individual single walled carbon nanotubes (SWNTs) by patterned aminosilane monolayer and its use for the fabrication of self-assembled nanotube transistors. This method brings a relevant solution to the problem of systematic connection of self-organized nanotubes. The aminosilane monolayer reactivity can be used to improve carrier injection and doping level of the SWNT. We show that the Schottky barrier height at the nanotube/metal interface can be diminished in a continuous fashion down to an almost ohmic contact through these chemical treatments. Moreover, sensitivity to 20 ppb of triethylamine is demonstrated for self-assembled CNTFETs, thus opening new prospects for gas sensors taking advantages of the chemical functionality of the aminosilane used for assembling the CNTFETs.
Single walled carbon nanotubes (SWNTs) have attracted an increasing interest since they provide unidimensional wires enabling the fabrication of a complete family of electronic devices. In particular, an individual semiconducting SWNT can be used as the channel of a field-effect transistor (CNTFET).1,2 Recent experimental3-8 and theoretical8-11 studies on CNTFETs have shown that most of them work as Schottky barrier transistors. Their switching characteristics are limited by the Schottky barriers at the metal/nanotube junction, bringing to the fore the crucial role of interfaces in such a transistor. By optimizing the nanotube/electrode interface3,7,12 and the gate coupling, device characteristics can be improved in both the on- and the off-state. Some CNTFETs13,14 already exhibit a level of performance comparable with state-of-the-art silicon MOSFETs at comparable geometry. In addition, SWNTs are also an ideal material for nanosensor fabrication: their one-dimensional structure and their high specific area make the conductance of semiconducting SWNTs highly sensitive to very small amounts of molecules.15 Given the promise of the CNTFETs, their optimization represents a key challenge for future applications. However, the possible use of carbon nanotubes as active elements in future nanoelectronics is closely related with a * Corresponding author. E-mail:
[email protected]. † CEA-DSM. ‡ TU Dresden. 10.1021/nl048032y CCC: $30.25 Published on Web 02/09/2005
© 2005 American Chemical Society
question of legacy/compatibility with the present information technology. Indeed, it is quite unlikely that a system based on a new technology consisting in architecture with completely random disposition of such devices could be introduced and accepted. Therefore, to fully take advantage of the unique electrical properties of SWNTs in device/circuit applications, it is very desirable to be able to selectively place themsfor connectionsat specific locations on a substrate with a low cost and high yield, self-assembly-based technique. Nowadays, the state of the art on this issue can be divided in two different classes of self-assembly methods: (i) in-situ CVD growth where the localization arises from the catalyst controlled positioning and (ii) post growth deposition on a substrate. In the latter case, the nanotubes are first grown, handled in solution, and subsequently positioned on the substrate thanks to a specific surface chemistry and functionalization. Obviously, the technique chosen for selective placement of the nanotubes must not degrade the electrical characteristics of the devices. We present here a method of chemical optimization of transistor performances that is not only compatible but turns out to be enhanced by the use of a patterned aminosilane monolayer as template for localized self-assembly of CNTFETs. Indeed, we discuss the key role of the nanotube/ substrate and nanotube/electrode interfaces on the device transport properties and how to improve the CNTFETs
Figure 1. AFM image (phase mode) of an individual SWNT deposited on an APTS pattern and connected in a back-gate transistor geometry.
performance by chemically modifying these particular interfaces. Moreover, we show that the aminosilane monolayer allowing the selective placement brings a new interface that can be used to chemically tune (by protonation or deprotonation) the device characteristics by improving carrier injection in the nanotube and changing the doping level of the SWNT. Finally, we demonstrate that the Schottky barrier height at the nanotube/metal interface can be decreased in a continuous fashion down to an almost ohmic contact through these chemical treatments. The SWNTs used in this work are self-assembled onto an e-beam patterned self-assembled monolayer (SAM) of (aminopropyl)triethoxysilane (APTS) formed by chemical vapor deposition16-18 that acts as a sticky patch to which the nanotubes bond. The SWNTs are then connected in a transistor geometry (Cr/Au source and drain electrodes, spaced by 100 to 500 nm) using the heavily doped and oxidized silicon substrate (200 nm of SiO2) as back-gate. Elaborating on pioneering works using an electrostatic interaction between the SWNTs and the SAM,19-21 we recently showed that a high yield of highly selective deposition of individual SWNTs in predefined areas of the substrate, and thus a high yield of CNTFETs fabrication, was possible using simultaneously a SWNT organic solvent dispersion and deposition on an optimized SAM.22-24 The SWNT raw material used in the present work was produced by a laser ablation technique at Dresden University and purified by an optimized soft acidic treatment.25 The purified SWNTs were dispersed by moderate sonication26 in N-methyl pyrrolidone (NMP), which proves the solvent of choice for a high yield of selective deposition onto the APTS patterns. Figure 1 displays a typical AFM image of an electrically connected self-assembled individual nanotube on an APTS stripe. For the sake of clarity we first compare the performances of devices made by this APTS self-assembly technique (called CNTFETs on APTS) with those of CNTFETs made by random deposition (called CNTFET on SiO2). ID(VGS) 452
Figure 2. Transfer characteristics ID(VGS) at VDS ) -1 V of two CNTFETs with a gate oxide thickness of 200 nm. Left: CNTFET made by random deposition on SiO2. Right: CNTFET made by the self-assembly technique. Insets show the same data in log-scale.
characteristics are measured at room temperature, in air, with the gate bias swept from the on- to the off-state.27 Figure 2 displays an example of the output characteristics of the two kinds of CNTFETs. We notice that both of them have very similar performances: on-state current up to ∼5 µA, onoff ratio of 4 orders of magnitude, transconductance of 0.4 to 0.5 µS, and sub-threshold slope S ) 2000 ( 300 mV/ dec. While some dispersion in these parameters is observed when many devices are prepared (due to their high sensitivity to the details of the nanotube/metal interface) no significant differences in performances are observed between the two types of fabrication methods when the comparison is extended to a large enough number of devices (>20). Under typical atmospheric conditions, the performances are set by the quality of the contacts and the gate efficiency (set by the oxide thickness) independently of the placement technique. In addition, we note that these performances are on equal footing with the best performances of CNTFETs of comparable geometry, reported in the literature. Thus, the chemical functionalization of SiO2 substrates brings a relevant solution to the problem of systematic connection of adsorbed nanotubes in a transistor geometry and is compatible with the production of high quality FETs. Moreover, in the following we show how we can take advantage of the APTS self-assembled monolayer to perform a chemical optimization of CNTFETs. Indeed, with respect to a CNTFET on SiO2, a CNTFET on APTS includes an additional, tunable, chemical interface (nanotube/APTS). The amino group of the APTS monolayer is protonable (pKa of protonation of N-propylamine to N-propylammonium equal to 10.2 in solution). Though the pKa value of a couple adsorbed on a surface is different from the one in solution, Bezanilla et al.28 have shown that a surface functionalized by APTS starts to be positively charged for pH < 10. As a consequence by exposing a CNTFET on APTS to acidic or basic molecules, it is possible to displace the equilibrium between the protonated form -NH3+ and deprotonated form -NH2 and thus modify the charge of the surface in the proximity of the nanotube. In this article we study this Nano Lett., Vol. 5, No. 3, 2005
Figure 3. Transfer characteristics ID(VGS) at VDS ) -0.2 V of a CNTFET on APTS before exposition to TFA vapors (black curve), after 8 min in the presence of TFA vapors (red curve) and 29 min after partial desorption of the TFA molecules (blue curve).
phenomenon by exposing the CNTFET on APTS either to acidic or to basic molecules, as separately described and discussed below. The acidic molecule used to study the physisorption effect onto CNTFETs on APTS is trifluoro-acetic acid (TFA) (pKa ) 0.3). TFA chemical treatment induces a protonation of the amino group of the APTS monolayer, i.e., nanotubes are adsorbed on a positively charged surface. TFA is also a strongly polar molecule (µ ) 2.26 D) and an oxidant.29 Finally, TFA is quite volatile, which enables its use in the gaseous phase. For this reason, the sample does not need to be rinsed with a solvent after its exposure (this way avoiding the modification of the chemical nature of the SAM), making the experiment reversible. Exposure to TFA is performed using a process similar to that of ref 12 in the case of CNTFETs on bare SiO2. A CNTFET on APTS connected to a measurement setup and enclosed in a sealed chamber is exposed to TFA vapors. The evolution of the electrical characteristics of the CNTFET is followed as a function of time during the adsorption of TFA molecules. Figure 3 displays typical electrical output characteristics of a CNTFET on APTS exposed to TFA vapor. Note that its initial characteristics (black curve, before TFA exposure) shows a sub-threshold slope S ) 1700 ( 150 mV/dec. During the exposure to TFA vapor, we observe a progressive right-hand shift of the ID(VGS) curve as well as an improvement of all the device parameters: the current in the onstate increases ≈3 times and the switching (the transition between the on- and the off-state) becomes very sharp as evidenced by the quite sizable decrease of the sub-threshold slope (S ) 175 ( 40 mV/dec), ≈10 times better compared to the initial curve. An optimal state is reached where the ID(VGS) characteristic (red curve) no longer evolves, approximately corresponding to a 40 V gate voltage right-hand shift. This state persists as long as the presence of TFA vapor is kept. At that point, if the TFA vapor is evacuated the Nano Lett., Vol. 5, No. 3, 2005
ID(VGS) curve progressively comes back to its initial state, as shown by the blue curve. At first, we note significant differences when comparing the behavior of a CNTFET on APTS compared to a CNTFET on SiO2 (not reported here, see ref 12): a substantial enhancement of the switching (optimum value of S ≈175 mV/dec for a CNTFET on APTS, two times better than S ≈360 mV/dec for a CNTFET on SiO2) and a more pronounced right-hand shift of the ID(VGS) characteristics followed by a more progressive improvement of the switching for CNTFETs on APTS. The single difference between the two kinds of CNTFETs is the presence of an APTS monolayer in the immediate proximity of the nanotube. When a CNTFET on APTS is exposed to TFA vapors, molecules are adsorbed onto the nanotube, the APTS SAM, and the metal electrodes. As discussed in ref 12 in the case of bare SiO2 substrates, the adsorption of TFA (weak electron acceptor) onto the nanotube wall has a minor impact (slight p-doping effect) that cannot be alone responsible for the pronounced right-hand shift of the curves observed on APTS treated substrates. Conversely, as demonstrated on CNTFETs on SiO2,12 the adsorption of TFA onto the nanotube/metal electrode interface was shown to result in the lowering of the Schottky barrier controlling hole injection and thus in the sub-threshold slope (S) improvement.6 This explanation clearly also applies to the present case since the metal/nanotube interface in the case of CNTFETs on APTS is similar to that of CNTFETs on SiO2. However, the optimization is clearly more significant in the case of CNTFETs on APTS. A possible explanation may be the presence of the APTS monolayer close to the nanotube/metal interface giving rise to either an increased density or a better local organization of the adsorbed TFA molecules. This may result in an increased dipole density and thus a more substantial Schottky barrier lowering, which leads to the better improvement of S. We note that the TFA chemical treatment of a CNTFET on APTS indeed enables to obtain a transistor with an excellent switching behavior. The S value of ≈175 mV/dec is close to the one predicted for an ohmic contact in a back-gate geometry with gate oxide thickness of 200 nm.6,30 At the TFA dose corresponding to the optimum S value (red curve), the pinning of the Fermi level at the metal/nanotube interface becomes very close to the valence band. The on-state then corresponds to direct injection through the valence band, while the off-state (vertical part of the curve) corresponds to thermal emission over the valence band.8 Beyond the modification of the contacts, the major effect in the present case is due to the reaction of the APTS SAM to the acidic treatment. Indeed, the acidic nature of the TFA displaces the equilibrium between the two forms NH2/NH3+ of the SAM. As a result, the terminations of the APTS monolayer are under the dominant form NH3+. Therefore the nanotubes are adsorbed on a positively charged compact monolayer, the NH3+ terminations of which operate like electron acceptors. Those terminations very close to the nanotube induce a strong p-doping that provokes the pronounced right-hand shift of the ID(VGS) curves. 453
These experimental findings make evident the key role played by the different interfaces of a CNTFET and show how chemistry can be used to optimize its transport properties. In particular, they show the strong influence of the APTS monolayer. This last point is strongly supported by a supplementary experiment where we chemically modify the APTS monolayer by a molecule that has a basic behavior. For comparison we also perform the experiment on CNTFETs on SiO2. The chosen molecule is triethylamine (TEA) (pKa ) 11.1). The reasons for this choice are dictated by its chemical and physical properties: (i) TEA chemical treatment leads to a deprotonation of the amino group of the APTS monolayer; (ii) this molecule enables to work with a gaseous phase as for TFA; (iii) TEA is also an electron donor, and we expect to observe an n-doping effect during its adsorption on the wall of the SWNT, as was reported by Dai et al. for butylamine.31 To carefully monitor this effect, it is suitable to use CNTFETs showing n-type or ambipolar characteristics. The ambipolar transistors needed in this experiment are obtained by thermal annealing treatment (T ≈ 120 °C) under vacuum conditions (P ≈ 10-6 mbar) during approximately 10 h. This is a well-known method for removing the oxygen adsorbed at the metal/nanotube contacts and obtaining ambipolar characteristics.3,5 The TEA vapors are controllably introduced in an evacuated vacuum chamber, which enables us to expose the devices to very small and controlled amounts of TEA. The evolution of the electrical characteristics of the CNTFETs is monitored as a function of the TEA dose in the vacuum chamber. Figure 4 displays typical electrical output characteristics of the two types of CNTFETs exposed to TEA vapors. On each graph, the black curve represents the initial n-branch of the ID(VGS) characteristics without TEA (p-branches are out of the displayed bias range shown32). In the case of a CNTFET on SiO2, the red curve shows the measured behavior after the introduction of only a few particles per billion (ppb) of TEA. Note that this very small quantity of TEA is sufficient to induce a detectable left-hand shift of the curve, and confirm the extreme sensitivity of SWNT to a small change of its gaseous environment. When the TEA concentration increases, the n-branch continues shifting to the left at constant S. Simultaneously, we observe that the current in the on-state increases. This experimental result is in good agreement with the theoretical predictions formulated by Heinze et al.9 in the case of an n-doping effect (e.g., using potassium). It is likely that the physisorbed TEA is reducing enough to induce an electron transfer from the TEA to the nanotube. For a CNTFET on APTS, the main observation is that the left-hand shift due to TEA exposition is greatly enhanced compared to the recorded behavior for a CNTFET on SiO2. Indeed, the gap between two consecutive branches increases and is systematically higher for CNTFET on APTS than the one measured for the CNTFET on SiO2. Note also that for a CNTFET on APTS the sub-threshold slope of the n-branch becomes ≈3 times better when the TEA dose increases: (1850 ( 160) mV/dec before TEA to (660 ( 60) mV/dec for a dose of 100 ppm. 454
Figure 4. Comparison of the adsorption influence of TEA molecules onto a CNTFET on SiO2 (a) and a CNTFET on APTS (b). VDS ) 1 V. (c) Evolution of the lateral shift of the transfer characteristics as a function of the TEA dose for a CNTFET on SiO2 (red) and on APTS (black). Lines are guide for the eyes.
To better visualize the differences between the two types of devices with respect to TEA adsorption, Figure 4c displays the evolution of the lateral shifts of the ID(VGS) characteristics as a function of the TEA dose (log scale). We first note that in both cases ∆VGS is proportional to the logarithm of the TEA concentration. Since VGS determines the carrier concentration in the transistor channel, ∆VGS is directly related to the change in the Fermi level position. This can be related Nano Lett., Vol. 5, No. 3, 2005
to the conventional relation that describes doping in semiconductors: EF(n) - EF(i) ) kbT‚ln(Nd/ni), where EF(n) and EF(i) represent the position of the Fermi level in a doped and intrinsic case respectively, Nd is the doping concentration, and ni the intrinsic carrier concentration. In the case of a CNTFET on SiO2, the electron donor nature of the TEA is sufficient to induce a transfer of electrons from the molecules directly physisorbed on the nanotube. This effect is also present in the case of a CNTFET on APTS but it accounts for a minor part of the lateral shift only (
