Enhanced electroluminescence in suspended carbon nanotube transistors
Although much progress has been made in the study of electronic transport properties of carbon nanotubes (CNTs), little work exists on their optoelectronic properties. Electrically excited infrared (1R) emission from single nanotube molecules has been recently demonstrated [1-3]. This was accomplished through the radiative recombination of electrons and holes injected from the opposite ends (source and drain) of an ambipolar Schottky barrier carbon nanotube field-effect transistor (CNTFET) [1]. In this novel molecular light source the CNT is in direct contact with the substrate, a fact that is expected to open non-radiative dissipation channels. Indeed the measured quantum efficiency was in the range of 10 -6-10 -7 [3]. In this work, we report the first electroluminescence results from suspended CNTFETs. We find not only an increase in quantum efficiency by 2-3 orders of magnitude, but a completely different dependence of the light emission on the electrical properties of the CNTFET. To fabricate a suspended CNTFET, we etched 2-10 μm wide trenches in 200 nm thick SiO 2 on Si wafers. The trenches extended through the SiO 2film and 2 μm into the Si substrate. CNTs with diameter around 2-3 nm were grown on the trenched-substrate by chemical vapor deposition. The growth process is similar to that in Ref. [4] with 3 nm Fe nanoparticles as catalyst and CH 4 as carbon source. Palladium source and drain electrodes were then patterned on CNTs with channel length between 30 and 60 μm. A highly doped silicon substrate was used as the back gate. A scanning electron micrograph of such a partially suspended CNTFET overlaid with film stack schematics is shown in Fig. 1. The channel length is 26 um, the width of the trench is 3.5 μm, and the nanotube diameter is 1.9 nm. We image the IR emission from the CNTFETs with a liquid nitrogen-cooled HgCdTe detector array mounted on the probe station that is used to measure their electrical properties. The lateral resolution of the detector is diffraction limited at about 2 μm. The electroluminescence spectrum determined by a set of bandpass filters is in agreement with that reported before [3]. In the following experiment, we collect emission on the high-energy side at 1.8 μm as a function of the gate bias. Fig. 2 shows a collected IR emission superimposed on an optical image of a CNTFET fabricated as described above. Unlike non-suspended CNTFETs whose emission can be spatially translated along the length of a CNT by changing the gate bias [2], emission from the partially suspended device is 'pinned' at the suspension even under a range of gate biases. Fig. 3 (a) shows transfer characteristics of the device while the emission is collected. The open circles and the solid circles show the hole and the electron current, respectively. Fig. 3(b) compares the emission intensity at the trench from the electron branch to that from the hole branch. We observe no detectible emission when the drain current is below certain threshold current (I th), with I th ∼ 2.4 μA for holes and I th ∼ 3.4 μA for electrons. At I > I th, the emission intensity increases exponentially with increasing current. There are two major differences of emission from suspended CNTFETs when compared with non-suspended devices: (1) the dominant emission from the suspended CNT occurs at unipolar regions (V g = -6 V for holes and V g = 1 V for electrons) and no appreciable emission collected at the ambipolar region where the current reaches minimum (V g = 2.6 V) [2]; (2) the minimum drain bias needed to observe emission is much smaller in suspended tubes (V ds ∼ 5 V) than that from non-suspended tubes (V ds ∼ 20 V) [3], yet the emission intensity is 2-3 orders magnitude stronger in the suspended tubes. In a suspended tube, reduced capacitive coupling from the Si back-gate to the suspended CNT modifies band bending of the tube, and the most band bending occurs at the suspension and SiO2 substrate interface where the dielectric constant was changed from 1 to 3.9. The resulting local high field at this interface contributes strongly to the generation of hot carriers in electroluminescence. The inset of Fig. 3b shows a schematic diagram of e-h pair generation by the relaxation of a single type (electrons in this case) of hot carriers. The different rates of emission intensity increase with drain current and the different I th for electrons and holes (Fig. 3b) could be due to the asymmetric interfaces formed at the two sides of the trench during fabrication. Unique to the 1-D system, the strong Coulomb interaction between the e-h pairs leads to a significant exici tonic binding energy, which can be as large as 1/3 of the CNT band gap, preventing the e-h pairs from dissociating under high field [5]. The collected quantum efficiency from suspended CNT is around 10 -3-10 -4, i.e., improved by 2-3 order of magnitudes from that of CNTs laying on planar substrates. Fig. 4a shows emissions collected with gate scan under other forms of locally enhanced field in addition to a trench: a local defect and at one of the metal-CNT contacts. Fig. 4b shows the drain current (squares) and the emission intensities at trench (diamonds), local defect (triangles) and contact (circles) sites vs. gate scan while the emission in Fig. 4a is collected. The comparison shows that the luminescence generated at the suspended site is over 10 times more efficient. © 2005 IEEE.
