Impact of the nanotube diameter on the performance of CNFETs

Abstract

Since the first carbon nanotube field-effect transistor (CNFET) was demonstrated in 1998, substantial effort has been put into improving the device performance. Individual CNFET devices with characteristics rivaling those of state-of-art Si MOSFETs have been obtained with proper scaling and optimized geometry [1,2]. However, published characteristics of CNFETs with comparable geometries do show a wide range of on-currents, Ion. This irreproducibility is a major obstacle that needs to be overcome to make nanotube devices a viable technology. The question to address in this context is: "What is the origin of this on-current variation?" In principle, the contact quality or intrinsic tube-to-tube variations may be responsible for the observed Ion irreproducibility. This study investigates 38 CNFETs and shows that nearly 3 orders of magnitude current variation can be explained in a comprehensive way by the diameter variation among nanotubes alone. This is the first systematic analysis that correlates the device performance with the nanotube properties quantitatively. It also shows that one can neglect the impact of the preparation on the contact quality to a large extend. CNFETs studied here use standard back-gate geometry as shown in Fig. 1. Laser ablation grown nanotubes dispersed on 10nm SiO2 are contacted by palladium (Pd) source/drain (S/D) with a 300nm separation. This S/D contact at the end of the channel material (nanotube) is the most common geometry used in current nanotube based nano-electronics. At a metal/semiconductor interface, depending on the line-up between the metal Fermi level and the valence/conduction band of the semiconductor, a barrier in general exists. This is why CNFETs are considered Schottky barrier (SB) devices [3,4,5,6]. It is usually more difficult to make a "good" contact to a wide-gap semiconductor. The band gap (Eg) effect on the current injection through the barrier in a CNFET is illustrated in Fig. 2. Assuming a constant work function (defined as the sum of the electron affinity and 1/2 Eg in the bulk nanotube) for all nanotubes, the barrier height, ΦSB, increases linearly with increasing Eg, In this picture, one can expect current variation in devices having different band gaps, which results according to tight binding theory from different tube diameters (Eg ∼ I/d). Fig. 3 shows the substhreshold characteristics of three Pd-CNFETs measured at V ds=0.5V. Similar hole injection is observed for all three devices at Vgs < 0, but the on-current differs by almost 2 orders of magnitude. We have recorded Ion at Vgs,-Vth = -0.5V for every device and have generated an Ion distribution plot, as shown in Fig. 4 (black dots). To correlate the current with the nanotube diameter, we characterized 78 nanotubes by TEM and plotted the diameter distribution also in Fig. 4 for comparison with the current distribution. The diameters are found to range from 0.6nm to 1.5nm with an average diameter around 0.9nm. Both of the curves are normalized to obtain an integral of 1. By comparing the integration values, we can now correlate the device on-current with the nanotube diameter. Fig. 5 plots the on-currents as a function of the nanotube diameters (black triangles). The most important information obtained from this graph is that, Ion exhibits larger variation for small diameter nanotubes than for large diameter tubes. Nanotubes grown by other synthesis schemes encounter the same diameter variation. To show the universal applicability of our analysis, we have extended Ion (d) to larger diameters and show the comparison with 3 data points from other publications [2,7,8]. These 3 devices used CVD grown nanotubes which have larger diameters. As apparent from Fig. 5, the 3 data points follow nicely the projected trend (dash line). Note: that data point "1" is slightly above our curve due to the stronger gate control provided by 8nm HfO2 used in the device and point "3" has slightly smaller Ion due to the longer channel length (>2μm). This consistent trend between two different nanotube sources provides first evidence for our statement: the observed current variation results from different nanotube diameters. In an SB-CNFET, on a logarithmic scale, current injection through the SB is inversely proportional to ΦSB. With the linear dependence between ΦSB and Eg as discussed above, one can expect log(Ion) ∼ - Eg ∼ -1/d Fig. 6 shows this linear dependence which provides extra evidence for the validity of our analysis. In summary, we have presented the first statistical analysis of the Ion dependence in a CNFET on the nanotube diameter. The different current values reported for state-of-art CNFETs can be consistently explained as a result of the d-variation. Our study clearly shows smaller current variation occurring in CNFETs with larger diameter nanotubes, which is a crucial finding for the choice of nanotube type in device applications. In order to ensure proper device behavior and high on-currents, it is desirable to operate with tubes exhibiting d∼1.7nm. © 2005 IEEE.