نقش دوپینگ در ترانزیستورهای نانولوله ای کربن با شکاف های بین سورس/درین Role of Doping in Carbon Nanotube Transistors With Source/Drain Underlaps

I. INTRODUCTION UNDERSTANDING of electrostatics [1]–[۴], transport mechanisms [5]–[۸], scaling behavior [9]–[۱۴], and performance [15]–[۲۲] of carbon nanotube field-effect transistors (CNTFETs) has been rapid. Notable breakthroughs are the discovery of zero-Schottky-barrier contacts [23]–[۲۸], integration of logic gates, a static RAM, a ring oscillator [29], [30], and the large scale integration of CNTs with Si [31]. Performance improvement especially in the on state of CNTFETs by chemically doping the nanotube to n-type has been demonstrated experimentally [32], [33]. Doping of both single-wall and multiwall nanotubes by either electron donors or electron acceptors has been reviewed [34]. As-grown p-type nanotube devices have been converted to n-type devices by controlled doping [35], [36]. The application of doped nanotubes as single-electron transistors has been discussed [37]. Theoretical studies on CNTFETs with doped nanotube contacts (C-CNTFETs) have been performed using single band [38] and full band calculations [39], [40]. Recently different issues in modeling C-CNTFETs have been discussed [41]. Originally motivated by bioassembly of CNTFETs, we studied CNTFETs with ultrashort metal gates and source/drain underlaps to understand the transport physics and performance [5], [6], [42]. The effects of dielectric constant and thickness and geometry were investigated [14], [22]. It soon became apparent that these designs had excellent performance metrics and that their applicability went beyond the concerns of bioassembly. We have found, as a rule of thumb, that the device with the least capacitance ‘wins’ in terms of speed performance [14], [22]. As with any rule of thumb, there are exceptions such as the asymmetric device with its figures of merit shown in Table III of [6], but the rule generally points one in the right direction. For the case of CNTFETs, we have found the capacitance to be dominated by the fringing fields directly from metal gate to metal source/drain, the equivalent of the overlap capacitance in traditional Si FETs. The underlaps reduce this capacitance. The underlaps reduce the electric field in the drain and thus reduce the interband tunneling in the drain which is the cause of the ambipolar leakage current and reduced on/off current ratios. The underlaps reduce the gate to drain capacitance and thus the effect of the drain potential on the potential under the gate. Therefore, the underlaps improve both dc and ac device performance. We are not the first to discover the advantages of underlaps. Similar device geometries with doped source/drain and undoped underlaps have been previously considered for silicon FinFETs [43], [44]. An underlap value of 10 nm has been reported to reduce short-channel effects and to obtain optimal performance [45], [46]. Recently Stanford and IBM groups have studied CNTFETs with source/drain underlap geometry to minimize parasitic capacitance [25], [47]. In our previous studies, the CNTs were undoped. In this work, we investigate the effects of doping on the same model device geometry. The n-type doped carbon nanotube transistors have been studied experimentally by Javey et al.. [33] and Radosavljevic et al. [32] for CNTs with diameters 1.4 to 1.6 nm that we study here. Both the groups reported high on current and larger transconductance values with doped carbon nanotubes. Javey et al. observe an optimal doping level in terms of on/off current ratio. The off current increases significantly at doping levels higher than the optimal value and the on/off current ratio degrades. To understand the physics of doping effects on the performance of CNTFETs, especially on the off-state, and to design an optimal device in terms of high on/off current ratio we consider two types of contacts in this study. (i) Zero Schottky-barrier contacts with lightly doped source/drain underlaps which we will designate as SB-CNTFETs and (ii) heavily doped CNT contacts which we will designate as C-CNTFETs following Knoch et al. [38]. With a 50 nm CNT and a 5 nm gate, the SB-CNTFETs have optimal on/off current ratios for a doping level of 10 dopants per atom. To obtain MOSFET-like contacts in a 50 nm CNT, we need a doping concentration value of 5 10 or higher dopants per atom. At these doping concentrations the direct source to drain tunneling dominates the leakage current. This reduces the on/off current ratio in C-CNTFETs by orders of magnitude. We propose a 100 nm SB-CNTFET with a gate length of 10 nm and an optimal doping level of 10 dopants per atom as the optimal device. This device has an on/off current ratio of 6 10 , an inverse subthreshold slope of 62 mV/dec, a switching delay time of 32 fs, and has almost unipolar characteristics over the entire range of gate bias used in this study.