Emerging Silicon and Non-Silicon Nanoelectronic Devices:
Opportunities and Challenges for
Future High-Performance and Low-Power Computational Applications
(Invited Paper)
Opportunities and Challenges for
Future High-Performance and Low-Power Computational Applications
(Invited Paper)
ABSTRACT
Several key emerging nanoelectronic devices, such as Si nanowire field-effect transistors (FETs), carbon nanotube FETs, and III-V compound semiconductor quantum-well FETs, are assessed for their potential in future high-performance, low-power computation applications. Furthermore, these devices are benchmarked against state-of-the-art Si CMOS technologies. The two fundamental transistor benchmarking metrics utilized in this study are (i) CV/I versus LG and ii) CV/I versus ION /IOFF. While intrinsic device speed is emphasized in the first metric, the tradeoff between device speed and off-state leakage is assessed in the latter. For high-performance and low-power logic applications, low CV/I and high ION /IOFF values are both required. Based on the results obtained, the opportunities and challenges for these emerging novel devices in future logic applications are highlighted and discussed.
I. INTRODUCTION
According to Moore's Law, the number of transistors per integrated circuit doubles every 24 months, and it has been the guiding principle for the semiconductor industry for over 30 years. The sustaining of Moore's Law, however, requires continued transistor scaling and performance improvements. The physical gate length LG of the Si transistors used in the 90 nm logic generation node is ~ 50 nm. It is projected that transistor LG will reach ~ 10 nm in 2011. By way of innovation in silicon technology, such as strained-Si channels [1, 2], high-κ/metal-gate stacks [3�5], and the non-planar Tri-gate CMOS transistor architecture [6], CMOS transistor scaling and performance will continue at least until the middle of the next decade. Recently, a lot of interest generated has been generated and good progress has been made in the study of novel silicon and non-silicon nanoelectronic devices, including Sinanowire field-effect transistors (FETs) [7�11], carbon-nanotube FETs (CNTFETs) [12�18], and III-V compound semiconductor
quantum-well FETs (QWFETs) [19, 20], in the capacity of future computation applications. These devices hold promise as candidates for integration with the ubiquitous silicon platform in order to enhance circuit functionality while simultaneously enabling the extension of Moore's Law well into the next decade and beyond. In this work, two fundamental device metrics, namely (i) CV/I versus LG and (ii) CV/I versus ION /IOFF, are used to benchmark these emerging nano-electronic devices vis-à-vis state-of-the-art Si CMOS transistors with regard to high-performance, low-power logic CMOS-like applications. These benchmarking metrics and
corresponding methodologies have previously been described in great detail [21, 22]. While the first metric highlights the intrinsic speed of devices, the latter permits an assessment of the tradeoff between device intrinsic speed and off-state leakage. Data from our own research devices and also from literature were used in this study. The merits and potential shortcomings of these emerging devices will be discussed. Figure 1 shows the images of some of the novel research transistors discussed in this work.
quantum-well FETs (QWFETs) [19, 20], in the capacity of future computation applications. These devices hold promise as candidates for integration with the ubiquitous silicon platform in order to enhance circuit functionality while simultaneously enabling the extension of Moore's Law well into the next decade and beyond. In this work, two fundamental device metrics, namely (i) CV/I versus LG and (ii) CV/I versus ION /IOFF, are used to benchmark these emerging nano-electronic devices vis-à-vis state-of-the-art Si CMOS transistors with regard to high-performance, low-power logic CMOS-like applications. These benchmarking metrics and
corresponding methodologies have previously been described in great detail [21, 22]. While the first metric highlights the intrinsic speed of devices, the latter permits an assessment of the tradeoff between device intrinsic speed and off-state leakage. Data from our own research devices and also from literature were used in this study. The merits and potential shortcomings of these emerging devices will be discussed. Figure 1 shows the images of some of the novel research transistors discussed in this work.
II. EMERGING P-CHANNEL NANOELECTRONIC
DEVICES
DEVICES
The room temperature CV/I versus LG comparison of conventional Si transistors, Si nanowire transistors, and CNT transistors with p-channels is shown in Figure 2. The data indicate that CNTs exhibit significant CV/I improvement when compared to conventional Si devices. This improvement is due primarily to the mobility enhancement in CNTs. By contrast, the CV/I characteristics of Si nanowire devices and conventional Si devices are similar. A recent report suggests that, fundamentally, there is no reason to expect Si nanowire transistors to have higher channel mobility than standard planar Si devices at room temperature [23]. For example, TCAD simulations have shown that phonon scattering increases, and
hence the phonon-limited mobility decreases, at room temperature for devices containing Si nanowires with diameters less than 15 nm [23]. Additionally, experimental studies reveal that while phonon scattering in Si nanostructures is suppressed at low temperatures, phonon scattering limited transport is indeed prevalent at room temperature, limiting effective channel mobility, as shown in Figure 3. Hence, at room temperature, Si nanowires with dimensions of interest for scaling do not exhibit transistor performance enhancement when compared to conventional planar Si architectures, as shown in Figure 2. The p-channel CV/I versus ION /IOFF characteristics of the CNTFET are shown in Figure 4. Included in this figure are data from conventional planar Si and non-planar Tri-gate Si transistors for comparison. Despite the observation that CNTFETs exhibit high intrinsic speed (CV/I) performance, as shown in Figure 2, they in fact suffer from a low ION /IOFF ratio. This low ratio is attributed to a high IOFF for the CNTFET, as shown in Figure 5, which in turn is due to the existence of ambipolar leakage [15, 21, 22]. The ambipolar leakage is a consequence of metal-CNT Schottky contacts, which are used instead of standard implanted or diffused p-n junctions. It is
anticipated that the use of standard p-n junctions will eliminate ambipolar leakage and improve the ION /IOFF ratios of CNTFETs. It is noted, however, that for high-performance and low-power logic applications, both low CV/I and high ION /IOFF values are required.
DEVICES
Figure 6 shows the room temperature CV/I versus LG comparison of conventional Si transistors, CNT transistors, and III-V (InSb) compound semiconductor QW transistors [19, 20] with n-channels. In comparison with conventional Si devices, InSb transistors exhibit significantly larger n-channel intrinsic speed (CV/I), a benefit of higher channel mobility and lower utilized supply voltage VCC (0.5 V). The increased channel mobility also translates to a highfrequency gain in InSb transistors, as shown in Figure 7 [20]. In this case, the dc CV/I data is directly correlated to the ac cutoff-frequency fT data. The n-channel CNT devices in Figure 6 all show degraded CV/I performance compared to conventional Si n-channel devices. This
phenomenon can possibly be explained by considering that a suitable n-type workfunction metal that forms a stable interface with CNTs has yet to be demonstrated. Upon resolution of this issue, a high performance n-channel CNTFET is anticipated based on the symmetry of the conduction and valence bands for CNTs [24]. In Figure 8, n-channel CV/I versus ION /IOFF characteristics are shown for CNTFETs with chemically-doped junctions [18] and for InSb QWFETs. Also shown, for the sake of direct comparison, are conventional planar Si and non-planar Tri-gate Si transistors. The use of chemically-doped junctions in CNTFETs [18], as opposed to metal-CNT Schottky junctions, suppresses the ambipolar leakage conduction and reduces IOFF, thus improving the ION /IOFF ratio. However, the resulting CV/I performance is still degraded, possibly due to increased external parasitic resistance of the doped junctions. Nevertheless, this chemically-doped junction approach is indeed a major advancement for CNTFET research [18]. Interestingly, the n-channel InSb QWFET also exhibits a low ION/IOFF ratio, as shown in Figure 8. This phenomenon is a consequence of high gate leakage, as exhibited in Figure 9, due to the low barrier height at the Schottky metal-semiconductorjunction. The low barrier height arises from (a) Fermi-level pinning at the metalsemicondctor
interface and (b) the use of a narrow-bandgap semiconductor. It is predicted that the use of a gate dielectric between the metal gate and the III-V device layers will eliminate such Schottky gate leakages and improve the ION /IOFF ratio [20].
IV. CONCLUSIONS
In this paper, we have identified the merits and potential shortcomings of various emerging nanoelectronic devices with respect to future logic applications. Specifically, we have shown that (a) Si nanowires offer no transistor CV/I performance advantage over conventional Si transistors at room temperature, likely due to the significant role played by phonon scattering at room temperature, and that (b) both p-channel CNTFETs and n-channel QWFETs exhibit impressive CV/I gain when compared against conventional Si transistors, but they suffer from degraded ION/IOFF ratios, a result of ambipolar conduction and Schottky gate leakage,respectively. Based on this study, we anticipate that upon solving the off-state leakage
problems, high-mobility devices such as CNTFETs and III-V QWFETs have the potential to enable high-performance logic applications with very low supply voltage VCC (e.g. below 0.5 V).
problems, high-mobility devices such as CNTFETs and III-V QWFETs have the potential to enable high-performance logic applications with very low supply voltage VCC (e.g. below 0.5 V).
Cesar Hernandez
19.502.806
CRF
Referencias
No hay comentarios:
Publicar un comentario