Random Dopant Variation in Junctionless nanowire Transistors Nima Dehdashti Akhavan, Isabelle Ferain, Pedram Razavi, Ran Yu and Jean-Pierre Colinge Tyndall National Institute, University College Cork, Ireland. Email:
[email protected] Abstract In this article we study the influence of random dopant variation in junctionless nanowire transistor using numerical modelling. Simulations have been carried out by three-dimensional quantum simulator based on Non-equilibrium Green’s function formalism. The simulations reveal that junctionless nanowire transistor suffers less variation in subthreshold region due to random dopant distribution compare to the conventional inversion mode counterpart. This improved behaviour arises from uniform distribution of ionized dopant in the channel under the gate and in the source/drain regions.
Introduction As the gate length of transistors is shrunk down to nanometer scales, the control of doping profile in the channel region becomes problematic and an unavoidable variation of various transistor parameters takes place due to the random positioning of a finite number of doping atoms in the channel region. Even in the absence of extrinsic process variations, the number and position of doping atoms are subject to stochastic variations due to the physics of ion implantation, doping diffusion and other processes involved in the doping of silicon devices. This phenomenon poses serious challenges in scaling inversion mode (IM) transistors below the 22-nm node mainly due to diffusion of impurities from highly doped source/drain junctions into the undoped or moderately doped channel region. Here we demonstrate that the recently proposed junctionless nanowire transistor (JNT) with uniform doping in source, channel and drain regions exhibits less drain current variation in the presence of randomly positioned dopants than IM devices[1].
Simulation and device design A three-dimensional (3D) quantum simulator based on the Non-equilibrium Green’s function (NEGF) formalism has been used to extract the physical parameters [2]. We have considered two types of structures: (a) JNT device with uniform doping of 1020 cm-3 and (b) IM device with S/D doping of 1020 cm-3 and undoped channel. Both devices are rectangular gate-all-around (GAA) nanowire with: S/D length LSD=10nm, channel length LG=10nm, gate oxide thickness TOX=1nm and cross-section of 3nm×3nm. The simulations have been carried out considering discrete dopant atoms in both structures. In the JNT device, to the channel contains 12 discrete donor atoms [3,4]. In the IM device we have assumed that 3 donor atoms have diffused 2nmfrom the S/D regions into channel region, which corresponds to typical doping gradient predicted by the ITRS for the 10-nm technology node [5].
Results and Discussion All simulations were been carried out at VDS=0.1 V. Figure 1 shows the cross-section of the JNT and IM devices under consideration. Figure 2 schematically represents the imperfectly steep doping
978-1-61284-760-3/11/$26.00 ©2011 IEEE
gradient at S/D junctions in the IM device. The problem disappears when using the JNT concept due to absence of doping concentration gradients. Figure 3 shows an example of random donor atom distributions for IM and JNT Figure 4 shows the 3D electron density profile of a JNT at VG=VTH. It is clear that the donor impurities give rise to local electron concentrations [6]. Figure 5 shows the energy-position resolved density of states (DOS) and current density for JNT at threshold, together with the first subband profile. The creation of local DOS by donor atoms is clearly visible. The current spectrum is dominated by tunnelling between impurityinduced local DOS pockets, which means that electron-phonon scattering can be neglected in the presence of such an impurity distribution profile [6]. In Figure 6 the IDS-VGS curves have been plotted for both the IM and the JNT devices. The curves with symbols (● or ■) represent the simulation with continuous doping profiles (ND =0 cm-3 in the IM device channel and ND=1020 cm-3 in the JNT). The other curves correspond to the simulations with discrete donor atoms. The JNT has negative VTH while the IM has positive one because we use Φ MS=0V in both devices. 30 different random discrete doping atom distributions were used to simulate both IM and JL devices. In order to compare the effect of random distribution we have considered subthreshold current variation at VGS=VTH-0.2V. The JNTs show 20% less variation of average subthreshold current than the IM devices. In an actual fabrication process one can even achieve better performance for JNT in terms of current variation by introducing an annealing step which makes the random dopant profile distribution even more uniform along the device, however the latter needs further theoretical investigation and modelling.
Conclusion Using 3D atomistic numerical simulation we showed that JNTs suffer from less subthreshold current variation than IM device due to absence of doping concentration gradients. The JNT can be considered as a potential competitor to IM multigate devices as the gate length is shrunk down to few nanometers.
Acknowledgments This work was supported by the Science Foundation Ireland grants 05/IN/I888 and 10/IN.1/I2992, the European project SQWIRE under Grant Agreement No. 257111 and the European Community (EC) Seventh Framework Program through the Network of Excellence Nano-TEC under Contract 257964.
References [1] J.-P. Colinge et al., Nature Nanotech., 5-3, pp. 225-229 (2010) [2] A. Afzalian et al., J. Comput. Electron., 8, pp 287-306, (2009) [3]Y. Li et al., Journal of Applied Physics 102, 084509 (2007) [4] A. Martinez et al., Proceedings ULIS, pp. 42-45, (2011) [5] D.-H. Moon, et al., Jpn. J. Appl. Phys. 49, 104301, (2010) [6] N. Dehdashti et al., Proceedings of EUROSOI, pp. 79-80 (2011) [7] N. Dehdashti et al., JAP, 108-3, pp.034510-034510-8(2010)
Source
Gate
Drain
Inversion-mode short channel length
Source
Figure 1: Schematic representation of the JNT (top) and IM (bottom) devices structures considered in this work.
Gate
Drain
Junctionless short channel length
Source
Gate
Drain
Inversion-mode ultra-short channel length Source
Gate
Drain
junctionless ultra-short channel length
Figure 2: In the IM device the imperfectly steep doping gradient at S/D junctions randomizes the effective gate length (top). The uniform doping concentration in a JNT eliminates that problem (bottom).
Figure 3: Example of random donor atom distributions in IM and JNT devices.
Figure 4: Three-dimensional electron density profile in JNT device at VGS=VTH. The pockets of higher electron concentrations around the vicinity of donor impurity can easily be observed.
Figure 5: Energy-position resolved DOS (top) and current density (bottom) and the first subband profile in JNT at VGS=VTH. It can be seen that the current is dominated by tunneling.
Figure 6: IDS-VGS curve for IM and JNT devices. The curves with symbols (● or ■) represent the simulation with continuous doping profiles (ND =0 cm-3 in the IM device channel and ND=1020 cm-3 in the JNTs). The other curves correspond to the simulations with discrete donor atoms.
