Optimization of FIBMOS Through 2 D Silvaco ATLAS and 2 D Monte Carlo Particle-based Device Simulations

Focused Ion Beam MOSFETs (FIBMOS) demonstrate large enhancements in core device performance areas such as output resistance, hot electron reliability and voltage stability upon channel length or drain voltage variation. In this work, we describe an optimization technique for FIBMOS threshold voltage characterization using the 2D Silvaco ATLAS simulator. Both ATLAS and 2D Monte Carlo particle-based simulations were used to show that FIBMOS devices exhibit enhanced current drive capabilities when compared to normal MOSFETs. It was also found that the device performance is very much dependent upon the FIB implant profile. High and narrow doping of the FIB implant leads to high drain current and low hot carrier reliability, whereas low and wide doping gives rise to lower drain current and higher hot carrier reliability.


INTRODUCTION
Focused Ion Beam (FIB) implantation has recently been proposed for channel engineering in high-performance MOSFET fabrication. Shen and his co-workers [1] fabricated FIBMOS devices with a special narrow doping region implanted with FIB on the source side of the channel. They showed that this novel device structure, schematically shown in Figure 1, exhibits higher output resistance, reduced hot electron degradation, more stable threshold voltage (Vr) upon device scaling, and higher operation frequency. However, the rather asymmetric doping profile in this device prevents one from using the classical expressions for the threshold voltage since it is a rather complex function of the step doping width and density. In this work, we first describe the optimization technique that Kang et al. [2] developed for FIBMOS threshold voltage characterization, based on 2D device simulations and 3D Vr contour mapping. It enables one to design FIBMOS devices with a certain Vr and best performance in consideration of the drain current, hot-electron degradation, Vx stability and maximum operation frequency.
Using both 2D Silvaco ATLAS energy balance model and 2D Monte Carlo particle-based simulations, we also examine hot-electron reliability of FIBMOS devices. This is a severe problem in normal deep-submicrometer MOSFETs in which the high substrate doping, used to prevent the punch-through effect, leads to large electric fields and enhanced impact ionization. For n-channel MOSFETs, the electrons generated via the impact ionization process tend to be injected into the gate oxide (leading to a threshold voltage shift and degradation of the channel mobility) or are injected into the drain. The generated holes are swept into the substrate, thus giving rise to substrate leakage current and enhanced impact ionization due to the forward biasing of the source/substrate junction. Both the 2D Silvaco ATLAS and the 2D Monte Carlo particle-based simulator show that the built-in electric fields at the source side of the channel, due to the presence of the FIB implant, lead to enhanced current-drive capabilities of the FIBMOS device when compared to normal MOSFETs. Our 2D Monte Carlo particle-based simulations also show that (due to the relatively low electric field at the drain end of the channel) the average electron energy in FIBMOS devices is low which, in turn, significantly reduces the probability for impact ionization to occur.

THRESHOLD VOLTAGE CHARACTERIZATION
As already discussed in the introduction, FIBMOS devices have an asymmetric channel doping profile which influences the device performance. For example, the threshold voltage is not only a function of the step doping density, but it is also a function of the step doping width. To understand the relationship between the doping profile of the FIBMOS device and the threshold voltage, we simulated a large number of FIBMOS devices with various implant step widths and doping densities. The Silvaco device simulator ATLAS was used for this purpose [2]. The extracted threshold voltages were then mapped on a plane of step width and step doping density. In other words, three-dimensional contour plots for the threshold voltage were generated with the step width and the step doping density on the horizontal and vertical axis, respectively. The extracted threshold voltages were then rearranged for contour plots. A sample contour plot for 0.25tm gate-length FIBMOS devices is shown in Figure 2. From these contour plots, it is possible to find the combinations of step doping width and doping density that give the desired threshold voltage. In Table I, we list four different combinations of parameters that lead to the best device performance. One should note that the four combinations of doping parameters do not result in identical device performance, even though the threshold voltage is the same. The drain current is reduced as one goes from high and narrow to low and wide doping profile. These are important features of FIBMOS. Therefore, if one needs high output current and device reliability is less important, high and narrow doping is best. Step size [nm]  Step doping width (nm) 60 80 100 120 Step doping density  [3]. Intravalley scattering is limited to acousitic phonons. For the intervalley scattering, we include both g-and f-phonon processes. It is important to note that, by group symmetry considerations, the zeroth-order low-energy f-and g-phonon processes are forbidden. Nevertheless, three zeroth-order f-phonons and three zeroth-order g-phonons with various energies are usually assumed [4]. We have taken into account this selection rule and have considered two highenergy f-and g-phonons and two low-energy f-and g-phonons. The high-energy phonon scattering processes are included via the usual zerothorder interaction term, and the two low-energy phonons are treated via a first-order process [5]. The first-order process is not really important for low-energy electrons but gives a significant contribution for high-energy electrons. The low-energy phonons are important in achieving a smooth velocity saturation curve, especially at low temperatures. The phonon energies and coupling constants in our model are determined so that the experimental temperature-dependent mobility and velocity-field characteristics are consistently recovered [6]. At present, impact ionization, surfaceroughness and Coulomb scattering are not included in the model. However, these simplifications do not prevent us from examining the overall performance enhancement of FIBMOS devices with respect to normal MOSFETs. The gate length of both the regular MOSFET and the FIBMOS device being investigated equals 0.25 gm. The source and drain extension is 50 nm, and the junction depth is 36 nm. The standard device has a substrate doping density of 1018 cm -3. The doping density of the source and drain regions of this device is 1019cm -3. The substrate doping of the FIBMOS device equals 1016 cm-3, the length of the one-step FIB region is 70 nm, the depth of the FIB region is 36 nm, and the doping of this region is 1.4 x 10 TM cm -3. The source and drain doping of the FIBMOS device is identical to the one used for the standard device.
The conduction band edge, for applied bias V-V and Vz 0.8 V, is shown in the top panel of Figure 3. One can clearly see the presence of a potential barrier near the source end of the channel, as a result of the high doping of the FIB region. Also note that the width of the barrier region is less than 40nm, which suggests that source to drain tunneling might take place in this device structure, thus leading to enhanced off-state leakage currents. The x-component of the electric field (electric field component along the channel) is shown on the bottom panel of Figure 3  and drain end of the channel lead to a reduction in both the average drift velocity and the average electron energy, thus eliminating the problem of substrate leakage currents due to impact ionization. The characteristics of the regular device are quite different. The drift velocity is relatively low near the source end of the channel which, as shown later, leads to a factor of four smaller drain current when compared to the drain current of the FIBMOS device. On the other hand, the average electron energy peaks at the drain end of the channel and is about a factor of four higher than the average electron energy in the FIBMOS device. This gives rise to hot-carrier degradation in regular MOSFETs, and can be reduced by the introduction of the lightly doped drain (LDD) regions.
The transfer and the output characteristics of the FIBMOS device and the regular MOSFET, obtained with the 2D particle-based simulator, are shown in Figures 5(a) and 5(b), respectively. Also shown in Figure 5(b) are the output characteristics of both the regular and the FIBMOS device obtained with the Silvaco ATLAS simulator [2]. Assuming    to series resistance effect [7]. This, in a way compensates for the fact that Coulomb and surface-roughness scattering are not being considered in the present model. The source and drain doping used in the Silvaco ATLAS simulations equals 10 = cm -3, thus leading to significantly smaller source and drain series resistance.

CONCLUSIONS
We have proposed and used an optimization technique for threshold voltage extraction of a one-step FIBMOS device. Using 3D contour plots for a range of step doping width and doping density, we have determined the doping profiles that set the threshold voltage at the targeted value for best device performance. We find that if the system requirements are high current drive or lowpower operation, one needs to use narrow FIB implants with high doping. On the other hand, if the system requires long device lifetime, then wide and low-doping profiles for the FIB implant need to be used. Both the 2D Monte Carlo particlebased simulations and the Silvaco ATLAS simulations suggest performance enhancement of FIBMOS devices with respect to normal MOSFETs. An increase in the drain current of about a factor of four at the highest drain voltage was found for the FIBMOS device when compared to the regular MOSFET device. Similar trends have been observed in Ref. [8].