Improving subthreshold MSB-EMC simulations by dynamic particle weighting Carlos Sampedro ∗ , Francisco G´ amiz ∗ , Andr´ es Godoy ∗ , Raul Val´ ın ‡ , and Antonio Garcia-Loureiro § ∗ Dep. de Electr´ onica y Tecnolog´ ıa de Computadores, Universidad de Granada, 18071 Granada (SPAIN). Email: csampe@ugr.es ‡ College of Engineering, University of Swansea, Swansea (UK) § Departamento de Electr´ onica y Computaci´ on, Universidade de Santiago de Compostela, Santiago de Compostela 15782 (SPAIN) Abstract— The study of current and futures nanodevices de- mands a special focus on the subthreshold regime for switching and power consumption considerations. MSB-EMC simulators represent one of the best options for the study of ultimate CMOS devices offering the most detailed description of carrier transport. However, several issues derived from the charge discretization process and the statistical nature of the technique limit the application to subthreshold regime. This paper presents a method for statistical enhancement including dynamical calculation of electron-particle equivalent (EPE) values with position and bias dependence in order to extend the application of MSB-EMC simulators to subthreshold regime in a feasible way from a CPU and memory requirements point of view. I. I NTRODUCTION As semiconductor devices are aggressively scaled down in the search for improved performance and lower power consumption, the semiconductor industry must face important challenges arising from the use of new geometries and materi- als at the nanoscale in order to fulfill the requirements given by the ITRS for the forthcoming 14/11 nm nodes [1]. To accom- plish such extreme scaling, it is necessary to control Short Channel Effects (SCEs) and to enhance transport properties including arbitrary channel orientations or strain materials to increase the device performance. At this point, standard bulk- MOSFET technology cannot provide good enough solutions for sub-22 nm nodes due to the limited control of SCEs and variability problems coming from a highly doped channel [2]. Two are the mainstream Silicon based options to reduce SCEs based on novel device structures: the use of multiple gate devices (MuGFETs) and the extension of planar technology by means of SOI. The first option, thanks to an outstanding SCEs control [3], [4], is able to extend the end of the roadmap by means of different candidates (FinFETs, Trigate and Gate-All- Around). The second option tries to take advantage of the ben- efits provided by SOI devices. More precisely, Extremely Thin Fully Depleted SOI devices (ET-FDSOI or simply ETSOI) are chosen thanks to the extra control over SCEs provided by the thin silicon channel, the buried oxide (BOX) and their simpler fabrication process compared to planar bulk architectures and, of course, 3D devices. This fact allows a reduction in the overall cost and an almost straightforward layout transfer from bulk to SOI [5] in opposition to the implementation of MuGFET technology, where a complete redefinition of the fabrication flow and the introduction of new processing steps are mandatory. Within this framework, the use of advanced device sim- ulation tools offers several advantages for the development of upcoming technological nodes. On the one hand, it is possible to predict the performance of different architectures and technological choices. On the other hand, the development stage can be reduced in terms of cost and time. Another important advantage is the possibility of studying the impact of each technological booster separately to explain experimental results and to determine which one is the most effective in improving the device performance. Depending on the required accuracy, the computational re- sources and time available to perform the simulations, different approaches from classical to full quantum could be considered. However simple tools based on drift diffusion models should not been employed since confinements effects are of special importance. At the opposite end of the spectrum, full quantum simulators based on numerical solutions of the Schr¨ odinger equation or the Non-Equilibrium Green’s Functions theory (NEGF) have also been developed [6]. The introduction of scattering in the simulations involves a very high computa- tional cost and for this reason, only simplified models can be used in practical quantum simulations [7]. In this scenario, Multi-Subband Ensemble Monte Carlo sim- ulators (MSB-EMC) [8] represent one of the best options for the study of ultimate CMOS devices offering the most detailed description of carrier transport, catching the main quantum effects and showing balanced computational cost and memory needs. However, at subthreshold regime the main limitation of this method becomes of special relevance. The study of device characteristics below threshold voltage has became of special interest in the last years due to the exponential grown of mobile application where stand-by power consumption is of paramount importance. The stochastic nature of MC tech- niques limits the maximum accuracy in current calculations due to the inherent statistical noise. This fact, that can be reduced increasing the number of flights in one-particle ap- proaches, is difficult to be dealt with when the self-consistency between the electrostatics and the particle ensemble must be kept. The quantization introduced in the charge density in the conversion process from a continuum to a particle-based description adds rounding errors which are non-negligible 276 978-1-4673-5736-4/13/$31.00 ©2013 IEEE