Andrea Ghetti

Field acceleration for oxide breakdown-can an accurate anode hole injection model resolve the E vs. 1/E controversy?

A simple model, based on the concept of Anode Hole Injection, explains a number of puzzling measurements of oxide lifetime as a function of applied voltage. We provide systematic explanations of these measurements, and explore its implications for gate oxide reliability.

Giant Random Telegraph Signals in Nanoscale Floating-Gate Devices

The magnitude of a random telegraph signal (RTS) in nanoscale floating-gate devices has been experimentally investigated as a function of carrier concentration. Discrete current switching, which is caused by a single trap, has been found to be almost one order of magnitude higher with respect to what was predicted by the classical theory of carrier number and correlated mobility fluctuations. Nevertheless, the trap signature well fits the typical SiO2 trap spectroscopy. In addition, the rigid shift between the transfer curves related to filled- and empty-trap state, together with the normalized current fluctuation dependence on the channel carrier density, suggests that a pure number fluctuation is the correct theoretical interpretative framework. Thus, we propose a possible physical explanation for such a giant RTS on the basis of a quasi-1-D current filamentation.

Injection efficiency of CHISEL gate currents in short MOS devices: physical mechanisms, device implications, and sensitivity to technological parameters

This paper analyzes MOSFET gate currents in the so-called channel initiated secondary electron injection regime (CHISEL). A Monte Carlo model of the phenomenon is validated and then extensively used to explore CHISEL scaling laws. Results indicate that, compared to conventional channel hot electron injection (CHE), CHISEL exhibits a weaker dependence on channel length and a larger sensitivity to short channel effects. These results are confirmed experimentally and exhaustively explained with the help of simulations; furthermore, some of their possible detrimental consequences on the programming efficiency of CHISEL based flash cells are analyzed. Finally, the impact of channel doping, oxide thickness, and junction depth on CHISEL efficiency has been explored, and guidelines to maintain high injection efficiency in short devices are derived.

Low-voltage hot electrons and soft-programming lifetime prediction in nonvolatile memory cells

This paper reports experiments and Monte Carlo (MC) simulations of flash memory cells at the typical bias conditions of read operations (high V/sub GS/ and low V/sub DS/) leading to the soft-programming phenomenon. Comparing experiments with simulations we first show that, differently from the previously reported case of homogeneous injection experiments, efficient energy gain mechanisms must be invoked to explain the order of magnitude of gate (I/sub G/) and substrate (I/sub B/) currents at low voltage. Second, the voltage scaling behavior of the soft-programming lifetime is analyzed and the validity of usual extrapolation techniques to evaluate this parameter is addressed.

Bias and temperature dependence of homogeneous hot-electron injection from silicon into silicon dioxide at low voltages

In this paper, new homogeneous hot-electron injection data at 300 K and 77 K is provided covering applied voltages from well below to well above the Si-SiO/sub 2/ barrier height, and a wide range of oxide fields. We found that, in contrast to the MOSFET case, homogeneous injection shows two different regimes for accelerating voltages below and above the barrier height. A simple interpretation of the data is proposed, and supported by Monte Carlo (MC) simulations of the injection experiment. Essentially, the two regimes are the signature of a marked transition between an electron population mostly heated by the electric field, and a tail population created by additional but less efficient energy gain mechanisms, leading to a sharp transition in the carrier distribution function. The details of the bias and temperature dependence of injection are then interpreted as the combined effect of tunneling and carrier distribution. Furthermore, possible implications on MOSFET gate currents are briefly discussed.

Comparative Simulation Study of the Different Sources of Statistical Variability in Contemporary Floating-Gate Nonvolatile Memory

For the first time, a comprehensive comparative study of the impact of different sources of statistical variability in nonvolatile memory (NVM) has been carried out using the 3-D numerical simulation of large statistical ensembles and approaches based on the impedance-field method. Results of the threshold voltage variability in a template 32-nm floating-gate NVM subject to random discrete dopants (RDD), line edge roughness, oxide thickness fluctuations, polysilicon granularity, and interface trapped charge (ITC) are presented. The relative impact of each source of statistical variability has been highlighted, with RDD being identified as the dominant source and ITC as the next most dominant source. Based on the simulation of statistical samples of 1000 microscopically different devices, the shape and spread of the statistical distribution associated with each individual and combined sources of variability have been found to significantly be different from a normal distribution, particularly within the tails that may have significant implications for design and yield. Finally, an ensemble of 59 000 devices is used to characterize the combined impact of all sources of variability.

Shallow Trench Isolation Edge Effect on Random Telegraph Signal Noise and Implications for Flash Memory

Random telegraph signal (RTS) noise is of increasing concern for sub-100-nm flash memories. To quantitatively study the shallow trench isolation (STI) edge effect, NMOS devices with and without STI edges in the channel area are designed and analyzed. The significant impact of STI on width scaling is demonstrated and quantified. It is shown that the noise induced by STI edges dominates the RTS noise for smaller device sizes and is caused by an increase in the number of trapping sites at the STI edges. By rounding the STI corner, the number of stress-induced traps can be significantly reduced.

Atomic migration in phase change materials

During normal operation of Phase Change Memory (PCM) cells active materials undergo very high electrical and thermal stresses that cause a motion of the different atoms leading to composition variation which has a fundamental impact on performance and reliability. In order to address this issue we introduce here a comprehensive 3D physical model for mass transport in chalcogenide materials. In addition to the driving force for atom diffusion coming from concentration gradient and electric field, the model also accounts for the effect of temperature gradient and phase segregation. This new diffusion model is coupled with a calibrated electro-thermal-phase change model, thus providing a unified framework for the self-consistent simulation of both the electro-thermal and the phase/material change problems. The model is applied to the study of different types of PCM cells showing good agreement with experiments and demonstrating in particular the fundamental role played by the temperature profile.

Impact of Nonuniform Doping on Random Telegraph Noise in Flash Memory Devices

This paper presents a thorough numerical investigation of the effect of nonuniform doping on random telegraph noise (RTN) in nanoscale Flash memory devices. For a fixed average threshold voltage, the statistical distribution of the RTN fluctuation amplitude is studied with nonconstant doping concentrations in the length, width, or depth direction in the channel, showing that doping increase at the active area corners and retrograde and δ-shape dopings appear as the most promising profiles for RTN suppression. In particular, the improvements offered by retrograde and δ-shape dopings increase the more the high doping regions are pushed far from the channel surface due to a more uniform source-to-drain conduction during read. Finally, the suppression of RTN by engineered doping profiles is correlated with the reduction in cell threshold voltage variability.

Assessment of accuracy limitations of full band Monte Carlo device simulation

In this paper we investigate some of the numerical approximations involved in the development of Full Band Monte Carlo (FBMC) programs for semiconductor (silicon) devices. In particular, we focus on how the accuracy in describing the Full Band silicon structure affects quantities such as the density of states (DoS), scattering rates, velocity field curves, etc.