W. Goes

The Paradigm Shift in Understanding the Bias Temperature Instability: From Reaction–Diffusion to Switching Oxide Traps

One of the most important degradation modes in CMOS technologies, the bias temperature instability (BTI) has been known since the 1960s. Already in early interpretations, charge trapping in the oxide was considered an important aspect of the degradation. In their 1977 paper, Jeppson and Svensson suggested a hydrogen-diffusion controlled mechanism for the creation of interface states. Their reaction-diffusion model subsequently became the dominant explanation of the phenomenon. While Jeppson and Svensson gave a preliminary study of the recovery of the degradation, this issue received only limited attention for many years. In the last decade, however, a large number of detailed recovery studies have been published, showing clearly that the reaction-diffusion mechanism is inconsistent with the data. As a consequence, the research focus shifted back toward charge trapping. Currently available advanced charge-trapping theories based on switching oxide traps are now able to explain the bulk of the experimental data. We give a review of our perspective on some selected developments in this area.

The time dependent defect spectroscopy (TDDS) for the characterization of the bias temperature instability

We introduce a new method to analyze the statistical properties of the defects responsible for the ubiquitous recovery behavior following negative bias temperature stress, which we term time dependent defect spectroscopy (TDDS). The TDDS relies on small-area metal-oxide-semiconductor field effect transistors (MOSFETs) where recovery proceeds in discrete steps. Contrary to techniques for the analysis of random telegraph noise (RTN), which only allow to monitor the defect behavior in a rather narrow window, the TDDS can be used to study the capture and emission times of the defects over an extremely wide range. We demonstrate that the recoverable component of NBTI is due to thermally activated hole capture and emission in individual defects with a very wide distribution of time constants, consistent with nonradiative multiphonon theory previously applied to the analysis of RTN. The defects responsible for this process show a number of peculiar features similar to anomalous RTN previously observed in nMOS transistors. A quantitative model is suggested which can explain the bias as well as the temperature dependence of the characteristic time constants. Furthermore, it is shown how the new model naturally explains the various abnormalities observed.

A two-stage model for negative bias temperature instability

Based on the established properties of the most commonly observed defect in amorphous oxides, the E′ center, we suggest a coupled two-stage model to explain the negative bias temperature instability. We show that a full model that includes the creation of E′ centers from their neutral oxygen vacancy precursors and their ability to be repeatedly charged and discharged prior to total annealing is required to describe the first stage of degradation. In the second stage a positively charged E′ center can trigger the depassivation of Pb centers at the Si/SiO2 interface or KN centers in oxynitrides to create an unpassivated silicon dangling bond. We evaluate the new model to experimental data obtained from three vastly different technologies (thick SiO2, SiON, and HK) and obtain very promising results.

Switching oxide traps as the missing link between negative bias temperature instability and random telegraph noise

Due to the ongoing reduction in device geometries, the statistical properties of a few defects can significantly alter and degrade the electrical behavior of nano-scale devices. These statistical alterations have commonly been studied in the form of random telegraph noise (RTN). Here we show that a switching trap model previously suggested for the recoverable component of the negative bias temperature instability (NBTI) can more accurately describe the bias and temperature dependence of RTN than established models. We demonstrate both theoretically and experimentally, that the recovery following bias temperature stress can be considered the non-equilibrium incarnation of RTN, caused by similar defects. We furthermore demonstrate that the recoverable component is solely constituted by individual and uncorrelated discharging of defects and that no diffusive component exists. Finally it is highlighted that the capture and emission times of these defects are uncorrelated.

Recent advances in understanding the bias temperature instability

Our understanding of the bias temperature instability (BTI) has been plagued by disagreements related to measurement issues. Although even in the early papers on BTI the existence of recovery was acknowledged and discussed [1, 2], for unknown reasons this had little impact on the way we used to think about the phenomenon until recently [3–7]. Even after the re-discovery of recovery [3], it took a few years until it was fully appreciated that any measurement scheme conceived so far considerably interferes with the degradation it is supposed to measure, often accelerating its recovery. Nonetheless, this experimental nuisance has led researchers to think in more detail about the problem and has thus opened the floodgates for fresh ideas [6–11]. Some of these ideas together with the experimental data supporting them are reviewed in the following.

An energy-level perspective of bias temperature instability

Many recent publications discussing the stress and recovery behavior of bias temperature instability (BTI) have suggested the existence of two components contributing to the phenomenon. One of these components was found to be quickly relaxing while the other was only slowly relaxing or even permanent. Curiously, although the most likely suggested mechanisms are the generation of interface states and the capture of holes into pre-existing traps, there is no agreement on which mechanism corresponds to which component and both possibilities have been suggested. Alternatively, other groups have suggested evidence that BTI is dominated by a single mechanism, and used the reaction-diffusion (RD) model to describe the degradation. However, RD theory cannot explain the recovery and related intricacies of the phenomenon. We present a new modeling framework based on the various possible energetic configurations of the system and tentatively assign these levels to the hydrogen binding/transport levels in an amorphous oxide. We investigate the possibility that the often observed recoverable and permanent components are in fact two facets of a single degradation mechanism proceeding as a series of steps. We finally subject the model to various experimental data (DC, AC, duty-factor, negative and positive stress, mixed stresses) which are all well reproduced by the model.

Advanced characterization of oxide traps: The dynamic time-dependent defect spectroscopy

An accurate understanding of oxide traps is essential for a number of reliability issues, including the bias temperature instability, hot carrier degradation, time-dependent dielectric breakdown, random telegraph and 1/f noise. Recent results have demonstrated that hole capture and emission into oxide traps in pMOS transistors are more complicated than the usually assumed Shockley-Read-Hall-like process. In particular, both charging and discharging proceed via a non-radiative multiphonon (NMP) mechanism involving metastable defect states. The existence of these metastable states can be demonstrated by extending the previously introduced time-dependent defect spectroscopy (TDDS) to a more general dynamic case by employing AC stresses and precisely timed discharge pulses during recovery. Application of AC stresses clearly reveals a frequency-dependence of the effective capture time, which confirms the existence of an intermediate metastable state. Application of pulses during recovery, on the other hand, allows extraction of the effective emission time also in depletion as well as in accumulation, thereby clearly revealing a metastable switching state. While all investigated traps show a frequency-dependent capture time constant, suggesting them to be of the same microscopic origin, we find two different kinds of emission behavior, namely fixed positive and switching traps. We finally demonstrate that our multi-state NMP model perfectly captures both cases.

On the volatility of oxide defects: Activation, deactivation, and transformation

Recent studies have clearly shown that oxide defects are more complicated than typically assumed in simple two-state models, which only consider a neutral and a charged state. In particular, oxide defects can be volatile, meaning that they can be deactivated and re-activated at the same site with the same properties. In addition, these defects can transform and change their properties. The details of all these processes are presently unknown and poorly characterized. Here we employ time-dependent defect spectroscopy (TDDS) to more closely study the changes occurring at the defect sites. Our findings suggest that these changes are ubiquitous and must be an essential aspect of our understanding of oxide defects. Using density-functional-theory (DFT) calculations, we propose hydrogen-defect interactions consistent with our observations. Our results suggest that standard defect characterization methods, such as the analysis of random telegraph noise (RTN), will typically only provide a snapshot of the defect landscape which is subject to change anytime during device operation.

Hydrogen-related volatile defects as the possible cause for the recoverable component of NBTI

The recently suggested time-dependent defect spectroscopy (TDDS) has allowed us to study the recoverable component of NBTI at the single-defect level. To go beyond our previous efforts, we have performed a long-term TDDS study covering also the kilo-second time window. We found that even in this extended window NBTI recovery is due to a collection of first-order reactions. In particular, there is no trace of a diffusion-limited process as assumed in the reaction-diffusion model. Most intriguingly, the responsible traps show various degrees of volatility, that is, they can disappear and reappear. Our observations lend strong support to the idea that the recoverable component of NBTI is due to hydrogen-related defects which are active when a hydrogen atom is at the defect site and inactive when not.

Gate-sided hydrogen release as the origin of "permanent" NBTI degradation: From single defects to lifetimes

The negative bias temperature instability (NBTI) in pMOS transistors is typically assumed to consist of a recoverable (R) and a so-called permanent (P) component. While R has been studied in great detail, the investigation of P is much more difficult due to the large time constants involved and the fact that P is almost always obscured by R. As such, it is not really clear how to measure P and whether it will in the end dominate device lifetime. We address these questions by introducing a pragmatic definition of P, which allows us to collect long-term data on both large and nanoscale devices. Our results suggest that (i) P is considerably smaller than R, (ii) that P is dominated by oxide rather than interface traps and therefore (iii) shows a very similar bias dependence as R, and finally (iv) that P is unlikely to dominate device lifetime. We argue that a hydrogen-release mechanism from the gate-side of the oxide, which has been suspected to cause reliability problems for a long time [1-6], is consistent with our data. Based on these results as well as our density-functional-theory (DFT) calculations we suggest a microscopic model to project the results to operating conditions.