Characterization of dynamic NBTI by ultra-fast charge pumping and ultra-fast switching method
Teo, Andy Zhiqiang
Date of Issue2015
School of Electrical and Electronic Engineering
Till very recently, the most well-known model used to explain the NBTI phenomenon is the reaction-diffusion (R-D) model, which describes the evolution of Si/SiO2 interface states (Nit) contributing to NBTI based on a hydrogen transport mechanism. However, there has been much on-going discussion regarding the commonly observed instantaneous threshold voltage (Vth) shift during stress and relaxation phase. Several believe that this instantaneous Vth shift is solely explained by the generation and passivation of Nit at the Si/SiO2 interface based on the framework of the R-D model. On the other hand, some thinks that NBTI has two mechanisms, involving both hole trapping (Not) and Nit generation, which are used to explain the fast and slow Vth shift. In this work, an objective and systematic analysis was carried out to challenge the validity of the R-D model versus the two-mechanism NBTI model through dynamic NBTI stress/recovery testing. Further understanding of the proposed two-mechanism NBTI framework as well as its implication on NBTI will be discussed. Two fast measurement methodologies with the capability to characterize ΔVth and ΔNit within 100 ns measurement time were developed, namely the ultra-fast switching method (UFS) and its variant, ultra-fast switching charge pumping (UFS-CP). The hardware setup and working principle are presented as well. Dynamic NBTI experimental results measured by UFS have revealed a specific fast responsive cyclical |ΔVth| component which can be generated during stress and fully recovered upon relaxation, regardless of the stress and recovery intervals. This behavior, observed over many stress/recovery cycles, is inconsistent with the R-D model, which stipulates that interface state relaxation should decrease progressively with stress/recovery cycling. Further insights through an Arrhenius experiment revealed two distinct degradation mechanisms, evident by a small and large activation energy (~45 meV and ~110 meV). The relatively small energy activation energy of 45 meV suggests a cyclical |ΔVth| component arising from a hole trapping/detrapping mechanism (via tunneling). The stronger thermal activation of 0.11 eV was perceived as a more energy-intensive process involving the breaking of Si-H bonds for creation of Nit. Detailed testing revealed different time dependence for hole trapping and Nit generation, implying that these two mechanisms are not coupled. The hole trapping/detrapping phenomenon, perceived by many to result in the rapid cyclical |ΔVth| evolution in dynamic NBTI, may be ascribed to the well known oxygen vacancy defect (E' centers) and its variants (E'δ centers) in the SiO2 gate dielectric. As evident from the first principal calculation, the shallow energy state of E'δ center is calculated to be 0.5 eV to 1 eV above the SiO2 valance band, which favors the condition for a rapid charge exchange and short tunneling constant in a ultra-thin gate oxide. However, one must also consider the wide spread of the defect energy state and tunneling time constant in an amorphous nature gate oxide. Further evidence from a positive relaxation gate voltage revealed another important class of hole traps, classified as deep level hole trap (DLHT). A DLHT is ascribed to a defect located very close to the interface and as a consequence, their energy states are “pinned” by the Si/SiO2 band discontinuity. Due to the higher (deeper) energy state, a charged DLHT requires a much longer emission time (which could take up to several days) for it to be discharged. However, the recovery of DHLT can be accelerated by a positive gate bias. Through a combination of dynamic NBTI testing at high and low voltages, experimental results show that the number of hole traps that can be activated under a given stress condition is fixed, although this number can vary with the exact stress voltage applied. Even when the applied stress resulted in an increase of stress induced leakage current (SILC) or bulk oxide trap density, the number of hole traps activated under dynamic NBTI testing did not increase but exhibited a decrease instead. Analysis of |ΔVth| measured immediately upon the termination of stress and after a substantial recovery, shows interestingly a faster increase of the latter. As the |ΔVth| measured at the end of stress probes the total positive oxide trapped charge while that measured after relaxation only probes the unrecovered portion of the positive oxide trapped charge, the results show that the decrease in the number of hole traps activated during the dynamic NBTI test is a consequence of a portion of the hole traps being transformed into a more permanent form. This transformation is shown to always correspond to the increase of SILC, i.e. when the transformation is absent, so is the increase in SILC absent, implying that the transformation is responsible for the generation of more permanent bulk traps that are responsible for the observed SILC. The transformation is consistent with the irreversible structural change that may occur at an E' center upon the capture of a hole. Further presumption that the R-D model can still be used to explain the Nit generation in the two-mechanism NBTI model was also ruled out. UFS-CP measurement of dynamic NBTI shows that the amount of ΔIcp induced in a given stress interval is completely removed in the following recovery interval, which disagrees with the notion of a hydrogen transport model that depicts the amount of recovery to decrease progressively with stress/recovery cycling. Instead, it was found that stress induced ΔNit comprises of two distinct characteristics coupled together; (i) one is a consistent fast recoverable Nit component which is spontaneously recovered upon termination of stress (ii) while the other is the permanent Nit component. The fast recoverable Nit component, identified as a parasitic signal which originates from the Not was inevitably captured in the UFS-CP measurement. The other permanent component, identified as the generation of Nit, is shown to be an irreversible process, whereby the generation of Nit is solely dependent on stress period and the generated defects show no apparent recovery even after an extremely long recovery period (up to 50 days). A separation methodology is proposed to separate the parasitic oxide trap signal from as-measured signal in UFS-CP, yielding a power-law time exponent of ~0.27 for ΔNit. Based on the phenomenological observation that NBTI is determined by a cyclic hole trapping/detrapping mechanism and that interface-state generation is relatively permanent, a systematic separation approach is outlined to decouple both components to evaluate the impact on NBTI. In UFS methodology, the time dependence of hole trapping during stress is precisely determined through dynamic NBTI, and then subtracted from the overall degradation of the first cycle to yield the time dependence of Nit generation. Nit generation is shown to exhibit power-law time dependence with an initial exponent of ∼0.5, which subsequently decreases to a steady value of 0.25 after ∼1000 s at the stress condition studied (oxide field ∼10 MV/cm). This evolution is shown to be consistent to that obtained via the charge-pumping method, confirming the underlying principle of the approach. Lastly, the impact of hole trapping at operating condition is examined for the first time. A device was pre-stressed for an extremely long time (up to 4.5 days), whereby further generation of Nit is insignificant. The impact of hole trapping is examined over a broad range of stress voltages and extrapolated to operating bias condition of 1.2V for the first time.
DRNTU::Engineering::Electrical and electronic engineering::Microelectronics