General framework about defect creation at the interface Cite as: J. Appl. Phys. 105, 114513 (2009); https://doi.org/10.1063/1.3133096 Submitted: 07 January 2009 • Accepted: 11 April 2009 • Published Online: 09 June 2009 C. Guerin, V. Huard and A. Bravaix ARTICLES YOU MAY BE INTERESTED IN Hot-carrier degradation caused interface state profile—Simulation versus experiment Journal of Vacuum Science & Technology B 29, 01AB09 (2011); https:// doi.org/10.1116/1.3534021 Dielectric breakdown mechanisms in gate oxides Journal of Applied Physics 98, 121301 (2005); https://doi.org/10.1063/1.2147714 Passivation and depassivation of silicon dangling bonds at the Si/SiO2 interface by atomic hydrogen Applied Physics Letters 63, 1510 (1993); https://doi.org/10.1063/1.110758 J. Appl. Phys. 105, 114513 (2009); https://doi.org/10.1063/1.3133096 © 2009 American Institute of Physics. 105, 114513 三种Si-H键的打破方式 JOURNAL OF APPLIED PHYSICS 105, 114513 共2009兲 General framework about defect creation at the Si/ SiO2 interface 在Si/SiO2系统中研究,有HK时候呢? C. Guerin,1,2,a兲 V. Huard,1 and A. Bravaix2 1 ST Microelectronics, Crolles, 850 rue Jean Monnet, 38926 Crolles, France 2 IM2NP-ISEN, UMR CNRS 6137,Maison des Technologies, place G. Pompidou, 83000 Toulon, France 共Received 7 January 2009; accepted 11 April 2009; published online 9 June 2009兲 This paper presents a theoretical framework about interface state creation rate from Si–H bonds at the Si/ SiO2 interface. It includes three main ways of bond breaking. In the first case, the bond can be broken, thanks to the bond ground state rising with an electrical field. In two other cases, incident carriers will play the main role either if there are very energetic or very numerous but less energetic. This concept allows one to physically model the reliability of metal oxide semiconductor field effect transistors, and particularly negative bias temperature instability permanent part, and channel hot carrier to cold carrier damage. © 2009 American Institute of Physics. 关DOI: 10.1063/1.3133096兴 I. INTRODUCTION The existence of silicon dioxide plays a key role in the everlasting Si-based semiconductor industry. Since the mid1980s, the continuous scaling down of the dielectric thickness reaches dimensions in actual nodes close to atomic level 共a 1.5-nm-thick dielectric is only 5 atom thick兲. In addition, at the Si/ SiO2 interface, both electronic and structural properties are changed. As a result, silicon dangling bond 共SDB兲 defects are intrinsically present resulting in electrically active interface traps 共Nit兲, also called Pb centers.1 In order to minimize the effect of the electrically active states due to SDBtype defects, silicon devices are exposed to hydrogen through thermal annealing. The goal is that the H atoms will passivate these interfacial defects by forming Si–H bonds. Most of reliability degradation modes occurring in the front-end part of the process are thought to be related to the dissociation of the Si–H bonds, such as negative bias temperature instability 共NBTI兲, hot carrier 共HC兲, and timedependent dielectric breakdown 共TDDB兲 generally explained by hydrogen release at Si/ SiO2 interface.2–4 Though all these degradation modes are related to the same initial SDB-type defect, many physical explanations have been proposed to explain the experimentally observed time-dependent hydrogen depassivation, ranging from direct electronic excitation 共DEE兲, vibrational ladder climbing to diffusion-limited mechanisms. We report here a general framework to explain hydrogen depassivation at the Si/ SiO2 interface based on the last published results on Si–H bond local environment. This work will explain several reported results in literature including various degradation modes focusing particularly here on our new HC data obtained in the state-of-the art complementary metal oxide semiconductor 共CMOS兲 technology nodes 共40 nm兲 using medium to ultrathin gate oxides 共Tox = 1.3– 5 nm兲. II. ELECTRONIC AND VIBRATIONAL MODE PROPERTIES OF SI–H BOND Understanding the Si–H bond properties at the atomic level is required to control and fabricate MOS devices with a兲 Electronic mail: chloeguerin@hotmail.com. 0021-8979/2009/105共11兲/114513/12/$25.00 both high performances and reliability. Since some of these properties are seldom experimentally accessed, simulations such as first-principles molecular dynamics method are powerful tools for investigating and understanding atomic-level phenomena. The aim of this paragraph is to propose a short review concerning the different bond states under carrierinduced excitation, including both recent simulated and experimental results. The hydrogen atom transfer is viewed as a potential barrier crossing problem. The crossing of such an EB barrier by thermal activation has been invoked in explaining the motion of individual atom with the tip of a scanning tunneling microscope 共STM兲.5 In the case of STM desorptions, the underlying microscopic mechanism linked to EB reduction has been related to vibrational excitation by inelastic scattering of incident carriers,6 i.e., a vibrational excitation of the substrate-adsorbate bond. A key ingredient in this context is the anharmonic coupling between the adsorbate and the incident carrier via an adsorbate-induced resonance state.7 This resonance is usually modeled by a Lorentzian density of states 共DOS兲 a, centered at the energy a interacting two continua of carrier levels representing the incident carriers and the substrate7–9 共Fig. 1兲. In a MOS transistor case, the incident carriers are channel carriers 共electrons for n-MOS or holes for p-MOS兲, tunneling gate carriers or substrate carriers. When carriers are electrons, this resonance concerns the Si–H 6* state.9,10 When they are holes, it is the 5 state.8 The precise shape of the DOS is not Lorentzian but it has been shown to range from 1.5 to 7 eV.8,9 The energy acquired by carriers is represented by a shift between the Fermi levels of the two continua 共Fig. 1兲. This carrier energy can be transmitted to the bond by inelastic scattering through the resonance. Then the energy is dissipated by bond localized phonon modes, i.e., adsorbate vibrational modes. The two main Si–H bond vibrational modes are known to be stretching and bending modes11 共Fig. 1兲. Stretching mode corresponds to a path where the adsorbate is moved away from silicon toward an interstitial site. Bending vibrational mode corresponds to a path where the adsorbate rotates around Si toward a neighboring bond center site.12,13 Each mode is characterized by a bond breaking energy EB, a vibrational mode energy ប, and a relaxation time e. These three pa- 105, 114513-1 © 2009 American Institute of Physics 114513-2 J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix Substrate Adsorbate Incident carriers Bending H Stretching E εa ρa TABLE II. Si–H bond stretching or bending vibrational mode wave numbers from several references. Availabe vibrational mode lifetimes and transversal/lateral and optical/acoustical 共T/L and O/A兲 phonon wave numbers are also mentioned in bracket. FIG. 1. 共Color online兲 Schematic model of inelastic electron scattering via an adsorbate induced resonance at an energy a with a DOS a near the Fermi level of the substrate. The energy transfer from carriers through resonance excites bending or stretching bond vibrational modes. After Refs. 7 and 8. rameters allow a full determination of the adsorbate characterization since the bond breaking is a trade-off between the energy brought by bond excitation and the relaxation process linked to e. A. Bond breaking energy Several teams studied theoretical Si–H bond dissociation energy, thanks to calculations based on density-functional theory. The first step of their work is to model atom cells including a dangling bond passivated by an hydrogen. Then, the H atom is moved away from the Si and the required additional energy is monitored. Results from several references have been reported in Table I for stretching and bending pathways.11–14 It is worth noticing that some of the calculations are done for a Si–H bond in silicon bulk, whereas others concern Si–H at the Si/ SiO2 interface. However, all the data related to one mode are coherent. In consequence, we set the stretching path energy at 2.5 eV and the bending path energy at 1.5 eV. Reference Site Stretching 共eV兲 Bending 共eV兲 12 13 Bulk Si Bulk Si 2.5 ⬃3.6 14 11 共111兲Si/ SiO2 interface 共111兲Si/ SiO2 interface 3.3 ⬍3 共SiO2兲 2.2 共Si兲 1.5 1.75 共path II兲 1.9 共path III兲 1.9 1.5 Phonon 共cm−1兲 2100 共0.8 ns兲 1838–2145 共1.6– 295 ps兲 BC 1998 共7.8 ps兲 HV¯VH 2072 共295 ps兲 650 640 共8 ps兲 TO 463 52 21 15, 17, and 19 Simulation Simulation Silicium 20 Silicium 817 共12.2 ps, 5 K兲 53 Si/ SiO2 592 54 Si/ SiO2 2083 TA⬃ 150 LA⬃ 340 LA 357 TA 150 LO 517 TO 460 TO 468 LO 435 626 B. Vibrational mode energy The presence of light atoms such as hydrogen in a crystalline solid gives rise to localized vibrational modes with wave numbers 共i.e., energy兲 above the phonon bands of the solid.15 The infrared absorption spectroscopy 共IRAS兲 allows measuring the wave number of a Si–H bond excitation mode by detecting absorption line for each mode.15 Wave number values from several studies are regrouped in Table II. The reported values for a given mode are rather identical in spite of various result sources 共simulation or experiment measurement done on different H-related defects兲. As for the energy modes, we set the stretching wave number at 2083 cm−1 and the bending wave number at 610 cm−1. Knowing these wave number values, the energy of the vibrational mode ប can be deduced from the relation 1 eV⬅ 8.065 5 ⫻ 103 cm−1.16 It leads to បs = 0.25 eV for stretching mode and បb = 0.075 eV for the bending mode. C. Bond vibrational mode lifetime A second term can be extracted from an absorption line: the lifetime of the associated vibrational mode e. In fact, e is related to the full width at half maximum ⌫ by ⌫= TABLE I. Si–H bond dissociation energy following stretching or bending mode from several references. Si–H bonds are placed either in bulk Si or at the Si/ SiO2 interface. Bending 共cm−1兲 Source εF+ energy εF Stretching 共cm−1兲 Reference 1 . 2ce 共1兲 Nevertheless, due to the presence of inhomogeneous broadening, ⌫ value can be underestimated. A more precise value of e can be extracted after IRAS by the transient bleaching spectroscopy technique. In the following it consists of the decrease in the transmission coefficient from the wave number line measured by IRAS.15 The lifetime values have been also noted in Table II when given by the reference. Measurements have been done between 5 and 10 K. Contrary to the wave number, the vibrational mode lifetime is extremely dependent on the defect structure. Since previous work17 argued that the Si–H bond of the divacancy binding 114513-3 J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix two H atoms 共HV · VH共110兲兲 resembles that of the H-passivated Pb centers at the Si/ SiO2 interface, we set e at 295 ps for the stretching mode and at 10 ps for the bending mode. Finally, we need to extend the knowledge on e from cryogenic toward high temperatures. A high wave number excited mode decays into a path of vibrational modes of the lattice, i.e., phonons, whose wave numbers are smaller than the wave number of the excited mode. That is why we also reported the phonon wave numbers in Table II for transversal, longitudinal, acoustic, and optical 共T, L, A, O兲 modes. The decay of a vibrational mode by a decay channel constituted by several phonons is governed by the following equation:18 we = = 1 = 2 兺 兩Gi兩2 f i e i with fi exp共ប/kBT兲 − 1 Ni 兿 j=1关exp共ប j/kBT兲 − 1兴 , 共2兲 where the total energy decay rate we is the sum of the rates of all the decay channels. Gi is the coupling strength of the channel, f i is its temperature dependence, is the frequency of the bond vibrational mode, j is the phonon frequency, and Ni is the total number of phonon required for decay. Furthermore, we know that the wave number of the bond vibrational mode is equal to the sum of the phonon wave numbers and that channels with low Ni are likely to dominate the decay.15 The decay channel of the stretching mode has been modeled by Lüpke19 with five accepting modes with frequency 兵521, 521, 343, 343, 343其. Concerning the bending mode, the same study has only been carried out by Sun et al.20 but only on the 817 cm−1 mode, whereas experimental measurements at the Si/ SiO2 interface show that the Si–H bending mode is around 610 cm−1. However, the author also showed that the vibrational lifetimes follow a universal frequency-gap law, i.e., that the decay time increases exponentially with increasing decay order 共Ni兲. From this law, we suggest that the bending mode only needs a two orders of magnitude decay, it can be related to the sum of transversal optical 共TO兲 共460 cm−1兲 and transversal acoustic 共TA兲 共150 cm−1兲 phonon modes. In this paragraph, we pointed out values we chose for the three characteristic parameters of the Si–H bond for the stretching and bending modes. Our choices have been supported by a wide bibliographic study concerning both simulations and experimental results. We will show that these values are useful for our general Si–H bond breaking model, in agreement with the experimental results. III. A GENERAL MODEL FOR SI–H BOND BREAKING When the SDB is passivated, the system is in the ground state. The atom transfer corresponds to the emission of the H atom to a mobile transport state. This transfer occurs in two steps: first, the barrier EB needs to be reduced, and then the H atom can overcome this barrier to the neighboring transport state via thermal emission.21 We already mentioned in the Thermal emission EB Carrier-induced excitation Transport state Ground state Thermal emission Oxide field E! EB Transport state Ground state FIG. 2. 共Color online兲 Schematic of the energy levels involved during a bond breaking. If the barrier EB is reduced either artificially by excitation of the bonding electron 共top兲 or by the oxide field 共bottom兲, H can be emitted to a mobile transport state. previous paragraph a way of reducing EB artificially, thanks to carrier-induced excitation. This excitation allows carriers to move on a higher energetic level and then to be easily emitted. By the second way, EB is really reduced due to the application of an electric field which has a direct effect on the ground state.22,23 These two scenarios are schematically sketched in Fig. 2 and detailed in the two next subsections since they offer basic explanations for several MOS field effect transistor 共MOSFET兲 degradation modes. Part A is devoted to the electric field EB reduction, whereas part B deals with carrier-induced excitation. A. NBTI: Oxide field driven NIT creation Our recently proposed NBTI model24 considers two components in the degradation: a permanent part due to 共fast兲 interface traps and a recoverable one related to 共slow兲 or border traps. It is based on reaction-limited 共RL兲 model and hole trapping mechanism into the gate insulator. Data have demonstrated for the first time without ambiguity that Nit does not get passivated 共even after 1 week of recovery兲 and that the induced Nit is the main contributor to the permanent part. Besides, data also provide additional proof that NBTI transient effects are uncorrelated to the Nit creation and that the recoverable part is linked to oxide hole traps introduced by process steps as nitridation but not to generated traps during NBTI stress.24 As the topic of this study is Nit creation, we will only focus on the permanent part. The plot of Nit variation under NBTI stress as a function of stress time presents several time dependences.25–27 The first part, which follows a linear time dynamic, is related to RL mechanism. Nit creation is then expressed by dNit = kF共N0 − Nit兲, dt 共3兲 where kF is the forward reaction rate and N0 is the initial Si–H bond number. Considering short stress time, Nit is negligible compared to N0, then Nit is linearly generated with stress time. The slope is equal to kFN0. Knowing N0, it is possible to experimentally determine the forward reaction rate kF. It was observed on various oxide interface qualities 114513-4 J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix Distribution of dissociation energies g(E B) 0.08 H 0.06 Eox Oxide Si-H bond length d=1.48Å Si Bulk 0.04 Electronegativity: Si=1.8, H=2.1 Local field Eloc=(1+χ/3)Eox EBm=EBm0-d.Z.q.cos²θ 0.02 Eox1 Eox2<Eox1 Z.q=0.2q 0 0.6 0.8 1 *Eloc FIG. 5. Bending mode related dissociation path allows calculating the Si–H dipole moment. 1.2 Dissociation Energy EB (eV) FIG. 3. Fermi-derivative distribution of dissociation energies for two oxide fields. B. Channel hot carrier and channel cold carrier: Energy driven NIT creation 1. Modeling 共i.e., number of available bonds兲 that the forward reaction rate kF is dispersed over several decades; it is expressed as24 冉 kF = kF0 exp 冊 − EB共Eox兲 . kT 双井模型 共4兲 This experimental observation is in contradiction with the reaction-diffusion 共RD兲 model which considers, as main assumption, that kF is constant.28 This kF dispersion related to EB dispersion is modeled according to a Fermi-derivative function, where is the distribution dispersion 关Eq. 共5兲兴 共Fig. 3兲, 1 g共EB, 兲 = 冉 exp EBm − EB 冋 冉 冊 EBm − EB 1 + exp 冊册 2 共5兲 . The median dissociation energy EBm for a given oxide field is shown to follow a linear relationship with oxide field 共Fig. 4兲. The median dissociation energy in the absence of oxide field is thus estimated to be equal to 1.5 eV, which is in close agreement with the estimated value of the Si–H bending pathway. Our physical model takes into account the existence of an electrostatic dipole Zq in between Si and H. It assumes that the Si–H bond is bended by the vertical oxide field due to the dipole, which, in turn, reduces EB 共Ref. 11兲 共Fig. 5兲. The slope of the oxide field linear dependence is directly related to the Si–H bond dipole moment Zq. The dipole moment value is estimated about Zq = 0.2q with our set of experiments, in close agreement with theoretical values found in literature.29 a): DEE σ* Pexc Si EBmo = 1.5 eV 1.6 Pcou b): SVE * σ Pexc Si Si H Pemi N H ћω EB Si H σ EES 1.8 H σ !a): EES 2 Mean dissociation energy EBm (eV) In our physical framework, the Nit creation related to carrier-induced excitation can be divided into two main groups depending on the energy of the incident carrier: either it has enough energy to excite the adsorbate resonance state or it does not. Each main group can also be divided in two other subgroups; it leads to have four Si–H bond breaking ways 共Fig. 6兲. Let first introduce group 1, linked to high energetic carriers. With sufficient energy, incident carriers might excite 共i.e., have an excitation probability Pexc兲 the resonance. In spite of a fast excitation decay, a fraction of the atoms can survive on the excited resonance state long enough to induce a direct bond breaking. This first desorption pathway is called DEE7,30 关case 1共a兲兴. The Si–H bond breaking rate Rb is defined for DEE by the product of the carrier energy distribution function 共CEDF兲 共i.e., the carrier number at a given energy兲 by the excitation probability of these carriers. The CEDF integral over the energy is macroscopically linked to the current I, i.e., in relation to the channel drain current Ids for HC mode, gate current Igs for TDDB, tip current for STM, giving Pexc Si 1.4 Pcou σ* Pemi H EB Si H σ 1.2 p = -0.056 1 !b): MVE Pemi σ* 0.8 0.4 0.2 σ EBm = EB m0 - p.Eox 0 0 5 Oxide field (MV/cm) Pexc Si 电场越高,势垒越低,Kf越大 0.6 10 FIG. 4. Linear dependence of median dissociation energies with oxide field. H ћω EB Si H I: Id if HC, Ig if TDDB FIG. 6. 共Color online兲 Schematics of the four Si–H bond breaking modes related to carrier-induced excitation: DEE, SVE, assisted or not by EES, and MVE. Pexc, Pcou, and Pemi are the probabilities of resonance excitation of the Nth level occupation induced by coupling and of thermal emission from the Nth level, respectively. 114513-5 J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix RbDEE:Bond breaking rate for DEE RbDEE = IPexc . I: Id if HC, Ig if TDDB Pexec:Excitation Probability 共6兲 If the resonance decays from the resonance excited state before the bond breaking occurs, the energy stored in the resonance state is transferred via an anharmonic coupling to low-energy modes associated with a particular motion of the adsorbate, in competition with fast vibrational relaxation by bulk phonons emission. The anharmonic coupling is described by a coupling probability Pcou between the resonance state and the localized vibrational modes. Assuming that only the highest vibrational level, named N, yields to hydrogen desorption, the probability Pcou relates to the occupation probability of the last level N induced by the decay of the resonance state. This case is related to a multistep vibrational excitation in a single jump 共or single carrier兲 also called single vibrational excitation 共SVE兲 共Ref. 31兲 关case 1共b兲兴. In the case of SVE, the variation in the occupation density of the Nth level 共nN兲 with stress time t is defined as a competition between the probability of the ground state 共n0兲 excitation by anharmonic coupling and the decay to the N − 1 vibrational level. Since the last vibrational level is part of the continuum of scattering states just above the barrier, Si–H bond is further described as Si*H* and the nN variation with stress time t is given by Pemi ⬇ t = t exp共− Eemi/kBT兲. Finally, the Nit creation relates to the probability to emit H to transport states from the latest excited level H*, which can be written as ⌬Nit = 关H*兴t. 共12兲 Combining Eqs. 共8兲 and 共12兲 gives the total number of intern0是整个的Nit的数量 face traps ⌬Nit, ⌬Nit = 冑 n0IPexcPcou 0.5 t . we 共7兲 注意这时候n=0.5 共13兲 At this point, it is important to notice that the time dynamics is expected to be a power law with an exponent of 0.5. This equation stands as long as generated Nit is negligible compared to the whole population of Si–H bond n0. At longer stress times and/or larger degradation, saturation is expected with progressive decrease in n0. From Eqs. 共13兲, the Si–H bond breaking rate RbSVE can be directly related to the device lifetime , defined as the time to reach a given degradation criterion, i.e., a given amount of interface traps ⌬Nit,crit with ⌬Nit,crit = 共RbSVE兲0.5 with dnN = IPexcPcoun0 − we关Si*兴关H*兴, dt 共11兲 n0IPexcPcou . RbSVE ⬀ we 共14兲 Lamba是一个常数,见Eqn(9) 共15兲 我们相当于把Pexec*Pcou 用Vg和Vds来表述了 where we defined in part II as the decay rate between two vibrational levels. As the typical time range of interface traps creation is relatively long 共seconds or greater兲 compared to the bond relaxation time 共picoseconds兲, we consider that the ensemble of bonds reaches a steady state distribution very quickly. In this case dnN IPexcPcoun0 = 0 → 关Si*兴 = ⌬Nit = . dt we关H*兴 共8兲 In the last level, the bond is almost dissociated but it can still easily be reformed by excitation decay. The H atom will be dissociated from the Si only if it is transmitted to transport states. The probability of H emission Pemi from the Nth level exists, while the excitation on the Nth level is sustained. The excited state can emit the H with a probability per unit time, where corresponds to a thermal activated emission of H over a small barrier 共Eemi兲 from the excited state, with an attempt frequency ,21 = exp共− Eemi/kBT兲. 共9兲 Then, Pemi = 1 − exp共− t兲. 共10兲 Here, Nit creation is experimentally observed to be a monotonous phenomenon spreading over large time scale. That is why we consider that stress time t is much shorter than the emission time −1, such that t Ⰶ 1, which allows to rewrite Eqs. 共9兲 and 共10兲 as After defining Rb for two high energetic carrier cases, Si–H breaking rate with lower energetic carrier is detailed in group 2. Several authors32–34 applying Monte Carlo to n-MOSFETs have predicted that electrons heated by electron-electron scattering 共EES兲 should induce a second knee in the electron CEDF. This EES phenomenon dominates the high CEDF energy tail and extend it beyond qVds 共with Vds as the drain-source voltage drop兲 until 2qVds. In fact, this interaction results in an exchange of energy between two electrons, thus promoting one of them toward higher energy 共encircled electron in Fig. 6兲. This process may allow this promoted electron to have enough energy to excite the resonance whereas it was not possible without EES. Finally, this case 2共a兲 is equivalent to case 1共b兲 except for the fact that the CEDF is no longer proportional to I but I2,35 giving RbEES ⬀ n0I2 PexcPcou . we 共16兲 The last case 关group 2共b兲兴 is dominant when the carrier energy is not high enough to excite the resonance even after EES. The vibrational modes can be directly excited by multiple incident carriers with low energy. Each carrier contributes to the multiple-step vibrational ladder climbing until reaching EB where each ladder level is separated to ប. This phenomenon is called multiple vibrational excitation 共MVE兲. It is a competing process since it requires that the excited vibrational state lifetime e is not too short compared to the average time between subsequent carrier scattering events. As a consequence, the MVE degradation is supposedly strongly correlated to current I 共i.e. the number of electrons 114513-6 J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix “hitting” the bond per second兲. H desorption from H passivated Si surfaces STM has been related to MVE due to a strong dependence on current and voltage.10,30 A model that has been proven successful in explaining the STM data is the truncated harmonic oscillator model.10,36–38 It defines the excitation rate Rexc as the excitation to the highest bound vibrational state N. Considering that each vibrational state has the same energy, N is defined as N= EB . ប 共17兲 dn0 = P dn 1 − P un 0 , dt 共18兲 dnn = Pd共nn+1 − nn兲 − Pu共nn − nn−1兲. dt 共19兲 If we consider a steady state distribution, 冉 冊 dnn Pu nn = 0 → Pn = = dt n0 Pd n . 共20兲 For the last N level, N + 1 level does not exist and Eq. 共19兲 gives dnN = Pu PN−1 − Pd关Si*兴关H*兴 = PunN−1 − Pd⌬Nit关H*兴 dt 共21兲 and from Eqs. 共12兲, 共20兲, and 共21兲, the total Nit creation is defined as 冑 冉 冊 n 0 Pu Pd N t0.5 . 共22兲 From Eq. 共22兲, the Si–H bond breaking rate RbMVE can be directly related to the device lifetime , defined as the time to reach a given degradation criterion ⌬Nit,crit = 共RbMVE兲0.5 with RbMVE ⬀ n0 冉 冊 Pu Pd 共23兲 N . 共24兲 There are two terms in each Pu and Pd probabilities:39 one term is purely dependent on the excitation/decay of vibrational modes induced by the lattice 共described by the decay rate we兲 and the other one is related to the stimulation and emission of vibrational modes by incoming electrons, which could either excite 共in Pu兲 or de-excite 共in Pd兲 a phonon mode 共supposed with a same rate兲 which is equal to SMVE共Ids / q兲, giving Pu = we exp共− ប/kBT兲 + SMVE共Ids/q兲, 共25兲 共26兲 The SMVE function has been named the “inelastic tunneling probability” in the case of an STM study,36 i.e., the scattering rate of the incident carrier with a vibrational mode. Since a vibrational mode is similar to a localized phonon mode, a linear electron-phonon coupling can be used to describe SMVE.8 This electron-phonon interaction is available in literature since widely used by Monte Carlo simulations.32,40 Finally, the breaking rate RbMVE can thus be defined as RbMVE Here Pn is the probability of occupation of the nth level, whereas Pu and Pd are the probabilities up and down 关up for the excitation from the nth to the 共n + 1兲th level or down for the decay from the nth to the 共n − 1兲th level兴. The variation in occupation density nn of each level is defined as Eq. 共18兲 for the first level and Eq. 共19兲 for level 2 to level N, ⌬Nit = Pd = we + SMVE共Ids/q兲. = n 0 冤 SMVE 冉 冊 冉 冊 − ប Ids + we exp q k BT Ids SMVE + we q 冉 冊 冥 EB/ប 冉 冊 exp − Eemi . k BT 共27兲 It is worth noticing that at low energy 共 ⬍ 4 eV兲, it has been calculated that the main contribution to RbMVE is due to an event where one electron gives only one energetic quanta ប to the bond, in comparison with events where several ប could be given at higher energies.8,9 In this paragraph, we have proposed a general theoretical framework considering the various pathways to excite a Si–H bond by carriers. We have derived defects generation rate for each configuration which are all related to intrinsic properties of the bonds. In parallel, a thorough bibliographic study has been led to reach an updated status on the values of these intrinsic properties based on both latest simulations and experimental results. 2. Experimental validation The aim of this paragraph is to validate our model of Nit creation after carrier-induced excitation on experimental results. The defect generation rate has already been shown to be independent on the way the carriers acquire energy, by combining data from channel hot electron 共CHE兲, Fowler– Nordheim tunneling, direct tunneling, and substrate hot electron.41 Since hot carrier degradation represents a severe challenge to control in advanced CMOS technologies and allows in a more flexible way to change independently carrier number or energy, we have made the choice to use hot carrier stress experiments to investigate the validity of our energy-driven theory. The Nit creation rate Rb is obtained by measuring the MOS device lifetime 共兲 for fixed parameter degradation and by taking it inverse 共1 / 兲. The criterion is 10% of reverse saturated drain current reduction 共⌬Idsat / Idsato Rev兲 throughout the paper or 1 k⍀ m of absolute change in inverse transconductance 共W / gmmax-lin兲 when specified. Because high gate voltage 共Vgs 艌 Vds兲 can also induce electron trapping in medium to thick gate insulators,42,43 we focus on ultrathin 共1.7 nm兲 to thin gate oxides 共3 nm兲 with a waiting time before measurements 共tw = 60 s兲 in order to stay mostly sensitive to 共fast兲 interface traps.44 Moreover, low noise frequency measurements do not evidence 共border兲 oxide traps creation in our samples tested under the same HC stress conditions, which points out that ⌬Nit represents the main con- J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix Lifetime τ (a.u.) 4 1.0E+04 10 8 1.E+08 10 0.9V 1V 1.1V 1.21V 1.4V 1.6V 1.8V 2V 1/(τ x Ids), 1/(τ x Ids²) (a.u) Vds = 6 1.0E+06 10 2 1.0E+02 10 L=0.04µm 0.8V<Vgs <2.8V 0 1.0E+00 10 0 1+ 冑 2共Vg − Vtat兲 1+ mEcL , 3 4 5 6 Edom (eV) 共28兲 where Vo is a halo-based potential, Ec is the critical field for velocity saturation, and m is a coefficient related to the body effect. When Vgs increases at a fixed Vds, Ids and Vdsat increase which leads to a VEFF decrease. That is why Fig. 7 can be roughly divided in two parts depending on the energy impact: the high and low energy ranges. a. Channel hot carrier (CHC) modes (SVE and EES) and STM data. Concerning the high energy range where the reso- nance can be excited by a single carrier 共SVE and EES兲, the ratios Rb / Ids and Rb / Ids2 are related to the product PexcPcou according to Eqs. 共15兲 and 共16兲 RbSVE ⬀ PexcPcou Ids Thin oxide, SVE mode Thin oxide, EES mode Thick oxide, SVE mode Thick oxide, EES mode Nit generation rate Sit 2 with Vdat 2共Vg − Vtat兲/m 2 10 1.E+02 0 tribution in the damage mechanism. By studying the HC degradation in our 65 nm node devices covering, a wide Vgs stress range per Vds stress, we have evidenced three regimes45 similar to a previous team.35 However, we showed that the third part of their EES-based model cannot explain our experimental dataset.45 At this time, we proposed MVE phenomenon as another explanation than EES for the device lifetime decrease at high Vgs. In this paper, we report data collected on 45 nm node device using a wider range of Vgs and Vds stressing voltages than in previous studies.35,45 It allows reaching higher Ids values 共Fig. 7兲. Figure 7 shows that the Vds influence 共i.e., carrier energy兲 is important a low Ids, whereas it is much reduced at higher Ids. This behavior change can also be explained in terms of energy variation. The defect generation rate Rb is dominant at a given energy Edom which can be determined by considering an effective potential VEFF from the drain to channel pinch-off point,46 = 4 1.E+04 10 0.002 FIG. 7. 共Color online兲 Evidence of two energy ranges: Vds dependence is strong at moderate Ids 共SVE and EES兲, whereas it is reduced at higher Ids 共MVE兲 共n-MOS, Tox = 1.7 nm兲. The degradation criterion is 10% of 共⌬Idsat / Idsato Rev兲. VEFF = V0 + Vds − Vdat 6 1.E+06 10 10 1.E+00 0.001 Ids/W(A/µm) 共29兲 FIG. 8. SVE and EES modes: the whole dataset 共for both thin 共1.7 nm兲 and thick 共5 nm兲 oxides兲 follows a single generation rate function Sit. Equation 共32兲 is adjusted to experimental data 共line兲. This product is rebuilt experimentally by plotting 1 / 共Ids兲 and 1 / 共Ids2兲 versus their respective Edom, considering that Edom is close to the energy of the knee 共i.e., Edom = qVEFF兲 for SVE, whereas it is around 2qVEFF for EES.33 Our experimental dataset shown in Fig. 8 has been adjusted by a single empirical law similar to the one used to describe the impact ionization,40,45 PexcPcou ⬀ B共E − EB兲 pit . 共31兲 Here, Pit is around 11 and EB = 1.5 eV. It is worth noticing that EB is the same than that found with NBTI data, and if its value is modified with the same dispersion than the one obtained with NBTI results, it is still possible to fit the data as well. This desorption yield equation, which is proportional to the Nit number generated per carrier, allows describing STM data as well.10,30,38 The STM yield shows a threshold at a sample bias around 6 V 共Fig. 9兲. This has been explained as the threshold between DEE at high voltage and MVE at lower voltage because of the high drain and voltage dependence of the lowest voltage part. Figure 9 shows that the yield dependence in the low voltage range is perfectly modeled by our PexcPcou function 关Eq. 共31兲兴 using the same power law and Nit threshold energy as for a HC degraded transistor 共11 and 1.5 eV, respectively兲. The first conclusion is that this voltage and current dependent part can be explained by SVE and/or EES. Thus, MVE is not reached on this figure because the energy is too high even at 4 V and the threshold happens between DEE and SVE. The second point concerns the DEE part. According to Eq. 共6兲, the yield Desorption yield (atoms/electron) 114513-7 -5 1.E-05 10 -6 1.E-06 10 -7 1.E-07 10 -8 1.E-08 10 -9 1.E-09 10 STM data Sit -10 1.E-10 10 and 3 4 5 6 7 8 9 10 Sample voltage (V) RbEES ⬀ PexcPcou . Ids2 共30兲 FIG. 9. The low voltage part of STM desorption yield for hydrogen at 300 K is model by our Sit function. STM data are from Ref. 38. 114513-8 J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix 8 1.E+08 10 6 1.E+06 10 Vds =1.6V -40°C 0.8V<Vgs <2.8V 25°C τ (a.u.) 1.E+06 10 4 1.E+04 10 Lifetime Lifetime τ (a.u.) 6 2 1.E+02 10 0 1.E+04 10 T 2 1.E+02 10 W/L=10/0.04 Vds =1.6V W/L=10/0.06 W/L=10/0.05 W/L=10/0.045 W/L=10/0.04 W/L=10/0.035 0 1.E+00 10 125°C 4 0.6V<Vgs <2.4V 0 10 1.E+00 0.001 Ids/W (A/µm) 0 0.002 0.001 Ids/W (A/µm) 0.002 FIG. 10. 共Color online兲 Evidence of a pure single Ids dependence in MVE mode independent on Vgs and L in contrast with SVE and EES mode behavior 共n-MOS, Tox = 1.7 nm兲. FIG. 12. 共Color online兲 Device lifetime decrease with increased temperature in MVE mode 共n-MOS, Tox = 1.7 nm兲. RDEE / I is proportional to Pexc which means that Pexc is independent of the carrier energy 共Fig. 9兲. Finally, the energy dependence of PexcPcou is only due to Pcou. To draw a parallel with SMVE, this occupation probability of the last level N after coupling 共Pcou兲 can also be defined as Sit, the scattering rate of the incident carrier directly with the last level to be able to create a Nit, The L effect on Ibs / Ids is still visible even at high Ids. It means that energy impact does not disappear totally at high Ids and that degradation increase should be due to a different mechanism. Figure 12 also points out another nonordinary behavior: the high current degradation is activated with an increased temperature. It is not the case for classical CHC mode as seen on thicker 32 Å oxide in Fig. 13: at low Ids, HC degradation is reduced when temperature increases, whereas it is the contrary at high Ids. T effects and high current mode distinction 共Fig. 14兲 are also similar on p-MOS excepted that CCC mode is dominant at smaller Ids than in n-MOS, as holes gain less energy than electrons. We checked that p-MOS data de not interfere with NBTI degradation since they have been collected at 25 ° C and for a maximum 兩Vgs兩 ⬍ 兩Vds兩 with a waiting time before measurements 共tw = 60 s兲. Therefore we are cautious with respect to the possible influence of NBTI damage which is known to be relative to a recoverable part attributed to the balance between hole trapping 共E⬘ center兲 and discharging from the tunneling distance of the substrate 共and gate兲.24,26,44 To go more deeply into CCC mode modeling, we collected data with Vgs much larger than Vds 共Vgs 艌 Vds + 1 V兲. The Idlin degradation is extracted after 20 000 s of stress instead of the lifetime extraction for a given criterion 共as 10%兲 which could be approximate for small Vds where the degradation is smaller than the criterion. In terms of bond breaking rate equation 关Eq. 共27兲兴, the beginning of the MVE range corresponds to Sit共Ids / e兲 ⬇ we exp共−ប / kBT兲. All parameters Sit ⬀ B共E − EB兲 pit . 共32兲 This Sit function is valid when one single carrier interacts with a bond; it merges both SVE and EES mode data, generally named CHC modes. b. CCC mode (MVE). In the highest Ids range in Fig. 7, carrier energy is typically even lower. Nevertheless, the degradation is worse than in higher energy range. In analogy with CHC mode, this mode is named channel cold carriers 共CCC兲 mode. Experimental HC lifetimes acquired for one Vds but several channel lengths 共L兲 show large channel length dependence at low Ids, whereas this dependence is almost gone at higher Ids 共Fig. 10兲. To check if it is not due to a specific energy behavior, Ibs / Ids is followed for the same stress conditions than in Fig. 10 共Fig. 11兲. The ratio Ibs / Ids gives an energy indicator since we have40,46 Ibs ⬇ Sii共qVEFF兲 ⬀ 共E − ii兲 pii , Ids 共33兲 where Sii is the scattering rate for impact ionization, ii is the impact ionization energy threshold 共 ⬇ 1.1 eV兲, and pii ⬇ 4.5. -5 1.E-05 10 W/L=10/0.16; 25°C W/L=10/0.12; 25°C W/L=10/0.16; 125°C W/L=10/0.12; 125°C Vds =1.6V 3 1.E+03 10 Lifetime τ (a.u.) 0.8V<Vgs <2.8V Ibs/Ids 4 1.E+04 10 T 2 1.E+02 10 -6 10 1.E-06 1 1.E+01 10 W/L=10/0.06 W/L=10/0.05 W/L=10/0.045 W/L=10/0.04 W/L=10/0.035 -7 1.E-07 10 0 V ds =3V 1.E+00 10 0 0.001 Ids/W (A/µm) 0.002 FIG. 11. 共Color online兲 Impact ionization ratio for the same conditions as in Fig. 10 showing that L effect on Ibs / Ids persists at high Ids 共n-MOS, Tox = 1.7 nm兲. T 0.5V<V gs <3.5V 0 0.0005 0.001 Ids/W (A/µm) 0.0015 FIG. 13. 共Color online兲 Classical HC thermal activation in n-MOS Tox = 3.2 nm for high energetic carrier related modes 共SVE and EES兲 but same thermal activation than found in MVE mode in n-MOS, 1.7 nm 共Fig. 12兲 共lines are guides for the eyes兲. 114513-9 J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix Bond breaking rate R bMVE (s ) pMOS, 25°C 11 1.E+11 10 -1 Vds = 7 1.E+07 Lifetime τ (a.u.) 10 -1.21V -1.4V -1.6V -1.8V -2V 5 1.E+05 10 3 1.E+03 10 W/L=10/0.04 0.6V<|Vgs |<2V 1 1.E+01 10 0 6 1.E+06 10 1 1.E+01 10 -4 1.E-04 10 -9 1.E-09 10 experimental data bending model stretching model -14 1.E-14 10 0.0005 -7 1.E-07 10 0.001 -5 1.E-05 10 |Ids/W| (A/µm) -1 1.E-01 10 共34兲 When we substitute it in Eq. 共27兲, the model is in good agreement with the data 共Fig. 17兲. In this figure, we also added the same data get at two others temperatures, showing an increase in the bond breaking rate RB with a temperature increase. It allows determining the thermal activation barrier Eemi as 0.26 eV since it is the only unknown parameter. Even if we cannot claim rather H is emitted toward Si or SiO2, this value is in the same range than previous results on H diffusion 共0.2– 0.56 eV兲.47–50 Finally, a simplified equation of RbMVE 关Eq. 共27兲兴 is rewritten including the temperature effect as the term we exp共−ប / kBT兲 is neglected compared to SMVE共Ids / e兲 while this last term is negligible compared to we finally leading to TABLE III. Chosen parameters for Si–H bond at the Si/ SiO2 interface from Tables I and II. Parameters Stretching Bending EB 共eV兲 q 共eV兲 we 共 = 1 / e兲 共ps−1兲 Phonon decay path 2.5 0.25 共or 2072 cm−1兲 1 / 295 共5 K兲 2L0 + 3LA 共2 ⫻ 521+ 3 ⫻ 343 cm−1兲 1.5 0.075 共or 610 cm−1兲 1 / 10 共5 K兲 TO+ TA 共460+ 150 cm−1兲 1 ⬀ RbMVE MVE 冋 冉 冊册 ⬀ 共Vds − ប兲0.5 Ids W 冉 冊 EB/ប exp − Eemi . k BT 共35兲 That can be simplified as ប = 75 meVⰆ qVds giving for a given temperature 1 MVE 冋 冉 冊册 0.5 ⬀ RbMVE ⬀ Vds Ids W EB/ប 共36兲 . The validity of this relationship is presented 共Fig. 18兲 where experimental data obtained from device lifetime extraction under MVE mode 共MVE兲 are plotted as a function of the drain current 共Ids / W兲 showing a weak Vds dependence 共left兲, while this effect disappears when one plots the product EB/2ប MVEVds on the ordinate scale 共right兲. The experimental data give an exponent EB / ប = 17.5 corresponding to EB = 1.4 eV and ប = 80 meV in good agreement with the theoretical values obtained in the range of 1.4 eV⬍ EB ⬍ 1.6 eV and 75 meV⬍ ប ⬍ 80 meV, i.e., with exponents ranging between 21 and 17.5. c. Complete Hot Carriers to Cold Carriers modeling. A simple equation can also be associated to the two other HC modes. Because Sit function has been related to 共Ibs / Ids兲m,35,46 Eqs. 共29兲 and 共30兲 give 1 SVE ⬀ Ids -1 -1 in Eq. 共27兲 have been defined for both bending and stretching modes in the first section and summed up in Table III. Figure 15 shows that experimental data are in agreement with bending mode whereas stretching mode cannot relevantly explain our results. In this figure, the electron-phonon scattering rate SbMVE has been set at 10−5, which explains the difference between the slopes of the data and the model. In fact, the SMVE depends slightly on the electron energy 共 ⬃ E0.5兲 as shown by Monte Carlo simulations,40 i.e., on the apply voltage qVds 共Fig. 16兲. Hence, SMVE is obtained by a similar expression than Sit and Sii 关Eqs. 共32兲 and 共33兲兴 giving FIG. 15. Comparison between experimental data 共25 ° C兲 and bond breaking rate model 关Eq. 共27兲兴 using stretching and bending vibrational mode parameters of Table III 共SMVE is kept constant兲. Electron-phonon scattering SMVE (e s ) FIG. 14. 共Color online兲 Evidence of a single Ids dependence for MVE mode in p-MOS with the use of a waiting time 共tw = 60 s兲 in order to remain mostly on the permanent damage 共Nit兲 for the sake of comparison to n-MOS 共Fig. 7兲 with Tox = 1.7 nm. SMVE ⬀ 共Vds − ប兲0.5 . -3 1.E-03 10 Ids/W (A/µm) 冉 冊 Ibs Ids m 共37兲 , -4 1.E-04 10 -5 1.E-05 10 -6 1.E-06 10 -7 1.E-07 10 0 1 2 3 4 5 Electron energy (eV) FIG. 16. Electron-phonon scattering rate SMVE simulated by Ref. 40 using Monte Carlo. J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix 6 1.E+06 10 TABLE IV. Summary of experimental parameter extractions from Eq. 共39兲 obtained for the best fits of n-MOS and p-MOS devices from the previous figures 共Figs. 18 and 19兲. 25C 75C 125C -1 Bond breaking rate R bMVE (s ) 114513-10 4 1.E+04 10 2 1.E+02 10 0 1.E+00 10 n-MOS p-MOS W/L = 10/0.04 0.6V<Vds <1.4V 0.0015 0.002 0.0025 Ids/W (A/µm) FIG. 17. 共Color online兲 Comparison between experimental data at three temperatures and bond breaking rate model 关Eq. 共27兲兴 using bending vibrational mode parameters of Table III and SMVE of Fig. 16. EES 2 ⬀ Ids 冉 冊 Ibs Ids m 共38兲 . As a consequence, a full modeling of cold to hot carrier degradation is determined by device lifetime extraction 共兲 for all the stressing conditions by assuming that all three degradation modes compete in parallel, which gives a general equation as 1 / = 兺i1 / i, = KSVE 冉 冊冉 冊 Ids a1 W Ibs 1 m + KEES Ids a1 a2 a3 Eemi 共eV兲 2.7 1.3 1 1 2.5 3 17.5 12 0.26 0.26 1.8V<Vgs <2.4V -2 1.E-02 10 0.001 1 m Device 冉 冊冉 冊 Ids W a2 Ibs Ids m a3/2 + KMVEVds 冉 冊 冉 冊 Ids W a3 − Eemi exp , k BT 共39兲 where KSVE, KEES, and KMVE are three constants determined in their respective dominant mode. We have validated this modeling for various gate-oxide thickness 共Tox = 5, 3.2, and 1.7 nm兲 under a large set of voltages conditions 共Vgs and Vds兲, temperatures and device geometries 共W / L兲, both on n-MOS and p-MOS. Table IV summarizes all the parameters extracted from the previous figures using Eq. 共39兲 for both n-MOS and p-MOS. The complete modeling explains also the change in the damage behavior through the gate-length dependence 关Fig. 19共a兲兴 where at small Ids it originates from the Ids and Ibs gate-length dependence, while at higher Ids the lifetime becomes only dependent on Ids magnitude on not on L anymore. The influence of temperature is explained by the temperature dependence of each degradation modes Figure 19共b兲 in ultrathin gate-oxide 共1.7 nm兲 n-MOS. Under EES corresponding to mode 2, temperature effect is included into the intrinsic temperature dependence of both currents Ids and Ibs 共so Ibs / Id兲 while entering into MVE regime 共mode 3兲; the activation energy is Eemi = 0.26 eV according to a large set of experimental data and previous results on H diffusion.47–50 Equation 共39兲 leads to the definition of an age function calculated by the same way than proposed in Ref. 51, 8 1.E+08 10 Vds =1.6V Vds = 6 1.E+06 Lifetime τ (a.u.) 10 4 1.E+04 10 0.9V 1V 1.1V 1.21V 1.4V 1.6V 1.8V a Lifetime τ (a.u.) 0.8V<Vgs <2.8V 2 1.E+02 10 0 1.E+00 10 6 1.E+06 10 4 1.E+04 10 2 1.E+02 10 W/L=10/0.06 W/L=10/0.05 W/L=10/0.045 W/L=10/0.04 W/L=10/0.035 0 1.E+00 10 0 0.001 0 0.002 0.001 Ids/W (A/µm) 0.002 a V ds = 6 1.E+06 10 4 1.E+04 10 2 1.E+02 10 b 0.9V 1V 1.1V 1.21V 1.4V 1.6V 1.8V W/L=10/0.04 Vds =1.6V 6 1.E+06 10 Lifetime τ (a.u.) τ x VdsEB/(2ħω) (a.u.) Ids/W (A/µm) 0.6V<Vgs <2.4V 4 1.E+04 10 T - 40°C 2 1.E+02 10 slope = -17.5 0 25°C CHC mode CCC mode complete model 0 1.E+00 10 0 0.001 0.002 Ids/W (A/µm) FIG. 18. 共Color online兲 共a兲 A weak Vds effect is observed on the experimental dependence of the MVE mode extracted from Fig. 7 which is explained by the simplified relationship of RbMVE 关Eq. 共36兲兴. 共b兲 Once device lifetime EB/2ប is plotted as MVE. Vds as a function of Ids / W, a single and universal degradation dependence is observed independent of the gate length in 40 nm n-MOS devices 共VDD = 1.1 V, Tox = 1.7 nm, T = 25 ° C兲. 125°C 10 1.E+00 0 0.001 Ids/W (A/µm) 0.002 b FIG. 19. 共Color online兲 共a兲 Comparison between the complete modeling 共lines兲 and the experimental data of Fig. 10 共symbols兲 showing a clear change in the gate-length dependence between mode 2 共EES兲 and mode 3 共MVE兲. 共b兲 Comparison between experimental data of Fig. 13 and the complete modeling obtained for three temperatures in L = 40 nm n-MOS with Tox = 1.7 nm. 114513-11 J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix ∆(1/gmmax) () 1000 8 0.5 t nMOS, Tox = 32nm, 25°C 100 10 1 1,E-04 10-4 1,E-02 10-2 1,E+00 100 1,E+02 102 Age FIG. 20. Degradation of the peak transconductance ⌬共1 / gm兲 in linear mode as a function of the age modeling with Eq. 共40兲 and plotted for 30 different stress conditions 共distributed in SVE, EES, and MVE modes兲 showing a single power law trend 共L = 0.15 m, Tox = 3.2 nm, and T ° = 25 ° C兲. Age = 冋 冉 冊冉 冊 冉 冊冉 冊 冉 冊 冉 冊册 Ids t = t KSVE W a3/2 + KMVEVds Ids W a1 Ibs Ids a3 exp m + KEES Ids W − Eemi k BT . a2 Ibs Ids m 共40兲 The Age function allows merging degradations of the three modes on the same figure 共Fig. 20兲. This figure confirms that the time dependence follows a power law with an exponent of 0.5, for all stressing conditions. IV. CONCLUSIONS Interface state creation at Si/ SiO2 has been related to several mechanisms depending on the stress conditions and the carrier energy. NBTI data are explained by a reduction in the bond energy with the oxide field. This barrier is equal to 1.5 eV without oxide field. The same threshold energy is found under hot carrier damage, whereas the underlying mechanism is totally different. HC mode is related to the excitation of bond vibrational modes induced by carriers. If a carrier has enough energy to excite the adsorbate resonance state, its energy may be transfer by coupling and lead to SVE. If single carriers have not enough energy to reach the resonance state, they can cooperate through EES and then induce SVE. The third HC mode happens when carrier energy is so low that the resonance is not possible even after EES. In this case, carriers may induce a step by step vibrational ladder climbing if the current is high enough 共MVE兲. A modeling of the bending MVE mode, based on a review of electronic and vibrational mode properties of the Si–H bond, has been applied to a wide HC stress experimental dataset. Finally, a complete equation taking into account the three HC modes 共SVE, EES, and MVE兲 has been proposed, it enables an accurate description of the CHC to CCC modes including the temperature effect in actual technology nodes. Y. Nishi, Jpn. J. Appl. Phys. 10, 52 共1971兲. D. J. Di Maria and J. W. Stasiak, J. Appl. Phys. 65, 2342 共1989兲. J. H. Stathis, IEEE Int. Reliab. Phys. Symp. Proc. 2001, 132. 4 J. Suñé and E. Y. Wu, Tech. Dig. - Int. Electron Devices Meet. 2005, 388. 5 N. D. Lang, Phys. Rev. B 45, 13599 共1992兲. 6 R. E. Walkup, D. M. Newns, and Ph. Avouris, Phys. Rev. B 48, 1858 共1993兲. 7 G. P. Salam, M. Persson, and R. E. Palmer, Phys. Rev. B 49, 10655 共1994兲. 1 2 3 K. Stokbro, C. Thirstrup, M. Sakurai, U. Quaade, B. Yu-Kuang Hu, F. Perez-Murano, and F. Grey, Phys. Rev. Lett. 80, 2618 共1998兲. 9 K. Stokbro, B. Y.-K. Hu, C. Thirstrup, and X. C. Xie, Phys. Rev. B 58, 8038 共1998兲. 10 T.-C. Shen, C. Wang, G. C. Abeln, J. R. Tucker, J. W. Lyding, Ph. Avouris, and R. E. Walkup, Science 268, 1590 共1995兲. 11 C. Kaneta, T. Yamasaki, and Y. Kosaka, Fujitsu Sci. Tech. J. 39, 106 共2003兲. 12 C. G. Van de Walle and R. A. Street, Phys. Rev. B 49, 14766 共1994兲. 13 B. Tuttle and C. G. Van de Walle, Phys. Rev. B 59, 12884 共1999兲. 14 B. Tuttle, Phys. Rev. B 60, 2631 共1999兲. 15 M. Budde, G. Lüpke, C. Parks Cheney, N. H. Tolk, and L. C. Feldman, Phys. Rev. Lett. 85, 1452 共2000兲. 16 N. W. Ashcroft and N. D. Mermin, Solid State Physics 共Saunders College Publishing, Philadelphia, 1976兲. 17 M. Budde, G. Lüpke, E. Chen, X. Zhang, N. H. Tolk, L. C. Feldman, E. Tarhan, A. K. Ramdas, and M. Stavola, Phys. Rev. Lett. 87, 145501 共2001兲. 18 A. Nitzan and J. Jortner, Mol. Phys. 25, 713 共1973兲. 19 G. Lüpke, X. Zhang, B. Sun, A. Fraser, N. H. Tolk, and L. C. Feldman, Phys. Rev. Lett. 88, 135501 共2002兲. 20 B. Sun, G. A. Shi, S. V. S. Nageswara Rao, M. Stavola, N. H. Tolk, S. K. Dixit, L. C. Feldman, and G. Lüpke, Phys. Rev. Lett. 96, 035501 共2006兲. 21 R. Biswas, Y.-P. Li, and B. C. Pan, Appl. Phys. Lett. 72, 3500 共1998兲. 22 J. W. McPherson, J. Appl. Phys. 99, 083501 共2006兲. 23 J. R. Jamesona, W. Harrison, P. B. Griffin, J. D. Plummer, and Y. Nishi, J. Appl. Phys. 100, 124104 共2006兲. 24 V. Huard, C. Parthasarathy, N. Rallet, C. Guerin, M. Mammase, D. Barge, and C. Ouvrard, Tech. Dig. - Int. Electron Devices Meet. 2007, 797. 25 H. Kufluoglu and M. A. Alam, Tech. Dig. - Int. Electron Devices Meet. 2004, 113. 26 V. Huard, M. Denais, and C. Parthasarathy, Microelectron. Reliab. 46, 1 共2006兲. 27 S. Chakravarthi, A. T. Krishnana, V. Reddya, and S. Krishnan, Microelectron. Reliab. 47, 863 共2007兲. 28 M. A. Alam and S. Mahapatra, Microelectron. Reliab. 45, 71 共2005兲. 29 R. Biswas and Y.-P. Li, Phys. Rev. Lett. 82, 2512 共1999兲. 30 Ph. Avouris, R. E. Walkup, A. R. Rossi, T.-C. Shen, G. C. Abeln, J. R. Tucker, and J. W. Lyding, Chem. Phys. Lett. 257, 148 共1996兲. 31 H. Ueba, Appl. Surf. Sci. 237, 565 共2004兲. 32 M. V. Fischetti and S. E. Laux, Tech. Dig. - Int. Electron Devices Meet. 1995, 305. 33 P. A. Childs and C. C. C. Leung, J. Appl. Phys. 79, 222 共1996兲. 34 J. D. Bude, T. Iizuka, and Y. Kamakura, Tech. Dig. - Int. Electron Devices Meet. 1996, 865. 35 S. E. Rauch, G. LaRosa, and F. J. Guarin, IEEE Trans. Device Mater. Reliab. 1, 113 共2001兲. 36 B. N. J. Persson and Ph. Avouris, Surf. Sci. 390, 45 共1997兲. 37 K. Hess, L. F. Register, B. Tuttle, J. Lyding, and I. C. Kizilyalli, Physica E 共Amsterdam兲 3, 1 共1998兲. 38 E. T. Foley, A. F. Kam, J. W. Lyding, and Ph. Avouris, Phys. Rev. Lett. 80, 1336 共1998兲. 39 W. McMahon, K. Matsuda, J. Lee, K. Hess, and J. Lyding, Technical Proceedings of the 2002 International Conference on Modeling and Simulation of Microsystems, San Juan Marriott Resort & Stellaris Casino, Puerto Rico, 576 共2002兲. 40 Y. Kamakura, H. Mizuno, M. Yamaji, M. Morifuji, K. Taniguchi, C. Hamaguchi, T. Kunikiyo, and M. Takenaka, J. Appl. Phys. 75, 3500 共1994兲. 41 D. J. DiMaria, J. Appl. Phys. 87, 8707 共2000兲. 42 B. Doyle, M. Bourcerie, C. Bergonzoni, R. Benecchi, A. Bravaix, K. R. Mistry, and A. Boudou, IEEE Trans. Electron Devices 37, 1869 共1990兲. 43 D. Vuillaume and A. Bravaix, J. Appl. Phys. 73, 2559 共1993兲. 44 A. Bravaix, G. Goguenheim, V. Huard, M. Denais, C. Parthasarathy, F. Perrier, N. Revil, and E. Vincent, Microelectron. Reliab. 45, 1370 共2005兲. 45 C. Guerin, V. Huard, and A. Bravaix, IEEE Trans. Device Mater. Reliab. 7, 225 共2007兲. 46 S. E. Rauch III and G. La Rosa, IEEE Trans. Device Mater. Reliab. 5, 701 共2005兲. 47 A. Van Wieringen and N. Warmoltz, Physica 共Amsterdam兲 22, 849 共1956兲. 48 E. Cartier, D. A. Buchanan, J. H. Stathis, and D. J. Dimaria, J. Non-Cryst. Solids 187, 244 共1995兲. 114513-12 B. Tuttle, Phys. Rev. B 61, 4417 共2000兲. B. L. Sopori, K. M. Jones, X. Deng, R. Matson, M. Al-Jassim, and S. Tsuo, Photovoltaic Specific Conference 1991 共unpublished兲, p. 833. 51 P. L. Lee, M. M. Kuo, K. Seki, P. K. Ko, and C. Hu, Tech. Dig. - Int. 49 50 J. Appl. Phys. 105, 114513 共2009兲 Guerin, Huard, and Bravaix Electron Devices Meet. 1988, 134. C. G. Van de Walle and W. B. Jackson, Appl. Phys. Lett. 69, 2441 共1996兲. 53 Z. Chen, J. Guo, and P. Ong, Appl. Phys. Lett. 83, 2151 共2003兲. 54 S. Watanabe and Y. Tamura, Fujitsu Sci. Tech. J. 39, 84 共2003兲. 52