Bachelor Thesis Estimation of Interface and Oxide Defects in Direct Contact High-k/Si Structure by Conductance Method Toru Kubota Department of Electrical and Electronic Engineering School of Engineering Tokyo Institute of Technology Supervisor: Prof. Hiroshi Iwai February, 2011 February, 2011 Abstract of Bachelor Thesis Estimation of Interface and Oxide Defects in Direct Contact High-k/Si Structure by Conductance Method Supervisor: Prof. Hiroshi Iwai Department of Electrical and Electronic Engineering 07_09508 Toru Kubota Since the scaling of MOSFETs utilizing SiO2 as gate dielectric has reached its limit, high-k materials have been studied as gate dielectrics. For high-k materials such as Hf-based MOSFETs, an SiO2 interfacial layer is intentionally formed at the high-k/Si interface to improve electric characteristics. Besides, La2O3 gate dielectric is a promising high-k material because it shows fairly nice electrical characteristics by forming La-silicate layers at the interface. Since the dielectric constants of the silicates are larger than that of SiO2, the equivalent oxide thickness scaling can be further extended by achieving a direct contact of high-k with the Si substrate. However, the interface properties of La-silicate and Si substrate are still unknown. Moreover, a method to further improve the interfacial properties is not clarified yet. In this thesis, the interfacial properties of La-silicate/Si structures were evaluated by conductance method and modeled to create an equivalent circuit. The equivalent circuit includes capacitances and conductances representing the interface states and the parasitic conductance caused by the leakage current through the high-k. By comparing the spectra obtained from reference SiO2/Si MOS capacitor, additional peaks at frequency range of 100 Hz was detected with La-silicate/Si structure. The origin of the signals can be interpreted as the defects in the silicate layers adjacent to the interface as their capture and emission time constants are different from those for interface states in SiO2/Si structure. Based on the above findings, an equivalent circuit model of MOS capacitor including the contribution from oxide defects is newly developed. In consequences, a good agreement between the measured and calculated data can be obtained. It is shown from this study that the quantitative study on interface traps and oxide traps is very effective in improving the interfacial properties of MOS devices. i Contents 1 Introduction 1.1 Introduction of high-k materials --------------------------------------- 1 1.2 Issue in high-k gate dielectrics ------------------------------------------ 2 1.3 La2O3 as high-k gate dielectrics ---------------------------------------- 3 1.4 Issue in La2O3 gate dielectric -------------------------------------------- 4 1.5 Purpose of this study ------------------------------------------------------ 6 2 Experiments 2.1 Experimental principles 2.1.1 SPM cleaning and HF treatment ----------------------------------- 7 2.1.2 Formations of dielectrics 2.1.2.1 Formation of SiO2 dielectric by thermal oxidation --------- 7 2.1.2.2 Deposition of La2O3 dielectric by MBE ------------------------ 8 2.1.3 RF magnetron sputtering -------------------------------------------- 9 2.1.4 Patterning of resist by photo lithography ----------------------- 10 2.1.5 Dry etching by RIE -------------------------------------------------- 10 2.1.6 PDA and PMA in F.G. ambient ------------------------------------ 12 2.1.7 Deposition of Al by vacuum evaporation ------------------------ 12 2.2 Experimental procedure ------------------------------------------------- 13 2.3 Estimation of interface state density by conductance method 2.3.1 Statistical analysis of GP/ω ------------------------------------------- 16 2.3.2 Analysis of capacitance ----------------------------------------------- 19 3 Results and Discussions in SiO2/Si structure 3.1 Contribution of SiO2/Si structure to small signal voltage --------- 20 3.2 Energy distribution of Dit ------------------------------------------------ 21 ii 4 Results and Discussions in La2O3/Si structure 4.1 Interface state at the high-k/Si interface ------------------------------ 25 4.2 Influence of leakage current on the determination of GP/ω ------- 25 4.3 Proposal of a new model ------------------------------------------------- 29 4.4 Application of a new model to the measurement -------------------- 31 5 Conclusions ----------------------------------------------------------------- 34 Appendices A. The derivation of calculation changing equivalent circuit model to simplified circuit ----------------------------------------------------------- 35 B. The derivation of calculation changing measured circuit to simplified circuit ---------------------------------------------------------- 35 C. The derivation of calculation changing measured circuit to equivalent circuit model ------------------------------------------------ 37 References --------------------------------------------------------------------- 38 Acknowledgements --------------------------------------------------------- 40 iii 1 Introduction 1.1 Introduction of high-k materials Metal-oxide-semiconductor field effect transistor (MOSFET) is one of the most principal elements for very large-scale integration (VLSI) technology. It is used in a lot of electronic equipment, and we use it every day. It is not too much to say that our modern life consists of VLSI technology. The high density, performance, and low consumption electricity of MOSFETs was progressing with the scaling of those device sizes. Now, most of the gate dielectrics used are silicon dioxide (SiO2) because it works as layers which have high insulation, passivation, and diffusion barrier [1]. However, as device scaling progresses, physical thickness of SiO2 gate dielectrics becomes thin, so that leakage current flows because of the quantum mechanical tunnel effect. Leakage current causes larger electric consumption in the device. Therefore, scaling of MOSFETs with SiO2 gate dielectrics reach its limit. High-k materials have been studied to break the limitation instead of SiO2 dielectrics. High-k gate dielectrics have high dielectric constants, so we can get same capacitance in the larger physical thickness. Because of this reason, leakage current can decrease compared with SiO2 gate dielectrics. It is shown in Fig. 1.1. It is introduced that the index which expresses scaling of high-k gate dielectrics is equivalent oxide thickness (EOT). EOT can be given by EOT = ε SiO 2 t OX , ε high −k (1.1) where εSiO2, εhigh-k are each dielectric constant of SiO2, high-k, and tox is the physical thickness. Therefore, the study of high-k dielectrics is very important for device scaling in the future. 1 Gate Gate SiO2 S same EOT and capacitance D High-k S D Si-substrate Si-substrate Fig. 1.1 High-k gate dielectrics become larger thickness when they have the capacitance as same as SiO2, and leakage current can decrease. 1.2 Issue in high-k gate dielectrics It is effective to introduce high-k gate dielectrics in order to realize the low EOT. However, it is considered that fixed oxide charge generates in high-k gate dielectrics because the metal diffuses from gate to those [2]. Fixed oxide charge causes bad influences [3] . Fixed oxide charges work as scattering center, which is called coulomb scattering. Coulomb scattering makes mobility of carriers in surface inversion channel decrease. This is one of the most serious problems. So, the method that is considered in order to solve this problem is forming SiO2 interfacial layer at the interface between high-k gate dielectrics and Si-substrate as shown in Fig. 1.2. SiO2 interfacial layer is necessary to improve electrical characteristics. However, we are demanded to achieve EOT = 0.5nm in 2015 as shown in Fig. 1.3 from ITRS 2009 [4]. So, it is difficult for high-k gate dielectrics formed SiO2 interfacial layer to achieve EOT = 0.5nm. Thus, we need to select the high-k gate dielectric that can directly contact Si-substrate. Gate High-k SiO2 Si-substrate Fig. 1.2 It is difficult to decrease EOT because of SiO2 interfacial layer. 2 1.1 EOT (nm) 1 Planar Bulk MOSFET 0.9 0.8 0.7 0.6 0.5 0.4 2009 2011 2013 2015 YEAR Fig. 1.3 EOT versus YEAR plots in planer bulk MOSFETs from ITRS 2009. We are demanded to achieve EOT = 0.5nm in 2015. 1.3 La2O3 as high-k gate dielectrics Lanthanum oxide (La2O3) is one of the promising high-k gate dielectrics. The relations between band gap and dielectric constants, k are shown in Fig. 1.4 [5]. The dielectrics generally have trade-off relations between band gap and dielectric constants. Fig. 1.4 shows that La2O3 has a high dielectric constant (εr = 23.4) and broad band gap (Eg = 5.6 eV) together. La2O3 Fig. 1.4 Band gap versus dielectric constants, k plots. La2O3 has a high dielectric constant and broad band gap together. 3 In addition, La2O3 has one more advantage. After the device used La2O3 as gate dielectric is annealed, it is formed La-silicate layer at the interface between La2O3 and Si-substrate [6]. It is shown in Fig. 1.5. La-silicate has a higher dielectric constant that is from 8 to 14 than that of SiO2. So, La2O3 gate dielectric is effective to decrease EOT in the future. From these discussions, La2O3 is used as high-k gate dielectrics in this study. Gate Gate Annealing La2O3 La2O3 La-Silicate Si-substrate Si-substrate Fig. 1.5 After annealing, La-silicate formed at the interface between La2O3 and Si-substrate, which has a higher dielectric constant that is from 8 to 14. 1.4 Issue in La2O3 gate dielectric La2O3 can contact Si-substrate directly without SiO2 interfacial layer. However, it is one of the problems that the interface state at the surface between La2O3 and Si-substrate is worse than that of SiO2. The reason is the metal diffuses into La2O3 dielectric easily than SiO2. Interface state causes next two phenomena that have bad influences, so it is important to decrease interface state density (Dit). First, interface trap generates additional capacitance because the charge density trapped to interface with the change of surface potential changes. This capacitance affects capacitance-voltage (C-V) characteristics, which is one of the indicators expressed the performance of MOSFET. Therefore, it is important to use device processes which make Dit minimize in order to do good C-V characteristics that have reproducibility. Second, interface trap generates leakage current in gate diode structures because it works as generation and 4 recombination centers, or promotes band to band tunneling processes. Therefore, we must minimize Dit in order to realize high performance and reliability. However, we need to consider the method estimating Dit in the devices with any dielectric first. Conductance method is one of the methods estimating Dit. It is a way which we replace MOS capacitor with equivalent circuit models and calculate Dit. Fig. 1.6 shows equivalent parallel conductance versus frequency plots. This figure means that the local maximum of each curve indicates the magnitude of Dit (see in paragraph 2.3). We can see the peak of each curve in Fig. 1.6. The curve of SiO2 gate dielectric has one peak, but that of La2O3 gate dielectric has two peaks in this figure. So, peaks of the curve used La2O3 gate dielectric mean that interface state and other energy level are measured. The peak of other energy level is much larger than that of interface state, and it appears at low frequency region. Therefore, the peak of interface state at low frequency is concealed by that of other energy level, and the problem interface state cannot be measured is generated. (x 10-6) 3 (x 10-9) 5 La2O3 系列2 系列 Gp /ω (F/cm2) 2.5 SiO2 系列1 系列 4 2 3 1.5 2 1 E-Ei = 0.12 eV Vamp = 100 mV 0.5 1 0 0 101 102 103 104 105 106 107 Frequency (Hz) Fig. 1.6 Equivalent parallel conductance over angular frequency (GP/ω) versus frequency plots. Solid dots are La2O3 gate dielectric, open dots are SiO2 gate dielectric. 5 1.5 Purpose of this study The purpose of this study is to estimate interface and oxide defects by conductance method, and to compare SiO2 and La2O3 dielectrics. Chapter 1 shows the background and purpose of this study. In chapter 2, experimental principles, procedure, and estimate method are described. In chapter 3, the results in SiO2/Si structure are reported. In chapter 4, the results in La2O3/Si structure are reported, and compared with SiO2/Si structure. Finally, the conclusions of this study are summarized in chapter 5. 6 2 Experiments 2.1 Experimental principles 2.1.1 SPM cleaning and HF treatment In fabrication processes, it is important to clean the surface of Si-substrate because particles and metal ions affect the performance, reliability, and yield of the devices greatly. One of the effective methods is sulfuric acid hydrogen peroxide mixture (SPM) cleaning and hydrofluoric acid (HF) treatment. They remove particles and metal ions, and make the surface of Si-substrate clean. The liquid used in SPM cleaning is mixed liquid of H2O2 and H2SO4 (H2O2:H2SO4 = 1:4), which has strong oxidation. Si-substrate is soaked into the liquid, then particles and metal ions are separated from Si-substrate because of its oxidation. At that time the surface of Si-substrate is also oxidized and formed chemical oxide. In order to eliminate chemical oxide, Si-substrate is soaked into 1% HF. Then the devices with clean surface are completed. Fig. 2.1 shows schematic illustration of SPM cleaning and HF treatment. Particle Metal ion SiO2 Chemical oxide Chemical oxide SiO2 SiO2 SiO2 SiO2 Si-substrate Si-substrate Si-substrate SPM cleaning H2O2+H2SO4 (1:4) SiO2 HF treatment 1%HF Si-substrate Fig. 2.1 Schematic illustration of SPM cleaning and HF treatment. 2.1.2 Formations of dielectrics 2.1.2.1 Formation of SiO2 dielectric by thermal oxidation The most general method formed SiO2 dielectric is thermal oxidation [7]. It is oxidized while flown O2 in high temperature, which is typically about 1000oC. 7 Therefore, SiO2 dielectric is formed at the surface. This method takes many times, but forms high quality and reliability SiO2 dielectric because the position of surface after thermal oxidation moves inside. Heat O2 Si-substrate Fig. 2.2 Schematic illustration of the oxidation furnace. Thermal oxidation SiO2 SiO2 SiO2 SiO2 Si-substrate SiO2 Si-substrate Surface after thermal oxidation Fig. 2.3 SiO2 dielectric is formed by thermal oxidation. 2.1.2.2 Deposition of La2O3 dielectric by MBE The method used in order to deposit La2O3 is Molecular beam epitaxy (MBE) in this study. Fig. 2.4 shows schematic illustration of MBE [8]. First, the inside of chamber is done in ultra high vacuum in order to make the orbit of evaporated molecular beam stable. Then, accelerated electron beam (E-beam) emitted from hot-filament is bended by the magnetic field, and it hits the source material. The place hit by E-beam is evaporated because of its heat. The generated molecular beam from there splashes for Si-substrate, and then La2O3 dielectric is deposited on the wafer. 8 Substrate Molecular beam Evaporated source by heat Accelerated E-beam Source material Fig. 2.4 Schematic illustration of MBE. 2.1.3 RF magnetron sputtering Radio frequency (RF) magnetron sputtering is one of the deposition methods. Schematic illustration is shown in Fig. 2.5. First, the high voltage is applied between target and substrate, and Ar is flown into the chamber. This voltage difference causes high electric field. Therefore, that makes Ar become the state of plasma, and Ar is ionized. Ar ions are attracted to the target because electrons are accumulated in the target. Ar ions collide with the target, then atoms of the target are emitted and the film is deposited. Substrate Ar+ Atom of target material target material plasma Fig. 2.5 Schematic illustration of RF magnetron sputtering. 9 2.1.4 Patterning of resist by photo lithography Resist patterning is the method to eliminate only parts of required gate metal. It is used positive resist in this study, which places exposed to light are melted. First, resist is applied on the gate electrode. In order to make resist uniform, Si-substrates applied resist are revolved by the spinner. After that, the substrates are heated to fixate the resist, which is called pre-bake. And, the substrates are adjusted to photo mask to do resist patterning. After they are exposed to light, they are soaked into the developer. In consequence, only resist uncovered with the mask is melted by the developer as shown in Fig. 2.6. Finally, they are heated to fixate the resist, too, and this is called post-bake. Therefore, only required resist is remained. Light Resist Gate electrode Mask Gate dielectric SiO2 SiO2 Si-substrate SiO2 Exposure and Developing SiO2 Si-substrate Fig. 2.6 Schematic illustration of photo lithography 2.1.5 Dry etching by RIE Reactive ion etching is (RIE) is one of the methods to etch the films [9]. It is similar to RF magnetron sputtering, but it is used by not only physical but also chemical reaction. The gate electrodes used in this study are tungsten (W) and titanium nitride (TiN). The gas etching them is sulfur hexafluoride (SF6), and the gas etching resist is O2. First, the gas used to etch the film becomes the state of plasma as same as RF magnetron sputtering. SF6 and O2 ionized radicals with high reaction, cations, electrons, and so on. Chemical reaction occurs when radicals are absorbed on the substrates, and 10 then the compound is separated from the substrates because of its volatility. Cations collide with the substrates because of electric field, which is physical etching. It is shown in Fig. 2.7. The advantage of RIE can etch to the same as the shapes of mask because of anisotropic etching. The reason is that cations are incident vertically. It is shown in Fig. 2.8. SF6 or O2 e- Radical e- e- e- e- plasma Cation Substrate Fig. 2.7 Schematic illustration of RIE. Radical Cation SF6 O2 Resist SiO2 SiO2 Si-substrate Gate electrode Gate dielectric SiO2 Gate electrode patterning SiO2 Si-substrate SiO2 Resist removal SiO2 Si-substrate Fig. 2.8 Schematic illustration after etching. 11 2.1.6 PDA and PMA in F.G. ambient Post deposition annealing (PDA) and post metallization annealing (PMA) are very important processes to fabricate high performance devices. These two methods are different from annealing timing, but same as the principle. The structures and properties of silicon that has periodicity terminate at the surface between Si and gate dielectrics. So, many dangling bonds that cause interface state are generated there. PDA and PMA are used in order to terminate them. After it is formed a vacuum, forming gas (F.G.) (N2:H2 = 97:3) is flown and the substrates are heated at 420oC for 30min or 800oC for 2sec in this study. Thus, forming gas annealing (F.G.A.) is effective to decrease Dit. It is shown Fig. 2.9. H2 N2 SiO2 Annealing SiO2 SiO2 SiO2 H H H Si Si Si Si Si Si Si Si Si-substrate Si-substrate Fig. 2.9 Schematic illustration of forming gas annealing (F.G.A.). 2.1.7 Deposition of Al by vacuum evaporation Finally, Al electrode is deposited on the back side of substrates. In this study, vacuum evaporation method is used to deposit Al as shown in Fig. 2.10. Al is placed on W boat, and current is flown in the circuit connected with W boat. Large current is flown, and then Al is vaporized by joule heat because the boiling point of Al is lower than the melting point of W. Therefore, vaporized Al is deposited on the substrates. 12 Back side of substrate Heated Aluminium W boat Fig. 2.10 Schematic illustration of vacuum evaporation method. 2.2 Experimental procedure Fig. 2.11 shows schematic illustration of MOS capacitors fabrication processes. First, the fabrication of MOS capacitors with SiO2 gate dielectric is explained. SiO2 (10nm) is deposited on the Si-substrates after SPM cleaning and HF treatment by thermal oxidation, and W (60nm) is deposited by RF magnetron sputtering. After W is pattered by photo lithography and RIE, PMA at 420oC for 30min is conducted. Finally, Al (50nm) is deposited on the back side of substrates, and then they are measured. Next, the fabrication of MOS capacitors with La2O3 gate dielectric is explained. La2O3 (3nm) is deposited by MBE, and Si (1.5nm) is deposited by RF magnetron sputtering. In order to form La-silicate layer, PDA at 800oC for 2sec is conducted here. And then, TiN (60nm) is deposited as a gate electrode by RF magnetron sputtering. After that PMA and Al (50nm) deposition are conducted as same as MOS capacitors with SiO2. 13 n&p-Si (100) substrate Particle Metal ion SiO2 SiO2 Si-substrate SPM cleaning and HF treatment SiO2 SiO2 Si-substrate in-situ SiO2 (10nm) deposition by thermal oxidation La2O3 (3nm) deposition by MBE SiO2 SiO2 La2O3 SiO2 SiO2 Si-substrate Si-substrate W (60nm) deposition by RF magnetron sputtering Si (1.5nm) deposition by RF magnetron sputtering W SiO2 SiO2 SiO2 Si SiO2 SiO2 SiO2 Si-substrate Si-substrate PDA in F.G. at 800oC for 2sec SiO2 La-silicate SiO2 Si-substrate TiN (60nm) deposition by RF magnetron sputtering TiN SiO2 La-silicate Si-substrate 14 SiO2 Photo lithography TiN W SiO2 SiO2 SiO2 SiO2 Si-substrate La-silicate SiO2 Si-substrate Gate electrode and resist etching by RIE PMA in F.G. at 420oC for 30min TiN W SiO2 SiO2 SiO2 SiO2 Si-substrate La-silicate SiO2 Si-substrate Al deposition by vacuum evaporation method TiN W SiO2 SiO2 SiO2 SiO2 Si-substrate La-silicate Si-substrate Al Al Measurement Fig. 2.11 Schematic illustration of experimental procedure. 15 SiO2 2.3 Estimation of interface state density by conductance method 2.3.1 Statistical analysis of GP/ω In this study, conductance method is used as estimating Dit. Conductance method is a way which we replace MOS capacitor with equivalent circuit models and calculate Dit. Fig. 2.12(a) shows an equivalent circuit model of MOS capacitor [10], where COX is the oxide capacitance per unit area, CS is the silicon capacitance per unit area, Rit and Cit are the resistance and capacitance components per unit area related to interface trap, and Gt is tunnel conductance per unit area related to leakage current. On the other hand, the measurement circuit is shown in Fig. 2.12(b), where Cm and Gm are the measured capacitance and conductance per unit area. And the equivalent parallel circuit is shown in Fig. 2.12(c), where CP and GP are the equivalent parallel capacitance and conductance per unit area. The reason why it is changed like that can be expressed that GP has only interface trap information not including CS when the circuit is converted. CP and GP are given by CP = CS + GP ω = qDit 1 + (ωτ it ) qωτ it Dit 1 + (ωτ it ) 2 2 , , (2.1) (2.2) where Cit = qDit, and τit = CitRit (see detail calculation in Appendix A). Dit is interface state density and τit is interface trap time constant here. Eq. (2.2) becomes symmetrical in ωτit because dividing GP by ω is done. These two equations are for interface traps with a single energy level in the band gap. However, interface traps actually are continuously distributed. Therefore, these two equations are rewritten as [11] CP = CS + GP ω = qDit ωτ it [ tan −1 (ωτ it ) , ] qDit 2 ln 1 + (ωτ it ) . 2ωτ it (2.3) (2.4) These two equations are for interface traps with the continuum level. It is more effective for Eq. (2.4) than (2.3) to calculate Dit because Eq. (2.4) has only parameters related to interface traps without including CS. 16 On the other hand, the capacitance and conductance are measured from Fig. 2.12(b). So, we need to change Fig. 2.12(b) for (c), and compare with Eq. (2.4). Then, GP/ω is calculated as (see detail calculation in Appendix B) ω (Gm − Gt )C OX 2 = . ω (Gm − Gt )2 + ω 2 (C OX − C m )2 GP COX COX Cm Gt Gm Gt Rit CS (2.5) CP GP Cit (a) (b) (c) Fig. 2.12 Equivalent circuits of MOS capacitor; (a) an equivalent circuit model of the MOS capacitor, (b) the measured circuit, (c) the simplified circuit of (a). The values calculated from Eq. (2.2), (2.4) and (2.5) are shown in Fig. 2.13 respectively. We see it is generated serious error between continuum level and measurement data. It is considered that the reason is surface potential fluctuations. This is generated by inhomogeneities in oxide charge and interface charge [12]. So, we consider surface potential fluctuations, but it is difficult to analyze the experimental data. Therefore, we assume that surface potential fluctuations follow normal distribution, and then Eq. (2.4) becomes ∞ [ ] q ⌠ Dit 2 = ln 1 + (ωτ it ) P(ψ S )dψ S , 2 ⌡−∞ ωτ it ω GP P(ψ S ) = ( ψ −ψ exp − S 2 S 2σ 2πσ 2 1 17 ) 2 , (2.6) (2.7) where ψ S is the surface potential, ψ S and σ are mean and standard variation of the surface potential respectively. So, the larger σ is, the lower values of the peak and broader width of the curves are. Values calculated from these equations are shown solid line in Fig. 2.x. The parameters that is used as the variables are Dit, τit, σ, and Gt in Eq. (2.6) and (2.7) at this time. These values in Fig. 2.13 are Dit = 8.2 x 1010 eV-1cm-2, τit = 3.1 x 10-5 s, σ = 0.041, and Gt ≈ 0. The integration of Eq. (2.6) is calculated that Dit and τit are determined by fitting GP/ω curves first. Then, Dit and τit between determined values are interpolated by spline interpolation method, which is found the values between determined values by solving polynomicals of three dimensions. We see measured and statistic data are close values in this figure. Therefore, it is important for accurate measurements to consider surface potential fluctuations. Gp /ω (x10-9 F/cm2) 7 Single level Continuum level Measurement Statistic 6 5 4 3 2 1 0 102 103 104 105 106 107 Frequency (Hz) Fig. 2.13 Equivalent parallel conductance over angular frequency (GP/ω) versus frequency plots. Dashed line is values used by single level, alternate long and short dash line is values used by continuum level, dots are measured values, solid line is statistic based on normal distribution. 18 2.3.2 Analysis of capacitance The problems in paragraph 1.4 cause measurement values not to be able to be fitted by GP/ω. Because we need to discuss other parameters, we notice measured capacitance and discuss that. Therefore, we change Fig. 2.12(c) to (b), and measured capacitance is given by (see detail calculation in Appendix C) GP 2 COX + C P (COX + C P ) ω Cm = . 2 GP 2 + (COX + C P ) ω (2.8) Eq. (2.8) is assumed continuum level. Similarly, surface potential fluctuations follow normal distribution, so that Eq. (2.8) becomes ∞ G P 2 ⌠ C OX ω + C P (C OX + C P ) Cm = P(ψ S )dψ S . 2 GP 2 + (C OX + C P ) ω ⌡−∞ (2.9) And, substituting Eq. (2.3) and (2.4) into Eq. (2.8) and (2.9), we obtain Fig. 2.14. Thus, we can fit by using Cm, too. 34 Continuum level Measurement Statistic Cm (nF/cm2) 32 30 28 26 24 22 20 102 103 104 105 106 107 Frequency (Hz) Fig. 2.14 Measured capacitance (Cm) versus frequency plots. Statistic is similar to measured capacitance. 19 3. Results and discussions in SiO2/Si structure 3.1 Contribution of SiO2/Si structure to small signal voltage AC voltage is applied to the device, and it can be measured Dit by conductance method. The reason is that electrons are not captured to interface state and emitted from there in the case of DC voltage. It is used sine wave as power sources in this experiment. So, we need to examine the dependence of small signal voltage and effect influenced Dit measurement. Fig. 3.1 shows GP/ω in each small signal voltage. We see that the curve of larger small signal voltage becomes the smaller local maximum and broader width of that. So, it is considered the change of the small signal voltage works as same as parameter σ varies. However, even if the small signal voltage is small, measurement curves do not agree with continuum level curves. And in Fig. 3.1, frequency of each curve at the local maximum is different. The reason is explained in Fig. 3.2 and Eq. (2.5). GP/ω in each energy level is shown in Fig. 3.2. In this figure, as the energy level becomes high, the peak of curve becomes high and the frequency shifts the right hand. Because GP/ω is followed by Eq. (2.5), when σ is large, it is deeply related to GP/ω of other energy level. Therefore, the peaks of curves shift the right hand. Gp /ω (x10-9 F/cm2) 7 small signal voltage 6 5 5 mV 50 mV 100 mV down 4 3 2 up 1 0 102 103 104 105 106 107 Frequency (Hz) Fig. 3.1 GP/ω versus frequency plots in each small signal voltage. 20 Gp /ω (x10-9 F/cm2) 8 E-Ei = 0.02 eV 0.16 eV 0.11 eV 0.07 eV 7 6 0.20 eV 5 4 3 2 1 0 102 103 104 105 106 107 Frequency (Hz) Fig. 3.2 Equivalent parallel conductance over angular frequency (GP/ω) versus frequency plots when small signal voltage is 50 mV. 3.2 Energy distribution of Dit Dit is distributed in the band gap of silicon. When Dit is measured by conductance method, the place of measured interface state is equal to Fermi level. So, we can determine the place of interface state from Fermi level and surface potential. Because we cannot determine surface potential directly, it is necessary to determine from the relation between gate bias surface potential. Although the relation is expressed a complicated equation, we can make it easy by using depletion approximation. The reason we use depletion approximation is that most of Dit is measured and it is easy to analyze in depletion region. Although it is measured in weak inversion region, it is difficult to analyze [13]. The relation between surface potential and gate bias is given by Vg = − 2ε Si qN D ψ S C OX − ψ S + V fb (n-type silicon), 21 (3.1a) 2ε Si qN A ψ S Vg = C OX + ψ S + V fb (p-type silicon), (3.1b) where εSi is silicon permittivity, q is electronic charge, ND and NA is donor and acceptor impurity density in bulk silicon, and Vfb is flat-band voltage. Surface potential is determined by solving Eq. (3.1a) or (3.1b) as ψ S . The values of ψ S is defined that it is plus values when the silicon band is bends downward and minus values when the silicon band is bends upward. And Fermi level is given by N E f = kT ⋅ ln D ni (n-type silicon), (3.2a) N E f = − kT ⋅ ln A (p-type silicon), ni (3.2b) where k is Boltzmann’s constant, T is absolute temperature, ni is intrinsic carrier density, and the mid gap energy Ei is replaced as Ei = 0 after this. From these equations, we get energy distribution of Dit as shown in Fig. 3.3. In this figure, we see Dit is decreasing as energy approaches the mid gap. And, we also see next two characteristics. First, the ranges that can be measured peaks of curves are not equal between p-type and n-type. Actually, the measured ranges of p-type and n-type are 0.11 eV and 0.19 eV. The reason is considered τit is different in each type. τit is given by [14] τn = qψ S exp − (n-type silicon), vth _ nσ n n0 kT τp = qψ S exp (p-type silicon), vth _ pσ p p 0 kT 1 1 (3.3a) (3.3b) where vth_n and vth_p are electron and hole thermal velocity, σn and σp are electron and hole capture cross section, n0 and p0 are density of free electrons and holes at thermal equilibrium. In these parameters, n0 and p0 are almost same because they can be replaced with impurity densities at room temperature. The relation to σn and σp are not found, but we see vth_n is larger than vth_p [15]. Therefore, the terms except exponential term in Eq. (3.3b) are larger than that of Eq. (3.3a), so that τp changes more largely than τn when each ψ S in Eq. (3.3a) and (3.3b) changes for same amount. Because of 22 this reason, GP/ω of p-type silicon shifts more largely than n-type silicon, and it is considered that the range of measurement is narrow in the p-type silicon. Second, we see the energy that is minimized Dit is not mid gap in this figure. It is considered that the bulk impurity densities are not accurate values because the Si-substrate in this study is used with a given resistivity. Fig. 3.4 shows C-V curves of the measurement data and ideal data. We see they do not agree in depletion region from -0.1 V to 0.4 V. In depletion region, the contribution of the depletion layer capacitance is the largest of all regions. Also, it depends on the impurity density. Therefore, Fig. 3.4 indicates the impurity density is not correct. Thus, if the impurity densities have errors, Fermi level changes, and it makes energy distribution change, too. The errors of impurity densities especially influence the fitting by measured capacitance. Dit (x 1010 eV-1cm-2) 14 12 10 p-type Si 8 6 n-type Si 4 2 0 -0.3 -0.15 0 0.15 0.3 E-Ei Fig. 3.3 Interface state density versus energy plots. Mid gap in silicon band gap means E-Ei = 0. Dit is decreasing as energy approached the mid gap. 23 Capacitance (µ µF/cm2) 0.4 Measurement Ideal 0.3 0.2 0.1 W/SiO2/n-Si Not agree @100kHz Nd = 3x1015 0 -0.5 0 0.5 1 1.5 2 Gate Voltage (V) Fig. 3.4 Capacitance versus gate voltage plots. Measurement data do not agree with ideal C-V curve in depletion region. 24 4 Results and Discussions in La2O3/Si structure 4.1 Interface state at the high-k/Si interface It is found that GP/ω curves in La2O3/Si structure are different from those of SiO2/Si structure as shown in Fig. 4.1, which shows GP/ω versus frequency plots in each scale. Each small signal voltage and energy is different in order to make the figures easy to see. Anyway, we can see two peaks from each curve, and each value of two peaks is different about one digit in these figures. This characteristic is not seen in the case of SiO2 gate dielectric. Compared with Fig. 3.2, it is considered that peaks on high frequency region seen in Fig. 4.1(b) are probably caused by interface state. In this chapter, considerations about these peaks are shown. 30 5 Gp /ω (x10-7 F/cm2) E-Ei Vamp = 25 mV -0.1 eV 25 0.05 eV 0.10 eV 0.15 eV 0.20 eV 4 0 eV 0.1 eV 20 0.2 eV E-Ei 3 15 2 10 1 5 Vamp = 100 mV 0 0 101 102 103 103 104 105 106 Frequency (Hz) (a) (b) Fig. 4.1 GP/ω curves have two peaks with different values; (a) low frequency region, (b) high frequency region. 4.2 Influence of leakage current on the determination of GP/ω In chapter 3, we discuss SiO2 gate dielectric, and then GP/ω are calculated as Gt ≈ 0 in Eq. (2.5). This reason is that SiO2 dielectric has thick membrane (10 nm), so leakage 25 107 current can be almost ignored. However, we must consider the influence of leakage current in La2O3/Si structure because the thickness of La2O3 is 3nm. By the way, it is found that EOT of this sample is 0.88nm and flat-band voltage (Vfb) is -0.69V from the measurement of C-V characteristic. So, we need to consider values of tunnel conductance (Gt). Fig. 4.2 shows the relation between leakage current per unit and gate voltage. Seeing overall, the shape of measured plots is similar to exponential function. However, negative voltage region has particularly errors compared with exponential approximation. The range from -1.1 to -0.7 is measured by conductance method. Most of this range is depletion region because Vfb is -0.69V. And Gt is determined by leakage current differentiated by gate voltage. The relation between Gt and gate voltage is shown in Fig. 4.3, where the values calculated by measurement are distinguished from those calculated by approximation (Gt calculated by measurement is called measured Gt, and calculated by exponential approximation is called approximated Gt in this paper). Errors of Gt between measured plots and exponential approximation in region B are largely than those of leakage current in region A. Thus, it is difficult to estimate Gt. Leakage Current (x10-3 A/cm2) A 60 0 50 aMeasurement 40 bapproximation -0.5 Exponential aMeasurement Exponential -1 30 bapproximation -1.5 20 -2 A 10 0 -2.5 -10 -3 -2 -1.5 -1 -0.5 0 0.5 1 -1.1 -1 -0.9 -0.8 -0.7 Gate Voltage (V) Fig. 4.2 Leakage current per unit area versus gate voltage plots. Measured plots have errors compared with exponential approximation in region A. 26 Tunnel Conductance (x10-3 S/cm2) B 160 4 140 Exponential 120 2 60 1.5 1 B 20 系列2 系列 approximation 2.5 80 40 Exponential 3 bapproximation 100 Measurement 系列1 系列 3.5 aMeasurement 0.5 0 0 -2 -1.5 -1 -0.5 0 0.5 1 -1.1 -1 -0.9 -0.8 Gate Voltage (V) Fig. 4.3 Tunnel conductance per unit area (Gt) versus gate voltage plots. Errors between measured plots and exponential approximation are large in region B. Here, these two types of Gt are compared based on GP/ω in Eq. (2.5). First, GP/ω calculated by measured Gt is shown in Fig. 4.4. Although we can see a symmetry curve at E-Ei = 0.05eV, the others are not symmetry. In addition, the values of them become zero in low frequency region. Generally, values of GP/ω cannot become zero. Therefore, the probability that measured Gt is not correct is high. Next, GP/ω calculated by approximated Gt is shown in Fig. 4.5. Compared with Fig. 4.1, it is found that they are almost same. Therefore, it is considered the effect of leakage current is small when approximated Gt is used. There is another reason leakage current is small. C-V characteristic of this sample is shown in Fig. 4.6. If leakage current is large, the phenomenon capacitance decreases in accumulation region is seen, but the phenomenon does not occur from this figure. Because of the above reasons, it is considered approximated Gt is proper. 27 -0.7 Gp /ω (x10-7 F/cm2) 3.5 E-Ei 3 0.05 eV 0.10 eV 0.15 eV 0.20 eV 2.5 2 1.5 1 0.5 Vamp = 25 mV 0 102 103 104 105 106 107 Frequency (Hz) Fig. 4.4 GP/ω calculated by measured Gt. Most values become zero below 1kHz. 30 5 E-Ei Gp /ω (x10-7 F/cm2) 0 eV E-Ei Vamp = 25 mV -0.1 eV 25 0.05 eV 0.10 eV 0.15 eV 0.20 eV 4 0.1 eV 20 0.2 eV 3 15 2 10 1 5 Vamp = 100 mV 0 101 102 0 103 103 104 105 Frequency (Hz) (a) (b) Fig. 4.5 GP/ω calculated by approximated Gt; (a) low frequency region, (b) high frequency region. 28 106 107 Capacitance (µ µF/cm2) 3 100kHz 1MHz 2.5 not decrease 2 1.5 1 0.5 0 -2 -1.5 -1 -0.5 0 0.5 1 Gate Voltage (V) Fig. 4.6 Capacitance does not decrease in accumulation region. Therefore, leakage current is small. 4.3 Proposal of a new model In paragraph 4.3, it was found that the peaks appearing in low frequency region were not related to leakage current. So, we must consider other reasons. According to Nicollian and Brews [16], there are two types of trap in the silicon. One is interface traps located at or near the interface. The other is bulk traps, which are distributed uniformly throughout the Si-substrate. Moreover, they reported bulk traps are mainly measured in strong inversion region, and not measured in depletion region unless the value of Dit is less than 1010 eV-1cm-2. However, the peak in low frequency region even at E-Ei = 0.2eV is seen in Fig. 4.5(a). This means the peak is measured in depletion region too. So, the probability that GP/ω curves are caused by bulk traps is small. Therefore, it was considered that the peaks were caused by traps inside the oxide. Schematic illustration expressed how electrons are captured and emitted is shown in Fig. 4.7. Electrons transited from conduction band are trapped by traps near the surface. It is the interface state. And it is considered electrons transited from conduction band 29 are trapped by defects inside the oxide. It is supposed this oxide trap causes GP/ω curves. There are two reasons about this phenomenon. First, the peaks appear in not high frequency region but low frequency region. This means electrons cannot follow in the case of high frequency. Generally, capture and emission time constant is short when electrons can follow high frequency. Therefore, the place traps exist is near conduction band. On the contrary, in the case of low frequency, the place traps exist is far from conduction band. Second, the values of frequency in appearing the local maximum hardly change at any energy level. From Fig. 4.5(b), it is found that values of frequency in appearing the local maximum change about two digits when energy level changes 0.15 eV. On the other hand, they do not change even one digit in spite of changing 0.30 eV in Fig. 4.5(a). Therefore, because capture and emission time constant does not depend on surface potential much, it is considered electrons moves for horizon direction for a long time in Fig. 4.7. By the way, it is considered the equivalent parallel circuit in Fig. 2.12(a) is changed to that in Fig. 4.8. Here, Cot is generated capacitance when electrons are captured and emitted by the states inside the oxide, and Rot is generated losses then. It is supposed that oxide traps are located near the surface. And it is considered that the systems of capture and emission are also similar to those of interface traps. Therefore, Cot is the capacitance generated by changing the surface potential. And then, Eq. (2.6) is rewritten as ∞ [ ] q ⌠ Dit 2 = ln 1 + (ωτ it ) P(ψ S )dψ S ω 2 ⌡−∞ ωτ it GP ∞ [ ] q ⌠ Dot 2 + ln 1 + (ωτ ot ) P(ψ S )dψ S , 2 ⌡−∞ ωτ ot (4.1) where Dot is state density inside the oxide and τot is oxide trap time constant. In other words, if this assumption is correct, GP/ω curves are expressed the sum of interface traps and oxide traps. Thus, GP/ω curves are separated into interface trap and oxide trap. Therefore, they may be able to be described accurately, and we can fond interface state density even if any oxide is used. Also, because these peaks in low frequency 30 region are not always measured, it is considered we can remove them by proper processes. La2O3 Si EC Ef Interface trap EV Oxide trap Fig. 4.7 Energy-band diagram of interface trap and oxide trap. COX Gt CS Rit Rot Cit Cot Fig. 4.8 A new equivalent parallel circuit model of MOS capacitors. 4.4 Application of a new model to the measurement In order to confirm whether the model supposed in paragraph 4.3 is correct, calculation curves compared with measured curves as shown in Fig. 4.9. Calculation curves in Fig. 4.9(a) are based on the second term in Eq. (4.1), and these curves do not agree with measured plots even if values of σ are adjusted. It is found difference 31 between two is half breath. So, the assumption like Eq. (4.1) is probably a mistake, and it is considered the state caused by oxide defects is single energy level. Therefore, Eq. (4.1) is rewritten correctly as ∞ [ ] qωτ ot Dot q ⌠ Dit 2 = ln 1 + (ωτ it ) P(ψ S )dψ S + . 2 2 ⌡−∞ ωτ it ω 1 + (ωτ ot ) GP (4.2) Curves calculated from the second term in Eq. (4.2) are shown in Fig. 4.9(b). As a result, good agreement can be gained. Errors between calculation and measured curves around 1 kHz are caused by interface traps. Because of this reason, oxide defects are not influenced well by surface potential fluctuations. Fig. 4.10(b) shows curves calculated from the first term in Eq. (4.2). These curves are caused by interface state. And, it is found calculation agree with measurement in the region more than 10 kHz. On the contrary, errors between calculation and measured curves in the region less than 10 kHz are caused by oxide traps. Thus, it is found distribution of oxide defects have single energy level. Also, measured GP/ω curves are expressed by the sum of interface traps and oxide traps. Gp /ω (x10-7 F/cm2) 30 30 Statistic fitting 25 25 20 20 15 15 10 10 5 0 101 5 Open dots : Measurement Solid lines : Calculation 102 Single level fitting Open dots : Measurement Solid lines : Calculation 0 103 101 102 Frequency (Hz) (a) (b) Fig. 4.9 Calculation curves compared with measured curves caused by oxide traps. (a) Calculation curves are based on statistic model. (b) Calculation curves are based on single level model. 32 103 Gp /ω (x10-7 F/cm2) 5 Statistic fitting 4 Open dots : Measurement Solid lines : Calculation 3 2 1 0 103 104 105 106 107 Frequency (Hz) Fig. 4.10 Calculation curves based on statistic model compared with measured curves caused by interface traps. 33 5 Conclusions In this study, interface state and oxide defects are estimated by conductance method. In consequence, peaks of GP/ω curves caused by interface state are measured in SiO2/Si structure. However, the signal responding low frequency is measured in high-k/Si structure. It is considered they are caused by defects inside the oxide. Because the traditional equivalent parallel circuit model is considered only interface traps, new one is supposed in order to estimate that in detail. As a result, separating GP/ω curves into interface state and oxide defects is succeeded. It is considered that the quantification of interface traps and oxide traps is very effective in order to discuss interfacial properties of MOS devices. 34 Appendices A. The derivation of calculation changing equivalent circuit model to simplified circuit The admittance except COX and Gt in Fig. 2.12(a) is given by Ya = jωC S + = jω C S + = jω C S + 1 Rit + 1 jωC it jωC it 1 + jωτ it jωqDit (1 − jωτ it ) 1 + (ωτ it ) 2 , (A.1) where we replace Cit = qDit, and τit = CitRit. And, the admittance except COX and Gt in Fig. 2.12(c) is given by Yc = jωC P + G P . (A.2) The real part and imaginary part in Eq. (A.1) and (A.2) are compared respectively, so that we can obtain CP = CS + GP ω = qDit 1 + (ωτ it ) qωτ it Dit 1 + (ωτ it ) 2 2 , . (A.3) (A.4) B. The derivation of calculation changing measured circuit to simplified circuit First, Equivalent parallel circuit is partially replaced impedance and admittance in order to explain simply like Fig. B.1. Because the total impedance and admittance of these two circuits are equal each other, the equations are given by Yt = 1 + Gt = Gm + jωC m , Z0 Z0 = 1 1 . + jωC OX G P + jωC P 35 (B.1) (B.2) We obtain next equation from Eq. (B.1) and (B.2). 1 1 1 = − G P + jωC P Gm − Gt + jωC m jωC OX (B.3) The reciprocal of Eq. (B.3) is G P + jωC P = = 1 1 1 − Gm − Gt + jωC m jωC OX jωC OX (Gm − Gt + jωC m ) jωC OX − (Gm − Gt + jωC m ) =− =− jωC OX (Gm − Gt + jωC m ) (Gm − Gt ) − jω (COX − C m ) jωC OX ⋅ jω [C m (G m − Gt ) + (C OX − C m )(Gm − Gt )] (Gm − Gt )2 + ω 2 (COX − Cm ) 2 + jω L ω 2 (Gm − Gt )COX 2 = + jω L . (Gm − Gt )2 + ω 2 (COX − C m )2 (B.4) Because the amount of calculation is much, the calculation of imaginary part was omitted. The real part in Eq. (B.4) is compared, so that we can obtain GP ω = ω (Gm − Gt )COX 2 . (Gm − Gt )2 + ω 2 (COX − C m )2 (B.5) Z0 COX Gt CP Cm Gm GP Yt Fig. B.1 Equivalent parallel circuit is partially replaced impedance and admittance in order to explain simply. 36 C. The derivation of calculation changing measured circuit to equivalent circuit model First, we change Fig. 2.12(c) to (b) contrary to Appendix B. We want information of Cm, so the calculation of real part is omitted. In Fig. B.1, we obtain next equation from Eq. (B.1) and (B.2). Gm + jωC m = 1 1 1 + G P + jωC P jωC OX + Gt = jωC OX (G P + jωC P ) + Gt jωC OX + (G P + jωC P ) = jωC OX (G P + jωC P ) + Gt G P + jω (C OX + C P ) = [ ]+ G +L jωCOX G P + ω 2 C P (COX + C P ) 2 G P + ω 2 (C OX + C P ) 2 2 t (C.1) The imaginary part in Eq. (C.1) is compared, so that we can obtain GP 2 COX + C P (COX + C P ) ω Cm = . 2 GP 2 + (COX + C P ) ω 37 (C.2) References [1] Y. Taur, T. H. Ning: “Fundamentals of Modern VLSI Devices Second Edition”, p.98, CAMBRIDGE UNIVERSITY PRESS, New York (2009). [2] S. Saito, K. Torii, M. Hiratani, T. Otani: ”Improved theory for remote charge scattering limited mobility in metal oxide semiconductor transistors”, Appl. Phys. Lett., Vol.81, No.13, p.2391 (2002). [3] Y. Taur, T. H. Ning: “Fundamentals of Modern VLSI Devices Second Edition”, p.101, CAMBRIDGE UNIVERSITY PRESS (2009). [4] ITRS, “INTERNATIONAL TECHNOLOGY ROADMAP FOR SEMICONDUCTORS 2009 EDITION FRONT END PROCESSES”, http://www.itrs.net/Links/2009ITRS/2009Chapters_2009Tables/2009_FEP.pdf [5] J. Robertson: “Interface and defects of high-k oxides on silicon”, Solid-State Electronics, Vol.49, Issue3, pp.283-293 (2005). [6] J. A. Ng, et al.: “Effective mobility and interface-state density of La2O3 nMISFETs after post deposition annealing”, IEICE Electronics Express, Vol.3, No.13, pp.316-321(2006). [7] K. Tsutsui: “Yoku wakaru denshi debaisu”, p.143-156, Ohmsha (1999). [8] T. Watanabe: “Denshisen jochaku · teikou kanetsu jochaku sochi”, http://www.khlab.msl.titech.ac.jp/~www/facilities/EBeamEvaporator.html. [9] K. Nishioka: “Hannosei ion etching wo mochiita Si to SiO2 no etching”, http://www-eng.kek.jp/meeting09/proceedings/pdf/h21gp404.pdf. [10] Dieter K. Schroder: “Semiconductor Material and Device Characterization 3rd Edition”, p.347-350, Wiley Interscience, New York (2006). [11] Nicollian & Brews: “MOS Physics and Technology”, p.200, Wiley Interscience, United States of America (2003). [12] E. H. Nicollian and E. Goetzberger: “The Si-SiO2 Interface – Electrical Properties as Determined by the Metal-Insulator-Silicon Conductance Technique”, Bell System Journal, Vol.46, p.1055 (1967). [13] Nicollian & Brews: “MOS Physics and Technology”, chapter 5, Wiley Interscience, United States of America (2003). 38 [14] M. Inoue, J. Shirafuji: “a.c. Conductance Measurement of MOS Diodes Degraded by FN Injection”, Institute of Electronics, Information, and Communication Engineers, Technical Report of IEICE, SDM94-154, p.101 (1994-11) [15] Y. Taur, T. H. Ning: “Fundamentals of Modern VLSI Devices Second Edition”, p.26, CAMBRIDGE UNIVERSITY PRESS (2009). [16] Nicollian & Brews: “MOS Physics and Technology”, p.111, Wiley Interscience, United States of America (2003). 39 Acknowledgements First of all, the author would like to express gratitude to his supervisor Prof. Hiroshi Iwai. The author could gain the results of this study because of his continuous encouragement and advices. He also gave the author the opportunity to attend the conference. The experience is precious for the author’s life. The author also deeply thanks Prof. Takeo Hattori, Prof. Kenji Natori, Prof. Nobuyuki Sugii, Prof. Akira Nishiyama, Prof. Kazuo Tsutsui, Associate Prof. Parhat Ahmet, and Assistant Prof. Kuniyuki Kakushima for giving him useful advices and great help whenever he faced difficult problems. The author also thanks the members of Iwai Lab. for their friendship, active many discussions and many of encouraging words. The author also thanks secretaries, Ms. Matsumoto, Ms. Nishizawa, and Ms. Karakawa for supporting him and the members of Iwai Lab. Finally, the author would like to thank his parents Akira and Chieko and his brother Naoki for endless support and encouragement. 40