Status and Prospects For VUV Ellipsometry (Applied to High K and Low K Materials) N.V. Edwards Advanced Products Research and Development Laboratory, Semiconductor Products Sector, Motorola Inc. Abstract. The recent commercialization of Vacuum Ultraviolet spectroscopic ellipsometry (VUV SE) instruments means that it is now possible to routinely perform SE measurements at wavelengths below 190 nm. This new capability has obvious implications for lithographic work but also for the characterization of other materials of importance to the Si industry. These are materials that are nominally transparent at long wavelengths but that possess unique absorption signatures in the VUV, such as newly emerging high-k gate materials (e.g. Al2O3, HfO2, ZrO2, Y2O3) and low k materials (porous SiO2, organo-silicate glasses), as well as more familiar dielectrics (e.g. SiOxNy, Si3N4, SiOF, and TEOS). We provide a review of recent progress and a critical assessment of the capabilities of VUV SE with respect to a selected examples of these materials, with special emphasis on low k and high k materials. These capabilities include increased access to unique VUV spectral features as a means of tuning process parameters and increased ability to determine the thickness of thin films grown on Si. We also address the initial challenges that had to be overcome in order to develop optical constants at short wavelengths and to enable this sort of materials characterization. nominally transparent dielectrics and in terms of increased sensitivity for measuring thin films. We will conclude with a critical assessment of measurement capability for high and low k dielectric materials. For high k materials, we will assess the ability of the technique to simultaneously determine optical constants and film thickness for the very thin films required by increasingly shrinking device dimensions. The sensitivity of ellipsometric fitting routines to the presence of interface layers is a necessary part of this discussion, and the ability to measure band gaps of new gate materials when good values for optical constants are obtained will be discussed as well. For low k materials, a special emphasis will be given to issues concerning the ability to measure porosity and electron density, in the context of the complicating presence of low index inclusions. Further, we hope to elucidate the origins of the issues, as the limitations that we encounter are less a consequence of working in a new spectral region but are rather more endemic to SE measurements and data analysis routines in general. INTRODUCTION Scope of Work With the economic impetus provided by 157nm lithographic applications, Vacuum Ultraviolet Spectroscopic Ellipsometry (VUV SE) has made a recent transition from a synchrotron-based characterization technique more suitable for basic research to a workhorse for industrial problem solving. For this to occur, significant challenges associated with instrumentation and data analysis had to be solved. The work presented here is not simply a general review of progress toward this transition; rather, it will also be a critical assessment of the capabilities of the technique with respect to key microelectronic applications. In other words, in the context of advanced gates and interconnects, what can be measured and what cannot? During the course of answering these questions, a good portion of the evolution of the technique will be discussed by default. Toward this end we will provide a quick introduction to ellipsometry, followed by a discussion of the initial challenges faced by those first using the technique in an industrial setting. We will demonstrate the advantages of VUV SE with respect to increased access to unique spectral features for Introduction to Ellipsometry We are perhaps most familiar with ellipsometry in its incarnation as an in-line metrology CP683, Characterization and Metrology for ULSI Technology: 2003 International Conference, edited by D. G. Seiler, A. C. Diebold, T. J. Shaffner, R. McDonald, S. Zollner, R. P. Khosla, and E. M. Secula © 2003 American Institute of Physics 0-7354-0152-7/03/$20.00 723 k 2 1 ε 1+ 2 (2νεσ )2 1 = 2 2 ε1 Real(Dielectric Constant), HfO2 2.0 3.0 1.5 2.5 1.0 2.0 0.5 1.5 0 2 4 6 8 Photon Energy (eV) Index of refraction ’n’ 2.8 (b) 0.60 Al2O3 n k 2.4 2.0 1.6 0 0.0 10 0.40 0.20 300 600 900 1200 Wavelength (nm) 1500 0 1800 ‘K’ (2νεσ )2 2.5 ε1 ε2 Extinction Coefficient 1 ε 1+ 2 1 = 3.5 ε2 n 2 (a) 4.0 Imag(Dielectric Constant), tool, where index of refraction n and sample thickness d are delivered at 633nm. Ellipsometry is far more versatile than this single application, however. It can be operated in static or dynamic mode, at single wavelengths or over wide spectral ranges. Traditionally, spectroscopic ellipsometry (SE) has been performed from 1770 to approximately 190 nm, and in static mode, it is heavily used for material diagnostics, e.g., determination of band gap, alloy composition, porosity and strain.1,2 Measurements in dynamic mode range from such diverse applications as the control or monitoring of semiconductor thin film growth, etching processes and deposition of proteins on semiconductor surfaces. See Refs. 1 and 2 for a an excellent introduction to SE capabilities and system configurations. SE is an optical characterization technique that is used to determine the complex reflectance ratio ρ = rp rs-1 where rp and rs are the complex reflectances of light polarized parallel (p) and perpendicular (s) to the plane of incidence, respectively. From ρ, optical constants are determined. These are the quantities required by device designers, crystal growers, and manufacturing engineers: index of refraction n, extinction coefficient k, and dielectric function ε. These are related to one another by the following: FIGURE 1. Optical functions in the spectral range associated with traditional spectroscopic ellipsometry. (a) Real (ε1) and imaginary (ε2) parts of the dielectric function for a monoclinic hafnia film. (b) Real (n) and imaginary parts (extinction coefficient k) of the index of refraction for a c-plane oriented Al2O3 (ordinary component). polycrystalline and structurally inhomogeneous materials.2 With SE, film thicknesses can be determined, properties of buried interfaces and surfaces can be studied, and measurements can be performed in any transparent ambient. Until recently, however, none of this could be done in an industrial setting at wavelengths below 190nm—which marks the beginning of the spectral range commonly referred to as the “Vacuum Ultraviolet.”3 + 1 − 1 where ν and σ are conductivity of the material and the frequency of the impinging plane wave of radiation, respectively. It should be pointed out that n and ε are complex quantities, where n = n + ik and ε = ε1 + iε2. When optical functions (i.e. optical constants given as a function of energy or wavelength) are reported, it is customary to give both the real and imaginary part of ε or n, as shown in Fig. 1 (a) and (b), respectively. For in-line metrology tools, values at 633nm (or ~2 eV) are typically given. Here the extinction coefficient k is 0 for the majority of materials measured with these instruments, given that most common dielectrics and wide bandgap materials are transparent at this wavelength. Consistent with the classic definition of ε from freshman physics (where ε is related to the dipole moment per unit volume), the sensitivity of SE to the presence of long range order on the scale of ~10 Å to 100 Å means that the technique is a non-destructive means of measuring the density of amorphous, The Industrial Need for Measurements in the VUV Regardless of system configuration, ellipsometry had been traditionally performed in air and with quartz optical elements, both of which are absorbing at spectral wavelengths below 190 nm. The exception was the Ultra-High Vacuum SE system at the BESSY I synchrotron in Berlin, which had provided a means for researchers to relate newlymeasured optical features in the VUV with the electronic bandstructure of novel wide bandgap materials.4 Routine, high throughput determination of short wavelength optical constants for materials of interest to the Si industry, e.g., photoresists, ARC layers, high-k gate materials, photomasks, passivation layers, or interlayer dielectrics—either at accuracies demanded by manufacturing and production environments or with the large, widely varying sample sets needed to properly tune process parameters—had 724 associated with microscopic polarization phenomena into the macroscopic sample properties of interest: thin film thicknesses, optical constants, compositions, porosities, etc. In other words, we cannot use ellipsometry to directly measure much of the information of interest to industrial customers. An ‘optical model’ is needed to bridge the gap. This model is an ideal mathematical representation of the sample that allows us to calculate its polarization state change in terms of physical properties like thickness, refractive index, and composition. For each sample, ρ is calculated to match the experimental data, using multilayer mathematical models that contain parameters, such as layer thickness or composition, which are independent of the parameters varied in the measurements. Model parameters and their confidence limits are then obtained by linear regression analysis with the commercial software packages that accompany ellipsometric instruments.3 A cartoon of this process is given in Fig. 2. Here, a general representation of an ellipsometer is given in the top portion of the figure, with χi and χf representing the initial and final polarization states, respectively. ρ can also be determined from χ: ρ = rp rs-1 as well as ρ = χi/ χf. How ε is calculated from ρ is more complicated, however. A brief explanation of this process will also explain why there were initial challenges associated with obtaining optical constants in the VUV. not been achieved due to a lack of instrumentation feasible for use in an industrial setting. 3 Entrance Optics χi ϕ Sample Detector Exit Optics χf χ→ρ→ε 131 to 1770 nm Experimental Data 80 300 ExpΨ-E 65° Exp∆-E 65° 60 200 Ψ in degrees ∆ in degrees 40 100 20 0 0 0 2 4 6 8 Photon Energy (eV) -100 10 Model Sample Properties: d, n, k, ε composition roughness bandgap porosity FIGURE 2. Schematic diagram of ellipsometer (top) and the ellipsometric modeling process (bottom). The need for short wavelength optical constants by those designing optical elements for 157 nm lithography provided the economic motivation for commercial development of such instrumentation. Optical constants at 157nm are required for the design of photoresists and pellicles, but also for simulations that optimize (either minimizing or maximizing) reflectivity for the design of ARCs and other mask components for increased inspection contrast at multiple wavelengths.5 The recent introduction of commercial VUV SE instruments means that it is now possible to routinely perform spectroscopic ellipsometry measurements at wavelengths below 190 nm. The development of enclosed, nitrogen-purged systems and the use of deuterium lamp sources and MgF2 optical elements have extended the measurement range of combined IR/ VIS/ VUV systems to span from as low as 131 nm to as high as 1770 nm, with relatively fast data acquisition times on the order of several hours.3 This, however, did not guarantee that accurate optical constants and materials diagnostics readily followed from such measurements as a matter of course. 3-phase model: ambient εa overlayer εo d↕ a) substrate εs overlayer εo b) ε substrate εs = ε s + 4 π id λ n a ε s (ε } <ε> } Source εs )( − ) − s ε o ε o ε a ε s − sin ε ε o (ε s − ε a ) a 2ϕ INITIAL CHALLENGES FIGURE 3. (a) Schematic diagram of the pseudodielectric function. (b) On left, what model assumes (mathematically sharp interfaces). On right, what is encountered in reality (interface layers, contamination, surface roughness). The expression for the pseudodilectric function in the three-phase model is given below. Introduction to SE Data Analysis Advances in instrumentation aside, reducing data to determine optical constants in the VUV at the throughputs and accuracies demanded in industrial settings was initially very difficult. This had less to do with challenges specific to the VUV and more to do with the inherent nature of SE. The general challenge of ellipsometry is to convert measured quantities In the case of multilayer samples, either an accurate mathematical description of the optical dispersion of each layer must be generated or optical constants must be known for each layer before a good 725 model can be constructed. 6 We will see that this refers to intentionally deposited layers as well as unintentional layers, such as interface layers, surface roughness and contamination. This follows from the basic definitions of ε. In the two–phase model2 (where the two phases are defined as the substrate and ambient), we can write an expression for the dielectric function of the substrate in terms of the experimental parameters χi, χf and the angle of incidence φ: 1− ρ ε s = sin ϕ + sin ϕ tan ϕ 1 + ρ 2 2 εs is the foundation for the entire ellipsometric modeling process. In an industrial characterization laboratory, where most materials submitted for analysis are multilayer samples consisting of materials grown on Si substrates, the lack of optical constants for Si in the VUV presented a major stumbling block that had to be removed before analysis of the materials of interest could proceed. Since the optical properties of Si at lower spectral energies are well understood, it was possible to follow the example of previous workers to obtain VUV optical constants. After the work of Herzinger, et al.,7 a thermal oxide series was generated on Si, consisting of nine samples, ranging in nominal oxide thickness from 20 Å to 2200 Å. VUV SE measurements were performed from 131 nm to 1770 nm at angles of incidence ranging from 40° to 75°. The resulting optical constants for Si are shown in Fig. 4 and agree well with the prior work performed at lower energies.8, 9, 10 The inset is a magnification of the VUV region, where a new critical point was discovered. It likely corresponds to a direct, band-to-band transition between the X1V-X1C valence and conduction bands, respectively. 11 2 2 In ellipsometric terms, the mechanism for accounting for other layers is via the three-phase model.2 This consists of a simple substrate/ overlayer/ ambient system where mathematically sharp interfaces are assumed, as shown in Fig. 3a. Reality often consists of that shown in Fig. 3b: interface layers, oxides, roughness and other non-idealities. The full expression for the three phase model is given in Fig. 3, where the three phases are analyzed as two, resulting in the combination of εo and εs into the pseudodilectric function <ε>. The pseudodilectric function is a weighted average of the heterostructure layers penetrated by the measurement, which of course is a function of wavelength. Hence, in the modeling process, interface layers become more important for long wavelength measurements and surface roughness becomes more important for short wavelength measurements. Since penetration depths of most semiconductors are on the order of several hundred Ångstroms in the VUV,2 this can seriously complicate the measurement of thin films, as we will see in the upcoming discussion. At any rate, we introduce the concept of the three-phase model to show how overlayers are accounted for in calculations of the dielectric function and in doing so, to emphasize that models must account for all heterostructure layers present—whether intentional or unintentional-–or risk the introduction of large uncertainties into the determination of optical constants. Indeed, if the model does not accurately describe all of the layers present in the sample, then the interpretation of the data is seriously compromised—even if the data themselves are of the highest quality.3 pseudodielectric function 2.8 Si 50 2.3 New Critical Point: X1v-X1C transition 30 1.8 7.2 ε1 10 7.7 8.2 ε2 Motorola11 -10 Aspnes8 Herzinger7 Jellison9 Yasuda10 -30 0 2 4 6 8 10 energy in eV FIGURE 4. VUV optical constants for Si (black curve) shown in the context of prior work at lower spectral energies. The inset shows a magnification of the VUV spectral region where a new transition corresponding to a direct, band-to-band transition between the X1V-X1C valence and conduction bands was discovered. (Originally commissioned by Semiconductor Fabtech and printed in Edition 18, 2003.) These optical constants were obtained with: (1) the Gaussian-Broadened Polynomial Superposition (GBPS) parametric dispersion model of Herzinger and Johs 12 to represent the optical dispersion of the Si layer; (2) a Tauc-Lorentz model13 (which is typically used to describe the optical dispersion of amorphous materials) to represent a Si-SiO2 interface layer; and (3) a Tauc-Lorentz model to represent the SiO2 layer. The Need for VUV Optical Constants for Si The complexity of building ellipsometric data analysis models that represent the multiplicity of layers present in actual samples also strongly emphasizes the importance of understanding the optical dispersion of the substrate layer. Knowledge of 726 samples and a common fit was performed. In Fig. 6 a and b we show the results of the common fit for the thinnest and thickest sample, respectively, to illustrate the validity of the assumption that the optical constants for Si and SiO2 are largely the same across the sample set. (Fits for samples of intermediate thickness are shown later in the discussion, in the context of increased sensitivity to film thickness for the measurement of thin films.) While it is reasonable to grant that the interface layer may vary for samples of different thickness, grown by under slightly different deposition conditions, we did not have sensitivity to this degree of variation in the samples. Indeed, an interface layer of 7 Å was initially assumed for all of the samples, but this coupled thickness was eventually released in the fitting and a coupled thickness value of 9.4 was obtained. Further details will be discussed elsewhere, as this work is still in progress. The mathematical formalism for these is described further in Refs. 12 and 13, but most are part of the standard recipes included in commercial ellipsometric data analysis software (or can be readily programmed). In this case, a multi-sample analysis was performed. This type of analysis is predicated on the assumption that the optical constants for the materials are the same in each sample, independent of thickness. The data analysis is performed simultaneously for all samples, with the optical constants of each material coupled in all models for all of the samples. Thus it is possible to reduce strong correlations that often occur between fit parameters.12 SiO2: Tauc-Lorentz oscillator Amp= 40.024, En= 10.643, C= 0.72608, Eg= 7.5258 Pole 1: Pos= 13.167, Mag= 94.386 Pole 2: Pos= 0.135, Mag= 0.0127 E1 offset= 1.263 Interface Layer: Tauc-Lorentz oscillator Amp= 158.67, En= 10.643, C= 0.72608, Eg= 7.5258 Interface Layer Pole 1: Pos= 13.167, Mag= 94.386 Pole 2: Pos= 0.135, Mag= 0.0127 Si: Parameterized Semiconductor Layer 2.00 FIGURE 5. Additional details concerning fit parameter for data ellipsometric models for analysis of Si and SiO2 VUV SE data. (a) Data: ε1, blue ε2, green Model: red Thinnest Sample: SiO2 Int. Layer 9.4 Å 100 (b) Si Substrate Ψ in degrees Int. Layer 9.4 Å 200 60 100 40 1.80 0.040 1.70 0.030 1.60 0.020 1.50 0.010 2 4 6 Photon Energy (eV) 8 0 300 600 900 1200 Wavelength (nm) 1500 0.000 1800 Since SiO2 is also an important material for the microelectronics industry—and as it is a byproduct of the analysis used to obtain Si optical constants—these optical constants are shown in Fig. 7. Note the absorption edge near 160 nm; this edge cannot be seen with traditional IR/ VIS/ UV ellipsometry. 3 0 20 0 0 ∆ in degrees 2189.3 Å 300 80 Thickest Sample: 0.050 FIGURE 7. VUV optical constants for SiO2: index of refraction n (solid) and extinction coefficient k (dashed). (Originally commissioned by Semiconductor Fabtech and printed in Edition 18, 2003.) Si Substrate SiO2 1.90 1.40 7.5 Å 0.060 n k SiO2 Extinction Coefficient k Index of refraction n E1 offset= 1.5705 -100 10 FIGURE 6. Sample fits for thin (a) and thick (b) SiO2 samples grown on Si substrates. The fit was a by-product of a multisample analysis that involved the simultaneous fitting of nine thermal oxide samples of various thickness. ADVANTAGES AND APPLICATIONS OF VUV SE Additional information about fit parameters is given in Fig. 5. Here the parameter values for the Tauc-Lorentz oscillators used for the interface and SiO2 layers are shown. All parameters in the interface layer of a given sample were coupled to the corresponding parameter in the SiO2 layer of the same sample, with the exception of the Amplitude and E1 offset. (This was done in order to allow the index of the interface layer to vary.) As well, all of the parameters were subsequently coupled for the nine With the optical constants of the substrate and its oxide determined, it became possible for us to analyze a wide variety of materials grown on Si substrates. These materials span a wide range of industrial applications—in addition to the obvious lithographic work for which the VUV technique is so well suited. We have selected a few examples designed to illustrate the capabilities of VUV SE compared to standard IR/ VIS/ UV systems. Since our own system is a combined IR/ VIS/ VUV system, 727 constants to distinguish between nominally similar organo-silicate glasses in order to correlate optical and mechanical properties for interlayer dielectric applications; 14 to correlate optical and electrical properties for SiNx films for MIM capacitor applications15; and for the development of novel ARC layers to enable the inspection of reticles for 157nm and EUV lithography.5 An example from the latter application is given in Fig. 8. Here the extinction coefficient k vs. wavelength is plotted for a selection of SiOxNy films of different compositions (as determined by Rutherford Backscattering16) deposited on Si. At wavelengths above ~200 nm, all of the samples are transparent. Accordingly, k = 0.17 In the VUV, the optical properties of the films are all very different and composition dependent (the red labels provide a summary of the RBS results). Not surprisingly, films without nitrogen have optical properties similar to SiO2. As the oxygen and nitrogen concentration increases, optical structure is introduced into the extinction coefficient lineshape. The greater extinction coefficient seen for these films reflects the fact that the these films are more absorbing, as absorption coefficient α is related to the extinction coefficient k by the expression k = (λ/ 4π) • α.17 Clearly one must be careful to select the appropriate nitrogen and oxygen composition of SiOxNy films if they are to be used for 157nm or other lithographic applications.3 spanning a spectral range of 131 nm to 1770 nm, we obviously view the two systems as complementary. The addition of the VUV capability, however, has some distinct advantages. Extinction Coefficient k 0.5 SiOx Ny 0.4 0.3 Equal Parts Si, N, O 0.2 Si: mid 30% O: high 40% N: high teens 0.1 No Nitrogen 0 100 150 200 250 300 Wavelength (nm) FIGURE 8. Extinction coefficients in the VUV for a series of SiOxNy films of varying compositions. Compositional information (as determined by RBS) is given by the red labels. Note the large differences seen in the VUV between the films. At longer wavelengths, the films are transparent and indistinguishable. (Originally commissioned by Semiconductor Fabtech and printed in Edition 18, 2003.) Increased Access to Unique Spectral Features Many of the materials used in Si processing are nominally transparent at long wavelengths: SiOxNy, organo-silicate glasses, Si3N4, SiOF, and TEOS are familiar examples. Additionally, many of the new high-k gate materials under consideration by the semiconductor industry—e.g., Al2O3, HfO2, ZrO2, Y2O3 –are wide bandgap materials and can be included in this category as well. This is quite convenient for inline optical metrology, as it allows for the use of relatively simple mathematical formalisms to build optical models and recipes for use at 633nm. However, this also means that it can be very hard to distinguish between films of the same material grown under slightly different conditions, based on what can often be very small changes of index of refraction at and near 633nm. For these materials, gaining access to the absorption behavior in the VUV gives us a foothold for additional materials diagnostics; ie. we now have a way to distinguish between films that otherwise appear very similar when measured with a standard IR/ VIS/ UV system or in-line metrology tool. 3 We have found this newly available access to VUV absorption behavior useful for tuning process parameters to address mechanical, electrical and optical performance issues of blanket depositions of many materials used throughout the CMOS process. Indeed, VUV SE has been used to determine optical 10 HfO2 8 ε1 6 monoclinic 4 amorphous 2 0 0 2 4 6 8 10 8 6 ε2 amorphous 4 monoclinic 2 0 0 2 4 6 8 10 Energy in eV FIGURE 9. The real and imaginary part of the dielectric function ε = ε1 +ε2 are plotted vs. energy for an as-deposited HfO2 film (blue curve) and for two HfO2 films that have been annealed post-growth (black and yellow curve). In the VUV, large differences between the annealed and as-grown films emerge. At lower energies, the films are transparent and indistinguishable. (Originally commissioned by Semiconductor Fabtech and printed in Edition 18, 2003.) 728 We observe similar behavior for HfO2, potentially of interest for high-k gate applications. Optical constants for three HfO2 films are shown in Fig. 9.18 The real and imaginary parts of the dielectric function ε = ε1 + i ε2 are plotted vs. energy for an asdeposited film (blue curve) and for two films that have been annealed post-growth (black and yellow curve). Just below 6 eV (above ~200nm) the material is transparent, deducible from the fact that ε2= 0 (analogous to k=0).17 Note that the differences in ε1 are very small in this region, which makes it difficult to distinguish between the films by optical means. In the VUV, large differences between the annealed and asgrown films emerge: the as-grown film lacks the optical structure apparent for the other films, important because these features are related to direct band-toband electronic transitions identifiable in HfO2 band structure. X-ray diffraction analysis19 indicated that the as-grown film was in fact amorphous while the annealed films were of the monoclinic phase, meaning that the relative lack of optical structure in the one case can be accounted for by differences in electronic structure expected for an amorphous vs. crystalline film. Thus we have a means to differentiate between HfO2 films grown under different conditions and for interpreting the electronic structure of the different phases of HfO2 that often emerge with different growth conditions in these films.20 100 Model Fit Data 65° Data 70° Data 75° 185 Å Al2O3 80 Ψ in degrees Si Substrate 60 40 20 0 0 2 4 6 Photon Energy (eV) 8 10 FIGURE 10. Ellipsometric data (green dotted lines) and model fit (red solid line) for a 185 Å thick Al2O3 film, at three angles of incidence. The spectra for this relatively thin film are dominated by a single interference oscillation near 6.7 eV, highlighted with the red arrow. The location of this oscillation in the VUV illustrates the extra capability that the instrument provides for measuring thin films on absorbing substrates. (Originally commissioned by Semiconductor Fabtech and printed in Edition 18, 2003.) with the red arrow. The location of this oscillation in the VUV illustrates the extra capability that the instrument provides with respect to measuring thin films on absorbing substrates. If a film is sufficiently thick to build up one cycle of interference, we can exploit the classic conditions for thin-film interference to determine optical thickness of the sample, assuming that the film is transparent. In the three-phase model, the phase θ is given by 2k0⊥d, where wavevector k=2π/λ.6 For one cycle of interference (i.e. one interference oscillation), the phase change ∆θ is given by 2π = ∆θ = (4πnd/ hc) ∆E, where E is spectral energy, h is Planck’s constant, and c is the speed of light. Since ∆E = -(hc/ λ2) ∆λ, we can substitute the latter equation into the former and determine that λ2 /∆λ = -2nd. Note that for a thin film, the quantity defining the spectral width ∆λ of the interference oscillation will be relatively large, as thick films produce more oscillations in the same spectral energy range than a thin film. Therefore, neglecting optical dispersion and angle of incidence effects, we observe that for small values of film thickness d, interference—or that which enables us to determine the optical thickness of the film— will occur for small values of λ. We demonstrate in Fig. 11, which features ellipsometric data (data shown in green; model fit in red) vs. energy for a series of increasingly thick SiO2 films grown on Si. For the thinnest film (95Å shown in Fig. 6a), the interference oscillation is located just outside of our spectral range; only the low energy edge is visible near 9 eV (cf. red arrow). This film is too thin to determine n and d independently; here we used a multi-sample analysis approach to Increased Sensitivity to Film Thickness Accompanying the increased access to the VUV spectral range is an increased ability to determine film thickness for thin films deposited or adsorbed on substrates. The more rigorous arguments justifying this premise are based on the solution of the Fresnel reflectance expressions for a three-phase model (i.e. ambient/ film/ substrate)21 for the optical thickness (nd) of the sample in terms of the ellipsometric parameter ρ and angle of incidence φ. From this analysis, the concept of a thickness period Dφ is developed, where shorter wavelengths will move through one complete period for smaller values of film thickness, since the thickness period is directly proportional to spectral wavelength. The reader is referred to Refs. 22 and 23 for a more complete treatment. This principle can also be demonstrated empirically. In Fig. 10 we show ellipsometric data (green dotted lines) and model fit (red solid line) for a 185 Å thick Al2O3 film, a material of interest for highk gate applications. The ellipsometric angle Ψ is plotted vs. energy at three angles of incidence. The spectra for this relatively thin film are dominated by a single interference oscillation near 6.7 eV, highlighted 729 one of the monoclinic hafnia films18 discussed previously. When optical constants are obtained for a material, it is possible to use them in conjunction with the Si substrate optical constants to simulate ellipsometric data. This was done to simulate the equivalent data for a “series” of films deposited on Si, all of varying thickness but with the same optical constants. We do so in order to determine a projected range for the minimum thickness required to determine n and d independently for the hafnia film, as per our earlier discussion. The results of the simulation are shown in Fig. 12. Here ellipsometric angles (Ψ) were simulated (assuming a Si substrate) to create a series of curves analogous to the data series in Fig. 11. In this case, the interference oscillation of interest arises near 6.5 eV for a 50 Å film deposited on Si. In other words, for simulated thicknesses less than 50 Å one cycle of interference has not yet been completed. Since projected thicknesses for hafnia gates are in the 40 to 50 Å range, for this particular hafnia film we are unable to determine n and d independently (or could barely do so in the best case). This casts doubt on the ability of VUV SE to measure the thickness of thin hafnia gate material unless the optical constants are previously determined. reduce correlations between fit parameters. We are, however, able to de-correlate d and n for a SiO2 145 Å film on Si. In Fig. 11b, for a 145 Å film the entire interference oscillation is contained within our spectral range, exhibiting a maximum at 9eV. The 500 Å film (11c) is sufficiently thick so as to generate two interference oscillations. The first oscillation is now located near 5 eV, well within the measurement range of standard IR/ VIS/ UV systems. The extended spectral range of VUV instruments therefore provide an increased ability to determine the thickness of thin films grown on Si. 3 (a) Ψin degrees 80 95 Å SiO2 60 * Si Substrate Model Fit Data 40° 45° 50° 55° 60° 65° 70° 75° 40 (b) Ψin degrees 20 0 80 145 Å SiO2 60 Si Substrate 40 20 60 50 500 Å SiO2 40 20 0 0 *All model fits include a 9.4 Å interface layer 2 HfO2 40 Si Substrate Psi (c) Ψin degrees 0 80 4 6 8 Photon Energy (eV) 10 Ang 20 Ang 30 30 Ang 20 40 Ang 50 Ang 10 60 Ang 0 10 0 2 4 6 energy in eV FIGURE 11. Ellipsometric data (data shown in green; model fit in red) vs. energy for a series of increasingly thick SiO2 films grown on Si: (a) 95 Å; (b) 145 Å ; (c) 500 Å . The interference oscillation that is located just outside of the VUV spectral range moves into view (cf. red arrow) as the SiO2 film thickness increases. This enables the determination of film thickness. (Originally commissioned by Semiconductor Fabtech and printed in Edition 18, 2003.) 8 10 Projected gate thickness 40-50Å FIGURE 12. Three-phase model calculations of ellipsometric data for hafnia films grown on Si substrates. ∆d= 10 Å. Note that the interference oscillation needed to allow for the independent determination of n and d occurs at only 50 Å. However, one could argue that this problem should be almost trivial to solve, after the initial research on optical constants are performed on thicker films. After all, this is the approach taken with in-line metrology tools for the measurement of thin films of SiO2 on Si. As well, a multi-sample analysis could be undertaken to aid in this fundamental work, similar to work done in the Si/ SiO2 system. There are two strong objections to this line of thinking for hafnia. First, hafnia films have been observed to change phase with thickness.19 The assumption that optical constants are the same for films of varying thickness both invalidates the main premise of the multisample analysis and prevents the STATUS AND PROSPECTS Critical Issues for High- k Materials The previous discussion is a convenient segue into the topic of critical issues for determining optical constants and film thickness of new high k dielectric materials—where very often the thickness of the films required for gate applications are insufficiently thin to allow for the independent determination of n and d. We illustrate with the optical constants obtained for 730 facile application of in-line metrology tool recipes predicated on the analysis of thicker films. Second, because the films under consideration are relatively thin, both the interface layer between film and substrate and surface roughness layer becomes a significant portion of the measured film stack in the VUV (recall the argument that penetration depths of most semiconductors are on the order of a few hundred Ångstroms in the VUV). This means that obtaining optical constants, even for relatively thick films (i.e. on the order of several hundred Ångstroms) is nontrivial. We illustrate by examining the fitting procedure for the previously discussed monoclinic hafnia film and then by demonstrating the sensitivity of VUV optical features to growth parameters, which further complicates the application of standardized recipes for data analysis. Interface Layer Si Substrate 207 Å 2Å 1 mm 80 300 MSE= 1.6956 60 100 40 0 0 2 4 6 Photon Energy (eV) 8 2 4 6 8 10 12 14 16 Interface Layer Thickness in Ang. FIGURE 14. Results of uniqueness fitting algorithm. This involves fixing the fit parameter under evaluation to a range of physically reasonable predefined values, where fits (with other parameters not fixed) are performed at each value. The relative Mean Squared Error is plotted vs. the fit parameter under evaluation. Here, the minima in the curve indicates that the optimal interface layer for the fit shown in Fig. 13 is 2 Å. 0 20 1.01 1 0.99 0.98 0.97 0.96 0.95 0.94 0.93 0.92 0.91 0 200 ∆ in degrees Model Fit Exp Ψ-E 65° Exp Ψ-E 70° Exp Ψ-E 75° Model Fit Exp ∆-E 65° Exp ∆-E 70° Exp ∆-E 75° 100 Ψ in degrees 9Å Relative MSE Surface Roughness Hafnia Lack of uniqueness in fitting of course arises because of correlation between fit parameters that result in “good” fits that yield multiple and nonequivalent n and d combinations for roughly similar Mean Squared Errors. In short, during the course of analyzing ellipsometric data obtained on hafnia films, we have discovered that the most relevant question seems to be the degree of sensitivity to the presence of both interface and surface roughness layers. This degree of sensitivity appears to vary according to the degree of optical structure in the data,18 which we will see varies with growth and postgrowth processing parameters. We are currently working to quantify the extent of this correlation. -100 10 FIGURE 13. Ellipsometric angles Psi (green) and delta (blue) vs. energy for a 207 Å thick monoclinic hafnia film. Fit to the data is shown in red. To verify the results obtained in Fig. 13, a fit uniqueness algorithm25 was employed. This involves fixing the fit parameter under evaluation to a range of physically reasonable predefined values, where fits (with other parameters not fixed) are performed at each value. The quality of the fit is then evaluated at each fixed parameter value in order to determine the degree of sensitivity to the test parameter. That is, the relative Mean Squared Error is plotted vs. the fit parameter under evaluation. In Fig. 14, this was done to evaluate the sensitivity to the thickness of the interface layer for the fit shown in Fig. 13. Here we see a distinct minima for an interface layer of 2Å, for this particular sample. A similar minima was observed for the surface roughness layer thickness value for this fit as well. However, for other samples in the series, we were not sensitive to the parameters under test, and a flat curve was obtained for a similar plot (graphed on a similar scale). Possible sources of this variation are currently under investigation. Ellipsometric angles Ψ and ∆ for a nominally 200 Å thick film are shown in Fig. 13. Data are given in green (Ψ) and blue (∆); the fit to the data is shown in red. This is same fit used as a basis for the simulation in Fig. 12. The data were fit with a general oscillator model in a commercial software environment, with a 2 Å interface layer (a TaucLorentz oscillator model, similar to the SiO2 layer but with different fit parameters) and a 9 Å surface roughness layer (50 % voids in the EMA approximation).24 Details of the fit will be given in more detail elsewhere. We see from the Mean Squared Error (MSE= ~1.7) and from visual inspection that the fit is quite good; however, with these relatively thin high k dielectric films, the challenge is not to obtain a good fit, but rather to obtain a unique fit. 731 8 ε1 ε2 Option 1 Option 2 8 6 Second direct transition 6 4 x Direct transition 4 2 0 Bandgap Onset of absorption due to indirect gap. The onset from the model is 5.02 eV 2 2 exciton before the bandedge 4 6 Photon Energy (eV) 0 10 8 207 Å thick the dielectric in green. The given in blue; 12 8 e1 T2 Real Part of the Dielectric Function vs. Energy 10 9 8 7 6 5 4 3 2 1 0 Real Part of the Dielectric Function vs. Energy 10 e1 FIGURE 15. Optical functions for the monoclinic hafnia film. The real part of function is shown in red; the imaginary part first interpretation of the observed features is the second is shown in red. Imag(Dielectric Constant), ε2 Real(Dielectric Constant), ε1 10 occurring near 7.5 eV (in red.)26 These features vary with growth and/ or post-growth parameters, as seen in Figs. 16 and 17, where with different growth and annealing temperatures, different phases of hafnia (most mixed phase combinations of monoclinic and tetragonal, according to X-ray diffraction measurements18) which results in different VUV optical structure. Preliminary attempts to correlate this optical structure with phase as determined by XRD are complicated by the presence of this mixed phase material, leading us to conclude that fundamental work in the VUV with bulk samples is necessary before we can understand the more complex thin film systems. To demonstrate, we refer to VUV measurements and analysis performed on another candidate high k material, Al2O3. 6 4 2 0 T3 0 2 4 6 8 10 energy in eV No anneal Increasing anneal temp 0 2 4 6 8 8 Imaginary Part of the Dielectric Function vs. Energy 10 7 Energy in eV 6 No anneal 5 Imaginary Part of the Dielectric Function vs. Energy 8 e2 Increasing anneal temp Monoclinic; all others mixed phase (M + tetragonal) with increasing percentage of monoclinic phase with increasing anneal temp 6 5 e2 4 3 7 4 3 2 1 0 0 2 4 6 8 10 Energy in eV 2 1 0 0 2 4 6 8 10 FIGURE 17. The real and imaginary part of the dielectric function ε = ε1 +ε2 are plotted vs. energy for an as-deposited HfO2 film (red curve) and for two HfO2 films that have been annealed post-growth (blue, pink, and yellow curves), grown at a different growth temperature than the films shown in Figs. 9 and 16. . In the VUV, large differences between the annealed and as-grown films emerge. At lower energies, the films are transparent and indistinguishable. Energy in eV FIGURE 16. The real and imaginary part of the dielectric function ε = ε1 +ε2 are plotted vs. energy for an as-deposited HfO2 film (pink curve) and for three HfO2 films that have been annealed post-growth (yellow, dark blue, and light blue curves), grown at a different growth temperature than the films shown in Fig. 9. In the VUV, large differences between the annealed and as-grown films emerge. At lower energies, the films are transparent and indistinguishable. In Fig. 18a, we show optical functions for bulk cplane Al2O3. There are several points to consider for this material. First, the more stable surface of Al2O3 eliminates concerns about surface preparation, save accounting for surface roughness. We do so both in the optical model used to fit the data and with complimentary AFM measurements. For this substrate, the RMS roughness was on the order of 5Å, averaged between substrate center and edge values. The real and imaginary parts of the pseudodielectric function (dotted lines) are plotted vs. photon energy. Our model fit (solid lines) and some IR/ VIS/ UV data from Yao and coworkers (dashed lines) are shown for comparison.27 Their user defined Cauchy fit is an After a reasonable fit to the data was achieved, optical functions ε = ε1 + iε2 can be obtained and are shown in red and green, respectively, in Fig. 15. The larger question of the interpretation of these data, however, is still unanswered. In the same figure we show two possible interpretations: option 1, which consists of an onset of absorption preceding a direct gap near 5 eV, with two higher lying direct transitions located near 6 and 7.5 eV, respectively (in blue), and option 2, where the feature in ε2 near 6 eV is interpreted as an exciton before the bandedge 732 excellent fit to the data, until roughly 6.5 eV. Of course, it is well-known that the Cauchy model is a (a) benchmark for the analysis of thin Al2O3 films on Si. We see from this and the preceding example the necessity for a thorough analysis of bulk high k materials before proceeding with more complex heterostructures. 29 (b) c-plane Al2O3 substrate 7.0 Index of refraction ’n’ <ε1> 5.0 2.8 this work: model fit this work: data prior art: Yao, et al. J.Appl. Phys., 1999 2.0 1.6 3.0 4.0 0.40 0.20 0 300 600 900 1200 Wavelength (nm) 1500 0 1800 Critical Issues for Low- k Materials ‘K’ 4.0 0.60 n k 2.4 Extinction Coefficient 6.0 From extensive investigations of low k dielectric materials in our laboratory, the following became apparent: while ellipsometry is sensitive to differences in relative electron density in low k films (recall the argument made about sensitivity to long range order made earlier), the difficulty in data analysis for these materials lies in the fact that multiple unknown microscopic and macroscopic material properties contribute to the averaged ellipsometric information. Sensitivity to these material properties is not the issue; rather, deconvoluting multiple correlated parameters during the data analysis process is the large challenge for low k materials. Even with the additional information that the VUV provides, static ellipsometric approaches are often not equal to the challenges posed to them by the processing engineers. We illustrate with a discussion involving organosilicate glasses (OSG). These materials are best described as C-doped silica, where it is assumed that the incorporation of CHx species introduces porosity into the SiO2-like matrix. This is of course one of the preferred means of achieving materials with dramatically lower static dielectric constants, hopefully without sacrificing robust mechanical properties. In contrast to high k materials, the interlayer dielectric thicknesses are safely within specifications that allow for the independent determination of n and d. Each layer is typically hundreds to thousands of Ångstroms thick, with the thickness of the entire ILD stack typically on the order of 1000 Å.30 This is fortunate, as these materials often shrink during the course of imaging by scanning electron microscope, meaning that the ability to determine n and d independently, by ellipsometric means, becomes crucial. Pore size and distribution, however, is the desired information for these materials. And while static ellipsometric measurements are certainly sensitive to changes in pore size and distribution, we maintain that it is unrealistic to extract meaningful (detailed) information about these parameters from such measurements—especially given that a slight adjustment of processing parameters can riddle the film with low index inclusions. These inclusions of course complicate the analysis further and prevent the facile application of in- line metrology recipes predicated on the assumption that the films are always <ε2> 3.0 2.0 rough surface 12.5 Å Gen. Osc. Model 1 mm 1.0 0 0 2 4 6 8 Photon Energy (eV) 10 FIGURE 18. (a) Pseudodielectric function data (dotted line) for Al2O3, from 0.7 to 9.5 eV. Our model fit, which involves a superposition of several oscillators is shown by the solid line, while the Cauchy fit of previous workers is given by the dashed line. (b) Optical constants index of refraction n (solid line) and extinction coefficient k (dashed) obtained to fit to the data in (a). (After Ref. 29, reprinted with permission) good representation of the transparent region of any optical spectrum. In the VUV the model fails, as expected because of the critical point located just above 9eV. To model both the transparent and absorbing regions, our fit was a complex superposition of two Lorentzians, one Gaussian, a pole and an ε1 offset term. (Details will be discussed elsewhere). The identity of this critical point can be deduced from previous VUV reflectivity work,28 where it was observed that there was an exciton associated with the O 2p nonbonding M0 bandedge, located just below the bandedge in spectral energy. Both are located at 9.2 and 9.5 eV at 325K, respectively. Our data are consistent with this interpretation; thus it is likely that the observed critical point is the excitonic transition just below the Al2O3 bandedge.29 However, optical constants obtained from VUV reflectivity are subject to a variety of errors, such as fluctuations in source intensity and issues surrounding Kramers-Kronig analysis that requires knowledge about optical constants beyond the measured spectral range. Because SE measures the ratio of two values, the data are not subject to such errors. Indeed, once a good fit to the ellipsometric data is achieved, it is possible to calculate optical constants without KK analysis. These constants are shown in Fig. 18b, where index of refraction n and extinction coefficient k are plotted vs. wavelength. The dramatic onset of absorption at 155nm (ie. the onset of non-zero k values) is associated with the excitonic critical point that we saw in ε1 and ε2. These data (and indeed the presence or absence of this feature) act as an excellent 733 identical to porous SiO2 (i.e. are transparent), as might be reasonable to expect. We illustrate the capabilities and limitations of static VUV SE measurements and data analysis for the characterization of low k organosilicate glasses with the following example from an industrial production environment. (See Ref. 14 for further details concerning samples and measurements). During the transfer of an OSG film growth process, we found that the films from the researchgrade reactor consistently had higher hardness.31 than those from the higher throughput production reactor, for nominally similar process parameters. Despite such a large difference in mechanical properties, the films appeared virtually identical when compared by XPS and also by TEM cross-section.31 Accordingly, spectroscopic ellipsometry (SE) was performed to see if any difference that could explain between the two sets of films could be observed that would explain the unusual differences in mechanical properties. 1.455 1.445 3.0E-05 a) b) 2.5E-05 wfr13 wfr14 wfr17 2.0E-05 k k n 1.435 1.425 wfr18 1.5E-05 wfr19 1.0E-05 1.415 wfr20 wfr21 5.0E-06 1.405 SiO2 0.0E+00 500 700 900 1100 350 400 450 500 550 Wavelength in nm FIGURE 20. OSG optical properties, data: in (a) index of refraction n vs. wavelength; in (b), extinction coefficient k vs. wavelength. Wafer 13, the representative sample from the research-grade reactor has a higher index of refraction (and thus lower relative porosity) than the production-grade samples. It also has a lower absorption edge (i.e. defined here as the onset of non-zero ‘k’ values) than the other samples, it is transparent to higher energies than the other films. (After Ref. 14, reprinted with permission) curve). The index of refraction scales as we would expect between the two extremes, with n decreasing with decreasing electron density/ increasing porosity. Extinction coefficients were also simulated from 138 to 1771nm (or 0.7 to 9.0 eV). The extinction coefficient for dense as well as porous (all ∆p values) SiO2 was zero. Recall that an extinction coefficient value of zero means that the material is transparent; i.e. that there was no absorption expected for SiO2 in this spectral range, whether porous or dense. Experimentally-derived values of n and k for the OSG films are given in Fig. 20 (a) and (b), respectively. It is immediately apparent that Wfr13 (squares), the representative research-grade sample with higher hardness, has different optical properties from both the production-grade samples (Wfrs. 14-21) and from SiO2 (dotted line). First, it has a higher index of refraction and lower onset of absorption (defined for our purposes as the onset of non-zero k values) than its production-grade counterparts. Compared to SiO2, we see in (a) that n for relatively more dense SiO2 is greater than for all of the OSG films (e.g. nSiO2=1.45 at 900nm). The introduction of voids with index n =1 into this matrix will decrease n, with nmatrix approaching nair as the void fraction increases, as we observed in Fig. 19. Thus n values of these OSG films are by comparison lower, ranging approximately between ~1.41 and 1.42 at 900nm. Accordingly, we can argue that Wfr13 with its high hardness and notably higher index of refraction is also more electronically dense than the other OSG films and is significantly less dense than SiO2. Unexpectedly, from the non-zero k values in (b), it is evident that the OSG films, unlike porous or dense SiO2, experience some FIGURE 19. Results of an effective medium approximation (EMA) calculation, simulating the effect of porosity on the index of refraction of SiO2. Porosity is introduced at an increment of ∆p=10%. Top curve: the index of SiO2 (0% porosity); bottom: air (100% porosity). As the porosity of the material increases, the index of refraction decreases. (After Ref. 14, reprinted with permission) We have argued that ellipsometry is sensitive to electron density and thus to local structural arrangement; therefore, differences in crystal structure, interface and surface properties, stoichiometry and relative porosity, if present, should manifest themselves in the optical data. 8,32 To investigate, we first simulated the effect of voids/ low index inclusions on the SiO2 index of refraction, shown in Fig. 19. Here pores of increasing void fraction (in increments of ∆p=10%) were introduced into a SiO2 matrix via an effective medium approximation (EMA) calculation.24 The index of refraction n as a function of wavelength was calculated for each case, ranging from 0% porosity (pure SiO2 ; top curve) to100% (air: bottom 734 After a thorough characterization of the materials in the VUV, it is possible to provide some degree of absorption in the spectral range under consideration. Here too Wfr 13 can be distinguished from its production-grade counterparts with its onset value. This is also indirect evidence for the presence of a combination of voids and other low index inclusions in all of the films, the identity and amount of which cannot be strictly verified by the available ellipsometric data. Model Fit Exp E 65° 60 Relative Porosity 40 ← Sellmeier → 20 10 520 520 480 480 440 H(GPa) 2.05 1.378 1.254 1.240 1.29 1.47 1.33 Stdev 0.20 0.085 0.091 0.089 0.13 0.30 0.15 Research 440 Production 400 10 0 0 9 9 8 8 7 7 6 6 Porosity Porosity(%) (%) wfr w13 w14 w17 w18 w19 w20 w21 1 1 1.2 1.2 1.4 1.6 1.8 2 1.4 Hardness 1.6 1.8 (GPa)2 Hardness (GPa) 2 4 6 Photon Energy (eV) 8 10 60 5 5 400 ← Cauchy / Urbach→ 80 Ψ in degrees 560 Absorption (nm) Tauc-Lorentz Ψ in degrees Absorption Onset (nm) 560 Tauc-Lorentz + 2 Gaussians 80 2.2 40 Tauc-Lorentz Model b 20 2.2 0 0 FIGURE 21. Ellipsometrically-obtained relative porosities and absorption onsets plotted vs. mechanical hardness for the two sets of samples. Mechanical hardness data is given in the inset table. The difference between the optical and mechanical properties of production-grade samples vs. the representative research-grade sample is readily apparent. (After Ref. 14, reprinted with permission) 2 4 6 Photon Energy (eV) 8 10 FIGURE 22. (a) Ellipsometric angle Ψ data (dashed line) vs. photon energy for Wfr 13 for a 65º angle of incidence. The optical model (solid line) that best fit the data was comprised of a Tauc-Lorentz and two Gaussian oscillators. The dashed vertical lines denote the spectral range of viability for other kinds of optical models. (b) The same ellipsometric data with a fit extrapolated into the VUV. Here we see that the single Tauc-Lorentz oscillator that fits so well at low energies cannot fit the data well above 6 eV. (After Ref. 14, reprinted with permission) Nonetheless, though these measurements cannot definitively yield pore size and distribution information—since ellipsometry is sensitive to net electron density only—they can serve as a very effective in-line diagnostic for processes yielding films with undesired electron densities. This is important because such properties should affect the mechanical behavior of a film (and presumably the static dielectric constant as well). This premise is supported in Fig. 21, where ellipsometrically-obtained relative porosities and absorption onsets for the films are plotted vs. mechanical hardness (see Ref. X for more detail). Though, again, we cannot determine pore size and distribution with the VUV SE data, these results indicate that ellipsometry is sufficiently sensitive to the differences in the local structure of the films conclude that the technique—even at longer wavelengths—has promise for the characterization of OSG films. If this degree of sensitivity at long wavelengths is possible, then in-line metrology is promising to identify potentially mechanically compromised materials. insight into challenges associated with the analysis of SE data of OSG films. In particular, we warn of complications that can arise during the course of the development of data reduction recipes because of the large difference that can exist between the onset of absorption and the onset of opacity for these materials. Lest the distinction between these onsets appear trivial, we reiterate that the correct choice of optical model applied over the correct spectral range is the means by which sample properties like thickness, relative porosity, n and k are extracted. We illustrate with an example: in Fig. 22(a) we have plotted the ellipsometric angle Ψ vs. photon energy at one angle of incidence for Wfr13. Below around 7eV, the data (dotted line) completely consist of interference oscillations, which is typical for transparent films grown on absorbing substrates. Above this energy is the onset of opacity, where the oscillations cease and 735 eV, where there clearly is none (cf. Fig 22b). In summary, then, it is important to note that for OSG films, the onset of absorption can be very different from the onset of opacity, because of the potential presence of low-index inclusions. For these particular films complementary Raman analysis indicated that amorphous Si clusters were present. In fact, it was possible to correlate the absorption edges obtained by SE and the a-Si cluster concentration, as shown in Fig. the film is completely absorbing. The model calculation (solid line) used to best fit the data was Absorption Edge (eV) 3.2 3 23.33 2.8 In short, for process transfer or the evaluation of new growth recipes, it is worthwhile to first evaluate the film over the broadest possible spectral range and to critically assess the confidence limits of model parameters to determine the suitability of the optical model. Moreover, blind application of standard oxide in-line metrology recipes applied to these materials, based on the assumption that they are merely porous SiO2, will result in spurious values for film thickness and electron density—and the link to mechanical behavior could be lost or distorted. In this way, though VUV SE does not have the capability to yield the desired pore size and distribution for low k films, we can see that it is very useful for the fundamental studies necessary for the development of in-line metrology recipes for these materials. 2.6 More a-Si 2.4 2.2 4.8 5 5.2 5.4 -1 5.6 5.8 -1 6 6.2 R(480cm )/R(800cm ) FIGURE 23. Ellipsometrically-obtained absorption edges of the OSG films plotted vs. a-Si cluster concentration, determined by Raman. comprised of a Tauc-Lorentz oscillator, which is typically used to describe the optical dispersion of amorphous materials,13 plus two Gaussian oscillators to fit the VUV absorption. (Further details will be given elsewhere.) With the VUV instrument, we can observe this edge for the very first time, and its presence in this spectral region is an obvious clue that the film is not structurally similar to (transparent) SiO2. Further, the vertical dotted lines in the figure define regions over which certain optical models are applicable, which is necessary to delineate because of this very small amount of absorption below the onset of opacity. Significant errors arise when these are applied outside of their region of applicability: e.g., the Cauchy and Sellmeier formalism should only be applied in the transparent region of the spectra and thus are common in-line recipes for measuring SiO2 optical properties. Since sample thickness is usually determined by applying one of these models in the transparent spectral region, thicknesses can be disastrously off-target (by as much as 1/3 in this case) if the model is mistakenly applied in a region where absorption is occurring (i.e. where k is actually nonzero). For this particular sample, this is at ~2.1 and 3.6 eV for the Cauchy and Sellmeier model, respectively. On the other hand, a single Tauc-Lorentz oscillator gives a beautiful fit for the data until just above 6 eV, which happens to cover the spectral range of most standard IR/ VIS/ UV instruments. While the sample thickness obtained with this model will only be offtarget by a percent or so, for this sample, the optical constants were distorted, since this fit mistakenly placed a large oscillator-related structure squarely at 8 VUV SE: status and prospects…. High k Low k Extract n and d independently for thickness required for process Extract n and d independently for thickness required for process Sensitive to interface layer and surface roughness in ‘thick’, single layer films Sensitive to density and surface roughness in ‘thick’, single layer films Sensitive to interface layer and surface roughness for multilayers Sensitive to pore size and distribution Legend: Requires invention/ potential showstopper Development required Solution known FIGURE 24. Status and prospects for VUV SE, as applied to the problems facing applications utilizing high k and low k dielectrics. CONCLUSIONS We have shown with the previous examples that VUV SE allows for increased access to unique spectral features and increased sensitivity to film thickness, meaning that the technique is of use not only for determining optical constants at 157nm (the obvious lithographic application) but also for analyzing materials from other stages of the device 736 fabrication process. Given the initial problems in reducing data to determine optical constants at the throughputs and accuracies demanded in industrial settings, a significant challenge for the industrial practitioner of VUV SE has been to develop the technique from its more esoteric origins as a research instrument to an industrially-viable diagnostic technique. Since most of these issues were rooted in the absence of optical constants for Si in the VUV, these types of problems have largely been solved. Indeed, when applied properly, VUV SE is not only practical for determining VUV optical constants and material properties but also a powerful instrument for industrial problem solving. However, other issues remain that are endemic to the general SE data analysis process. With respect to low k and high k materials, capability VUV SE is powerful, but (like any technique) not without its limits. In Fig. 24 we summarize the preceding discussion into a table capturing the status and prospects of the technique for these applications. 11 N.V. Edwards, “ VUV Ellipsometry of Si, SiO2, and Lowk Organo-Silicate Glasses,” SEMATECH Metrology Council Meeting, Austin, TX, 05 May, 2002. 12 B. Johs, J. A. Woollam, C. M. Herzinger, J. N. Hilfiker, R. Synowicki, and C. Bungay, Proc. SPIE CR72, 29 (1999). 13 G.E. Jellison and F. A. Modine, Appl. Phys. Lett. 69, 371 (1996). 14 N. V. Edwards, J. Vella, Q. Xie, S. Zollner, D. Werho, I. Adhihetty, R. Liu, T. E. Tiwald, C. Russell, J Vires, and K. H. Junker. Mat. Res. Soc. Symp. Proc. 697, P4.7.1 (2002). 15 N.V. Edwards, S. Zollner, J. White, D. Gajewski, Motorola Inc., unpublished. 16 H.G. Tompkins, R. B. Gregory, P.W. Deal, and S.M. Smith, J. of Vac. Sci. and Technol. A 17, 391 (1999). 17 H.G. Tompkins and W.A. McGahan, Spectroscopic Ellipsometry and Reflectometry: A User’s Guide, John Wiley and Sons, New York, 54-61 (1999). 18 R. Liu, N.V. Edwards, R. Gregory, D. Werho, E. Duda, J. Kulik, G. Tam, E. Irwin, X-D Wang, D. Triyoso, Motorola Inc., unpublished. 19 R. Liu, S. Zollner, P. Fejes, R. Gregory, S. Lu, K. Reid, D. Gilmer, B.-Y. Nguyen, J. Yu, R. Droopad, J. Curless, A. Demkov, J. Finder, and K. Eisenbeiser, Mat. Res. Soc. Symp. Proc. 670, K1.1 (2001). 20 J. Schaeffer, N.V. Edwards, R. Liu, D. Roan., B. Hradsky, R. Gregory, J. Kulik, E. Duda, L. Contreras, J. Christiansen, S. Zollner, P. Tobin, B-Y. Nguyen, R. Nieh, M. Ramon, R. Rao, R. Hegde, R. Rai, J. Baker, S. Voight, J. Electrochem. Soc., in print. 21 D.E. Aspnes in Optical Properties of Solids: New Developments, edited by B.O. Seraphin, North-Holland, Amsterdam, 799 (1976). 22 J. Hilfiker in Handbook of Ellipsometry, edited by H.G. Tompkins and E.A. Irene, Noyes Press, Park View, in print. 23 R. M. A. Azzam and N. M. Bashara, Ellipsometry and Polarized Light, Elsevier, Amsterdam, 283-293 (1987). 24 D. A. G. Bruggeman, Ann. Phys. (Leipzig) 24, 636 (1935). 25 C.M. Herzinger, private communication. 26 A. Demkov and N.V. Edwards, Motorola, Inc., unpublished. 27 H. Yao and C. H. Yan, Appl. Phys. Lett. 85, 6717 (1999). 28 R.H. French, D. J. Jones, and S. Loughin, J. Am. Ceram. Soc. 77[2], 412 (1994). 29 N. V. Edwards, O.P.A. Lindquist, L.D. Madsen, S. Zollner, K. Jarrendahl, C. Cobet, S. Peters, N. Esser, R. Liu, and D. E. Aspnes, Mat. Res. Soc. Symp. Proc. 697 (2002). 30 K. Junker, Motorola Inc., private communication. 31 J. Vella, Q. Xie, N.V. Edwards, J. Kulik and K. Junker, Mat. Res. Soc. Symp. Proc. 697, 6.25 (2002). 32 D. E. Aspnes, Thin Solid Films 89, 249 (1982). 33 R. Liu and N.V. Edwards, Motorola Inc., unpublished. REFERENCES 1 D.E. Aspnes, in Optical Properties of Solids: New Developments, ed. B.O. Seraphin (North-Holland, Amsterdam, 1976), p. 799. 2 D.E. Aspnes, SPIE Proceedings- Optical Characterization Techniques for the Semiconductor Industry 276, 188 (1981); 452, 60 (1983) 3 N.V. Edwards, “Materials Characterization with VUV SE,” Semiconductor Fabtech 18, 2003. 4 J. Barth, R. L. Johnson, S. Logothetidis, and M. Cardona, SPIE Proceedings - Soft X-Ray Optics and Technology 733, 265 (1986); S. Logothetidis, J. Petalas, H. M. Polatoglou, and D. Fuchs, Phys. Rev. B 46, 4483 (1992); T. Wethkamp, K. Wilmers, C. Cobet, N. Esser, W. Richter, O. Ambacher, M. Stutzmann, and M. Cardona, Phys. Rev. B 59, 1845 (1999); C. Cobet, K. Wilmers, T. Wethkamp, N. V. Edwards, N. Esser, and W. Richter, Thin Solid Films 364, 111 (2000). 5 J.R. Wasson, S. Han, N.V. Edwards, E. Weisbrod, W. Dauksher, P.J.S. Mangat and D. Pettibone, “Integration of Anti-reflection Coatings on EUV Mask Absorber Stacks,” Proc. of the SPIE, 22nd Annual BACUS Symposium on Photomask Technology, vol. 4889, in print. 6 D.E. Aspnes, Am. J. Phys. 50, 704 (1982); D.E. Aspnes, J. Mater. Education 7, 849 (1985). 7 C.M. Herzinger, B. Johs, W.A. McGahan, J.A. Woollam and W. Paulson, J. of Appl. Phys. 83, 3323 (1998). 8 D.E. Aspnes and A. A. Studna, Phys. Rev. B 27, 985 (1983). 9 G.E. Jellison, Jr., Opt. Mater. 1, 41 (1992). 10 T. Yasuda and D.E. Aspnes, Appl. Optics 33, 7435 (1994). 737
© Copyright 2025 Paperzz