Characterizing a CVD-Integrated Metrology System

James M. Holden Nanometrics, Sunnyvale, Calif. Martin J. Seamons Applied Materials, Santa Clara, Calif. -- Semiconductor International, 10/1/2000

With the upcoming transition to sub-130 nm lithography, copper/low-k and 300 mm diameter wafers, the semiconductor industry is ushering in a new set of production tools that must meet higher standards of cost efficiency. Simultaneously, the way in which metrology is used is becoming split between fast, integrated systems that perform the bulk of measurements in the fab and stand-alone metrology tools used mainly for off-line analysis, material characterization and recipe development for integrated measurement systems. The use of analysis techniques, such as the empirical dispersion models presented here, will provide a relationship between stand-alone systems and the integrated metrology system. With shrinking process windows, a viable path to wafer-to-wafer process control is needed. Such approaches also will help prevent misprocessing, especially of 300 mm wafers, which is prohibitively expensive.

New metrology approaches are gradually replacing past manufacturing practices that, due to broader process windows, required only periodic sampling of film properties in production after the process was "baselined" through several iterations on stand-alone metrology tools. By having the film metrology integrated with a process tool, immediate feedback on every production wafer is possible, allowing for quick detection of process deviations and, ultimately, wafer-to-wafer, closed-loop control.

Ultra-compact metrology system

Nanometrics designed a complete optical film metrology system that fits within a 10x12x12-in. box. We integrated the InTool system into the factory interface of an Applied Materials twin-chamber, single-wafer Producer CVD system. The metrology system consists of a spectroscopic reflectometer equipped with built-in reference and pattern recognition, an r-Q stage and a pre-aligner. The process tool controls the metrology tool as a part of the process recipe. Users can develop measurement programs off-line or access the metrology system's user interface via the process tool monitor. As follows, sophisticated film modeling software allows the user to determine the film thickness and refractive index of single- and multiple-layer films.

One possible dielectric stack in a typical dual-damascene process (Fig. 1) includes several layers of dielectric, which are deposited and then patterned and etched to define metal lines and vias to the lower metal layer. A thin top layer of PECVD silicon oxynitride acts as a dielectric antireflective coating (DARC) to improve CD control during patterning. It is necessary to precisely control the thickness and refractive index of this DARC. The etched pattern is then coated with a thin metal layer, usually a Ta/TaN barrier layer, filled with copper, then polished by CMP. Intermediate nitride layers act as barriers and/or etch stop layers.

We characterized the DARC films by spectroscopic ellipsometry, measuring the amplitude (y ) and phase (D) of the quantity,

where R_p and R_s are the Fresnel reflectivity coefficients for light polarized in the p-plane and the s-plane. Typically, y and D are measured as functions of wavelength, l, and angle of incidence, Q . For instance, Figure 2 shows the amplitude and phase for a typical silicon oxynitride ARC film over visible and ultraviolet wavelengths (l = 200-800 nm) for the three angles of incidence 65°, 70° and 75°.

If we know the optical constants, n and k, and structure of the substrate (in this case, crystalline silicon with a native oxide), then the film's thickness, refractive index and extinction coefficient can be determined for all wavelengths. Since Equation 1 generates two equations for every angle or wavelength, for m wavelengths and three angles there are 2x3xm=6m equations but only 2m+1 unknowns. We can then numerically solve Equation 1, with r written in terms of n(l), k(l) and film thickness, t.

1. A typical dual-damascene film stack includes dielectric ARCs (silicon oxynitride) to help control CD variations.

Parametric dispersion model

Another way to extract the n and k versus l values from phase and amplitude spectra is through dispersion modeling. A dispersion model describes the optical constants versus wavelength as a function of a few real valued parameters and l or, equivalently, photon energy, E = h\w = hc/l. We use the model of Herzinger and Johs, an improvement on the work of Kim and Garland et al, to directly analyze the spectroscopic ellipsometry data. The basic form for the complex dielectric constant in this model is

where the first term is summed over all conduction and valence bands and the second summation is for "core terms" with resonance at high photon energy. The function Jcv is the joint density of states times the absorption cross section for a single conduction-valence interband transition, and F is a complex valued Gaussian broadening function with decay factor scv. The convolution integral with the Gaussian broadening function guarantees consistency with the Kramers-Kronig causality relationships.

We fit the data of Figure 2 to the parametric dispersion model of Equation 2, using two terms in the first summation and one term from the second summation. This requires specification of 14 parameters, of which five need to be adjusted via an iterative Levenberg-Marquart algorithm to find the best fit to the data. Essentially, the parametric model uses physical principles to reduce the number of free parameters (unknowns) by tying them together with a smooth functional form.

Empirical dispersion model

Although the parametric dispersion model reduces the number of free parameters needed from 2m+1 to 5, it is still more general than necessary for a silicon oxynitride film produced for the purposes of an ARC. The parametric model we use for SiO_xN_y could actually fit a larger class of materials comprising most dielectrics. A model more specific to our particular film would require less data and would have fewer free parameters, such as y only instead of y and D, to uniquely determine n(l) and k(l).

Since SiO_xN_y is composed primarily of four elements - silicon, oxygen, nitrogen and hydrogen - it is natural to suppose that a model with two parameters similar to the x:y ratio that determines the stoichiometry could describe the system. We made a calibration set of wafers with optical constants that spanned the possible optical constants for SiO_xN_y ARCs and then devised a dispersion model that interpolates optical constants between the calibration nodes to fit the index of an unknown film.

A suitable set of calibration samples should be as identical to the target films as possible, yet span a reasonable range of optical constants to include films deposited when the deposition process strays from nominal - whether intentionally or not. Once we accurately and unambiguously determine the optical constants of the calibration films using y and D ellipsometry data, we can build the optical constants of any other ARC film within the "window" defined by the calibration set.

Figure 3a shows n(248 nm) versus k(248 nm) values for seven PECVD silicon oxynitride films. We deposited these films on two different Applied Materials CVD tools used for production deposition of ARC oxynitride films. Figure 3a provides reference values of SiO₂, Si₃N₄ and amorphous silicon films. The different n and k values of the oxynitride films were the result of varying the ratio of silane (SiH₄) flow to nitrous oxide (N₂O) flow. Figure 3b shows an expanded view of 3a. The first group of films (shown as triangles) was deposited under different process conditions than the second set of films (shown as triangles). The triangle group has a lower n for any k than the circle group. Since points labeled with a triangle are closer to the point representing SiO₂ in Figure 3, we call them oxide-like, whereas points labeled with a circle are nitride-like.

As we increased the fraction of SiH₄ in the process gases, the films contained proportionately greater amounts of hydrogen and/or silicon. We suspect that increased amounts of silicon or hydrogen atoms replace the oxygen or nitrogen atoms, allowing the silicon atoms to come into closer contact and making the material more like amorphous silicon, with n and k values that increase toward the node for amorphous silicon.

Figure 4 compares the n(l) and k(l) for two oxynitrides to a hydrogen-rich amorphous silicon. From group to group, the total flow of all process gases was slightly different, as were the deposition chambers. Changing the total flow rate appeared to alter the ratio of nitrogen to oxygen incorporated into the films.

One-parameter interpolation model

Figure 5a shows the refractive index for the three calibration samples labeled &#68; in Figure 3, and Figure 5b shows the extinction coefficient for the same set. The samples are labeled A, B and C, with A having been deposited with the lowest SiH₄:N₂O ratio and C deposited with the highest ratio. For purposes of the dispersion model, each of these films will be called "nodes" and have attached to it an index value, x. Sample A is given the value x = xA, sample B is given x = xB and sample C is given x = xC with the stipulation that x_AB< x_C. For an arbitrary sample, the index value is allowed to vary over the range of the nodes and slightly outside this range. Thus,

for some small value, &#100;, that defines the range outside the nodes to which the model can be meaningfully extrapolated. Given an arbitrary x between, say, x_AB, we can calculate the complex refractive index, N, in terms of the complex dielectric constant, &#949; = N2, as a function of the refractive indexes, N_A and N_B, of nodes A and B by

with the fraction given by

When x is between any two other nodes, a similar set of formulas applies. Note that when x is near a node, say A, x = xA, we have N = NA. When x is in the extrapolation region, say x = xA - &#100;, the fraction, f, is negative, and the line between nodes A and B is extended.

6. Specifying the refractive index at 248 nm does not unambiguously determine the extinction coefficient at 633 nm.

We can apply the one-parameter interpolation model to silicon oxynitride ARCs as long as the process conditions do not vary outside the one-parameter calibration set. Despite this limitation, the one-parameter model has the novel application of being able to predict values for n(248 nm) and k(248 nm) given only reflectance data in the visible wavelength range. This is because N(&#68;) is predicted over the entire ultraviolet spectral range by the single parameter x. Figure 6 plots the ultraviolet extinction coefficient, k(248 nm) against the visible refractive index, n(633 nm). As can be seen, if we focus on only one subset of samples, the ultraviolet k is a single-valued and well-defined function of the visible n. However, considering both subsets, specifying the n in the visible does not unambiguously determine the k in the ultraviolet.

Two-parameter interpolation model

We can build a two-parameter interpolation model when the calibration set includes more than one set of samples. As in the one-parameter interpolation model, each set of samples is assigned a node value, which we can label y. For the set labeled with a triangle, y is given the value y = yD, and for the set lableled with a circle, y = yo. The final index is:

and

where ND and NO are functions of the parameter x from the one-parameter model and y is allowed to range over

for some small z.

Application to copper films

In order to remove copper oxide from the surface of the metal, the electroplated copper substrate typically is reduced by exposure to hydrogen prior to nitride deposition. For characterization, we coated two copper wafers with nitride in the deposition chamber, omitting the treatment step for one wafer to determine the effect on the optical properties. We then measured the wafers by spectroscopic ellipsometry. The three unknowns were nitride thickness, refractive index of the copper and extinction coefficient of the copper. We fit the index of the nitride using the parametric dispersion model of Equation 2. As shown in Figure 7, we can then combine the n and the k of these two films into a one-parameter model.

Next we took the reflectance data from a 300 Å DARC film on silicon, and used the model to predict n, k and t values. Long-term repeatability was 1s, as we measured the same wafers 400 times over the course of 24 hours with a load/unload cycle between measurements. The 1&#115; reproducibility was less than 0.75 Å for thickness, 0.1% for refractive index and ~1% for the extinction coefficient.

Reflectance data from the multilayer film stack are depicted in Figure 8. Fit parameters in this measurement are thickness of each of the five layers, the value of x for the DARC and the value of x for the in situ treated copper. Rapid oscillations in reflectance versus wavelength are the result of constructive and destructive interference inside the thicker layers. The Table shows the dynamic reproducibility for the five-layer measurement. The 1s was 1.5 Å for thickness and less than 0.4% for n and k.

Repeatability (1s) of IMD Film Stack Parameters
	Nitride 1 Thickness (Å)	Oxide 1 Thickness (Å)	Nitride 2 Thickness (Å)	Oxide 2 Thickness (Å)	DARC Thickness (Å)	DARC n	DARC k
Mean	730.2	6534.1	1562.1	9039.0	734.8	2.182	0.607
Std. Dev.	1.7	1.7	1.1	1.1	1.6	0.001	0.002

Summary

Although shrinking process windows are demanding more overall measurements per wafer, the addition of integrated metrology will reduce the level of spot checking by stand-alone metrology systems. The advantages of integrated metrology - reduced handling and steps requiring a human operator, immediate feedback from the process tool, and wafer-to-wafer process control - will result in reduced scrap and increased overall equipment effectiveness for production. .

James Holden joined Nanometrics in 1997 and works there as a senior applications scientist. He has published several articles in the fields of optical spectroscopy and materials science, and has two patent applications related to semiconductor metrology. He holds a Ph.D. in physics from the University of Kentucky.

Martin Seamons, process engineering manager for the PECVD Dielectric Films Group at Applied Materials, has seven years experience developing dielectric deposition processes and equipment. He has a B.S. in chemical engineering from U.C. Berkeley and has performed post-graduate work at San Jose University.
Phone: 1-408-986-2768.
e-mail: martin_j_seamons@amat.com

Talkback

We would love your feedback!

Post a comment

» VIEW ALL TALKBACK THREADS

Related Content

• Topics
• Author

Related Content

 

By This Author

There are no other articles written by this author.

SPONSORED LINKS