KR20070032036A - Methods and apparatus for determining endpoint in a plasma processing system - Google Patents

Abstract

In a plasma processing system, a method of determining a process threshold is disclosed. The method includes exposing a substrate to a plasma process, including a process start portion, a substantial steady state portion, and a process end portion. The method also includes collecting the first data set during the substantial steady state portion; Generating a first statistical model comprising at least a statistical model component selected from the group consisting of a deviation component and an error component; And collecting the second data set. The method further includes generating a second statistical model comprising the statistical model component, wherein if the statistical model component of the first statistical model is substantially different from the statistical model component of the second statistical model, the process threshold is substantially Obtained.

Plasma processing systems, endpoint determination, process thresholds, statistical models

Description

METHODS AND APPARATUS FOR DETERMINING ENDPOINT IN A PLASMA PROCESSING SYSTEM}

Background of the Invention

FIELD OF THE INVENTION The present invention generally relates to substrate fabrication techniques, and more particularly to methods and apparatus for determining endpoints in plasma processing systems.

For example, in the processing of semiconductor substrates such as those used in flat panel display manufacturing or substrates such as glass panels, plasma is often used. As part of the processing of an example substrate, the substrate will be divided into a plurality of dies, or rectangular regions, each of which will be an integrated circuit. The substrate is then processed in a series of steps in which the material is selectively removed (etched) and deposited (deposited) to form electrical components on the substrate.

In an exemplary plasma process, the substrate is coated with a thin film of solidified emulsion (ie, photoresist mask, etc.) prior to etching. Thereafter, the area of the solidified emulsion is selectively removed, causing the components of the underlying layer to be exposed. The substrate is then placed in a plasma processing chamber on a substrate support structure that includes a monopolar or bipolar electrode, called a chuck or pedestal. A suitable etchant source then enters the chamber and forms a plasma that etches the exposed areas of the substrate.

1 shows a plasma processing system 150 including a chamber 100 with a pump 120 that maintains a low chamber pressure and discharges a process gas effluent. The chamber 100 is grounded like an upper electrode 104 which also operates as a showerhead type gas distribution system. RF power is supplied from the power source 101 to an electrostatic chuck (chuck) 108 located in the lower electrode assembly 106. The RF power source may include means for matching to plasma impedance by frequency tuning or variable impedance tuning in the matching network 145. RF electrical measurements are made using the probe 140 as a signal transmitted by the cable 141 to the process module controller 116. The plasma 102 is generated by supplying RF power to the chuck 108 to process the substrate 109. In this example system, the plasma 102 is confined between the chuck 108 and the electrode 104 by a confinement ring 103 that may regulate the pressure within the plasma 102. The confinement ring 103 can move to increase and decrease the space or gap between adjacent confinement rings, generally by use of a cam ring. In general, the gas distribution system 122 includes plasma processing gases (eg, C ₄ F _8, C ₄ F ₆ , CHF ₃ , CH ₂ F ₃ , CF ₄ , HBr, CH ₃ F, C ₂ F ₄ , N ₂ , O ₂ , Ar, Xe, He, H ₂ , NH ₃ , SF ₆ , BCl ₃ , Cl ₂ , WF ₆ And compressed gas cylinders).

During operation, plasma induced electromagnetic radiation (optical radiation) may be collected through window 110 and make an image on spectrometer 114 by lens 111 and fiber optic 112. The optical detector in spectrometer 114 transmits the spectral resolved radiation signal to signal processing controller 116 via signal cable 115.

Spectrometer 114 is preferably Ocean Optics, Inc. It may be a commercially available unit, such as a model S2000 manufactured by. Typically, a compact spectrometer will scatter and collect spectral signals over a wavelength range between about 200 nm and about 850 nm by an internal diffraction grating and optics, and an embedded CCD array with about 2048 pixels. In such a system, the optical resolution is typically about 1 nm. Optical emission spectra are collected during processing of the substrate at a sampling rate of about 1 to about 10 Hz.

In general, some forms of cooling system are coupled to the chuck to achieve thermal equilibrium once the plasma is ignited. In general, the cooling system itself consists of a cooling device that pumps the coolant through a cavity inside the chuck, and the helium gas pressurizes a small gap between the chuck and the substrate. In addition to removing the generated heat, helium gas also allows the cooling system to quickly control heat dissipation. In other words, increasing the helium pressure results in an increase in the heat transfer rate. In addition, most plasma processing systems are controlled by sophisticated computers that include operating a software program. In a typical operating environment, manufacturing process parameters (eg, voltage, gas flow mix, gas flow rate, pressure, etc.) are generally configured for a particular plasma processing system and specific method.

In a conventional substrate manufacturing method, known as dual damascene, the insulating layers are electrically connected by conductive plugs filling the via holes. Generally, openings are formed in an insulating layer, usually aligned with a TaN or TiN barrier, and then conductive materials (e.g., aluminum (Al), copper () that allow electrical contact between two sets of conductive patterns. Cu) and the like). This establishes electrical contact between two active regions on the substrate, such as the source / drain regions. The excess conductive material on the surface of the insulating layer is typically removed by chemical mechanical polishing (CMP). A blanket layer of silicon nitride is then deposited to cap copper.

There are three commonly used approaches for making dual damascene substrates: via-first, trench-first, and dual hard mask. In one example of the via-first method, the substrate is first coated with photoresist and the vias are then patterned lithographically. Next, the anisotropic etch cuts through the surface cap material and etches through the low-k layer of the substrate and stops at the silicon nitride barrier, just above the bottom metal layer. Next, the via photoresist layer is removed, the trench photoresist is applied and lithographically patterned. Typically, to prevent the lower partial vias from overetching during the trench etch process, a portion of the photoresist will remain at the bottom of the vias, or the vias may be covered by an organic ARC plug. The second anisotropic etch is then cut through the surface cap material and the low-k material is etched to the desired depth. This etch forms the trench. The photoresist is then removed and the silicon nitride barrier at the bottom of the via is very soft and opens to a low energy etch to prevent the underlying copper from sputtering into the via. As mentioned above, the trenches and vias are filled with a conductive material (eg, aluminum (Al), copper (Cu), etc.) and polished by chemical mechanical polishing (CMP).

Another method is trench-first. In one example, the substrate is coated with photoresist and a trench lithography pattern is applied. The anisotropic dry etch is then cut through the surface hard mask (again typically SiN, TiN, or TaN) and subsequently removed the photoresist. Another photoresist is applied over the trench hard mask, after which the vias are lithographic patterned. The second anisotropic etch is then cut through the cap layer and partially etched with the low-k material. This etch forms partial vias. The photoresist is then removed against the trench etch over the vias with the hard mask. The trench etch then cuts through the cap layer and partially etches the low-k material to the desired depth. The etch also stops on the final barrier located at the bottom of the via and simultaneously removes the via holes. The bottom barrier is then opened with a special etch.

The third method is a dual hard mask. This method combines oxidized etch steps, but requires two separate ILD (interlayer insulator) depositions between the nitride mask and the etch step. Low (via) insulators are deposited at the nitride etch stops on both the top and bottom. Top nitride is masked and etched to form a via hard mask. This requires a special nitride etch process. After that, a top (line) insulator is deposited. Finally, the trench mask is aligned with the via opening etched in the nitride, and both the trench and the via are etched in both layers of oxide in one etch step.

To facilitate the review, FIG. 2A shows an idealized cross-sectional view of the layer stack, representing the layers of the exemplary substrate, prior to the lithography step. In the discussion that follows, terms such as “above” and “below”, which may be used herein to discuss the spatial relationship between layers, do not always need to mean direct contact between related layers. You may. It should be noted that other additional layers may be present above, below, or in between. In addition, not all illustrated layers need necessarily be present, some or all of which may be replaced by other separate layers.

At the bottom of the layer stack, a layer 208 including a semiconductor is shown. Over the layer 208 a barrier layer 204 is typically disposed comprising nitride or carbide (SiN or SiC). The dual damascene substrate further includes a set of metal layers comprising M1 209a-b, typically comprising aluminum or copper. Above the barrier layer 204 is an intermediate dielectric (IMD) layer 206 comprising a low-k material (eg, SiOC, etc.) is disposed. Over the IMD layer 206, a cap layer 203, which typically includes SiO ₂ , may be disposed. Over the cap layer 203, a trench mask layer 202, which typically includes TiN, SiN, or TaN, may be disposed.

2B shows a somewhat idealized cross-sectional view of the layer stack of FIG. 2A after further photoresist layer 220 and BARC layer 212 are added.

2C shows a somewhat idealized cross-sectional view of the layer stack of FIG. 2B after photoresist layer 220 and BARC layer 212 have been processed via lithography. In this example, a photoresist mask pattern is created with the set of trenches 214a and 214b.

2D shows a cross-sectional view of the layer stack of FIG. 2C after trench mask layer 201 has been processed in a plasma system such that trenches 214a and 214b extend further to cap layer 203.

2E shows a cross-sectional view of the layer stack of FIG. 2D after photoresist layer 220 and BARC layer 212 are removed.

FIG. 2F shows the layer stack of FIG. 2E after the second photoresist layer 216 and the BARC layer 218 are disposed to create a second metal layer and vias connecting it to the first metal layers 209a and 209b. The cross section of the is shown.

2G shows a cross-sectional view of the layer stack of FIG. 2F after the photoresist layer is open and etching is performed to partially etch with IMD layer 206 to form vias.

2H shows that after the photoresist layer 216 and BARC layer 218 are removed, an additional etch process is performed to extend the trench to the desired depth, and etching through the via is stopped on the barrier layer 204. FIG. 2G is a cross-sectional view of the layer stack. FIG.

In FIG. 2I, the barrier layer 204 is etched by using, for example, CH ₂ F ₂ , CH ₃ F, and the like.

In FIG. 2J, a chemical mechanical polishing process is performed to polish the layer stack to the cap layer 203, wherein the conductive material (eg, aluminum (Al), copper (Cu), etc.) is present in the M1 metal material. Is arranged to contact.

One of the most important process steps during the plasma etch process is the endpoint. In general, an endpoint refers to a set, or range (eg, time) of values in a plasma process in which the process is considered complete. For example, when etching vias, it is important to determine when the barrier layer, such as SiN, should substantially penetrate in order to minimize the amount of etching to the underlying layer.

However, in such or other plasma processes, it is often difficult to monitor the process because process conditions may vary within the plasma processing system due to chamber residue accumulation, plasma damage to the chamber structure, and the like.

One common technique used within plasma processing systems is optical emission spectroscopy (OES). In OES, optical radiation from a set of selected chemical species (ie, radicals, ions, etc.) in a plasma processing system may be correlated with process thresholds, such as endpoints. That is, each type of species activated in the plasma processing chamber generally has a unique spectral signal, typically corresponding to a unique set of electromagnetic radiation wavelengths (generally between about 245 nm and about 800 nm). By monitoring for any other species or intensity of a particular wavelength not substantially produced by the plasma process itself, the process threshold can be determined by observing a change in the relative amount of a particular species in the plasma chamber.

For example, when SiO ₂ is etched using a CF-based etchant (eg, CF ₄ ), CO species are generally produced at a specific wavelength of about 483.5 nm. Likewise, when SiN is also etched with CF-based etchant, N species are generally produced at a specific wavelength of about 674 nm. Once suitable SiO ₂ or When the SiN material is substantially consumed, the corresponding wavelength of the calculated species generally drops, signaling that the process has reached an end point.

Referring now to FIG. 3, optical emission spectral snaps on a blanket oxide substrate (Ar / C ₄ F ₈ / CH ₂ F ₂ / O ₂ chemistry-6 kW / 50 mTorr) in which wavelength 304 is mapped to signal intensity 302. A simplified example of a shot is shown. In this example, CF ₂ 306 shows significant spectral emission for 275 nm and 321 nm. CO 308 shows significant spectral emission for 451 nm, 520 nm, 561 nm, and 662 nm. H 310 shows significant spectral emission for 656 nm. Ar 312, on the other hand, shows significant spectral emission for 750 nm.

However, current optical spectroscopy endpoint detection methods may have the problem that plasma optical radiation is sensitive to changes in chamber conditions. In some respects, these changes in plasma optical emission may be comparable to the expected changes used to trigger the endpoint call, thus resulting in a false endpoint call. Also, since only a small portion (typically less than 1%) of the total surface area may actually yield a signal change at the endpoint, the change may be difficult to detect in the presence of the background chamber OES signal. In addition, effective mission spectral analysis also becomes more difficult due to escalating conditions for substrates with sub-micron via contacts and high aspect ratios.

In view of the foregoing, there are preferred methods and apparatus for determining an endpoint in a plasma processing system.

Summary of the Invention

In an embodiment, in a plasma processing system, the present invention relates to a method of determining a process threshold. The method includes exposing a substrate to a plasma process that includes a process start portion, a substantially steady state portion, and a process end portion. The method also includes collecting the first data set during the substantial steady state portion; Generating a first statistical model comprising at least a statistical model component selected from the group consisting of a deviation component and an error component; And collecting the second data set. The method further includes generating a second statistical model comprising the statistical model component, wherein if the statistical model component of the first statistical model is substantially different from the statistical model component of the second statistical model, the process threshold is substantially Obtained.

In a plasma processing system, in one embodiment, the present invention relates to an apparatus for determining a process threshold. The method includes means for exposing a substrate to a plasma process comprising a process start portion, a substantial steady state portion, and a process end portion. The method also includes means for collecting the first data set during the substantial steady state portion; Means for generating a first statistical model comprising at least a statistical model component selected from the group consisting of a deviation component and an error component; And means for collecting the second data set. The method further includes means for generating a second statistical model comprising the statistical model component, wherein if the statistical model component of the first statistical model is substantially different than the statistical model component of the second statistical model, the process threshold is substantially Obtained.

These and other features of the invention will be described in detail in the detailed description below, in conjunction with the following figures.

Brief description of the drawings

The present invention is described by way of example and not by way of limitation, in the figures of the accompanying drawings which refer to like elements.

1 shows a simplified diagram of a plasma processing system.

2A-2J illustrate idealized cross-sectional views of a layer stack, representing a layer of an exemplary substrate.

3 shows a simplified illustration of optical emission spectral snapshots for a blanket oxide substrate.

4 illustrates a simplified process employing a statistical model used in a plasma processing system in which deviations are used to determine process thresholds (ie, endpoints, etc.), in accordance with an embodiment of the invention.

5 illustrates a simplified process of employing a statistical model used in a plasma processing system in which an error is used to determine a process threshold, in accordance with an embodiment of the invention.

6 shows a simplified diagram showing optical emission of CF ₂ relative to a substrate in a plasma processing system, in accordance with an embodiment of the invention.

FIG. 7 illustrates a simplified diagram showing that a set of hotelling T ² deviations is generated from a set of statistical models that includes a set of substantial steady state measurements and a set of process termination measurements, in accordance with an embodiment of the invention.

8 shows a simplified diagram illustrating that a set of q statistical errors is generated from a set of statistical models that includes a set of substantial steady state measurements and a set of process termination measurements, in accordance with an embodiment of the invention.

Detailed Description of the Preferred Embodiments

The invention will be described herein with reference to some preferred embodiments as depicted in the accompanying figures. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to one skilled in the art that the present invention may be practiced without some or all of these specific details. In other respects, well known process steps and / or structures have not been described in detail in order not to unnecessarily obscure the present invention.

Without wishing to be bound by theory, the inventors here believe that a statistical model of the plasma process can be used to determine process thresholds, such as etch endpoints. In general, many statistical analysis techniques may convert a set of measurements or samples into a statistical model that reasonably describes the observed measurements and, if possible, predicts the measurements.

The statistical model itself is a first set of components (sometimes called a deviation) that describes how the new sample follows the statistical model and a second set of components (often an error) that captures the deviation in the new sample that does not follow the statistical model. May be called). In an obscure method, a relatively more sensitive statistical model may be generated from a set of measurements during the portion of the plasma process with relatively small deviations. That is, deviations and errors in the statistical model may be relatively small. A new next measurement, which substantially increases the deviation or error, may signal a process threshold, such as an etch endpoint. In one embodiment, a statistical model is generated for each individual substrate, resulting in a substantially reduced sensitivity of process matching, plasma chamber matching, and process threshold detection caused by substrate matching. In another embodiment, the statistical model includes a set of restriction rings. In yet another embodiment, the statistical model comprises a low open area etch plasma process.

However, as described above, in such and other plasma processes, it is often difficult to monitor the process because process conditions may be dynamic within the plasma processing system due to chamber residue accumulation, plasma damage to the chamber structure, and the like. .

A common statistical technique used in dynamic environments is principal components analysis (PCA). PCA, a multivariate technique, can correlate a number of variables that are periodically measured and transformed into uncorrelated variables, or smaller sets of elements, that account for the major deviations in the data set. The PCA finds a combination of variables or elements that describe the main trends in the data set and represents each as a series of principal components. For example, PCA may be used to generate a factored model based on a set of electromagnetic emission spectra sequentially measured during a target etch step.

Once the PCA model is generated, the following measurements can then be compared to the PCA model to determine process thresholds such as endpoints and the like. An end point generally means a set, or range (eg, time) of values in the plasma process at which the process is considered complete. In general, the process engineer defines a range of measurements needed before a substantially representative PCA model can be generated based on information from the plasma process (eg, etch rate, etc.).

To increase the sensitivity of the statistical model, the model may be generated from the actual steady state period of the process. That is, most plasma processes generally consist of three stages: process start, steady state, and process end. During the process initiation phase, where pressure, power, and chemistry may exhibit significant transients prior to plasma stability, the set corresponding to the measurements is relatively high in deviation (for PCA generally measured by T ² statistics) and There will typically be an error (for PCA generally measured by Q statistics). After a certain time interval, typically after a few seconds, the process enters a steady state period where the next measurements tend to have relatively low deviations and errors. Finally, during the process termination phase, the set corresponding to the measurements again tends to have relatively high deviations and errors.

By generating an initial statistical model from a steady state set of measurements, the overall model deviation and error component is relatively small when compared to a model that includes both process start and steady state steps. Although traversing from the steady state phase to the process termination phase may have a minimum deviation in the OES signal, the PCA projection using the PCA model still requires a sufficient increase in the deviation and error to determine that a process threshold has been obtained. It can also be captured. Once a PCA model is determined from a steady state using a substantially specific number of principal components, the PCA projection method uses the eigenvalues and eigenvectors of the covariances obtained from the steady state step to terminate the step. PCA parameters (e.g., Q, T2, etc.) may be calculated at.

In US Pat. No. 5,288,367 a method has been proposed using an approach of principal component analysis where a particular wavelength of the emission spectrum is automatically determined and where the end point of the etching is detected based on the particular wavelength. According to this method, a specific wavelength can be determined automatically. However, unlike the present invention, the method includes a statistical model that includes all of the process start, steady state steps, and end steps. That is, as in the present invention, in contrast to the PCA projection onto the substantial PCA modeling and termination phase of the steady state portion of the process, from the beginning to the end of the process, the intensity for each measured spectrum is continuously tracked. And PCA modeled. In addition, US Pat. No. 5,288,367 is based primarily on a set of principal components to determine endpoints, as opposed to using deviations or errors in statistical models such as the present invention.

Mathematically, PCA is based on covariance of process variables or eigenvector analysis of correlation matrix. For a given data matrix X in m rows n columns, the covariance matrix of X is defined as:

Once the columns of X are mean centered (i.e., adjusted to have a zero mean by subtracting the mean of each column), they are automatically scaled (i.e. adjusted to zero mean and unit deviation by dividing each column by its standard deviation) Equation 1 gives the correlation matrix of X.

PCA analyzes the data matrix X as the sum of the products of the vectors t _i and p _i and the error matrix E:

While the p _i vector is the eigenvector of the covariance, the t _i vector is known as the score and contains information about how the samples are related to each other.

The first principal component (t ₁ p ^T ₁ ) is typically not used to determine the end point, and generally accounts for about 80% of the total deviation and averages due to drift window transmission caused by deposition of windows or the like. Follow the signal level change. The second principal component (t ₂ p ^T ₂ ), the third principal component (t ₃ p ^T ₃ ), and possibly the fourth principal component (t ₄ p ^T ₄ ), generally have a proportion of less than 20% of the overall deviation, It may be generally used to detect an endpoint. The remaining principal components generally contain noise and are therefore not commonly used for important patterning.

It is also possible to calculate the error, Q statistics for each sample. Q is simply the sum of the squares of each row of E (from Equation 2), for example, for xi, the i th sample in X:

e _i is the i th row of E, P _k is the matrix of the first k loading vectors held in the PCA model (each vector is a column of P _k ), and I is a unit matrix of appropriate size (n by n) . Therefore, when the PCA model is generated by the mth number of principal components from the steady state, Qj at the end step by PCA projection is expressed as follows:

The Q statistic is a measure of the amount of deviation in each sample not obtained by the m principal component maintained in the model from steady state. At the same time, it is a measure of the new amount of deviation in the termination phase as opposed to the steady state. As discussed above, by generating a PCA model from measurements in a steady state period and performing PCA projection onto an end step, the Q statistics may give a signal that crosses a process threshold, such as an end point.

The measure of deviation within the PCA model is given by the T ² statistics of the hotel ring. T ² is the sum of the standardized square scores defined as:

Where ti is the i-th row of Tk, and the matrix of k gets the vectors from the PCA model. Matrix l-1 is a diagonal matrix containing the inverse eigenvalues associated with the k eigenvectors (main components) held in the model. If the PCA model is generated by the m th number of principal components from steady state, then T _j ² at the end step by PCA projection Is expressed as:

Where P _m is the matrix of loading vectors of the PCA model from steady state. As described above, by generating the PCA model from the measured value in the steady state period, T ² by the PCA projection method at the end stage The statistics may give a signal that crosses process thresholds, such as endpoints.

Typical plasma processing system measurements that may be used with the PCA include: plasma species presence or concentration measured by optical emission, residual gas analyzer, optical absorption, etc., bias voltage of the substrate electrode, ESC DC current, and RF voltage, current, phase phase, and other electrical parameters such as associated harmonics, RF tuning frequency to match the plasma to generator impedance in a frequency tuning system, or RF tuning capacitance / inductance to match the plasma to generator impedance in a variable capacitor / inductor matching network.

For example, in endpoint determination, various aspects of the plasma process may be measured (eg, optical emission signal strength at wavelengths corresponding to a particular species, electrical measurements, etc.), and then statistics that may substantially determine the endpoint. Can be converted to a model.

  As mentioned above, endpoint determination is difficult for plasma processes aimed at etching relatively small open (unmasked) areas of the entire surface area of the substrate (eg, low open area etching, etc.). When using OES, this problem is because a small change in certain species can make it difficult to detect a corresponding signal change in the presence of a background signal from these species present in the plasma at any level before the endpoint. Worse. In particular, such perturbations in plasma optical radiation may be comparable to the expected perturbation used to trigger the endpoint call, thus causing false endpoint call occurrences.

Referring now to FIG. 4, in accordance with one embodiment of the present invention, a simplified process is employed that employs a statistical model used in a plasma processing system in which deviations are used to determine process thresholds (ie, endpoints, etc.). First, in step 402, for a substantially steady state step of the plasma process, OES spectral samples are collected. Next, an initial statistical model (eg, PCA, etc.) is generated. That is, an X-1 statistical model is generated that includes X-1 deviations and X-1 errors in step 404. Then, in step 406, additional OES spectral samples are collected. Then, a second statistical model is generated. That is, in step 408, an X statistical model is generated that includes the X deviation and the X error. In step 410, if the (before) X-1 deviation is not substantially smaller than the (after) X deviation, the process threshold is not reached and process monitoring continues according to X = X + 1 in step 414. That is, at step 406, additional OES spectral samples are collected again, and another statistical model is generated. In step 410, if the (before) X-1 deviation is substantially less than the (after) X deviation, then the process threshold is reached in step 412.

Referring now to FIG. 5, shown is a simplified process employing a statistical model used in a plasma processing system in which errors are used to determine process thresholds (ie, endpoints, etc.), in accordance with an embodiment of the present invention. First, in step 502, OES spectral samples are collected for a substantially steady state step of the plasma process. Next, an initial statistical model (eg, PCA, etc.) is generated. That is, in step 504, an X-1 statistical model is generated that includes X-1 deviations and X-1 errors. Then, in step 506, additional OES spectral samples are collected. Then, a second statistical model is generated. That is, in step 508, an X statistical model is generated that includes the X deviation and the X error. In step 510, if the (previous) X-1 error is not substantially smaller than the (after) X error, the process threshold is not reached and process monitoring continues according to X = X + 1 in step 514. That is, at step 506 additional OES spectral samples are collected again and another statistical model is generated. In step 510, if the (before) X-1 error is substantially smaller than the (after) X error, then in step 512, a process threshold is reached.

Referring now to FIG. 6, in accordance with one embodiment of the present invention, only about 0.8% of the surface area of the substrate is unmasked, and the plasma processing system (50 mT / 6 kW / Ar / C ₄ F ₈ / O ₂ process) is etched. A simplified diagram showing the optical emission of CF ₂ to the substrate is shown at. After about 70 seconds, a process endpoint appears at step 402. However, since the etched surface area is less than about 1% of the total surface area of the substrate, the corresponding detectable signal change at 260 nm wavelength is only about 0.5%.

Referring now to FIG. 7, in accordance with one embodiment of the present invention, a simplified representation is shown that a set of hotelling T ² deviations is generated from a set of statistical models that includes a set of substantial steady state measurements and a set of process termination measurements. It is also. As discussed above, an initial set of statistical models is generated from a set of steady state measurements. Thus, when compared to a model that includes both process start and steady state steps, the overall model deviation and error are relatively small. At 702, the intersection from the steady state phase to the process termination phase at about 80 seconds may substantially increase the deviation and error of the statistical model, signaling that a plasma process threshold, such as an endpoint, has been obtained.

Referring now to FIG. 8, there is shown a simplified diagram in which q statistical errors are generated from a set of statistical models comprising a set of substantial steady state measurements and a set of process termination measurements, in accordance with an embodiment of the invention. As discussed above, an initial set of statistical models is generated from a set of steady state measurements. Thus, when compared to a model that includes both process start and steady state steps, the overall model deviation and error component are relatively small. At 702, the intersection from the steady state phase to the process termination phase at about 80 seconds may substantially increase the deviation and error component of the statistical model, signaling that a plasma process threshold, such as an endpoint, has been obtained.

Although the invention has been described in several preferred embodiments, there are variations, substitutions, and equivalents falling within the scope of the invention. For example, although the invention has been described in connection with a plasma processing system (eg, Exelan ^™ , Exelan ^™ HP, Exelan ^™ HPT, 2300 ^™ , Versys ^™ Star, etc.) from Lam Research Corp., other plasma processing systems This may be used. In addition, the present invention may be used with substrates of various diameters (eg, 200 mm, 300 mm, etc.). In addition, a photoresist plasma etchant comprising gases other than oxygen may be used. In addition, there are a number of alternatives for implementing the method of the present invention.

In addition, other statistical analysis techniques, such as partial least squares (PLS), may be used. In addition, the set of measurements may include electromagnetic radiation, physical changes in the plasma processing system (eg, pressure, temperature, limit ring position, etc.), and RF changes (RF bottom power, RFB reflected power, RF tuning frequency, RF load). ), Phase error, RF power, RF impedance, RF voltage, RF current, and the like. The claimed invention may also be used to optimize the process model for other types of plasma processes in a plasma processing system.

Advantages of the invention include methods and apparatus for optimizing the determination of process endpoints in a plasma processing system. Additional advantages include optimizing the process model in the plasma processing system, generating statistical models that are more sensitive to process threshold determination, and dynamic generation of statistical models for each individual substrate. In this example, as shown in FIG. 6, the steady state portion is selected for 40 <t <50 seconds. Signal perturbation occurs at about t = 30-40 seconds (601) because of the limiting ring operation. If perturbation is expected during the steady state portion, such perturbations should be included in the first model set. For example, such perturbations may occur if the confinement ring is not fixed.

Although the exemplary embodiments and the best embodiments have been disclosed, variations and changes to the disclosed embodiments may be made within the spirit and spirit of the invention as defined by the following claims.

Claims (48)

In a plasma processing system, a method of determining a process threshold, Exposing the substrate to a plasma process comprising a process start portion, a substantial steady state portion, and a process termination portion; Collecting a first data set during the substantially steady state portion; Generating a first statistical model comprising at least a statistical model component selected from the group consisting of a deviation component and an error component; Collecting a second data set; And Generating a second statistical model comprising the statistical model component, If the statistical model component of the first statistical model is substantially different from the statistical model component of the second statistical model, the process threshold is substantially obtained. The method of claim 1, Wherein the first statistical model and the second statistical model comprise principal component analysis. The method of claim 1, Wherein the first statistical model and the second statistical model comprise partial least squares (PLS). The method of claim 1, And the plasma process is an etch process using an etchant. The method of claim 1, Wherein the process threshold is an endpoint. The method of claim 4, wherein The etchant is CF ₄ , process threshold determination method. The method of claim 4, wherein Etchant CHF ₃ Process threshold determination method. The method of claim 4, wherein The etchant is C ₄ F ₆ , the process threshold determination method. The method of claim 4, wherein Etchant C ₄ F ₈ Process threshold determination method. The method of claim 1, And the plasma process is a low open area etch. The method of claim 1, And the first data set and the second data set comprise optical emission. The method of claim 1, Wherein the first data set comprises optical emission signals collected at multiple confinement ring locations to include standard signal perturbations caused by optical collection aperture changes. The method of claim 1, Wherein the first data set and the second data set comprise electrical measurements in an RF transmission system. The method of claim 1, And the first data set and the second data set comprise plasma species present. The method of claim 1, And the first data set and the second data set comprise RF power. The method of claim 1, Wherein the plasma process is an insulating film etch. The method of claim 1, And the first data set and the second data set comprise chamber pressure. The method of claim 1, And the first data set and the second data set comprise an RF matching network tuning impedance. The method of claim 1, And the first data set and the second data set comprise an RF voltage measured on an RF transmission system. The method of claim 1, And the first data set and the second data set comprise a wafer DC bias voltage. The method of claim 1, Wherein the first data set and the second data set comprise impedances measured on an RF transmission system. The method of claim 1, And the first data set and the second data set comprise RF tuning frequencies. The method of claim 1, Wherein the first statistical model and the second statistical model comprise a limiting ring operation. In a plasma processing system, a method of constructing an in-situ substrate processing model, Exposing the substrate to a plasma process comprising a process start portion, a substantial steady state portion, and a process end portion; Collecting a first data set during the substantially steady state portion; Generating a first statistical model comprising at least a statistical model component selected from the group consisting of a deviation component and an error component; Collecting a second data set; And Generating a second statistical model comprising the statistical model component, If the statistical model component of the first statistical model is substantially different from the statistical model component of the second statistical model, the process threshold is substantially obtained. In a plasma processing system, an apparatus for determining a process threshold, Means for exposing the substrate to a plasma process, comprising a process start portion, a substantial steady state portion, and a process end portion; Means for collecting a first data set during the substantially steady state portion; Means for generating a first statistical model comprising at least a statistical model component selected from the group consisting of a deviation component and an error component; Means for collecting a second data set; And Means for generating a second statistical model comprising the statistical model component, And if the statistical model component of the first statistical model is substantially different from the statistical model component of the second statistical model, the process threshold is substantially obtained. The method of claim 25, And the first statistical model and the second statistical model comprise principal component analysis. The method of claim 25, And the first statistical model and the second statistical model comprise a partial least squares method. The method of claim 25, And the plasma process is an etch process using an etchant. The method of claim 25, And the process threshold is an endpoint. The method of claim 4, wherein And the etchant is CF ₄ . The method of claim 4, wherein The etchant is CHF ₃ Process threshold determination device. The method of claim 4, wherein And the etchant is C ₄ F ₆ . The method of claim 4, wherein And the etchant is C ₄ F ₈ . The method of claim 25, And the plasma process is a low open area etch. The method of claim 25, And the first data set and the second data set comprise optical radiation. The method of claim 25, And the first data set comprises optical emission signals collected at multiple confinement ring locations to include standard signal perturbations caused by optical collection aperture changes. The method of claim 25, And the first data set and the second data set comprise electrical measurements in an RF transmission system. The method of claim 25, And the first data set and the second data set comprise plasma species present. The method of claim 25, And the first data set and the second data set comprise RF power. The method of claim 25, And the plasma process is an insulating film etch. The method of claim 25, And the first data set and the second data set comprise chamber pressure. The method of claim 25, And the first data set and the second data set comprise an RF matching network tuning impedance. The method of claim 25, And the first data set and the second data set comprise an RF voltage measured on an RF transmission system. The method of claim 25, And the first data set and the second data set comprise a wafer DC bias voltage. The method of claim 25, And the first data set and the second data set comprise impedances measured on an RF transmission system. The method of claim 25, And the first data set and the second data set comprise an RF tuning frequency. The method of claim 25, And the first statistical model and the second statistical model comprise a confinement ring operation. In a plasma processing system, an apparatus for constructing an in-situ substrate processing model, Means for exposing the substrate to a plasma process, comprising a process start portion, a substantial steady state portion, and a process end portion; Means for collecting a first data set during the substantially steady state portion; Means for generating a first statistical model comprising at least a statistical model component selected from the group consisting of a deviation component and an error component; Means for collecting a second data set; And Means for generating a second statistical model comprising the statistical model component, And if the statistical model component of the first statistical model is substantially different from the statistical model component of the second statistical model, the process threshold is substantially obtained.

Images