JP2005217448A - Emission spectroscopic process apparatus and method for plasma process - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To sense a quick infinitesimal change of a separated spectrum with good reproducibility in a plasma process unit or the like. <P>SOLUTION: An emission spectroscopic process apparatus comprises a spectroscope 3 for spectrally separating an input light from the process unit, a photodetector 4 having a series of photodetecting elements each for detecting an amount of the separated input light at each wavelength, a first signal holding unit 10 for sequentially holding detecting signals of partial photodetecting elements adjacent in the series of photodetecting elements in a first period, an adder 11 for adding the detecting signals of the adjacent photodetecting elements of the photodetector including the detecting signals held by the adjacent partial photodetecting elements, and a second signal holding unit 12 for sequentially holding the added output of the adder, and further a signal process unit 9 for deciding a state of the process unit from an output of the second signal holding unit. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an emission spectral processing apparatus that spectrally divides light emitted from plasma or the like, converts each spectrum corresponding to a wavelength into an electrical signal by a light receiving element, and then performs signal processing on the signal to obtain a desired detection output. And a plasma processing method.

  There is known an emission spectral processing apparatus that separates light emitted from plasma, etc., converts each spectrum corresponding to the wavelength into an electrical signal by a light receiving element, and further performs signal processing on the converted signal to obtain a desired output. Yes. For example, Japanese Patent Application Laid-Open No. 2001-60585 discloses a process monitor apparatus using the following principal component analysis.

  That is, the electromagnetic radiation from the plasma chamber is input to a process monitor device including a spectrometer and a processor via an optical fiber or the like. The spectrometer spatially separates the plasma electromagnetic radiation based on wavelength (eg, via a prism or diffraction grating) and detects multiple spatially separated wavelength spectra, eg, by a 2048 channel CCD array. Then, a detection signal (that is, an emission spectroscopy (OES) signal) is generated. The generated OES signal is digitized (eg, via an analog-to-digital converter) and output to the processor for further processing. Thus, the electromagnetic radiation from the plasma is measured by a spectrometer and supplied to the processor in the form of a 2048 channel OES signal.

  The particular type of principal component analysis process performed by the processor is selected by a remote computer system, a manufacturing execution system, or the like. Instead of spectrometers, diffraction gratings, prisms, optical filters and other wavelength selection devices are used with multiple detectors (eg photodiodes, photomultipliers etc.) to provide the processor with information about multiple electromagnetic radiation wavelengths Also good. Note that the processor is coupled to the plasma etch controller via a control bus.

  A CCD is often used as a means for easily obtaining dispersed light as an amplitude signal of light for each wavelength. In a CCD in which a large number of light receiving elements are integrated, noise increases when a light receiving element with a small capacity is used to increase sensitivity, and sensitivity decreases when a light receiving element with a large capacity is used to reduce noise. For example, in a CCD with relatively high sensitivity (for example, ILX511, 2048 pixel CCD linear sensor manufactured by Sony), the signal-to-noise ratio (S / N ratio) when a saturated light amount is received is about 250, and the amount of received light decreases. The S / N ratio decreases in proportion to the square of the amount of received light. This is a common problem not only for CCDs but also for optical elements in which a large number of light receiving elements are integrated.

  A normal image sensor measures the average value or peak value of the amount of light received over the entire screen in response to fluctuations in the amount of incident light, and changes the degree of amplification of the CCD output signal or the CCD accumulation time based on this measured value. The gain is adjusted (see, for example, Japanese Patent Laid-Open No. 2000-324297, USP2001 / 0016053A1).

  On the other hand, in the plasma processing apparatus, the amount of incident light fluctuates significantly (about 10 times or more) due to the contamination of the chamber over time. Responding to this variation by changing the accumulation time is not preferable because it significantly changes the timing of the entire system. In addition, a spectral spectrum from plasma emission or the like in a plasma processing apparatus is a mixture of a high-intensity part having a plurality of sharp peaks and a low-intensity part that changes relatively gently with respect to the wavelength (see, for example, US Pat. 17A, or Japanese Patent Laid-Open No. 2001-60585-FIG. 3C). When such a spectral spectrum is received using a CCD, the S / N ratio in the low-luminance portion is greatly reduced if the amplification level of the CCD output signal is set so as not to saturate the sharp peak portion.

Conversely, when amplification is performed in accordance with the low luminance part, the peak part is saturated.

  The temporal change in the spectral spectrum emitted from the processing chamber of the semiconductor manufacturing apparatus indicates the change in the processing content in the processing chamber, and in recent years, the processing status has been estimated from the slight change. However, when a CCD or the like is used as a means for detecting a spectral spectrum, only a signal having a low S / N ratio can be obtained as described above. For this reason, at present, noise removal is performed by adding signals of the same wavelength many times. In this method, for example, if the signal-to-noise ratio is to be increased by one digit, it is necessary to add sampling data 100 times or more. This process takes several seconds to several tens of seconds for a normal CCD, and it is relatively difficult to detect a fast minute change (change of about 10% or less) of less than about 1 second, preferably 0.5 seconds or less. . In particular, it is very difficult to detect a fast minute change of less than about 1 second with good reproducibility in a low-luminance portion that changes relatively relatively with respect to the wavelength of a spectral spectrum such as plasma.

  The present invention has been made in view of these problems, and provides an emission spectral processing apparatus capable of detecting a minute change in a spectral spectrum with high reproducibility and a plasma processing method using the same.

  According to an example of the present invention, the present invention employs the following means in order to solve the above problems.

  A spectroscope that splits input light from the processing device, a light receiving unit that includes a series of light receiving elements for detecting the light quantity of the split input light for each wavelength, and a part of a series of light receiving elements adjacent to each other A first signal holding unit that sequentially holds detection signals of the light receiving elements in a first period, and detection signals of adjacent light receiving elements of the light receiving unit including detection signals of the held adjacent light receiving elements. An adder for adding, and a second signal holding unit for sequentially holding the addition output of the adder, and a signal processing unit for determining the state of the processing device based on the output of the second signal holding unit Prepared.

  Furthermore, in another example of the present invention, a set of a light receiving unit including a series of light receiving elements for detecting the amount of the input light separated for each wavelength, and a light emission wavelength specific to a preset light emitting substance. An adder for adding the detection signals of the corresponding light receiving elements, and a third signal holding unit for sequentially holding the addition output of the adder, and based on the output of the third signal holding unit, A signal processing unit for determining the state is provided.

  According to the present invention, it is possible to provide an emission spectral processing apparatus capable of detecting a fast minute change appearing in a spectral spectrum with good reproducibility.

  FIG. 1 is a diagram showing a first embodiment of the present invention. In the figure, plasma emission generated in the processing chamber of the plasma processing apparatus 1 is input to the spectroscope 3 via the optical fiber 2 and the slit. The spectroscope 3 splits the input light that has passed through the slit at different angles for each wavelength. The split light is input to a CCD (Charge Coupled Device) 4 including a series of light receiving elements (usually hundreds or thousands, assumed to be 2048 in the following description). In this way, the light receiving element (detector) at a specific position in the CCD 4 detects the intensity of the spectrum of the specific wavelength component in the incident light.

  The timing generation circuit 5 generates a CCD reset timing signal and a CCD transfer clock signal. The CCD reset timing signal determines the accumulation time of charges accumulated in the CCD, and the CCD transfer clock signal determines the transfer speed of the time series signal output from the CCD 4 in time series. In the following description, these signals are collectively referred to as a CCD drive signal 6. The CCD 4 is driven by the CCD driving signal 6, and the wavelength distribution of light during plasma emission is output as a time series signal at predetermined intervals. Next, this time-series signal is input to an amplifier circuit 7 having an offset adjustment and gain adjustment function. In the conventional system, the output of the amplifying circuit 7 is directly input to a signal processing device 9 composed of a CPU or the like via an analog-digital converter (hereinafter referred to as an AD converter) 8, The wavelength distribution, the time change of the light intensity for each predetermined wavelength, and the like are configured to be displayed on the display in the signal processing device 9.

  On the other hand, in the present embodiment, a plurality of (n, n ≧ 2) first time series output signals (obtained by sequentially operating adjacent CCDs) from the amplifier circuit 7 are obtained. 1 is stored in the signal holding circuit 10. The outputs of the plurality of signal holding circuits 8 and the output of the amplifier circuit 7 are added by the addition amplifier circuit 11 and transferred to the second signal holding circuit 12 at a predetermined timing. A plurality of signals corresponding to a plurality of adjacent different timings (a plurality of adjacent different wavelengths) are added in this way, output from the second signal holding circuit 12, and converted into a digital signal by the AD converter 8. Thereafter, the signal is input to the signal processing device 9.

  In this way, by adding n + 1 (n is the number of first signal holding circuits) signals by the first addition amplifier circuit 11, the S / N ratio of the output signal of the CCD 4 is reduced to 1 / (n + 1). In addition to the improvement to the square factor, the amount of data input to the AD converter 8 can be reduced to 1 / (n + 1). By sequentially performing this addition process for each adjacent CCD 4 (for every n + 1 adjacent CCDs), the influence of noise input to the signal processing device 9 can be greatly reduced.

  When the number n of the first signal holding circuits 10 is 8, 16, 32, the S / N ratio is improved to about 3, 4 times, and 6 times, respectively. An integrated circuit in which eight signal processing circuits are integrated in one standard-sized integrated circuit is already on the market, and the size of this portion of the circuit is not a problem.

  Since the amount of data is reduced by the analog addition processing by the addition amplifier circuit 11, it is difficult to analyze the incident light with high resolution (wavelength resolution) as it is.

  When high resolution (wavelength resolution) is also required, the output from the amplifier circuit 7 and the output from the second signal holding circuit 12 are input to the analog switch 13, and the high resolution (wavelength resolution) is obtained. When analysis is required, the analog switching circuit 13 is switched to the amplification circuit 7 side according to a command from the signal processing device 9, and the output from the amplification circuit 7 is directly input to the signal processing device 9 through the AD converter 8. You can do it. With this configuration, a single device switches between a mode with a low wavelength resolution but a high S / N ratio (high resolution) and a mode with a low S / N ratio but high wavelength resolution. Can do.

  FIG. 2 is a diagram showing another embodiment. A portion corresponding to the first signal holding circuit 10 in FIG. 1 is a first-stage signal holding circuit 10A including n (n ≧ 2) signal holding circuits (10-11 to 10-1n), m A second-stage signal holding circuit 10B having a number (m ≧ 2) of signal holding circuits (10-21 to 10-2m) is connected in cascade through an addition amplifier circuit 11-1. By doing so, the S / N ratio can be improved to [1/2] times ([n + 1) * (m + 1)] with (n + m) small signal holding circuits. For example, when n = 8 and m = 8, an improvement of nearly 9 times is obtained as the S / N ratio.

  FIG. 3 is a diagram showing still another embodiment. This spectroscopic processing apparatus can simultaneously analyze light emission from two processing chambers in the plasma processing apparatus 1. In order to simultaneously analyze the light emission from the two processing chambers, the optical fibers 2-1 and 2-2, the light amount adjusters 14-1 and 14-2, the spectrometers 3-1 and 3-2, the CCDs 4-1 and 4 -2, amplifier circuits 7-1 and 7-2, first signal holding circuits 10-1 and 10-2, addition amplifier circuits 11-1 and 11-2, and second signal holding circuits 12-1 and 12- Two sets of 2 are required. On the other hand, the circuit for the CCD drive signal 6 can be simplified by adding the same signal to the two CCDs 4-1 and 4-2 in common. For this reason, one timing generation circuit 5 and one AD converter 8 are sufficient.

  In order to digitize the two signals with one AD converter 8, the outputs of the second signal holding circuits 12-1 and 12-2 are multiplexed in time series by the time division multiplexing circuit 21. Then, the signal is input to the AD converter 8 via the analog switch 13. The outputs of the second signal holding circuits 12-1 and 12-2 are directly input to the analog switch 13, and are alternately selected according to a command from the signal processing device 9 and AD-converted, whereby the time division multiplexing circuit 21 is It can be omitted. For example, when n = 16, the amount of signal is reduced to 1/9. Therefore, even if AD conversion is performed alternately, the AD conversion speed is 1/8 as compared with the case of the conventional single optical input. Just do it.

  Further, the light emission from the four processing chambers in the plasma processing apparatus 1 can be analyzed simultaneously. Even in this case, since the AD conversion speed can be reduced as compared with the case of a single optical input, a low-cost AD converter or a low-cost signal processing device can be used. This is a great advantage when a plurality of optical input processes are simultaneously performed using the summing amplifier circuit 11 or 11-1, 11-2. When the measurement light from the processing device and the reference light are measured by different CCDs, when the CCD is driven at the same timing, the time difference of the data collection time at the corresponding wavelength between the two CCDs is reduced to 0. And the calculation using the measurement light and the reference light at each wavelength can be accurately performed. In particular, when light that changes frequently, such as plasma light, is used as measurement light and reference light, the advantage of driving the plurality of CCDs at the same timing is great.

  As described above, when a plurality of CCDs are operated at the same timing, the accumulation time of each CCD becomes the same, and it becomes difficult to adjust the sensitivity for each of the plurality of CCDs. In particular, when the levels of light input to a plurality of CCDs are significantly different, as shown in FIG. 3, between the plasma processing apparatus 1 and the optical fiber 2, in the optical fiber 2, or between the optical fiber 2 and the spectroscope 3. A light amount adjuster 14 is installed, and a command from the signal processing device 9 is converted into an analog amount via a light amount setting DA converter 22, and then the light amount adjuster 14 is controlled via a light amount controller 23. Just do it. The light amount adjuster 14 may be a liquid crystal element in which the amount of transmitted light varies depending on the applied voltage, or a diaphragm mechanism in which the size of the light aperture varies depending on the applied voltage.

  In the above, the improvement of the S / N ratio and the reduction of the data amount by analog addition using the addition amplifier circuit 11 and the like have been described.

  The improvement of the S / N ratio is further enhanced when the improvement by the digital processing in the signal processing device 9 is used in combination. In the following, the spectral processing apparatus of FIG. 3 that processes two light inputs will be described as an example.

  With an accumulation time of 25 milliseconds, 16 adjacent signals are added by the summing amplifier circuit 11 to 128 signals of the same channel (same wavelength) input every 25 milliseconds, and 128 analog signals for each wavelength are added. Get. The analog signal is converted into a digital signal by the AD converter 8 and then input to the signal processing device 9. In the signal processing device 9, the signal for each wavelength input every 25 milliseconds is added 16 times between adjacent wavelengths and for each sampling. Thereby, the signal which performed the average process between 16 adjacent wavelengths for every wavelength, and the signal which performed the sampling averaging 16 times can be obtained. Desired signal processing is performed based on these signals, and desired signals corresponding to two processing chambers having a significantly improved S / N ratio are obtained every 0.5 seconds. Furthermore, the end points of the processing in the two processing chambers of the plasma processing apparatus 1 can be found independently based on this signal.

  The improvement of the S / N ratio in this case is about 4 times by analog addition, about 4 times by wavelength addition in the signal processing device 9, and about 4 times by addition at every sampling point in the signal processing device, that is, 4 *. 4 * 4 = about 64 times improvement can be obtained.

  When the S / N ratio of the CCD 4 is 250 at full scale, the S / N ratio of 1/64 of the full scale optical signal is reduced to 250 / √64 = about 30. Thus, it can be recovered to about 30 * 64 = 1900. Even a minute fluctuation (for example, 1%) in a dark light signal of 1/64 of full scale can be separated into about 20 stages.

  In the above description, the quantization noise at the time of AD conversion is omitted. However, when the signal becomes a minute signal of about 1/64 of full scale, the noise cannot be ignored, and the S / N ratio of the signal due to noise at the time of AD conversion or the like. Will decline. For example, when converting to a digital signal by a 12-bit AD converter, the noise increases by about 1/3000 to 1/2000 of full scale including quantization noise and other circuit noise. In consideration of this, the S / N ratio of 1/64 of the full-scale optical signal is reduced to half or less of the above value.

  FIG. 4 is a diagram showing still another embodiment. In this spectral processing apparatus, the degree of amplification at the time of analog addition is changed for each wavelength. This can reduce the influence of quantization noise and circuit system noise during AD conversion in the low luminance region.

  When a gain setting command for the addition amplifier circuit 11 is input from the signal processing device 9 to the timing generation circuit 5, the timing generation circuit 5 sets the addition amplification circuit 11 (set to a gain of 1) after the CCD reset timing signal. The conversion of the gain setting AD converter 15 (preferably with a sample hold function) is started at the timing when the output signal of (there is) is stored in the second signal holding circuit 12 (the gain setting AD converter 15 is 8 A low-priced and small-sized digital signal that is converted into a digital signal of 4 bits or less (about 4 or 5 bits) is sufficient. The address corresponding to the wavelength is set in the address circuit 16 by the signal from the timing generation circuit 5, and the digital signal which is the output (signal magnitude information) of the gain setting AD converter 15 is stored in the corresponding address in the memory 17. Is memorized.

  When this operation is performed for one accumulation time of the CCD, information on the signal magnitude is accumulated in the memory 17 for 2048 / (n + 1) wavelengths. Next, when a data output command with gain is input from the signal processing device 9 to the timing generation circuit 5, the amplification degree of the addition amplifier circuit 11 is converted to information on the magnitude of the signal in the memory 17 after the CCD reset timing signal. Set accordingly. At this time, the gain of the summing amplifier circuit 11 is set via the gain setting circuit 18 for every 2048 / (n + 1) wavelengths.

  The relationship between the signal magnitude information and the gain of the summing amplifier circuit 11 may be set as follows.

Signal magnitude information (vs. full scale) Gain of summing amplifier 11 1) From 1/4 to 1 ... A times 2) 1/8 to 1 / Less than 4 ... 2A times 3) 1/16 to less than 1/8 ... 4A times 4) 1/32 to 1/16 Less than ...... 8A times 5) Less than 1/32 ... 16 times times A value is usually less than 1 (For example: 1 / (n + 1), where n is the number of first signal holding circuits).

  The spectral signal of the plasma emission usually does not fluctuate significantly during a single sample processing. For this reason, if the gain of the summing amplifier circuit 11 is set only once at the time of stable discharge in the initial stage of sample processing, it does not usually cause a problem. However, a region in which the upper quantization bit changes due to minute fluctuations in the analog signal is also included in the signal. For this reason, after setting the gain of the summing amplifier circuit 11, it is set with a margin so as not to saturate even if the signal at that wavelength increases (generally about 1.3 times or more) during one sample processing. Good.

  The gain setting data of the summing amplifier circuit 11 is converted into an analog signal by the gain output circuit 19 and the third signal holding circuit 20, and the analog switcher is time-sequentially at the same timing as the output of the second signal holding circuit 12. 13 is output. For this reason, the gain setting data can be read via the AD converter 8 according to a command from the signal processing device 9. As described above, the gain of the summing amplifier circuit 11 need only be set once during stable discharge in the initial stage of sample processing. Therefore, the gain data may be read only once during the stable discharge in the initial stage of sample processing.

  The signal processing device 9 outputs the output of the third signal holding circuit 20 for the time series gain setting data set in this way and the second time series output for each accumulation time during one process. By calculating between the same wavelengths using the output data of the signal holding circuit 12, the true value for each wavelength can be continuously calculated during the plasma processing.

  Note that when only a minute temporal change in the emission spectrum during one process is to be detected, the above calculation using the gain setting data in the signal processing device 9 is not necessarily essential. Further, in this example, the case where the gain of the addition amplifier circuit 11 is set only once at the time of stable discharge in the initial stage of the sample processing has been described. However, when the spectral intensity of the wavelength during the sample processing changes significantly. Alternatively, the gain of the summing amplifier circuit 11 can be reset during the sample processing.

  FIG. 5 is a diagram showing still another embodiment. Unlike the above example, this spectral processing apparatus performs AD conversion by performing only amplification without adding signals of different wavelengths. Of course, the S / N ratio of the low-brightness component can be improved by setting the gain of the amplifier circuit 7 for each different wavelength by the gain setting circuit 18. However, since the S / N of the analog signal itself is lower than in the case of FIG. 4, the effect is less than in the case of FIG.

  The emission spectral processing apparatus has been described above. When this processing apparatus is used, a minute and fast change in light emission during plasma processing can be detected at an early stage. For example, in a plasma processing apparatus used for gate etching processing of a semiconductor having a gate length of 0.1 μm or less, the thickness of a base insulating film to be processed is as extremely thin as several nm to 1 nm. For this reason, the plasma treatment step is completed with the film remaining several to several tens of nanometers before the etching target film is completely etched, and then the next plasma is processed under another condition having a high selectivity with the base. Processing steps need to be started.

  In order to measure the remaining film amount of the etching target film during the process, it is necessary to observe the interference light from the wafer. In this method, the change in light for each wavelength is about 0.1% to several%. Few. On the other hand, when the spectral processing apparatus is used, the S / N ratio of the signal can be greatly improved, and the plasma processing step can be stopped in response to an early response of 1 second or less. Etching can be performed at a gate length of 0.1 μm or less.

  On the other hand, when thousands of wafers are continuously etched, in order to see changes in the processing chamber, it is necessary to see changes in the amount of light with a high wavelength resolution, but high-speed response is not always necessary. do not need. In such a case, the output signal of the amplifier circuit 7 shown in FIG. 3 is selected by the analog switch 13 and input to the signal processing device 9 via the AD converter 8. That is, in such an application, since wavelength resolution is required, averaging between wavelengths is not performed, but averaging between a plurality of sampling data is performed. As a result, the S / N ratio during signal processing can be improved.

  For example, if an operation of collecting data once every 0.5 seconds and for each wavelength is performed for 1 minute during etching, 120 points of data can be collected for each wavelength. By averaging the collected data for each wavelength, the S / N ratio can be improved by √120 = 10.9 times. FIG. 3 shows detection of a minute change that does not require wavelength resolution but requires a response of 1 second or less, and detection of minute change that requires a wavelength resolution but requires a response of several tens of seconds. Both can be implemented with the single device shown.

  The present invention can also be applied to detection of a minute change of a light emitting component that requires a high-speed response of 1 second or less, or monitoring of a minute change of each wavelength component of light emission. Further, when the change amount of the change exceeds a predetermined value, an abnormal signal is issued, a warning is displayed, or the next process is stopped, so that abnormalities in the plasma process can be prevented in advance. .

  As described above, according to the present embodiment, a minute change (less than 10%) of the component wavelength during light emission during plasma processing is performed at a high speed at a timing of 1 second or less (preferably 0.5 seconds or less). It can be processed stably. In addition, a mode for processing minute changes in the spectrum for each wavelength during light emission during plasma processing at high speed and stably, and a change in spectrum for each wavelength during light emission during plasma processing for each adjacent wavelength. The mode determined with high resolution can be switched and used with one device according to the application.

Next, digital processing of an optical signal obtained from the CCD will be described. First, the characteristics of plasma emission during the plasma etching process will be described. In the plasma etching process in the vacuum processing chamber, Cl ₂ , HBr, CF ₄ , and C ₅ F ₈ gases are used as process gases (reactive gases). In order to further increase the ionicity of the plasma, Ar gas is used. These gases are decomposed by plasma into highly reactive Cl, Br, and F atoms (radicals). These radical gases are etching materials such as silicon (Si), polysilicon (Si), oxide film (SiO ₂ ), nitride film (Si ₃ N ₄ ), BARC (Back Anti-Reflection Coating), Pt, Fe, Etching proceeds by reacting with SBT (SrBi ₂ Ta ₂ O ₉ ) or the like to produce reaction products such as SiCl, SiCl ₂ , SiF, SiBr, C ₂ , Co, CN, PtCl, FeCl, TaCl. When the material to be etched disappears, that is, when the etching is finished, the reaction product is not generated and decreases, and the radical gas increases.

  Therefore, the emission spectrum intensity during the plasma etching process is classified into the following three types (1), (2), and (3).

(1) A spectrum due to a reaction product that decreases at the end of etching of the material to be etched.

(2) A spectrum due to radicals increasing at the end of etching.

(3) A spectrum of a substance that is not related to the etching reaction and does not change before and after etching.

  In the conventional end point determination by plasma emission, the time change of the emission intensity (for example, the spectrum by the reaction product) of a specific spectrum wavelength in the emission spectrum is used. However, as described above, a noise component exists in the emission spectrum signal from the CCD according to the signal intensity. Therefore, in the etching end point detection using the differential waveform of the emission spectrum signal, this noise component detects the end point. It was difficult.

  Hereinafter, an embodiment of the present invention capable of removing this noise component will be described with reference to FIGS. First, it is assumed that a signal obtained by digitizing an emission spectrum signal of wavelength λ from the CCD at time t by the AD converter 8 can be represented by an optical signal component i (λ, t) and a noise component δi (λ, t). Here, the noise component δi (λ, t) is electrical noise of the CCD or light fluctuation noise.

  This emission spectrum signal i (λ, t) + δi (λ, t) is stored in the digitized data holding circuit 910 in the signal processing device 9 for one hour, and is used for the emission spectrum classification set in the setting circuit 911 in advance. Using the mask function M (λ), the emission spectrum signal i (λ, t) + δi (λ, t) and the mask function M (λ) are added by the arithmetic circuit 912 over all the measured wavelengths λ.

  The integrated ΣM (λ) [i (λ, t) + δi (λ, t)] value for this wavelength λ is stored in the added value holding circuit 913, and the differential processing circuit 914 determines the time differential value of the emission intensity. An etching end point determination is performed by the differential value determination circuit 915 using the time differential value of the emission intensity.

  The processing flow at this time is as shown in FIG.

  First, in process 800, etching conditions such as a material to be etched and a process gas are input, and in process 801, a mask function M (λ) is set for the CCD wavelength. Next, in step 802, etching is started and sampling of the optical signal from the CCD is started. In step 803, optical signals i (λ, t) + δi (λ, t) for each wavelength λ of the CCD are acquired. Is done.

  Next, in process 804, an addition value ΣM (λ) [i (λ, t) + δi (λ, t)] at all wavelengths λ of the optical signal and the mask function M (λ) is calculated. In process 805, Based on this added value ΣM (λ) [i (λ, t) + δi (λ, t)], a time differential value of the light emission intensity at time t is obtained.

  In step 806, the time differential value is compared with a preset differential determination value, and the process returns to the optical signal i (λ, t) + δi (λ, t) acquisition processing 803 or in step 807. It is determined whether to set the end of the etching process and the end of the optical signal sampling.

  Here, in the integrated ΣM (λ) [i (λ, t) + δi (λ, t)] for this wavelength λ, Σ [δi (λ, t)] is a random noise, and therefore, many wavelengths Addition by λ brings the integrated value close to zero, and therefore noise can be removed by this integration.

  Next, a method for classifying the emission spectrum will be described. The emission spectrum from the wavelength λ of the CCD is classified into an optical signal that decreases before and after the etching of the plasma etching process, an optical signal that increases, or an optical signal that does not change. Can be determined by the following method.

  (a) Create a database of reactive gases and reaction products of the material to be etched in advance using a spectral library, and the wavelength belonging to the reactive gas from the database will be a wavelength that increases before and after the etching, and also belongs to the reaction product. Wavelengths are classified as wavelengths that decrease before and after the end of etching, and other wavelengths are wavelengths that do not change with time regardless of the reaction.

  The spectrum library at this time is described in the following document.

CRC Handbook of Chemistry and Physics, David R. Lide, CRC Press,
RWB Pearse and AG Gaydon,
“The Identification of Molecular Spectra”
John Wiley & Sons, Inc. 1976
(b) Perform sample wafer processing (etching of a wafer containing the same kind of material to be etched). With respect to the time change of the emission spectrum at the time of the process, a differentiation process is performed for all wavelengths, and classification is performed according to the value of the first derivative value before and after the etching is completed. As the differential processing method, the method described in JP-A-2000-228397 can be used. That is, it is classified into the following three types.

i.
A wavelength with a negative first-order differential value is a wavelength that decreases before and after etching (due to the reaction product).

ii.
A wavelength with a positive first-order differential value is a wavelength that increases before and after etching (due to radicals).

iii.
The wavelength with zero first-order differential value is a wavelength that does not change before and after the etching is completed (irrelevant to the reaction).

  (c) Perform sample wafer processing (etching of wafers containing the same kind of material to be etched). The principal component analysis is performed on the temporal change of the emission spectrum for all wavelengths during the processing, the spectrum of each component is obtained, and the spectrum is classified according to the spectrum value of each component.

  The principal component analysis is described in the following document, and the method can be used.

Shigeo Minami “Waveform Data Processing for Scientific Measurement”
CQ Publishing, p220-226, 1986
K. Sasaki, S. Kawata, and S. Minami,
“Estimation of Component Spectral Curves
from Unknown Mixture Spectra "
Appl. Opt. Vol. 23, p1955-1959, 1984
In this case, classification is performed based on the spectrum value of a certain component obtained by principal component analysis. For example, a wavelength having a negative spectrum value of a certain component is a wavelength that decreases before and after etching (due to a reaction product), and a wavelength having a positive spectrum value is a wavelength that increases before and after etching (radical). Therefore, a wavelength having a spectrum value of zero is classified as a wavelength that does not change before and after the completion of etching (irrelevant to the reaction).

  However, in this method, with respect to the positive or negative value of the spectrum value of a certain component, positive is not necessarily a reaction product and negative is not a radical.

  An operator M (λ) is introduced to specify groups for the groups classified into three types by the above method. For example, the wavelength λ where the emission intensity increases before and after etching is M (λ) = 1, the wavelength λ where the emission intensity decreases before and after etching is M (λ) = − 1, and the wavelength λ where the emission intensity does not change before and after etching is Let M (λ) = 1.

The results of the BARC (Back Anti-Reflection Coating) etching process to which this embodiment is applied are shown in FIGS. The etching process gas at this time is a mixed gas of HBr, CF ₄ , O ₂ , and Ar.

First, FIG. 8 shows an example of an emission spectrum before and after the completion of the BARC etching. In this etching, many emission spectra such as CN, Co, and C ₂ which are reaction products of the BARC material decrease before and after the etching. It can be seen that the emission spectra of OH and O are increased. The time variation of the emission spectrum was differentiated to classify the time variation of the emission spectrum, and the mask function M (λ) shown in the figure was determined.

  Next, FIG. 9 shows the integrated ΣM (λ) [i (λ, t) + δi (λ, t)] obtained for the wavelength λ using the mask function M (λ). Means that the average value of the emission spectrum intensity is about 1220 counts, and the reduction of the emission amount 1/100 means that the emission spectrum intensity is reduced and the average value is about 12.2 counts. From this figure, it can be seen that even when the amount of light emission is reduced to 1/100, the present invention can sufficiently remove the noise component and the temporal change in the light emission intensity can be accurately obtained.

  Next, an embodiment of the present invention in consideration of the property that the noise component δi (λ, t) from the CCD is inversely proportional to the intensity of the optical signal component i (λ) will be described. The CCD ( As described above, the S / N ratio of Sony ILX511) can be expressed as approximately 250√ (i (λ) / 4000), and the noise component is δi (λ, t) = 1/250 * √ (4000 * i (λ, t)). For example, the noise is δi (λ, t) = 16 (S / N = 250) when i (λ) = 4000, but δi (λ, t) = 0.8 when i (λ) = 10. (S / N = 12.5).

  Here, when the integration for λ is simply performed, the contribution of the light emission amount to the S / N ratio can be ignored.

  This can be achieved by normalizing the integration ΣM (λ) [i (λ, t) + δi (λ, t)] for λ described above with the noise component δi (λ, t). That is, if the integration processing 804 for λ in FIG. 8 is ΣM (λ) [i (λ, t) + δi (λ, t)] / [1/250 * √ (4000 * i (λ, t))]. good.

  Therefore, the processing flow according to this embodiment is as shown in FIG. 10, where the processing 814 is the normalization processing described above, and the other processing is the same as in the embodiment of FIG.

  Next, an embodiment of the present invention will be described in which the plasma emission fluctuation can be offset when the plasma emission changes such as abnormal discharge.

  First, the mask function M (λ) is set to the mask function M (λ) = 2 for the wavelength λ where the emission intensity increases before and after the etching, and the mask function M (λ for the wavelength λ where the emission intensity decreases before and after the etching. λ) = − 2, the wavelength λ at which the emission intensity does not change before and after the etching is set as a mask function M (λ) = 1, thereby distinguishing the wavelength λ at which the emission intensity does not change before and after the etching. The integrated ΣM (λ) [i (λ, t) + δi (λ, t)] for the wavelength λ is normalized by the optical signal of the wavelength λ whose emission intensity does not change.

  That is, as shown in the process flow of FIG. 11, a process 824 is provided instead of the process 804 in the process flow of FIG. 8, and first, A * ΣM (λ) [i (λ, t) + δi (λ, t)] / ΣM (λ ′) [i (λ ′, t) + δi (λ ′, t)] is calculated.

Here, ΣM (λ ′) [i (λ ′, t) + δi (λ ′, t)] is an integrated value at a wavelength λ ′ where the emission intensity does not change before and after etching, and the coefficient A is plasma etching. A = ΣM (λ ′) [i (λ ′, t) + δi (λ ′, t)] / ΣM (λ) [i (λ, t) + δi (λ, t) at an appropriate time t ₀ after the start of processing )] Value.

  According to this normalization process 824, the light emission fluctuation is canceled when there is a fluctuation such as abnormal discharge in the plasma light emission. Therefore, according to this embodiment, it is possible to perform an accurate and reliable end point determination. .

  Here, when a wavelength λ whose emission intensity does not change before and after the etching is not found, an integrated value for the wavelength λ at which the emission intensity increases before and after the etching may be used as a normalized value. An effect can be obtained.

  In this case, the wavelength λ in the integration of A * ΣM (λ) [i (λ, t) + δi (λ, t)] / ΣM (λ ′) [i (λ ′, t) + δi (λ ′, t)] “Addition” may be performed for a wavelength λ ′ at which emission intensity increases before and after etching.

It is a figure which shows one Embodiment of the emission spectral processing apparatus concerning this invention. It is a figure which shows other one Embodiment of the emission spectral processing apparatus concerning this invention. It is a figure which shows another one Embodiment of the emission spectral processing apparatus concerning this invention. It is a figure which shows another one Embodiment of the emission spectroscopy processing apparatus concerning this invention. It is a figure which shows another one Embodiment of the emission spectral processing apparatus concerning this invention. It is a block diagram of the signal processing unit in one embodiment of the present invention. It is a flowchart explaining the digital signal processing in one Embodiment of this invention. It is a figure which shows an example of the light emission spectrum waveform and mask function in one Embodiment of this invention. It is a figure which shows an example of the change waveform of the emitted light intensity in one Embodiment of this invention. It is a flowchart explaining the digital signal processing in other one Embodiment of this invention. It is a flowchart explaining the digital signal processing in other one Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Plasma processing apparatus 2 Optical fiber 3 Spectrometer 4 CCD
DESCRIPTION OF SYMBOLS 5 Timing generation circuit 6 CCD drive signal 7 Amplification circuit 8 AD converter 9 Signal processing apparatus 10 1st signal holding circuit 11 Addition amplification circuit 12 2nd signal holding circuit 13 Analog switch 14 Light quantity adjuster 15 AD for gain setting Converter 16 Address circuit 17 Memory 18 Gain setting circuit 19 Gain output circuit 20 Output holding circuit 21 Time division multiplexing circuit 22 Light quantity setting AD converter 23 Light quantity control circuit

Claims (8)

A spectroscope that separates input light from the processing device;
A light receiving unit including a series of light receiving elements for detecting the light amount of the split input light for each wavelength;
A first signal holding unit for sequentially holding detection signals of a part of adjacent light receiving elements in a series of light receiving elements in a first period;
An addition unit for adding detection signals of adjacent light receiving elements of the light receiving unit including detection signals of the held adjacent partial light receiving elements;
A second signal holding unit for sequentially holding the addition output of the addition unit;
An emission spectral processing apparatus, comprising: a signal processing unit that determines a state of the processing device based on an output of the second signal holding unit. In the description of claim 1,
The first signal holding unit includes a first stage signal holding unit that sequentially holds detection signals of a part of adjacent light receiving elements in a series of light receiving elements in a first period, and an addition of the holding unit An emission spectroscopic processing apparatus comprising a second-stage signal holding unit that sequentially holds an output in a second period longer than the first period. In any one of Claims 1 to 2,
The emission spectral processing apparatus according to claim 1, wherein the signal processing unit includes selection means for inputting one of the output from the addition unit and a detection signal for each light receiving element adjacent to the light receiving unit. In any one of Claims 1 thru | or 3,
The emission signal processing apparatus according to claim 1, wherein the first signal holding unit holds a detection signal of the input light amplified at a different ratio for each of a plurality of light receiving elements adjacent to the light receiving unit. In any one of Claims 1 thru | or 3,
The said processing apparatus is a plasma etching processing apparatus, The said emission spectral processing apparatus stops the etching process of the said plasma etching processing apparatus based on the addition output of the said addition part, The emission spectral processing apparatus characterized by the above-mentioned.   Separating the plasma light from the vacuum processing chamber, converting to a time-series analog electrical signal composed of different wavelength components at a predetermined period, adding between analog signals of different wavelength components, and adding a plurality of added A step of converting the signal into a digital quantity for each predetermined period, a step of digitally adding the converted plurality of added signals for each of a plurality of signals, and a predetermined plasma based on this signal A plasma processing method using an emission spectral processing apparatus, comprising: determining an end point of a processing step; and ending a predetermined plasma processing step.   Separating the plasma light from the vacuum processing chamber, converting to a time-series analog electrical signal having different wavelength components at a predetermined period, adding between analog signals of different wavelength components, and adding a plurality of added A step of converting a signal into a digital quantity at each predetermined period, a step of digitally adding the converted plurality of added signals for each of a plurality of signals, and each of the digitally added Adding a signal for each wavelength with respect to a wavelength corresponding to a light emission wavelength unique to a preset substance, determining an end point of a predetermined plasma processing step based on the signal, and a predetermined plasma processing step. And a step of terminating the plasma processing method using an emission spectral processing apparatus.   Separating the plasma light from the vacuum processing chamber, converting to a time-series analog electrical signal having different wavelength components at a predetermined period, adding between analog signals of different wavelength components, and adding a plurality of added A step of converting a signal into a digital quantity at each predetermined period, a step of digitally adding the converted plurality of added signals for each of a plurality of signals, and each of the digitally added Adding a signal for each wavelength with respect to a wavelength corresponding to a light emission wavelength specific to a preset substance, and adding or subtracting the digitally added signal for each substance corresponding to the substance, or Dividing, determining the end point of the predetermined plasma processing step based on this signal, and ending the predetermined plasma processing step And having a step of plasma processing method using a light-emitting spectral processor.

Images