JP2009032855A - Yield prediction method and yield prediction system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To precisely predict a yield of a highly integrated semiconductor device with high performance. <P>SOLUTION: In the middle of a manufacturing process for a semiconductor device, a membranous characteristics of the semiconductor device is measured, and the measured data is used to predict a yield of the semiconductor device. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a yield prediction method and a yield prediction system, and more particularly to a yield prediction method and a yield prediction system based on an in-line film quality characteristic evaluation in a semiconductor device manufacturing process.

  In recent years, competition for development of semiconductor devices has been severe, and generations must be changed in two to three years. In such an environment, there is a strong demand for improving the design completeness in a short period from development to mass production. In such a time schedule, it is important to create a system that identifies the cause of the failure of the semiconductor device and quickly improves the yield of the semiconductor device. In addition, in order to supply a semiconductor device stably, it is indispensable to predict the yield during the manufacturing process of the semiconductor device.

  As an example of a yield prediction method in the middle of the manufacturing process of such a semiconductor device, a highly accurate yield prediction in the in-line defect inspection process is conventionally performed by superimposing the design layout of the semiconductor device and the defect data obtained by the defect inspection apparatus. It is carried out. This conventional in-line defect inspection will be briefly described below (see, for example, Patent Document 1).

As shown in FIG. 15, in the conventional in-line defect inspection, the image data of the semiconductor device collected by the defect inspection apparatus 104 and the layout data 111 of the semiconductor device are overlapped to identify a portion that may be defective. The yield of the semiconductor device can be predicted by extracting, identifying the wiring failure mode type of the extracted defective candidate wiring, and determining the influence of the wiring failure mode type on the semiconductor device.
JP 2003-23056 A (particularly paragraphs 0008 and 0009)

  However, as a result of the complexity and sophistication of semiconductor device manufacturing methods that accompany high integration and high performance of semiconductor devices, the accuracy of the prediction results can be improved when yield prediction based on the above-described conventional inline defect inspection is used. For this reason, it has become impossible to sufficiently prevent frequent occurrences such as a decrease in yield due to a decrease in semiconductor manufacturing process margin.

  In view of the foregoing, it is an object of the present invention to make it possible to predict the yield of highly integrated and high performance semiconductor devices with high accuracy.

  In order to achieve the above object, the inventor of the present application has studied the cause of poor accuracy of the yield prediction result by the conventional inline defect inspection, and has obtained the following knowledge.

  In the conventional defect inspection method, it was possible to discriminate mainly disconnection or pattern collapse due to thinning of the wiring pattern due to the lens aberration of the stepper, or short-circuiting of the wiring pattern due to particles or the like.

  However, with miniaturization, in particular, factors that cannot be detected by a defect inspection apparatus, for example, an interlayer insulating film represented by a BPSG (boro-phosphosilicate glass) film, a USG (undoped silicate glass) film, or a TEOS (tetraethylorthosilicate) film Variation in moisture content in the film or in the film, variation in orientation or crystal structure of a metal film typified by a TiN film, Al film or Cu film, variation in stress of a substrate typified by a Si substrate, or trench structure It has been found that failure modes due to variations in stress concentration in the semiconductor device (hereinafter, such factors inherent in the components of the semiconductor device are collectively referred to as “film quality characteristics”) cannot be ignored. As a result, when the yield prediction method based only on the conventional defect inspection is used, the accuracy of the prediction result becomes insufficient.

  Accordingly, the inventor of the present application performs yield prediction with higher accuracy than the yield prediction by the conventional defect inspection apparatus by measuring the film quality characteristics involved in the manufacture of the semiconductor device in-line and performing, for example, multivariate analysis. The present invention has come up with a yield prediction method and a yield prediction system for semiconductor devices by in-line evaluation.

  Specifically, the first semiconductor device yield prediction method according to the present invention measures the film quality characteristics of the semiconductor device during the manufacturing process of the semiconductor device, and uses the measured data to obtain the yield of the semiconductor device. Predict.

  According to the second method of predicting the yield of a semiconductor device according to the present invention, the film quality characteristic of the semiconductor device is measured during the manufacturing process of the semiconductor device, and the measured data is stored in a database as graphic data (a And (b) performing multivariate analysis using the graphic data stored in the database, and predicting the yield of the semiconductor device using the analysis result obtained in the step (b). (C).

  In the second semiconductor device yield prediction method of the present invention, the step (a), the step (b) and the step (c) are performed for each of the plurality of film quality characteristics of the semiconductor device, and the plurality of film quality characteristics are obtained. A step (d) of predicting the final yield of the semiconductor device by integrating the predicted yield of the semiconductor device with respect to each of the semiconductor devices. Here, in predicting the final yield of the semiconductor device, the yield predicted by the conventional in-line defect inspection may be included.

In the second method for predicting a yield of a semiconductor device according to the present invention, when the step (d) for predicting the final yield is provided, in the step (d), each of the plurality of film quality characteristics is determined as a random defect factor or systematic. The semiconductor device is classified into failure factors and classified according to failure factor types (that is, the types of film quality characteristics), the final yield of the semiconductor device is Y, the yield of the semiconductor device due to the random failure factor is Yr, and the systematic of the semiconductor device Ys is a yield due to a defect factor, and YrC1, YrC2, YrC3,..., A plurality of film quality properties are predicted for each of the film quality characteristics classified as the random defect factor among the plurality of film quality characteristics. Are predicted for each of the film quality characteristics classified as the systematic failure factors. And the yield of the semiconductor device when the YsC1, YsC2, YsC3 ···,
Y = Yr + Ys
= (YrC1 * YrC2 * YrC3 ...) + (YsC1 * YsC2 * YsC3 ...)
The final yield Y of the semiconductor device may be calculated using the relational expression

  Here, the film quality characteristic classified as the systematic defect factor will be described in the third embodiment and the like, but the film quality characteristic classified as the random defect factor can be exemplified as follows.

  FIG. 16 shows an example of a result obtained by examining defects (particles) generated on a wafer when an SBT (strontium bismuth tantalate) film used for a ferroelectric capacitor is etched. As is apparent from FIG. 16, it can be seen that defects (particles) are detected over the entire wafer surface.

FIGS. 17A and 17B show the generation mechanism of particles when the SBT film of the ferroelectric capacitor is etched. As shown in FIG. 17A, a plug 52 is provided in a first insulating film 51 on a substrate (not shown), and a lower electrode 53 connected to the plug 52 is formed on the first insulating film 51. Has been. The lower electrode 53 is covered with the second insulating film 54 except for its upper surface. In this state, when the SBT film 55 formed on the lower electrode 53 is etched using a resist mask 56 with a Cl-based etching gas, a Bi-Cl reactant 57 adheres to the substrate surface. . After that, as shown in FIG. 17B, when the resist mask 56 is removed by ashing, the reaction product 57 is decomposed into Bi and Cl ₂ by the heat treatment at that time, and very fine particles 58 made of Bi crystals are formed. is generated (Bi-Cl → Bi + Cl 2 ⇒ Bi). As shown in FIG. 16, the particles 58 are generated on the entire surface of the wafer, adversely affecting the wiring contact in the subsequent process, and as a result, affecting the yield of the semiconductor device. As described above, the film quality characteristics of the SBT film are classified as random failure factors.

  In the above relational expression, the yield Yr due to the random defect factor is obtained by further integrating the yield due to the “random defect factor other than the film quality characteristic”, for example, the yield predicted by the conventional in-line defect inspection. May be.

  In the above-described relational expression, the yield Ys due to the systematic failure factor may be obtained by further integrating the yield due to the “systematic failure factors other than the film quality characteristics”. Examples of systematic defect factors other than film quality characteristics include the following examples.

  FIG. 18 shows the relationship between the electrical resistance of the source / drain impurity diffusion layer of 0.24 μm space and the dimensions after dry etching that define the active region of the semiconductor substrate. As can be seen from FIG. 18, as the size after dry etching that defines the active region of the semiconductor substrate increases, the electric resistance of the source / drain impurity diffusion layer increases in a positive linear function. It can be seen that there is a correlation between the electrical resistance of the diffusion layer and the size after dry etching that defines the active region of the semiconductor substrate.

  This is because the size of the source / drain impurity diffusion layer becomes smaller as the size after dry etching that defines the active region of the semiconductor substrate becomes thicker. Thus, it can be seen that the electrical resistance of the source / drain impurity diffusion layer varies when the dimensions after dry etching that define the active region of the semiconductor substrate vary.

  FIG. 19 shows the distribution of the electrical resistance (unit: Ω / μm) of the source / drain impurity diffusion layer in the wafer plane. As can be seen from FIG. 19, the electric resistance of the source / drain impurity diffusion layer varies within the wafer plane. From the data shown in FIG. 19 and the positive linear function relationship between the dimensions after dry etching defining the active region of the semiconductor substrate and the electric resistance of the source / drain impurity diffusion layer shown in FIG. The systematic failure in the wafer surface due to the electrical resistance of the layer is related to the variation in the wafer surface after dry etching that defines the active region of the semiconductor substrate, that is, the parameter other than the film quality characteristic. Recognize.

  As described above, the relationship between the electrical resistance of the source / drain impurity diffusion layer and the size after dry etching that defines the active region of the semiconductor substrate is an example showing the relationship between systematic failure and "parameter other than film quality characteristics". Become.

  In the first or second method for predicting the yield of a semiconductor device according to the present invention, the film quality characteristics of the semiconductor device include the amount of moisture desorbed or moisture in the insulating film, the crystal structure or orientation of the metal film, or the substrate or trench. It may be at least one of structural stresses.

  In the yield prediction method for the first or second semiconductor device of the present invention, the analysis method used for measuring the film quality characteristic of the semiconductor device is EDS, AES, XPS, Raman spectroscopy, TDS, X-ray diffraction, EBPS, or FT. It may be at least one of -IR.

  A yield prediction system for a semiconductor device according to the present invention includes a film quality characteristic evaluation unit that measures film quality characteristics of the semiconductor device and recognizes the measured data as graphic data during the manufacturing process of the semiconductor device, and the film quality characteristic evaluation. Film quality characteristic data storage means for storing the graphic data obtained by the means, multivariate analysis means for performing multivariate analysis using the graphic data stored in the film quality characteristic data storage means, and the multivariate analysis Yield predicting means for predicting the yield of the semiconductor device using the analysis result obtained by the means.

  In the yield prediction system for a semiconductor device according to the present invention, the yield prediction means predicts the final yield of the semiconductor device by integrating the predicted yield of the semiconductor device for each of the plurality of film quality characteristics of the semiconductor device. May be. Here, in predicting the final yield of the semiconductor device, the yield predicted by the conventional in-line defect inspection may be included.

  In the yield prediction system for a semiconductor device according to the present invention, the film quality characteristic of the semiconductor device is the amount of moisture desorbed or moisture in the insulating film, the crystal structure or orientation of the metal film, or the stress of the substrate or trench structure. There may be at least one.

  In the semiconductor device yield prediction system according to the present invention, the analysis method used to measure the film quality characteristics of the semiconductor device is at least one of EDS, AES, XPS, Raman spectroscopy, TDS, X-ray diffraction, EBPS, or FT-IR. It may be one.

  In the present invention, the “film quality characteristic” includes a characteristic of a member removed during the manufacturing process, a process condition having a correlation with a characteristic of a component of the semiconductor device, and the like.

  In the present invention, when predicting the yield using the film quality characteristic or the analysis result thereof, a direct or indirect correlation between the yield and the film quality characteristic obtained in advance may be used.

  According to the present invention, it is possible to perform a yield prediction with higher accuracy than a yield prediction by a conventional defect inspection apparatus by measuring a film quality characteristic involved in manufacturing a semiconductor device in-line and performing, for example, a multivariate analysis. Thus, it is possible to provide a yield prediction method and a yield prediction system for semiconductor devices by in-line evaluation.

(First embodiment)
Hereinafter, a yield prediction method for a semiconductor device according to a first embodiment of the present invention will be described with reference to the drawings.

  FIG. 1 is a cross-sectional view schematically showing a main part of a semiconductor device including a ferroelectric capacitor, to which the yield prediction method of this embodiment is applied. As shown in FIG. 1, a lower hydrogen barrier film 1, a lower electrode 2, a ferroelectric capacitor film 3 and an upper electrode 4 are sequentially formed on an active region 7 formed on an upper portion of a semiconductor substrate 6. . Here, a capacitor is composed of the lower electrode 2, the ferroelectric capacitor film 3, and the upper electrode 4, and an upper hydrogen barrier film 5 is formed so as to cover the upper surface and side surfaces of the capacitor. That is, since the ferroelectric is weak against reduction by hydrogen, the capacitor is completely covered with the lower hydrogen barrier film 1 and the upper hydrogen barrier film 5 that prevent hydrogen intrusion.

  A potential is applied to the ferroelectric film 3 through the lower electrode 2 and the upper electrode 4. Specifically, a potential can be applied to the lower electrode 2 from the outside through the lower hydrogen barrier film 1 and the active region 7. The upper electrode 4 has a potential from the outside through a conductor formed simultaneously with the lower electrode 2 (electrically separated from the lower electrode 2, not shown), the lower hydrogen barrier film 1 and the active region 7. Can be applied. With such a structure, a potential can be applied from the outside while the ferroelectric film 3 is completely covered with the hydrogen barrier film.

  In general, a ferroelectric film serving as a capacitive insulating film of a ferroelectric capacitor has a layered perovskite crystal structure. It is known that the ferroelectric film having such a structure is reduced by the influence of hydrogen during the semiconductor process, and as a result, the ferroelectric characteristics are deteriorated or lost. Therefore, in order to prevent reduction by hydrogen, a technique of covering a part or all of the periphery of the ferroelectric capacitor with a hydrogen barrier film is used.

  FIG. 2 shows an ozone BPSG film covering a ferroelectric capacitor whose surroundings are completely covered with a hydrogen barrier film (atmospheric pressure CVD (chemical CVD using TEOS as a material and ozone gas as an oxidizing agent). The inventor of the present application investigated the relationship between the TEOS flow rate (one of the “film quality characteristics” measured during the manufacturing process) and the yield of the ferroelectric capacitor (semiconductor device) during the formation of (obtained by vapor deposition). The results are shown. In FIG. 2, the scales of the vertical axis indicating the yield and the horizontal axis indicating the TEOS flow rate are arbitrary scales. 2 will be described in a third embodiment (FIG. 12) to be described later.

  As can be seen from FIG. 2, the yield of the semiconductor device monotonously decreases as the TEOS flow rate during the formation of the ozone BPSG film increases.

  As described above, even though the ferroelectric capacitor is completely covered with the hydrogen barrier film, as shown in FIG. 2, the TEOS in forming the ozone BPSG film covering the ferroelectric capacitor is formed. The inventor of the present application examined the cause of the decrease in the yield of the ferroelectric capacitor as the flow rate increased, and the following knowledge was obtained.

FIG. 3 shows the result of analysis of the ozone BPSG film by TDS (Thermal Desorption Spectroscopy). In FIG. 3, the horizontal axis indicates the TEOS flow rate (arbitrary scale), and the vertical axis indicates the amount of desorbed moisture (number of H ₂ O molecules) from the ozone BPSG film per unit area.

  FIG. 3 shows that the amount of desorbed moisture from the ozone BPSG film increases as the TEOS flow rate increases.

  4 relates to a hydrogen barrier film that covers the periphery of the ferroelectric capacitor shown in FIG. 1, and a hydrogen barrier film that covers the lower side of the ferroelectric capacitor and a hydrogen that covers the upper and side surfaces of the ferroelectric capacitor. The TEM image obtained when this inventor investigates a junction part with a barrier film | membrane by TEM (Transmission Electron Microscopy) is shown.

  FIG. 4 shows that the hydrogen barrier film covering the lower side of the ferroelectric capacitor and the hydrogen barrier film covering the side surface of the ferroelectric capacitor are not completely joined, and there is a gap between them. . If there is such a gap, hydrogen easily penetrates into the ferroelectric capacitor from this gap, and as a result, the characteristics of the ferroelectric capacitor deteriorates, and finally the yield as a semiconductor device decreases. I guess that.

  Therefore, as a result of further experiments by the inventor of the present application, it has been found that the size of the gap shown in FIG. 4 depends on the TEOS flow rate when the ozone BPSG film is formed. Specifically, the greater the TEOS flow rate when forming the ozone BPSG film, the greater the amount of moisture desorbed from the ozone BPSG film, and the excess oxidation amount of the side wall of the hydrogen barrier film covering the lower side of the ferroelectric capacitor. Will increase. As a result, as the TEOS flow rate increases, the excess oxidation region increases. Therefore, a gap between the hydrogen barrier film covering the lower side of the ferroelectric capacitor and the hydrogen barrier film covering the side surface of the ferroelectric capacitor widens, and hydrogen is It becomes easier to penetrate into the ferroelectric capacitor. Therefore, the characteristics of the ferroelectric capacitor are deteriorated, and the yield as a semiconductor device is finally reduced.

  As described above, according to the present embodiment, the relationship between the TEOS flow rate, which is the film quality characteristic of the ozone BPSG film, and the yield of the semiconductor device is shown by a linear curve (FIG. 2). Since the relationship with the amount of desorbed water from the BPSG film is also shown by a linear curve (FIG. 3), for example, the amount of desorbed water from the ozone BPSG film is analyzed using TDS, and the result is obtained in advance as shown in FIG. It can be seen that the yield of the semiconductor device can be predicted by applying the relationship shown in FIG.

  That is, according to the present embodiment, a semiconductor device having higher accuracy than the yield prediction by the conventional defect inspection apparatus by measuring and analyzing, for example, the film quality characteristics of the ozone BPSG film in-line during the manufacturing process of the semiconductor device. Yield prediction. As a result, by using the yield prediction result based on the film quality characteristics, it is possible to immediately take measures against defects in the manufacturing process of the semiconductor device during the manufacturing process of the semiconductor device. In addition, if it is found that a significant decrease in yield is unavoidable, useless manufacturing costs can be prevented by discontinuing the manufacturing process after the manufacturing process and discarding the semiconductor device during the manufacturing process. It becomes possible to do.

  Although not shown in this embodiment, instead of measuring the amount of desorbed water from the ozone BPSG film, the amount of OH bonds in the ozone BPSG film is measured using, for example, FT-IR (Fourier-transform Infrared Spectroscopy). By doing so, the same result as in the present embodiment can be obtained (in this case, the correlation between the TEOS flow rate and the OH bond amount is used).

(Second Embodiment)
Hereinafter, a yield prediction method for a semiconductor device according to a second embodiment of the present invention will be described with reference to the drawings.

  FIG. 5 is a cross-sectional view schematically showing a main part of a semiconductor device including a transistor, to which the yield prediction method of this embodiment is applied. As shown in FIG. 5, a gate electrode 11 made of polysilicon is formed on a semiconductor substrate 10. An LDD oxide film 13 made of a silicon oxide film is formed on a side surface of the gate electrode 11 via an LDD (Lightly Doped Drain) spacer 12 made of a silicon nitride film. A Co salicide 14 is formed on the upper surface of the gate electrode 11. The cap metal 15 covering the gate electrode 11, the LDD spacer 12, the LDD oxide film 13, and the Co salicide 14 is removed during the manufacturing process and does not remain in the final structure.

  In recent miniaturization processes, a silicide structure is employed for a transistor in order to obtain high-speed operation of a semiconductor integrated circuit. For this reason, when a disconnection of silicide occurs in the silicide structure, the wiring resistance is greatly increased, which is a major factor in yield reduction. In particular, it has been found that when a silicide disconnection occurs at the gate of a transistor, circuit delay, analog circuit resistance fluctuation, and the like occur, which greatly affects the operating characteristics of the semiconductor device.

  FIG. 6 shows a film thickness of a cap metal (TiN film as an example) used in a cobalt silicide manufacturing process (one of “film quality characteristics” measured during the manufacturing process) for a transistor having a cobalt silicide structure on the gate electrode. The present inventors have examined the relationship between the transistor and the yield of transistors (semiconductor devices). In FIG. 6, the scales of the vertical axis indicating the yield and the horizontal axis indicating the thickness of the TiN film are arbitrary scales.

  As can be seen from FIG. 6, the yield of the semiconductor device monotonously decreases as the thickness of the cap TiN film used in the cobalt silicide manufacturing process increases beyond a certain thickness.

  The inventor of the present application examined the cause of this, and obtained the following findings.

  FIG. 7 shows the result of analysis of the crystal orientation of the TiN film by the X-ray (XPS (X-ray Photoelectron Spectroscopy)) by the present inventor. In FIG. 7, the horizontal axis indicates the thickness (arbitrary scale) of the TiN film, and the vertical axis indicates the X-ray diffraction intensity (unit: cps (cycle per second)).

  From FIG. 7, as the TiN film thickness increases, the crystal component in the TiN (111) orientation increases, while the crystal component in the TiN (200) orientation hardly increases, and the columnar crystal component dominates in the TiN film. I understand that

  That is, as the TiN film thickness increases, as shown in FIG. 8A, the columnar crystals increase rapidly in the TiN film. As a result, in the transistor shown in FIG. 5, as shown in FIG. It is presumed that cracks are likely to occur at the edge portion of the gate electrode 11, and the oxygen barrier property at this portion is lowered to cause local deterioration of the cobalt silicide reaction. As described above, when local degradation of the cobalt silicide reaction occurs, silicide disconnection occurs, so that the resistance of the transistor greatly increases, and as a result, the yield of the semiconductor device decreases.

  As described above, according to the present embodiment, the relationship between the film thickness (one of the film quality characteristics) of the cap TiN film used in the cobalt silicide manufacturing process and the yield of the semiconductor device is a linear curve (FIG. 6). In addition, since the relationship between the TiN film thickness and the TiN film orientation is also shown in FIG. 7, the XPS analysis is used to analyze the TiN film orientation, and the result is obtained in advance. It can be seen that the yield of the semiconductor device can be predicted by applying the relationship shown in FIGS.

  That is, according to the present embodiment, a semiconductor device with higher accuracy than the yield prediction by the conventional defect inspection apparatus by measuring and analyzing, for example, the film quality characteristics of the cap TiN film in-line during the manufacturing process of the semiconductor device. Yield prediction. As a result, by using the yield prediction result based on the film quality characteristics, it is possible to immediately take measures against defects in the manufacturing process of the semiconductor device during the manufacturing process of the semiconductor device. In addition, if it is found that a significant decrease in yield is unavoidable, useless manufacturing costs can be prevented by discontinuing the manufacturing process after the manufacturing process and discarding the semiconductor device during the manufacturing process. It becomes possible to do.

  Although not shown in this embodiment, instead of analyzing the orientation of the TiN film, for example, by analyzing the crystal structure of the TiN film using X-ray diffraction, the same result as in this embodiment can be obtained. (In this case, the correlation between TiN film thickness and crystal structure is used).

  Further, measurement results of film quality characteristics other than those shown in the first and second embodiments, for example, Raman spectroscopic measurement results of stress at the bottom of an element isolation structure (STI (Shallow Trench Isolation)), nickel silicide EBSP (Electron) of the XRD (X-ray Diffractometry) measurement result or Raman spectroscopic measurement result of the crystal structure of nickel and silicon in the film, the crystal orientation distribution of the ferroelectric film, or the crystal orientation distribution of the metal film represented by the Cu film Even using the Back Scattering Pattern) measurement result or the like, it is possible to predict the yield of the semiconductor device as in the present embodiment.

FIG. 9 shows an example of a result of measurement of the crystal structure of nickel and silicon in the nickel silicide film by Raman spectroscopy. FIG. 9 shows the measurement results corresponding to the conditions (silicide formation conditions) A, C, and E, respectively. In FIG. 9, the horizontal axis represents the wave number (unit: cm ⁻¹ ), and the vertical axis represents the intensity (arbitrary scale).

From FIG. 9, it can be seen that the crystal structure and composition ratio of nickel and silicon in the nickel silicide film differ depending on the conditions. Specifically, the nickel silicide under the condition A has a NiSi structure. Further, as the condition B changes to D, the NiSi ₂ / NiSi ratio in the nickel silicide approaches 1, and the nickel silicide of the condition E has a complete NiSi ₂ structure. It is known that the resistance of the thin line portion in the semiconductor device is low when nickel silicide has a NiSi structure, and is high when nickel silicide has a NiSi ₂ structure. When the resistance increases, the yield of the semiconductor device decreases.

  As described above, the present invention examines film quality characteristics in-line, paying attention to the close relationship between various film quality characteristics measured and analyzed in-line and the yield of semiconductor devices. This is characterized in that the yield of the semiconductor device is predicted promptly.

  Furthermore, in the present invention, based on such yield prediction, measures can be taken immediately for defects in the manufacturing process of the semiconductor device during the manufacturing process of the semiconductor device. In addition, if it is found that a significant decrease in yield is unavoidable, useless manufacturing costs can be prevented by discontinuing the manufacturing process after the manufacturing process and discarding the semiconductor device during the manufacturing process. It becomes possible to do.

(Third embodiment)
Hereinafter, a yield prediction method and a yield prediction system for a semiconductor device according to a third embodiment of the present invention will be described with reference to the drawings.

  FIG. 10 is a diagram schematically showing a configuration of a yield prediction system according to the present embodiment, specifically, a yield prediction system using in-line evaluation of film quality characteristics of a semiconductor device.

  As shown in FIG. 10, the yield prediction system according to this embodiment includes a film quality characteristic evaluation unit 21 that measures film quality characteristics of a semiconductor device and recognizes the measurement data as graphic data during the manufacturing process of the semiconductor device. Defect evaluation means 22 for performing defect evaluation using the same pattern defect inspection apparatus, film quality database 23 for storing graphic data obtained by the film quality characteristic evaluation means 21, and defect data obtained by the defect evaluation means 22 ( Killer defect correlation data), multivariate analysis means 25 for performing multivariate analysis using graphic data stored in the film quality database 23, analysis results obtained by the multivariate analysis means 25, Yield for predicting the yield of semiconductor devices using defect data stored in defect database 24 Ri and a prediction unit 26.

  In the present embodiment, the film quality characteristic evaluation means 21 includes, for example, film quality characteristic evaluation means 21A, 21B, and 21C, and the film quality database 23 includes, for example, the film quality databases 23A and 23B according to the type of film quality characteristics to be evaluated. The multivariate analysis means 25 is composed of multivariate analysis means 25A and 25B. Moreover, the defect evaluation means 22 is comprised from defect evaluation means 22A and 22B according to the kind etc. of defect of evaluation object.

  In the present embodiment, the part of the film quality characteristic evaluation unit 21 and the defect evaluation unit 22 that performs analytical processing, the multivariate analysis unit 25, and the yield prediction unit 26 are, for example, the yield prediction system and the network of the present embodiment. It can be realized as a program that operates on a connected computer. Further, the film quality database 23 and the defect database 24 can be realized by data storage means such as a hard disk constituting the computer, for example.

  Moreover, in the film quality characteristic evaluation means 21 of this embodiment, for example, EDS (Energy Dispersive X-ray Spectroscopy), AES (Auger Electron Spectroscopy), XPS, Raman spectroscopy, TDS, X-ray diffraction (XRD), EBSP or FT- At least one of IR or the like is used.

  Further, the film quality characteristic evaluation means 21 of the present embodiment provides, as evaluation information, for example, the amount of desorbed hydrogen of the insulating film or metal film, the amount of moisture in the film, the atomic bonding state (crystal structure), the crystal orientation or the chemical composition ratio Alternatively, stress of the substrate or trench structure or the like is obtained, and data such as an X-ray reflection spectrum, an infrared absorption spectrum, or a Raman scattering vibration spectrum is obtained as an analysis result.

  Hereinafter, the yield prediction method of this embodiment will be described with reference to the flowchart of FIG.

  First, in step S1, film quality characteristic evaluation means 21 is used to acquire desired film quality characteristic data relating to a semiconductor device in the course of the manufacturing process, and the acquired film quality characteristic data is stored in the film quality database 23 as graphic data.

  Here, a method for recognizing graphic data in the present embodiment will be described. 12A to 12C show the results of TDS analysis of the ozone BPSG film formed using different TEOS flow rates. 12A to 12C, the horizontal axis indicates the temperature, and the vertical axis indicates the strength of the desorbed water amount. That is, FIGS. 12A to 12C show the temperature dependence of the amount of desorbed moisture during film formation. Note that conditions 1 to 3 in FIGS. 12A to 12C correspond to the TEOS flow rates of conditions 1 to 3 shown in FIG.

  12A to 12C, which are TDS analysis results under conditions 1 to 3 in order to make it easy to handle the data in the processing described later after acquiring the data as shown in FIGS. 12A to 12C. By connecting characteristic points in the graph, a simplified graph (figure) as shown in the schematic diagrams of FIGS. 13A to 13C is obtained. That is, the TDS analysis results under the conditions 1 to 3 shown in FIGS. 12A to 12C correspond to the figures 1 to 3 shown in FIGS. 13A to 13C, respectively.

  As can be seen from FIGS. 13A to 13C, the heights of the vertices (peaks) in the TDS analysis results under conditions 1 to 3 are different. Specifically, the peak of the TDS analysis result under condition 1 is the lowest, and the peak of the TDS analysis result under condition 3 is the highest. As described above, in step S 1, by connecting characteristic points in the acquired film quality characteristic data, the data is recognized as simplified graphic data, and the graphic data is stored in the film quality database 23.

  Next, in step S2, the multivariate analysis unit 25 performs multivariate analysis using the pattern recognition result obtained in step S1, that is, the graphic data stored in the film quality database 23.

  Here, a procedure for analyzing the yield of the semiconductor device using the film quality characteristic evaluation results obtained under different conditions will be described.

For example, when analyzing the yield of a semiconductor device using desorbed water amounts obtained under different TEOS flow rate conditions, a multivariate analysis technique based on pattern recognition by a PLS (Partial Least Square) method is used. Specifically, first, a measurement pattern of desorption moisture obtained in advance by the TDS method, for example, a measurement pattern of desorption moisture under three different TEOS flow conditions (flow rate T ₁ <flow rate T ₂ <flow rate T ₃ ) Accumulate in the database. Next, the desorbed water content of the sample to be analyzed is analyzed by TDS analysis, and the measurement of the desorbed water content accumulated in the database is performed for the measurement pattern (unknown figure A) of the desorbed water content obtained thereby. It obtains a difference between the respective patterns (known figure 1-3), the residual? x ² represented by the following (equation 1) as a condition to minimize seek TEOS flow T _a sample to be analyzed.

Σx _i ² = Σ (f (T _i ) −f (T _A )) ² (Formula 1)
However, in (Formula 1), Σ represents the sum, i represents an integer from 1 to 3, and f (T _i ) represents the measurement pattern of the desorbed water amount at the TEOS flow rate T _i (known figures 1 to 3). F (T _A ) represents the measurement pattern (unknown figure A) of the desorbed water content obtained for the sample to be analyzed.

Specifically, the desorption moisture amount measurement patterns (for example, FIGS. 14A to 14C) under _three TEOS flow rate conditions (flow rate T ₁ <flow rate T ₂ <flow rate T ₃ ) accumulated in the database in advance. The difference between the known figures 1 to 3) shown and the measurement pattern of the desorbed water content obtained for the sample to be analyzed (for example, the unknown figure A shown in FIG. 14 (d)) (difference at each horizontal axis coordinate) FIG. 14E shows a condition for minimizing the residual Σx ² expressed by the above (Expression 1) by performing multivariate analysis such as a least square method using the difference. a, it can be determined TEOS flow T _a sample to be analyzed.

Next, in step S3, the yield predicting means 26 predicts the yield of the semiconductor device using the analysis result obtained in step S2. For example, as described above, if the TEOS flow rate T _A of the sample to be analyzed is obtained, the semiconductor device can be obtained by applying the obtained TEOS flow rate T _A to the relationship shown in FIG. 2 as described in the first embodiment. The yield of the device can be predicted. Further, the final yield of the semiconductor device may be predicted by performing Steps S1 to S3 for each of the plurality of film quality characteristics of the semiconductor device, and accumulating the yield predicted for each film quality characteristic.

  Incidentally, the yield of the semiconductor device is divided into a yield Yr due to a random defect factor and a yield Ys due to a systematic defect factor.

  Originally, it is desirable that the “film quality characteristics” be uniform without depending on the position in the semiconductor substrate (wafer). However, in reality, the “film quality characteristics” are within the semiconductor substrate due to the characteristics of the manufacturing equipment. Often has a certain distribution. For example, there are variations in film quality characteristics between the vicinity of the center and the periphery of the wafer. In this case, there is a certain tendency that the number of defective semiconductor devices (that is, yield reduction) increases at the center of the wafer, for example. Such a defect is called a systematic defect.

  In addition, there is a factor that does not have a tendency in the characteristic distribution in the wafer as described above, such as a defect due to particles, and randomly causes a defect in the semiconductor device (that is, a decrease in yield). Such a defect is referred to as a random defect.

Therefore, in the present embodiment, each of the plurality of film quality characteristics used for predicting the final yield of the semiconductor device is classified into a random failure factor or a systematic failure factor, and is classified for each failure factor type (hereinafter referred to as a category). . Further, if the yield predicted for the film quality characteristic classified as a random defect factor is YrCn (n: natural number), and the yield predicted for the film quality characteristic classified as a systematic defect factor is YsCm (m: natural number), Since the yields YrCn and YsCm affect the final yield Y of the semiconductor device independently of each other, the final yield Y of the semiconductor device is calculated according to the following (formula 2) by integrating the yields YrCn and YsCm. Can do.
Y = Yr + Ys
= (YrC1 * YrC2 * YrC3 ...) + (YsC1 * YsC2 * YsC3 ...)
... (Formula 2)
For example, since the TEOS flow rate which is “film quality characteristic” is a systematic failure factor, the yield obtained using the TEOS flow rate can be treated as YsC1 constituting the yield Ys due to the systematic failure factor. Similarly, by obtaining other systematic failure factors as YsC2, YsC3,..., The yield Ys can be finally obtained. Further, the yield Ys may be obtained by further integrating the yield due to “systematic defect factors other than film quality characteristics”.

  Further, since the yield Yr due to the random defect factor is mainly determined by the yield reduction caused by the particles, the yield YrC1, YrC2, YrC3... Is finally obtained for each semiconductor device manufacturing process. Yield Yr can be obtained. Here, the defect evaluation necessary for calculating YrC1, YrC2, YrC3,... Is performed by, for example, the defect evaluation means 22 shown in FIG. 10, and the defect data (killer defect correlation data) obtained thereby is the defect database. 24, the yield predicting means 26 calculates the yield Yr using the defect data. Further, the yield Yr may be obtained by further integrating the yield due to “systematic defect factors other than film quality characteristics”.

  As described above, according to the present embodiment, the final yield Y of the semiconductor device is obtained by measuring various film quality characteristics involved in the manufacture of the semiconductor device and performing analysis such as multivariate analysis. The prediction can be made with higher accuracy than the yield prediction by the defect inspection apparatus.

  The present invention relates to a yield prediction method and a yield prediction system of a semiconductor device, and more particularly than a yield prediction by a conventional defect inspection apparatus by measuring a film quality characteristic related to the manufacture of a semiconductor device in-line and performing multivariate analysis or the like. This is very useful because it can predict the yield with high accuracy.

FIG. 1 is a cross-sectional view schematically showing a main part of a semiconductor device including a ferroelectric capacitor, to which a yield prediction method for a semiconductor device according to the first embodiment of the present invention is applied. FIG. 2 is a diagram showing a result of the inventor of the present invention examining the relationship between the TEOS flow rate when forming an ozone BPSG film covering a ferroelectric capacitor and the yield of a semiconductor device on which the ferroelectric capacitor is mounted. . FIG. 3 is a diagram showing the results of analysis of the ozone BPSG film using TDS by the present inventor. FIG. 4 shows a case where the inventor of the present application examines the junction between the hydrogen barrier film covering the lower side of the ferroelectric capacitor shown in FIG. 1 and the hydrogen barrier film covering the upper side and the side surface of the ferroelectric capacitor by TEM. It is a figure which shows the obtained TEM image. FIG. 5 is a cross-sectional view schematically showing a main part of a semiconductor device including a transistor to which a yield prediction method for a semiconductor device according to a second embodiment of the present invention is applied. FIG. 6 shows the relationship between the film thickness of the cap TiN film used in the cobalt silicide manufacturing process and the yield of the semiconductor device on which the transistor is mounted for a transistor having a cobalt silicide structure on the gate electrode. It is a figure which shows a result. FIG. 7 is a diagram showing the results of analysis of the crystal orientation of the TiN film using X-rays by the present inventor. FIG. 8A is a diagram showing how the columnar crystals are dominant in the thick TiN film, and FIG. 8B is a diagram in which cracks are generated at the edge portion of the gate electrode as a result of the rapid increase of the columnar crystals in the TiN film. It is a figure which shows a mode that it generate | occur | produced. FIG. 9 is a diagram showing the results of measurement of the crystal structure of nickel and silicon in the nickel silicide film by Raman spectroscopy. FIG. 10 is a diagram schematically showing a configuration of a yield prediction system for semiconductor devices according to the third embodiment of the present invention. FIG. 11 is a flowchart of a yield prediction method for a semiconductor device according to the third embodiment of the present invention. 12 (a) to 12 (c) are diagrams showing the results of TDS analysis of the ozone BPSG film formed by using different TEOS flow rates. FIGS. 13A to 13C are diagrams showing simplified graphs (figure) obtained by connecting characteristic points in the graphs of FIGS. 12A to 12C. FIGS. 14A to 14E are diagrams for explaining the processing contents of step S2 of the yield prediction method for a semiconductor device according to the third embodiment of the present invention. FIG. 15 is a diagram showing a configuration of a conventional defect inspection system. FIG. 16 is a diagram showing an example of a result of the inventor of the present application examining defects generated on a wafer when an SBT film used for a ferroelectric capacitor is etched. FIGS. 17A and 17B are diagrams showing the generation mechanism of particles when the SBT film of the ferroelectric capacitor is etched. FIG. 18 is a diagram showing a result of the inventor of the present application examining the relationship between the electrical resistance of the source / drain impurity diffusion layer and the size after dry etching that defines the active region of the semiconductor substrate. FIG. 19 is a diagram showing a result of the inventor of the present application examining the distribution of the electrical resistance of the source / drain impurity diffusion layer in the wafer surface.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Lower hydrogen barrier film 2 Lower electrode 3 Ferroelectric capacitor film 4 Upper electrode 5 Upper hydrogen barrier film 6 Semiconductor substrate 7 Active region 10 Semiconductor substrate 11 Gate electrode 12 LDD spacer 13 LDD oxide film 14 Co salicide 15 Cap metal 21 (21A 21C) Film quality characteristic evaluation means 22 (22A, 22B) Defect evaluation means 23 (23A, 23B) Film quality database 24 Defect database 25 (25A, 25B) Multivariate analysis means 26 Yield prediction means 51 First insulating film 52 Plug 53 Lower electrode 54 Second insulating film 55 SBT film 56 Resist mask 57 Reactant 58 Particle

Claims (10)

A method for predicting the yield of a semiconductor device,
A method of predicting a yield of a semiconductor device, comprising: measuring a film quality characteristic of the semiconductor device during a manufacturing process of the semiconductor device, and predicting a yield of the semiconductor device using the measured data. A method for predicting the yield of a semiconductor device,
A step (a) of measuring film quality characteristics of the semiconductor device during the manufacturing process of the semiconductor device and storing the measured data in a database as graphic data;
(B) performing a multivariate analysis using the graphic data stored in the database;
And a step (c) of predicting the yield of the semiconductor device using the analysis result obtained in the step (b). The yield prediction method for a semiconductor device according to claim 2,
The step (a), the step (b) and the step (c) are performed for each of the plurality of film quality characteristics of the semiconductor device, and the predicted yields of the semiconductor device are integrated with each other for the plurality of film quality characteristics. And a step (d) of predicting a final yield of the semiconductor device. The yield prediction method for a semiconductor device according to claim 3,
In the step (d), each of the plurality of film quality characteristics is classified into a random failure factor or a systematic failure factor and classified for each failure factor type,
The final yield of the semiconductor device is classified as Y, the yield due to the random defect factor of the semiconductor device is Yr, the yield due to the systematic defect factor of the semiconductor device is Ys, and the random defect factor among the plurality of film quality characteristics. YrC1, YrC2, YrC3..., The semiconductor device yield predicted for each of the film quality characteristics, and the semiconductor quality predicted for each of the plurality of film quality characteristics classified as the systematic failure factor. When the device yield is YsC1, YsC2, YsC3.
Y = Yr + Ys
= (YrC1 * YrC2 * YrC3 ...) + (YsC1 * YsC2 * YsC3 ...)
The final yield Y of the semiconductor device is calculated using the relational expression (1). In the yield prediction method of the semiconductor device according to any one of claims 1 to 4,
The film quality characteristic of the semiconductor device is at least one of a desorption moisture amount or moisture amount in the insulating film, a crystal structure or orientation of the metal film, or a stress of the substrate or trench structure. Device yield prediction method. In the yield prediction method of the semiconductor device according to any one of claims 1 to 5,
An analysis method used for measuring film quality characteristics of the semiconductor device is at least one of EDS, AES, XPS, Raman spectroscopy, TDS, X-ray diffraction, EBPS, or FT-IR. Yield prediction method. A system for predicting the yield of semiconductor devices,
During the manufacturing process of the semiconductor device, the film quality characteristic of the semiconductor device is measured, and the film quality characteristic evaluation means for recognizing the measured data as graphic data;
Film quality characteristic data storage means for storing the graphic data obtained by the film quality characteristic evaluation means;
Multivariate analysis means for performing multivariate analysis using the graphic data stored in the film quality characteristic data storage means;
A yield prediction system for a semiconductor device, comprising: a yield prediction unit that predicts a yield of the semiconductor device using an analysis result obtained by the multivariate analysis unit. The yield prediction system for a semiconductor device according to claim 7,
The yield predicting unit predicts a final yield of the semiconductor device by integrating the yield of the semiconductor device predicted for each of a plurality of film quality characteristics of the semiconductor device with each other. system. The yield prediction system for a semiconductor device according to claim 7 or 8,
The film quality characteristic of the semiconductor device is at least one of a desorption moisture amount or moisture amount in the insulating film, a crystal structure or orientation of the metal film, or a stress of the substrate or trench structure. Device yield prediction system. The yield prediction system for a semiconductor device according to any one of claims 7 to 9,
The analysis method used for measuring the film quality characteristic of the semiconductor device is at least one of EDS, AES, XPS, Raman spectroscopy, TDS, X-ray diffraction, EBPS, or FT-IR. Yield prediction system.