JP4551937B2 - Pattern formation method - Google Patents

Description

  The present invention relates to a pattern forming method for forming a desired pattern on a substrate to be processed. In particular, the present invention relates to a mask manufacturing method for manufacturing a mask based on a corrected pattern, and further to a mask having a corrected pattern. The present invention also relates to an LSI manufacturing method for forming an LSI pattern on a wafer.

  In manufacturing an LSI, first, a mask having an opening corresponding to an LSI pattern or a light shielding pattern is manufactured using a mask drawing apparatus or the like. Next, the pattern on the mask is transferred to the resist on the wafer using an optical stepper or scanner, and thereafter, a single layer pattern is produced through various processes such as development and etching. An LSI is manufactured by repeating such a pattern manufacturing process. Also in the manufacture of the mask, the mask is manufactured through several steps such as exposure of the resist on the mask by the mask drawing apparatus, development of the resist, and etching of COG (Cr On Glass).

  Currently, an electron beam exposure apparatus is mainly used for mask drawing, but light may be used in some cases. Further, in an apparatus for transferring a pattern on a mask onto a wafer, light is generally used, but a technique using an electron beam or an X-ray has been studied. In any case, even a pattern formation for only one LSI layer goes through various processes as described above.

  One of the problems found in LSI patterns produced through these processes or patterns on the mask is that “when viewed locally, each pattern is almost uniformly finished (locally, the design dimensions The pattern size changes gradually when viewed in the entire mask or inside the chip formed in the wafer (the difference from the design size changes gradually inside the chip) " It is.

  This is schematically shown in FIG. When the position within the chip on the wafer is close (local), the difference from the design dimension is small, but when the position within the chip is far (overall), the difference from the design dimension is large. Here, the error distribution shown in FIG. 1 is obtained by adding a pattern-dependent dimensional error and a position-dependent dimensional error as shown in FIG. In FIG. 2, (a) is a line & space pattern, (b) shows the dimensional change due to the pattern density, and (c) shows the dimensional change at the two-dimensional plane position.

  As a similar problem, there is a phenomenon called “fogging” that occurs when performing mask drawing with an electron beam exposure apparatus. Fogging is a phenomenon in which electrons are exposed to the resist on the mask, then bounce off the substrate, return to the upper part of the stage of the apparatus, and then reflect again to expose the resist. The result of this phenomenon is similar to that described above, and the dimensions gradually change in the order of several centimeters. Therefore, this is one of the causes of global pattern dimension deterioration on the mask.

  As a countermeasure for this problem, a technique for adjusting the irradiation amount for each place has been proposed (fogging correction). In this method, the irradiation amount for each place for correcting the dimensional variation is obtained in advance by a computer or the like, and the fogging phenomenon is corrected (accurately suppressed) by changing the irradiation amount based on the amount. Here, the dose is calculated by dividing the mask into small areas, calculating the density of the patterns therein, and using this. However, as described below, this fog correction method cannot be a sufficient solution to the problem of global dimensional variation.

  First, since this method is only a method for correcting fog in mask drawing by an electron beam exposure apparatus, it cannot be applied as it is when an LSI pattern is formed on a wafer.

  As a second problem, there are problems in accuracy and calculation time. In the electron beam exposure apparatus, in addition to the phenomenon of fogging, there is another dimension deterioration factor called a proximity effect. This is a relatively local dimensional degradation factor that affects the region of about 30 μm, but the dimensional error generated there is about 100 nm, which is much larger than the above-described global dimensional variation. When correcting this proximity effect, a method of changing the irradiation amount depending on the location is used.

  In other words, both the above-mentioned fog correction and proximity effect correction employ a method of correcting by changing the irradiation amount for each location. It is necessary to calculate the correct dose. In order to implement this, it is necessary to calculate the optimum dose for each area (for example, 1 μm × 1 μm) that is much smaller than the range covered by the proximity effect correction, and the range (for example, several cm) that is affected by the fogging. It is necessary to consider the influence of all patterns at the corners.

  At present, if only proximity effect correction is performed, the optimal dose is calculated using a dedicated circuit, but the time required for this is about one hour. The distance affected by the fogging is more than 30 times the distance affected by the proximity effect, and the calculation amount and the calculation time are affected by the area. Therefore, 30 × 30 times is about 1000 times. That is, in order to calculate the proximity effect correction and the fog correction at the same time, even if a dedicated circuit currently used for the proximity effect correction can be used, the time is 1000 hours, which is about two months. .

  Therefore, as a conventional fog correction method, in the fog correction, the irradiation effect is calculated only from the density of the pattern ignoring the result of the proximity effect correction, and in the proximity effect correction, the influence of the fog is ignored and the real time inside the drawing apparatus is ignored. After processing, the dose was calculated, and the final dose was calculated by combining the fog correction dose and the proximity effect correction dose, and correction was performed accordingly. If such a method is used, the mutual relationship and dependency between the proximity effect correction and the fog correction will be ignored, thereby causing an error. For this reason, an error occurs in the fog correction, and the effect on the global dimensional variation is only to suppress the error to ¼ at most. This is a value that does not satisfy the accuracy required at present and in the future.

  Thus, conventionally, in the manufacture of masks and LSIs, a plurality of manufacturing apparatuses are used and a plurality of processes are performed, and it is inevitable that global dimensional variations occur due to this. It has been difficult to accurately correct this global dimensional variation.

  The present invention has been made in consideration of the above circumstances, and the object of the present invention is to be able to accurately correct global dimensional variations that occur comprehensively during mask and LSI manufacturing, and to achieve higher accuracy. An object of the present invention is to provide a pattern forming method capable of manufacturing a mask and an LSI.

(Constitution)
In order to solve the above problems, the present invention adopts the following configuration.

That is, one aspect of the present invention is a mask manufacturing method for manufacturing a mask to be subjected to pattern exposure based on a predetermined design pattern, in which an average value of dimensional variation in a local region of the pattern is the local pattern. In order to correct the etching unevenness caused by the etching apparatus used for manufacturing the mask, the pattern dimension is changed depending on the position on the mask. The first global dimension variation characteristic fp (x, y) and the influence of the pattern density at a certain place (x, y) on the pattern dimensions in the vicinity are approximated by a function of distance, depending on the pattern density A second global dimension variation characteristic g (x, y) for changing the pattern dimension is obtained in advance, and is smaller than the global area when the mask is manufactured using the etching apparatus. The density of the LSI pattern existing in the small area is ρ (x, y), and the pattern correction amount Δ (x, y) in each small area is fp (x, y) and ∫g (x−x ′ , y−y ′) ρ (x ′, y ′) dx′dy ′, and corrects the dimensions of the design pattern for each of the small areas, and the desired mask is determined based on the corrected pattern dimensions. A pattern is formed.

Another aspect of the present invention is an LSI manufacturing method for producing a mask on the basis of a predetermined design pattern and forming a desired LSI pattern on the wafer using the mask. In order to correct the etching unevenness caused by the etching apparatus used for manufacturing the LSI, it is defined that the average value of the dimensional variation in the local region varies in a global region wider than the local region. The first global dimension variation characteristic fp (x, y) that changes the pattern dimension depending on the position in the chip or the chip group, and the pattern density at a certain location (x, y) are the pattern dimensions in the vicinity. A second global dimensional variation characteristic g (x, y) that changes the pattern dimension depending on the pattern density, which approximates the effect on the distance by a function, is obtained in advance, and When an LSI is manufactured using a chucking device, the density of LSI patterns existing in a small area smaller than the global area is ρ (x, y), and the pattern correction amount Δ (x, y) is obtained using fp (x, y) and ∫g (x-x ′, y-y ′) ρ (x ′, y ′) dx′dy ′, and the design pattern for each small region And a desired pattern is formed on the mask based on the corrected pattern dimension.

(Function)
In the present invention, the above-described global dimensional variation is corrected by correcting the dimension of the pattern in each region smaller than the standard of dimensional variation. As a result, high-accuracy correction that avoids interaction with other corrections such as proximity effect correction that occurs in the fog correction method can be realized. Further, global errors occurring in each device and each process can be corrected almost independently of each other. Further, the global dimensional error generated in each apparatus and each process can be separated into a contribution depending only on the position and a contribution depending on the pattern at a level where there is no practical problem. And it becomes possible to implement | achieve highly accurate correction | amendment by correct | amending each substantially independently.

  As a specific correction method, a global error generated in each device and each process and a global error generated through all processes are separated into a contribution depending on at least a position and a contribution depending on a pattern. The pattern is deformed locally using. This makes it possible to correct global dimensional variations.

  According to the present invention, characteristics of global dimensional fluctuations when using a group of devices used for forming a pattern on a mask, a wafer, or the like are examined in advance, and a pattern is formed on a sample using these groups of devices. When forming, for each region that is smaller than the standard distance at which a predetermined dimensional variation occurs, the dimensional variation characteristics are used to correct the dimensions of the pattern inside the region, and a desired pattern is formed based on this correction information. By doing so, it is possible to correct the dimensional variation that occurs globally on the inside of the mask or wafer, and the correction accuracy can be dramatically improved as compared with the conventional method.

  First, before describing embodiments of the invention, the basic principle of the present invention will be described.

  There are several possible causes for the above-described global dimensional variation. The list of confirmed and possible causes is as follows.

A) Steps up to mask fabrication 1) When the reticle is irradiated with electrons in the electron beam exposure apparatus, the electrons reflected by the reticle are reflected on the upper part of the exposure apparatus chamber, and the resist on the reticle is exposed again.

    2) Dependence on pattern density during etching. That is, the density of chromium dissolved in the etching solvent changes depending on the pattern density on the resist, and the chromium density in the solvent changes the dissolution rate of chromium. When the mask is dry-etched, a loading effect appears, and the etching rate changes depending on the pattern density. As a result, the chromium dimension changes depending on the location.

    3) There is a problem during development. That is, the same phenomenon as 2) may occur when developing a resist.

    4) There is a problem with the electron beam exposure apparatus. In other words, the reticle is not supported horizontally in the potential line optical device, and is slightly tilted, so that the dimensions change depending on the location.

    5) Even if the dimensions on the reticle are correctly formed, when the reticle is set on the stepper, it is not supported completely parallel to the wafer, so the dimensions vary from place to place on the wafer.

I) Processes until an LSI is fabricated on a wafer using a mask 1) When a pattern on a mask is transferred to a resist on a wafer by a transfer device such as a stepper or a scanner using the mask, Assembly errors and the like cause asymmetry of the optical system and cause dimensional variations depending on the location.

    2) After the step 1) -1), the resist is developed, but the dimension may vary depending on the pattern density and the like, as in 3) -3).

    3) When etching the base substrate using a resist as a mask, the size depends on the pattern due to the loading effect. Also, the dimensions vary depending on the location depending on the characteristics of the etching apparatus.

    4) When an LSI is manufactured by a damascene process or the like, pattern density dependency appears in a chemical mechanical polishing (CMP) process.

  The above-mentioned various causes can be considered, but common features are that the range of influence and the unit of the region where the dimensional variation occurs are about cm or more. In other words, when a region as small as several tens to several hundreds μm is viewed locally, the dimensional change may be regarded as uniform. Therefore, correction may be performed for each of such areas. This guarantees that “correction can be performed for each region smaller than the standard of the dimensional variation distance (distance at which a predetermined dimensional variation occurs)” of the method of the present invention.

  Furthermore, when this correction is performed, the dose is not changed for each place, but the dimensions are corrected for each figure. On the other hand, the proximity effect correction is performed by changing the dose. Therefore, there is no interaction between the correction of the global dimensional variation of the present invention and the proximity effect correction, and there is no interaction between correction items as in the conventional fog correction.

  Next, the above causes indicate that they can be corrected almost independently of each other. First, each factor and event itself may be considered to be almost independent from each other within the range of the first approximation. However, even if the events are independent, both affect the dimensional variation, so that mutual effects appear on the dimensions. This mutual effect is estimated below. As an example, all the global dimensional variations are (a) location-dependent dimensional variations due to unevenness of the etching device (during mask fabrication or wafer fabrication process) and (ii) density-dependent dimensions due to the mask drawing device. It is assumed that the fluctuation is caused by fluctuations, and further the fluctuation width is assumed to be 20 nm.

  If one of the effects (a) and (ii) is zero, the mutual effect is also zero. Therefore, the mutual effect between (a) and (ii) is the largest It can be considered that the effects of () and (i) are almost the same. Therefore, both the fluctuation range due to (a) and the fluctuation range due to (ii) are 10 nm. If expressed in ±, each becomes ± 5 nm.

  Here, it is assumed that p (x, y) is a dimensional variation caused by the etching at (a) at a certain location (x, y). However, −5 nm ≦ p (x, y) ≦ 5 nm. It is assumed that there is a line & space pattern with an irradiation part a μm, a non-irradiation part m-a μm, and a pitch m μm as a design on the mask in that place. In this case, depending on the location, the size of the irradiated portion is changed from a to a + p (x, y), and the size of the non-irradiated portion is conversely ma−p (x, y). Therefore, the density ρ changes from a / m to {a + p (x, y)} / m.

This rate of change is
{A + p (x, y)} / a = 1 + p (x, y) / a (1)
Thus, the relative error is p (x, y) / a.

  Here, the worst value of the relative error is considered. Since a is about the design dimension on the mask and is in the denominator, this lower limit value may be selected in order to investigate the worst value of the relative error. The design rule of the current state-of-the-art development product is 90 nm on the wafer and four times that on the mask, ie, 360 nm. This is adopted as a lower limit value of a and is set as a typical value. p (x, y) is a dimensional error that varies depending on the location. The upper limit value may be ± 5 nm. That is, the maximum value of the relative error is ± 5 nm / 360 nm. This is only ± 1.4% at best.

  The above-described ± 5 nm is adopted as the variation width of the dimension caused by the change in density. Since the relative error of density is 1.4%, the dimensional error caused by this can be considered as about 1.4% of ± 5 nm. This amount is only ± 0.07 nm. This value is sufficiently small and can be ignored.

Conversely, even if each is corrected independently, the error due to the independent correction is only ± 0.07 nm and may be ignored. That is, it can be understood that the correction may be performed independently. Generalizing this argument, the mutual influence of the two factors is
(Relative error due to (a))
× (Relative error due to (ii))
X It turns out that it is a design dimension grade.

The above discussion has been discussed as a correlation (positional) between position dependence and pattern dependence between different devices, but the discussion is the same for the same device. That is, position dependency and pattern dependency in a certain apparatus may be corrected independently of each other. Furthermore, if the positional dependency between two types of devices (for example, a mask drawing device and an etching device) is also discussed in the same manner as described above, the influence on each other is
(Relative error of position-dependent dimensional variation by device 1)
× (Relative error of position-dependent dimensional variation by device 2)
X It turns out that it is a dimension grade. That is, the pattern dependence between different devices may be considered independent as well.

  In the above, only two items contribute to the global dimensional fluctuation, but actually, other influences also contribute to a total fluctuation width of 20 nm. In this case, the individual effects are smaller than 10 nm as described above, so the influence between factors is even smaller than the above consideration. Therefore, the influence between factors may be ignored. In other words, there is no problem even if the dimensions are corrected independently for each factor. In addition, the “position-dependent error” that summarizes the factors and the “density-dependent error” that summarizes the factors may be independently dimensionally corrected.

  The above has been discussed mainly on masks, but is not limited to mask fabrication. The discussion and conclusion are the same in the LSI manufacturing process. The position dependency of the dimension may be read as the position dependency inside the chip on the wafer from the position dependency on the mask. The density dependency may be read as the pattern density dependency inside the chip on the wafer from the pattern density dependency on the mask. However, at this time, since the dimension on the wafer is 1/4 of the dimension on the reticle, the upper limit value 0.15 nm of the error on the mask is 0.04 nm on the wafer. The error is only about 0.04% compared with the 90 nm design dimension on the wafer of the newly developed product. That is, it can be understood from the above discussion that the position-dependent error and the location-dependent error can be corrected independently in the LSI manufacturing process.

  In the above, each error factor is considered as an independent event within the range of the first order approximation, but it can be similarly processed when there is some correlation between events. For example, as will be described in an embodiment described later, it is also possible to add a position dependency to the pattern dependency characteristic and process it in combination with a pure position dependency.

  When dimensional correction is performed in each small area, as shown in FIG. 3, an overlap (a) or gap (b) of a figure of about 10 nm occurs at the boundary of the area. This effect may appear as a dimension abnormality. Whether or not the dimension is abnormal depends on the characteristics of the resist, the process, and the beam resolution of the electron beam exposure apparatus. For example, when the beam resolution is much larger than 10 nm and the contrast of the resist is not so large, the influence of the overlap and the gap hardly appears and can be ignored. On the other hand, when the resolution is sufficiently small and the resist contrast is high, a dimensional abnormality occurs. If this abnormality cannot be ignored, the process of filling the gap generated between the regions, that is, the overlap removal may be performed by the CAD process. Such processing such as overlap removal is a technique that has already been performed and can be easily realized.

  The details of the present invention will be described below with reference to the illustrated embodiments.

(First embodiment)
In the present embodiment, for simplicity, the pattern density is used as an amount for characterizing the pattern.

  First, a pattern for measuring the location dependency, density dependency, and influence range of the dimensional variation is drawn in advance, and the dimension formed after the process is measured. FIG. 4 is a pattern for examining the dependency of the dimension on the mask on the position of the pattern, and a cross pattern having a width of 2 μm is arranged for every 1 mm for measurement.

  FIG. 5 is a pattern for examining pattern dependence, and the cross pattern for measurement is arranged at the same position as that of FIG. In addition, a line and space are arranged in the center. The line and space are set to a ratio of 1 mm: 1 mm (density 50%). On the other hand, the pattern of FIG. 4 may be considered to have a density of 0%. Therefore, as shown below, if a mask is prepared from both patterns and compared with the mask, information on density dependency can be taken in.

First, drawing, development, and etching are performed using the patterns of FIGS. 4 and 5, and then the dimensions of each cross pattern are measured. From the result thus obtained, the mask position dependency and the pattern dependency described above can be obtained. For example, let fp (x, y) be a dimensional error (difference between a design value and an actual measurement value) for each location obtained by measuring a mask manufactured from the pattern of FIG. Here, (x, y) represents a position on the mask. This data fp (x, y) is a characteristic of “size variation depending on location” generated in the above process. On the other hand, the dimensional error for each location obtained after measuring the mask produced from the pattern of FIG. 5 is defined as fd (x, y). The difference between the two diff (x, y) = fd (x, y) −fp (x, y) (2)
Is a dimensional error that occurs depending on the pattern density, and shows its location dependence. Based on this data, the dimensional variation characteristic depending on the pattern is obtained as follows.

  For simplicity, this diff (x, y) is considered as a result of density convolution at each location, and a function g (x, y) that is the core of convolution is approximated by a double Gaussian.

g (x, y) = θ × exp (−x ² / σ ² −y ² / σ ² ) (3)
That is,
diff (x, y)
= ∫g (x-x ', y-y') ρ (x ', y') dx 'dy' (4)
Think of it as a pattern. Here, the integration region is a pattern portion (beam irradiation portion at the time of mask drawing), and ρ (x, y) is a pattern density at a location (x, y). This function g represents the “pattern-dependent dimension variation characteristic” in the mask manufacturing process.

The function g can be obtained as follows. The function diff (x, y) is the following diff2 (x, y)
= ∫g (x-x ', y-y') ρ (x ', y') dx 'dy' (5)
It is sufficient to determine the parameters θ and σ of g such that the difference between diff (x, y) and diff2 (xy) is minimized.

For example, while changing the values of θ and σ,
∫ {diff (x, y) -diff2 (x, y)} ² dx, dy (6)
And θ and σ at which this value is minimized may be selected.

  Actually, it is not necessary to perform the above two-dimensional integration strictly. What is necessary is just to divide the mask into regions smaller than the standard of global dimensional variation and add the contribution for each region. For example, if the standard of dimensional variation is 2 cm, it is sufficient that the small area is 0.5 mm × 0.5 mm. Assuming that the center coordinates and pattern density of each small region are (xi, yi) and ρ (xi, yi) (0.0 to 1.0), the two-dimensional integration of equation (5) is It can be substituted by addition like.

diff2 (xi, yi) = Σg (xi-xj, yi-yj) ρ (xj, yj) Ds
... (7)
Here, dS is the area of a small region, 0.5 mm × 0.5 mm. Further, the sum may be obtained for a small region existing in a region sufficiently larger than the dimensional variation (for example, in a circle having a radius of 6 cm). The double integral appearing in the following discussion is also calculated by addition in units of such small regions unless otherwise specified.

  By using the function g (x, y) representing the “pattern dependent characteristics of global dimensional variation” obtained in this way, the dimensional error at any location of an arbitrary pattern can be expressed by the above equation (5). Can be calculated. Similarly to the above, the integration may be performed by obtaining the density ρ for each small region and adding the weight of g to obtain the sum.

  Next, a procedure for actually correcting an LSI pattern using the above results will be described with reference to FIGS. 6 shows the calculation of the pattern area density for each small region, FIG. 7 shows the correction of the dimensional variation due to location, FIG. 8 shows the correction of the dimensional variation due to the pattern, and FIG. 9 shows the correction of the dimensional variation (position dependent part and pattern dependent part). Of the sum).

  First, consider a mask on which an LSI pattern to be manufactured is placed. This is cut into an area sufficiently smaller than the above-mentioned standard of the distance of variation (here, 1 cm for simplicity). Here, the size is 500 μm × 500 μm.

  Next, the area of the LSI pattern existing inside each small region is calculated, and is defined as ρ (x, y). The reduction amount {Δ (x, y)} of the pattern in each small area is obtained as follows.

Δ (x, y)
= -Fp (x, y) -∫g (x-x ', y-y') ρ (x ', y') dx'dy '(8)
As mentioned above, the first term corrects for dimensional variations that are density independent and dependent only on location, and the second term corrects for dimensions that vary depending on pattern density.

  The time required for the above calculation is sufficiently short and does not matter. In order to obtain the area density for each small region, it is sufficient to have several tens of minutes to several hours with the current EWS (engineering workstation) of about 200 MHz. The convolution calculation using the area density calculation result is much shorter than this. This is because the size of the small region is 0.5 mm × 0.5 mm, and even if the mask size is 10 cm × 10 cm, only the 200 × 200 region exists and the amount of processing is small. Therefore, this convolution calculation is completed in a time of 1 second or less.

The process of deforming the pattern based on the reduction ratio for each area obtained in this way is performed as follows.
(A) First, the LSI pattern is decomposed corresponding to the small area.
(B) The pattern is reduced by a corresponding reduction amount for each small area.
(C) Next, after synthesizing the pattern after the above processing,
(D) Fill gaps between patterns.

  Here, (a) is a pattern cutting process, which can be easily realized. (B) and (c) are the same as the processes performed in the normal CAD process, and for example, a method known as a scan line method can be used. The gap filling in (d) is also a process performed in the CAD system.

  In the case of the present embodiment, the size of the generated gap is only about twice the maximum value of global dimensional variation. That is, it is at most about 20 × 2 nm. On the other hand, the dimension of the gap on the mask design is about 500 to 300 nm. That is, the gap generated by the above procedure is a size that cannot be generated in a general design, and therefore, such a gap can be easily identified and filled.

  The outline of the mask manufacturing process using the produced pattern data is shown in FIG. As a sample, as shown in FIG. 10A, a sample in which a COG film 12 is formed on a mask substrate 11 and a resist 13 is applied thereon is prepared. Data for the mask drawing apparatus is created from the pattern subjected to the global dimension correction as described above, and pattern writing is performed on the resist 13 by the mask drawing apparatus using this data (FIG. 10B). . Here, the mask drawing apparatus to be used is the same as that which first obtained basic data such as the location dependence of the dimensional variation. Alternatively, a drawing apparatus having similar machine characteristics is used. At this time, when an electron beam exposure apparatus is used as the mask drawing apparatus, necessary correction such as proximity effect correction is added.

  After the drawing, development is performed to form a pattern of the resist 13 (FIG. 10C). In the resist 13 on the mask obtained at this stage, global dimensional fluctuations caused by problems inherent to the electron beam exposure apparatus are almost corrected.

  Next, the COG film 12 is selectively etched by RIE or the like using the resist 13 as a mask (FIG. 10D). The etching method and apparatus at this time also use an etching apparatus that has taken basic data. Alternatively, a device having substantially the same characteristics is used. Thereafter, the resist 13 is peeled off (FIG. 10E).

  The mask produced by the above procedure is automatically subjected to various corrections, and a highly accurate mask can be obtained.

  As described above, according to the present embodiment, each device used for manufacturing a mask, a global dimensional error that occurs in each process, and a global dimensional error that occurs through all the processes are converted into contributions and patterns that depend only on the position. Separated into dependent contributions. Based on these correlations, the mask pattern is manufactured by correcting the dimensions of the design pattern in units of small regions where the dimensional variation is within an allowable value, and forming a desired pattern on the mask based on the corrected pattern dimensions. It is possible to accurately correct global dimensional variations that sometimes occur comprehensively, and to manufacture a mask with higher accuracy.

(Second Embodiment)
FIG. 11 outlines the LSI manufacturing process to which the present invention is applied. In the figure, 21 is a wafer as a base substrate, 22 is an insulating film, 23 is a resist, 24 is a base pattern, and 25 is an Al film. Although the process itself shown in FIG. 11 is a well-known method for forming an embedded wiring, this embodiment is characterized by manufacturing a mask used for forming a wiring pattern.

  First, process characteristics in several processes used in the LSI manufacturing process are measured and divided into position-dependent parts and pattern-dependent parts. This is started from the step of producing a mask using the patterns of FIGS. 4 and 5 as in the first embodiment. An electron beam exposure apparatus is used as the drawing apparatus. In the current measurement technique, the dimension measurement at the resist stage has a large error. Therefore, the measurement is performed after the chromium etching, and the characteristics are evaluated in the mask manufacturing process.

  After drawing, developing and etching the patterns of FIGS. 4 and 5, the dimensions are measured for each location, and as a result, R1 (x, y) and r1 (x, y) are obtained. Using these, as in the first embodiment, the location-dependent portion and the pattern-dependent portion are separated from the two types of data, and the function p1 (x, y) and the one-point function g1 (x, y) for each of them. Is obtained.

Next, using the mask prepared above, the pattern is transferred onto the wafer using a stepper, developed by a developing device, etched using a dry etching device, and then the pattern dimension formed on the insulating film is measured. . For each pattern, if the obtained results are R2 (x, y) and r2 (x, y), the overall dimensional variation of the pattern of FIG. 2 generated in this step is as follows.
ΔR2_1 (x, y) = R2 (x, y) R1 (x, y) (9)
For pattern dependent parts:
Δr2_1 (x, y) = r2 (x, y) r1 (x, y) (10)
Is obtained.

  From these ΔR2_1 and Δr2_1, ΔR2_1 is obtained as a position dependent portion and Δr2_1−ΔR2_1 is obtained as a pattern dependent portion in this step. In the same manner as in the first embodiment, from these, the kernel function g2 (x, y) of the pattern dependence term in this step can be obtained. Further, regarding the CMP process, if the same process, measurement, and analysis are performed, position-dependent characteristics and place-dependent characteristics in the CMP process can be obtained.

  After the process characteristics are obtained by the above processing, the subsequent processing is substantially the same as that of the first embodiment. First, the shortest dimension variation distance is selected from all the process characteristics (for example, 1 cm), and the smaller area size is selected (for example, 0.5 mm × 0.5 mm).

Next, the correction amount for each small area is calculated on the computer as follows. First, as in the first embodiment, the density ρ (x, y) of the pattern existing in each small region on the mask is calculated. The size correction amount inside each small area is
−p1 (x, y) −p2 (x, y) × p3 (x, y) (11)
For pattern dependent parts:
-∫ {g1 (x-x ', y-y') + g2 (x-x ', y-y') × g3 (x-x ', y-y')}
× ρ (x ', y') (12)
Can be calculated as Here, p1, p2, and p3 are position dependent functions of the mask manufacturing process, stepper transfer, etching process, and CMP process. Also, g1 (x-x ', y-y') + g2 (x-x ', y-y') x g3 (x-x ', y-y') is the core of each pattern-dependent function. .
The dimensions are corrected according to the correction amount obtained in this way, and pattern data can be obtained by removing overlaps or embedding gaps between figures if necessary, as in the first embodiment. Based on this data, a mask is manufactured using each device used for the characteristic evaluation. Then, by using this mask and performing the process of forming the buried wiring in the procedure shown in FIG. 11, the global dimensional degradation that occurs in each process is automatically corrected, and a highly accurate LSI pattern can be obtained. .

  There may be multiple types of devices used in each process. For example, there are several CMP apparatuses, and one may be selected and used as appropriate. If there is no significant difference in characteristics between CMP apparatuses, the above procedure may be used as it is. On the other hand, if there is a significant difference, the characteristic function of the process (CMP process) when the CMP apparatus is used is obtained for each CMP apparatus. For g3, the above procedure may be performed using a characteristic function when the device is used, for example, g3 ′.

(Modification)
In the second embodiment, an application method to an LSI manufacturing method called damascene has been described, but the present invention is not limited to this. FIG. 12 shows a manufacturing process different from damascene without CMP. In FIG. 12, 31 indicates a wafer as a base substrate, 32 indicates an Al film, 33 indicates a resist, and 34 indicates a pattern. Also in this case, the present invention can be applied by partially modifying the second embodiment.

  For example, in the second embodiment, the macro process may be considered as two types of (mask manufacturing process) and (transfer from stepper to dry etching), and the aluminum pattern may be directly processed by dry etching. The application of the invention method is linear, and dimensional variation characteristics are determined separately for each macro process in a position-dependent part and a pattern-dependent part, and based on that, a dimensional correction amount is obtained for each place, and the pattern is corrected. An LSI may be manufactured by manufacturing a mask using this.

  In the second embodiment and the above example, the pattern processing for one layer is divided into several processes, and the characteristics of each process are examined, and the dimension is corrected as the sum of the characteristics. However, it is also possible to examine the characteristics that have passed through all the processes and perform correction based on the characteristics. That is, two masks are manufactured using the patterns shown in FIGS. 4 and 5, and after completing the CMP process, the pattern formed on the wafer is measured, and the position dependent characteristics and the pattern dependent characteristics of the dimensions are collectively displayed. Various patterns may be corrected using the characteristic data.

  Thus, the present invention can be used without depending on the specific procedure and details of the LSI manufacturing process and the mask manufacturing process.

  In the first and second embodiments, for simplicity, the core of the density-dependent function is a single Gaussian. However, this may be the sum of two or more Gaussians, other functions may be used instead of Gaussian, and the sum of such functions may be used. In the above example, the Gaussian σ value in the x direction is the same as that in the y direction, and is described in an isotropic environment. However, when this isotropic property is shifted due to device characteristics or the like, Can reflect its anisotropy by changing the σ values in the x and y directions.

  The system of the present invention can be used in combination with local dimensional correction such as optical proximity effect correction. It is only necessary to take the above procedure in advance using the pattern (A) in which the dimensional error that appears anisotropically due to the optical proximity effect or illumination is corrected. When the reticle is manufactured, the pattern A is faithfully reproduced on the mask. In addition, when applied to an LSI manufacturing process, the pattern on the reticle does not faithfully reproduce the pattern A, and the size differs depending on the location, or the change corrects errors such as subsequent steppers. In the end, the desired dimensions are obtained on the wafer.

  Furthermore, although the electron beam exposure apparatus is used as the mask drawing apparatus in the above embodiment, a drawing apparatus using a light beam may be used. After the mask manufacturing process and the LSI manufacturing process, the position dependency and the pattern dependency exist in the deviation from the design value of the dimension on the LSI (or on the mask) as described above. Independence is as described in the above operation.

  That is, the method of the present invention can be applied regardless of the details of the equipment used therein, and the details of the equipment used are characterized by the "positional dependence of dimensional variation characterized for the whole process or the process using the equipment, and Absorbed in the “pattern dependency” information.

  Further, in the above embodiment, the state in which only one chip of LSI is drawn on one mask is discussed in mind, but the present invention can be applied even when a plurality of chips are drawn on one mask. Applicable. In the case of mask manufacturing, position-dependent dimensional variation correction is performed based on the position on the mask, and this may be combined with pattern-dependent correction. In the case of LSI manufacturing, a plurality of chips on a mask are grouped together, which is virtually regarded as a chip, and position-dependent dimensional variation correction is performed in the chip, which can be combined with pattern-dependent correction. It ’s fine.

  Furthermore, in the above embodiment, the pattern inside each small region is uniformly corrected with the correction value calculated for the small region. However, as shown in FIG. You may change the amount. By this, although it is nm order (or less), it can suppress that a dimension changes rapidly between each small area | region.

  Further, in the above, the position characteristic and the pattern characteristic are processed as independent. However, the present invention can be applied to a case where the independence of the correlation is slightly broken in one apparatus for some reason. In such a case, it is only necessary to incorporate the interaction between the position characteristics and the pattern characteristics. For example, this can be dealt with by adding place dependency to the convolution function g itself, for example, adding position dependency to θ.

Furthermore, in the above-described embodiment, the density is used as a feature quantity that depends on the pattern of dimensions. However, as used in the proximity effect correction, the center of gravity and area of the pattern within the small area may be used as the feature quantity. Alternatively, a pattern dimension existing inside the small area may be weighted and processed as a feature amount. For example, a dimension having an area of 1 μm ² or less may be accumulated with a weight of 0.0, a weight of 0.5 from 1 μm ² to 10 μm ^2, and a weight of 1.0 or more, and may be used as a feature amount.

  In the above embodiment, the dimensional correction amount is the sum of the position-dependent correction amount and the density-dependent correction amount, but the present invention is not limited to this. For example, in order to correct with higher accuracy, or to correct it when the correlation between position dependency and density dependency cannot be ignored, the following may be performed.

When the dimension correction amount at the location (x, y) is ΔL (x, y), the position-dependent dimension error and the density-dependent one are Δf (x, y) and Δg (x, y), respectively.
ΔL (x, y) = 1− {1 + Δf (x, y)} × {1 + Δg (x, y)}
It is also good. Or
ΔL (x, y) = aΔf (x, y) + bΔg (x, y) + cΔf (x, y) · Δg (x, y)
Alternatively, optimization may be performed by a process or an apparatus that uses parameters a, b, and c.

  Furthermore, it is also possible to feed back to the correction amount that the density of the original pattern changes by performing dimension correction as follows. That is, as described in the above-described embodiment, after performing dimension correction every 0.5 mm × 0.5 mm, performing overlap removal and gap embedding, the pattern density is obtained again, and the dimension correction amount for each area is calculated. Ask. If this is equal to or less than a predetermined value (for example, 1 nm) in all regions, the correction is terminated assuming that sufficient correction accuracy has been obtained. Conversely, if there is an area that exceeds the desired value, processing such as dimension correction and overlap removal is performed, and the above procedure is repeated.

  In short, the present invention can be implemented with various modifications without departing from the gist thereof.

The figure which shows the example of a local dimensional variation and a global dimensional variation. The figure which shows the example of the global dimension fluctuation | variation depending on each of a pattern and a position. The figure for demonstrating the clearance gap and overlap which generate | occur | produce after dimension correction | amendment, and the removal method. The figure which shows the pattern for investigating the position dependence of global dimensional variation. The figure which shows the pattern for investigating the pattern dependence of global dimension fluctuation | variation. The figure for demonstrating the calculation method of the area density of the pattern for every small area | region. The figure for demonstrating the correction method of the dimension variation by location dependence. The figure for demonstrating the correction method of the dimension variation by pattern dependence. The figure for demonstrating the correction method (total of a position dependence part and a pattern dependence part) of a dimension variation. Sectional drawing which shows an example of the mask manufacturing process in 1st Embodiment. Sectional drawing which shows an example of the LSI manufacturing process in 2nd Embodiment. Sectional drawing which shows an example of the LSI manufacturing process in the modification of this invention. The figure for demonstrating the example which changes a dimension correction amount depending on the position in a small area | region.

Claims (5)

A mask manufacturing method for manufacturing a mask to be subjected to pattern exposure based on a predetermined design pattern,
Defining that the average value of the dimensional variation in the local region of the pattern varies in a global region wider than the local region is defined as a global dimensional variation,
In order to correct the etching unevenness caused by the etching apparatus used for manufacturing the mask, a first global dimension variation characteristic fp (x, y) that changes the pattern dimension depending on the position on the mask and a certain place ( A second global variation characteristic g (x, y) that changes the pattern size depending on the pattern density, approximating the influence of the pattern density of x, y) on the surrounding pattern size by a function of distance. And in advance,
When producing a mask using the etching apparatus, the density of LSI patterns existing inside a small area smaller than the global area is ρ (x, y), and the pattern correction amount Δ (x , Y) is obtained using fp (x, y) and ∫g (x-x ', y-y') ρ (x ', y') dx'dy ', and the design is made for each small region. A mask manufacturing method comprising correcting a pattern dimension and forming a desired pattern on the mask based on the corrected pattern dimension. An LSI manufacturing method for producing a mask based on a predetermined design pattern and forming a desired LSI pattern on a wafer using the mask,
Defining that the average value of the dimensional variation in the local region of the pattern varies in a global region wider than the local region is defined as a global dimensional variation,
A first global dimension variation characteristic fp (x, y) that changes the pattern dimension depending on the position on the chip or the chip group on the wafer in order to correct the etching unevenness by the etching apparatus used for manufacturing the LSI. ) And the second global dimension variation characteristic that changes the pattern dimension depending on the pattern density, which approximates the influence of the pattern density at a certain location (x, y) on the pattern dimension in the vicinity thereof by a function of distance. g (x, y) is obtained in advance,
When manufacturing an LSI using the etching apparatus, the density of an LSI pattern existing in a small area smaller than the large area is ρ (x, y), and the pattern correction amount Δ (x , Y) is obtained using fp (x, y) and ∫g (x-x ', y-y') ρ (x ', y') dx'dy ', and the design is made for each small region. An LSI manufacturing method, wherein a pattern dimension is corrected, and a desired pattern is formed on a mask based on the corrected pattern dimension. A mask manufacturing method for manufacturing a mask to be subjected to pattern exposure based on a predetermined design pattern,
Defining that the average value of the dimensional variation in the local region of the pattern varies in a global region wider than the local region is defined as a global dimensional variation,
First global dimensional variation characteristic fp that changes the pattern dimension depending on the position on the mask in order to correct the dimensional variation caused by the resist developing apparatus or chemical mechanical polishing apparatus used for manufacturing the mask. (X, y) and a second global pattern that changes the pattern size depending on the pattern density, approximating the influence of the pattern density at a certain location (x, y) on the pattern size in the vicinity by a function of distance. Dimensional variation characteristics g (x, y) in advance,
When producing a mask using the apparatus, the density of LSI patterns existing in a small area smaller than the global area is ρ (x, y), and the pattern correction amount Δ (x, y) is obtained using fp (x, y) and ∫g (x-x ′, y-y ′) ρ (x ′, y ′) dx′dy ′, and the design pattern for each small region And a desired pattern is formed on the mask based on the corrected pattern dimension. An LSI manufacturing method for producing a mask based on a predetermined design pattern and forming a desired LSI pattern on a wafer using the mask,
Defining that the average value of the dimensional variation in the local region of the pattern varies in a global region wider than the local region is defined as a global dimensional variation,
In order to correct a dimensional variation by a resist developing apparatus or a chemical mechanical polishing apparatus used for manufacturing an LSI, a first pattern dimension is changed depending on a position on a chip or a group of chips on a wafer. Approximate the influence of the global pattern variation characteristics fp (x, y) and the pattern density of a certain location (x, y) on the surrounding pattern dimensions by a function of distance. A second global dimensional variation characteristic g (x, y) to be changed is obtained in advance;
When an LSI is manufactured using the apparatus, the density of LSI patterns existing in a small area smaller than the large area is ρ (x, y), and the pattern correction amount Δ (x, y) is obtained using fp (x, y) and ∫g (x-x ′, y-y ′) ρ (x ′, y ′) dx′dy ′, and the design pattern for each small region And a desired pattern is formed on the mask based on the corrected pattern dimension. An LSI manufacturing method for producing a mask based on a predetermined design pattern and forming a desired LSI pattern on a wafer using the mask,
Defining that the average value of the dimensional variation in the local region of the pattern varies in a global region wider than the local region is defined as a global dimensional variation,
At least one of the apparatuses used for correcting dimensional variations caused by a plurality of manufacturing apparatuses including at least one of an etching apparatus, a resist developing apparatus, and a chemical mechanical polishing apparatus used for manufacturing an LSI. For two manufacturing apparatuses, a first global dimension variation characteristic fp (x, y) for each apparatus that changes a pattern dimension depending on a chip or a position in a chip group on a wafer, and a certain place (x, The effect of the pattern density of y) on the surrounding pattern dimensions is approximated by a function of distance, and a second global dimension variation characteristic g (x, y) for each apparatus that changes the pattern dimensions depending on the pattern density. ) And in advance,
When manufacturing an LSI using the plurality of manufacturing apparatuses, the density of the LSI pattern existing in a small area smaller than the global area is ρ (x, y),
The first global dimensional variation characteristic fp (x, y) for each of the devices to be used, the second global dimensional variation characteristic g (x, y) for each device, and the LSI pattern density ρ (x, y). y) and the pattern correction amount Δ (x, y) in each small region from fp (x, y) and ∫g (x−x ′, y−y ′) ρ (x ′, y ′) dx′dy A method for manufacturing an LSI, wherein the design pattern dimension is corrected for each of the small regions, and a desired pattern is formed on the mask based on the corrected pattern dimension.

Images