JP7547185B2 - Information processing apparatus, exposure apparatus, and method for manufacturing an article. - Google Patents

Description

The present invention relates to an information processing device, an exposure device, and a method for manufacturing an article.

Conventionally, there has been a demand for correcting the fluctuation in imaging characteristics that accompanies the absorption of exposure energy in a projection optical system provided in an exposure apparatus.
It is also known that the fluctuations in the imaging characteristics of the projection optical system vary with changes in the density distribution of exposure energy in the projection optical system depending on exposure conditions such as the type of circuit pattern formed on the original and the effective light source distribution of the illumination light.

Patent document 1 discloses a method of performing correction by calculating correction coefficients for a prediction model for predicting changes over time in the imaging characteristics of a projection optical system according to such differences in exposure conditions.

JP 2009-32875 A

In recent years, it is known that in exposure apparatuses, the density distribution of exposure energy in the projection optical system changes widely as the number of originals on which circuit patterns of semiconductor devices are formed increases with the diversification of products such as semiconductor devices.
In order to deal with such a situation, if one were to use the method disclosed in Patent Document 1 to calculate correction coefficients of a prediction model for each of a huge number of exposure conditions, a significant amount of time would be required.

The present invention therefore aims to provide an information processing device that can efficiently predict the time-dependent changes in the imaging characteristics of a projection optical system installed in an exposure apparatus.

The information processing device of the present invention is an information processing device that predicts the time change in the imaging characteristics of a projection optical system provided in an exposure apparatus using a learning model obtained by machine learning, and is characterized in that the input data for the learning model is created by classifying at least one of the effective light source distribution in the exposure apparatus and the pattern drawn on the original in terms of anisotropy in cross sections parallel to the scanning direction and non-scanning direction .

The present invention provides an information processing device that can efficiently predict the change over time in the imaging characteristics of a projection optical system installed in an exposure apparatus.

1 is a schematic diagram of an exposure apparatus equipped with an information processing apparatus according to a first embodiment. 5A and 5B are diagrams showing examples of changes over time in the imaging characteristics of a projection optical system during exposure. FIG. 2 is a diagram showing an example of a circuit pattern drawn on a reticle. 3A and 3B are diagrams showing examples of effective light source distributions formed by an illumination optical system. 13A and 13B are diagrams showing characteristics of the density distribution of exposure energy in a projection optical system using X-dipole illumination. 13A and 13B are diagrams showing characteristics of the density distribution of exposure energy in a projection optical system using Y-dipole illumination. 4 is a flowchart of a process performed in the information processing device according to the embodiment. 5 is a schematic diagram showing how a correlation degree is calculated in the information processing device according to the embodiment. FIG. 11 is a block diagram of an exposure processing system in which an information processing apparatus according to a second embodiment is provided.

The information processing device according to this embodiment will be described in detail below with reference to the accompanying drawings. Note that the drawings shown below may be drawn at a scale different from the actual scale in order to facilitate understanding of this embodiment.

In the following description, the imaging characteristics include at least one of magnification, focus, distortion, astigmatism, spherical aberration, coma, and wavefront aberration, where the wavefront aberration is expressed as each term in the Zernike polynomial of the wavefront shape.
In the following description, the above-mentioned imaging characteristics may be collectively referred to as aberration.

In the following, the direction perpendicular to the substrate mounting surface of the substrate stage will be referred to as the optical axis direction (Z direction), the direction in which the substrate mounted on the substrate mounting surface is scanned will be referred to as the scanning direction (Y direction), and the direction perpendicular to the scanning direction in a plane parallel to the substrate mounting surface will be referred to as the non-scanning direction (X direction).

[First embodiment]
Conventionally, in the manufacturing process of semiconductor devices such as LSIs and VLSIs, an exposure apparatus is used to print a circuit pattern drawn on a reticle (mask) onto a substrate (wafer) coated with a photosensitive agent by projecting and exposing the circuit pattern.

In such exposure apparatus, it is known that optical characteristics such as the refractive index change with temperature due to heat generated when part of the energy of the exposure light is absorbed in the projection optical system as projection exposure is repeated.
Furthermore, if the exposure light continues to be irradiated onto the projection optical system for a long period of time, the imaging characteristics of the projection optical system will fluctuate, and there is a risk that a non-negligible deviation will occur in the line width resolution of the circuit pattern and in the alignment accuracy for accurately overlaying patterns across multiple processes.

For this reason, methods have been proposed to correct the variations in imaging characteristics that occur according to the energy of the exposure light irradiated onto the projection optical system.

For example, there is known an exposure apparatus that predicts fluctuations in such imaging characteristics (aberration) using a prediction model with variables such as exposure dose, exposure time, and non-exposure time, and performs correction by moving the substrate stage or the field lens of the projection optical system based on the prediction results.
Then, by predicting the fluctuations of each of the aberrations included in the imaging characteristics of the projection optical system, such as the fluctuations of magnification, focus, distortion, astigmatism, and wavefront aberration, using a prediction model, the fluctuations of each imaging characteristic can be corrected.

Further, in the past, when the amount of variation in the imaging characteristics of the projection optical system deviates from the prediction, a method has been known in which the correction coefficients of the prediction model are changed using the results of directly measuring the amount of variation in the imaging characteristics.
Furthermore, when using multiple reticles each having a different circuit pattern formed thereon, a method is known in which the deviation in the predicted value of the amount of fluctuation in the imaging characteristics of the projection optical system due to differences in the circuit patterns is predicted using a correction coefficient of a prediction model obtained for each reticle.

A method is also known in which correction is performed by calculating correction coefficients of a prediction model for predicting fluctuations in the imaging characteristics of the projection optical system depending on exposure conditions such as the type of circuit pattern formed on the reticle and the effective light source distribution of the illumination light.

In a lithography process included in a manufacturing process for semiconductor devices and the like, it is required to control the amount of variation in imaging characteristics due to exposure energy in a projection optical system within an allowable range.
In recent years, the number of types of reticles on which circuit patterns of semiconductor devices are formed has increased along with the diversification of products such as semiconductor devices. As a result, in the lithography process, exposure is performed while frequently changing reticles for each product lot.

For this reason, an operator of the exposure tool creates control parameters for setting the exposure conditions in the exposure tool depending on the type of reticle.
Such control parameters include, for example, a reticle ID for identifying a reticle, an illumination mode ID for identifying the illumination light setting, an exposure area irradiated with exposure light in the projection optical system, and an exposure amount.

In any of the above methods using a prediction model for predicting the fluctuation in the imaging characteristics of the projection optical system, it is necessary to calculate a correction coefficient for each exposure condition.
Therefore, it takes a lot of time to obtain correction coefficients for each of the many types of reticles that have been produced in recent years using such conventional methods.

Therefore, in this embodiment, the objective is to provide an information processing device that can efficiently predict the fluctuations in the imaging characteristics of the projection optical system for each exposure condition by adopting the configuration described below.

Figure 1 shows a schematic diagram of an exposure apparatus 1 equipped with an information processing apparatus 20 according to the first embodiment.

The exposure apparatus 1 is a lithography apparatus that exposes a substrate (wafer) to light so as to transfer a pattern formed on an original (reticle, mask) onto the substrate by a step-and-scan method.
The exposure apparatus 1 includes an information processing apparatus 20 according to this embodiment, a laser light source 101, a laser light source control device 102, a main control device 103, an illumination optical system 104, and an illumination optical system control device 108.
The exposure apparatus 1 also includes a projection optical system 114 , a wafer stage 116 , a wafer stage controller 120 , a reticle stage 123 , a reticle stage controller 124 , and a projection optical system controller 129 .

The laser light source 101 is filled with a gas such as KrF, and emits laser light in the far ultraviolet region, for example, with a wavelength of 248 nm.
A laser light source control device 102 controls the gas exchange operation in the laser light source 101, controls the wavelength stabilization, controls the discharge applied voltage, and the like.
The main controller 103 is connected to the exposure apparatus 1 via an interface cable, and generates commands to control the entire exposure apparatus 1 .

The illumination optical system 104 guides the beam emitted from the laser light source 101 to a reticle 113 placed on a reticle stage 123 .
Specifically, the beam emitted from the laser light source 101 is shaped into a predetermined shape (circular, annular, quadrupole, dipole, etc.) via the beam shaping optical system 126, and enters the integrator lens 105 to form a secondary light source.
The light beam from the secondary light source is directed by a condenser lens 107 having a function of changing the illumination distribution on a reticle 113, thereby Koehler-illuminating a variable slit 110.
The variable slit 110 has a mechanism capable of changing the opening width, and by controlling the opening width, the intensity distribution (illuminance distribution) of the slit-shaped light (exposure light) in the non-scanning direction is made uniform.

Between the integrator lens 105 and the condenser lens 107, an aperture stop 106 having a substantially circular opening is provided.
The illumination optical system control device 108 controls the diameter of the opening of the aperture stop 106, so that the numerical aperture (NA) of the illumination optical system 104 can be set to a desired value.
Furthermore, since the ratio of the numerical aperture of the illumination optical system 104 to the numerical aperture of the projection optical system 114 is the coherence factor (σ value), the illumination optical system control device 108 can set the σ value by controlling the aperture stop 106 of the illumination optical system 104.

A half mirror 111 is disposed on the optical path of the illumination optical system 104, specifically between the condenser lens 107 and the variable slit 110, and a portion of the exposure light illuminating the reticle 113 is reflected by the half mirror 111 and extracted.
A photosensor 109 is disposed on the optical path of the reflected light from the half mirror 111, and the photosensor 109 generates an output corresponding to the intensity (exposure energy) of the exposure light.
The output of the photosensor 109 is converted into exposure energy per pulse by an integrating circuit (not shown) that performs integration for each pulse emission of the laser light source 101 , and is input to the main controller 103 via the illumination optical system controller 108 .

A circuit pattern of a semiconductor element to be printed is formed on the reticle 113 , and the reticle 113 is illuminated by the illumination optical system 104 .
A variable blade 112 consisting of a light-shielding member that can move in two dimensions, i.e., in the XY plane perpendicular to the optical axis direction, is provided between the variable slit 110 and the reticle stage 123, and the irradiation area of the pattern surface of the reticle 113 is set arbitrarily.

The projection optical system 114 is positioned so that the exposure light that has passed through the reticle 113 is guided to a wafer 115 placed on a wafer stage 116, thereby projecting the pattern on the reticle 113 onto one shot area on the wafer 115 coated with photoresist.
When a portion of the pattern on the reticle 113 is reduced and exposed onto the wafer 115 at a reduction ratio β (for example, 1/4), the reticle stage 123 and the wafer stage 116 are scanned in opposite scanning directions with respect to the slit-shaped light at a speed ratio equal to the reduction ratio β.
Then, by repeating multi-pulse exposure using pulsed light emitted from the laser light source 101 , the pattern formed on the entire surface of the reticle 113 can be transferred onto a shot area on the wafer 115 .

The projection optical system 114 is provided with a field lens 127. The field lens 127 is held by a lens barrel 130, and the lens barrel 130, i.e., the field lens 127, can be moved in the optical axis direction by a driving mechanism 128 formed of air pressure, a piezoelectric element, or the like.
The projection optical system control device 129 controls the position of the field lens 127 on the optical axis, thereby improving the projection magnification and reducing distortion errors while suppressing deterioration of various aberrations of the projection optical system 114.

The wafer stage 116 holds the wafer 115 and can move in the optical axis direction (Z direction) of the projection optical system 114 and in the X and Y directions perpendicular to each other within a plane perpendicular to the Z direction.
The laser interferometer 118 can measure the position of the wafer stage 116 in the XY plane by measuring the distance between itself and a moving mirror 117 fixed to the wafer stage 116 .
A wafer stage control device 120 controls a driving mechanism 119 such as a motor based on the position of the wafer stage 116 measured by the laser interferometer 118, thereby moving the wafer stage 116 to a predetermined position within the XY plane.

In the exposure apparatus 1, the reticle 113 and the wafer 115 are positioned so as to have a predetermined positional relationship with each other.
Based on a synchronization signal from the main controller 103, the laser light source controller 102, the wafer stage controller 120, and the reticle stage controller 124 perform various controls for scanning exposure.
As a result, the circuit pattern formed on the entire surface of the reticle 113 is transferred onto the chip area of the wafer 115 .

Then, the wafer 115 is moved a predetermined amount in the XY plane by the wafer stage 116 based on the step-and-scan method, and the pattern formed on the reticle 113 is projected and exposed onto other chip areas of the wafer 115 in the same manner.

The information processing device 20 in this embodiment is a computer serving as an information processing means, and is capable of executing a learning program that can efficiently predict fluctuations in the imaging characteristics of the projection optical system 114 according to the exposure conditions in the exposure apparatus 1.
In the exposure apparatus 1, the information processing device 20 is configured to be able to send and receive data relating to the exposure apparatus 1 via the main control device 103 connected by an interface cable.
The information processing device 20 can control the amount of variation in the imaging characteristics of the projection optical system 114 for each exposure condition of the exposure apparatus 1 within an allowable range.

Next, we will explain a prediction model for the time change in the imaging characteristics of the projection optical system 114 used in the information processing device 20 according to this embodiment.

FIG. 2 shows an example of the imaging characteristic at a given image height of the projection optical system 114 caused by exposure by the exposure apparatus 1, that is, the change over time in the amount of aberration (aberration amount) F.
The aberrations referred to here include, for example, magnification, focus, distortion, astigmatism, spherical aberration, coma aberration, and wavefront aberration.
Also, at time _t0 , the aberration amount F is assumed to be the initial aberration amount _F0 , and the difference between the aberration amount F and the initial aberration amount _F0 , i.e., F- _F0, is defined as the aberration fluctuation amount ΔF, and generally the aberration fluctuation amount ΔF takes a different value for each image height.

As shown in FIG. 2, at time _t0 , the exposure operation begins when the exposure light emitted from the laser light source 101 passes through the projection optical system 114. As time t passes, the projection optical system 114 absorbs thermal energy from the exposure light, causing the amount of aberration F to vary.
Then, at time _t1 , when a predetermined time has elapsed from time _t0 , the aberration amount F reaches the aberration amount at maximum variation (hereinafter referred to as the maximum variation aberration amount) _Fm .

Thereafter, even if the exposure light passes through the projection optical system 114, the thermal energy absorbed by the projection optical system 114 and the thermal energy released by the projection optical system 114 reach an equilibrium state, so that the aberration amount F changes very little from the maximum variable aberration amount _Fm .
Then, when the exposure operation is stopped at time _t2 , the thermal energy absorbed by the projection optical system 114 is released, causing the aberration amount F to fluctuate again toward the initial aberration amount _F0 as time t passes, and at time _t3 , the aberration amount F reaches the initial aberration amount _F0 .

Here, when the amount of aberration at a given time is F _k , the amount of aberration F _k+1 after performing an exposure operation for a time Δt is approximated as shown in the following equation (1) using the maximum fluctuating aberration amount F _m and a time constant T _h .
F _k+1 = F _m - (F _m - F _k ) x exp (-Δt/T _h )...(1)
Similarly, when the amount of aberration at a given time is F _k , the amount of aberration F _k+1 after no exposure operation is performed for a time Δt is approximated by the following equation (2) using the initial amount of aberration F ₀ and a time constant T _c .
F _k+1 =F ₀ -(F ₀ -F _k )×exp(-Δt/T _c )...(2)

The time constants T _h and T _c in equations (1) and (2), respectively, are equivalent to the time constants in the heat transfer characteristics of the projection optical system 114 and are values specific to the projection optical system 114, and further have values that differ depending on the type of aberration.
In other words, in order to accurately predict the amount of variation in the imaging characteristics of the projection optical system 114 according to the exposure energy, i.e., the amount of aberration variation ΔF, it is necessary to obtain appropriate values by performing measurements for each projection optical system 114 and for each aberration.

Furthermore, the maximum variable aberration amount _Fm can be calculated using the aberration fluctuation amount K per unit light amount (unit exposure energy) and a parameter Q that determines the exposure energy incident on the projection optical system 114, as shown in the following formula (3).
_Fm =K×Q...(3)
The parameter Q is determined from conditions such as the exposure time, the exposure amount, the scanning speed, and the exposure area information.

As shown in FIG. 2, the time change in the aberration amount F of the projection optical system 114 can be modeled using a curve 201 based on the functions expressed by the above-mentioned equations (1) to (3) (hereinafter referred to as the "prediction model"), thereby making it possible to predict the time change in the aberration amount F according to the exposure energy.

From the above, in order to predict the time change in the imaging characteristics of the projection optical system 114 with high accuracy, it is necessary to appropriately determine the aberration variation amount K per unit exposure energy for each exposure condition.
This is for the following reason: the effective light source distribution formed by the illumination optical system 104 and the light diffracted according to the circuit pattern of the reticle 113 (diffracted light) change depending on the exposure conditions when exposure is performed. Furthermore, the irradiation area of the pattern surface of the reticle 113 controlled by the variable blade 112 changes depending on the exposure conditions when exposure is performed.
This causes a change in the energy density distribution of the exposure light incident on the projection optical system 114, and therefore changes the magnitude of the aberration fluctuation amount ΔF of the projection optical system 114 and its image height dependency.
Note that the image height dependency referred to here means that the amount of variation in the imaging characteristics (aberration), i.e., the amount of aberration variation ΔF, differs between the non-scanning direction (X direction) and the scanning direction (Y direction) on the imaging surface of the projection optical system 114.

As described above, in the information processing apparatus 20 according to this embodiment, correction coefficients such as Fm, F0, _{Th ,} and _Tc are calculated using the prediction model shown in equations (1) and (2) in relation to the temporal change in the imaging characteristics of the projection optical _system 114 according to the exposure energy.
The information processing device 20 according to this embodiment is characterized in that it uses machine learning to perform estimation in order to appropriately and efficiently predict the correction coefficients of the prediction model according to the combination of exposure conditions for product lots, which are becoming increasingly diverse in the lithography process.

Specifically, the characteristics of the density distribution of the exposure energy in the projection optical system 114 are modeled and quantitatively classified to create input data.
Next, the change over time in the imaging characteristics (aberration) of the projection optical system 114 during exposure is actually measured, and correction coefficients for the prediction model are calculated to generate teacher data.

In addition, learning data consisting of input data and training data is repeatedly created for each product lot, and machine learning is performed, that is, a learning model is created.
Then, by obtaining output data indicating the correction coefficients of the predictive model corresponding to the actual exposure conditions from the created learning model, it is possible to appropriately and efficiently predict the change over time in the imaging characteristics of the projection optical system 114 during exposure.

Next, we will explain a method for quantitatively classifying the characteristics of the density distribution of exposure energy in the projection optical system 114 from the drawing information of the circuit pattern formed on the reticle 113.

Figures 3(a) and (b) show an example of a circuit pattern drawn on the reticle 113.

Specifically, the line and space (hereinafter referred to as L&S) pitch (spacing) and the aspect ratio (HV ratio) are determined by actually measuring the circuit pattern drawn on the reticle 113 or by referring to the design data.
Here, the L&S pitch is a factor that determines the diffraction angle of light diffracted by the circuit pattern, and the HV ratio is a factor that determines the directivity of the diffracted light.

In the example shown in FIG. 3( a), a vertical L&S pattern (hereinafter, sometimes referred to as a vertical pattern) 301 having a vertical elongated shape and a horizontal L&S pattern (hereinafter, sometimes referred to as a horizontal pattern) 303 having a horizontal elongated shape are shown.
In this case, the vertical direction is the scanning direction (Y direction), and the horizontal direction is the non-scanning direction (X direction).

In the vertical pattern 301 , L&S are arranged in the horizontal direction so that the spacing between adjacent wirings is a pitch 302 .
Then, from the drawing information of the vertical pattern 301, the pitch 302 and the areas V1, V2, V3, V4 and V5 of each vertical wiring are calculated.

In addition, in the horizontal pattern 303, L&S are arranged along the vertical direction so that the interval between adjacent wirings is a pitch 304.
Then, from the drawing information of the horizontal pattern 303, the pitch 304 and the areas H1, H2, H3, H4 and H5 of each horizontal wiring are calculated.

FIG. 3B shows a rectangle 305 indicating an exposure area on the reticle 113 that is actually irradiated with exposure light.
The area of vertical pattern 301 that is actually irradiated with exposure light is within a rectangle defined by sides 306 and 307 of rectangle 305, and the sum of the areas of each vertical wiring contained within that exposure area, i.e., total area V, is calculated from areas V1 to V5.
Similarly, the area of horizontal pattern 303 that is actually irradiated with exposure light is within the rectangle defined by sides 306 and 308 of rectangle 305, and the sum of the areas of each horizontal wiring contained within that exposure area, i.e., the total area H, is calculated from areas H1 to H5.

Then, in order to quantify the directivity of the diffracted light due to the circuit pattern drawn on the reticle 113, the ratio between the horizontal wiring and the vertical wiring, that is, the HV ratio, is quantified using the following equation (4).
HV ratio=H/(H+V)...(4)

The information processing device 20 according to this embodiment classifies the directivity of the diffracted light due to the circuit pattern formed on the reticle 113 based on the above-mentioned L&S pitch and HV ratio using the table shown in Table 1 below.

For example, if the pitch of the vertical pattern 301 and the horizontal pattern 303 in the circuit pattern is 600 nm, the pattern is classified as "500 nm or more and less than 700 nm."
Furthermore, when the HV ratio of the vertical pattern 301 and the horizontal pattern 303 in the circuit pattern is 60%, it is classified as "50% or more and less than 75%".

Next, we will explain a method for quantitatively classifying the characteristics of the density distribution of exposure energy in the projection optical system 114 from the effective light source distribution formed by the illumination optical system 104.

Specifically, the light intensity in the effective light source distribution formed by the illumination optical system 104 is actually measured to obtain the XY intensity ratio of the illumination light.
1, the illumination light shaped into a predetermined shape (circular, annular, quadrupole, dipole, etc.) via the beam shaping optical system 126 in the illumination optical system 104 is measured by an effective light source measurement device 135 arranged on the wafer stage 116. This makes it possible to obtain the effective light source distribution of the illumination light.

FIG. 4A is a diagram showing how the effective light source distribution is obtained by the effective light source measurement device 135.
As shown in FIG. 4A, a very small pinhole 403 having a diameter of several tens of μm is disposed on an image plane 402 of an effective light source distribution 401 in the projection optical system 114 .

The illumination light that passes through the projection optical system 114 passes through a tiny pinhole 403 and is then irradiated onto a light receiving section 405 of the effective light source measurement device 135 with a light intensity distribution equivalent to the effective light source distribution 401, i.e., an effective light source distribution 404.
The light receiving unit 405 is, for example, a two-dimensional CCD sensor in which CCD pixels are arranged in a 512 x 512 grid pattern, and outputs an electric signal according to the brightness of the irradiated illumination light, thereby measuring the effective light source distribution 404 of the illumination light.

FIG. 4B shows a top view of the two-dimensional CCD sensor 405, illustrating a state in which an effective light source distribution 404 of illumination light having an annular shape is irradiated.
Here, the light intensity measured by a CCD pixel (a tiny point light source) located at positions X and Y is represented as A(X, Y), and when A(X, Y) = 0, it is assumed that the CCD pixel is not illuminated with illumination light.

The light receiving area of the two-dimensional CCD sensor 405 is divided by four diagonal lines in the directions of 45 degrees, 135 degrees, 225 degrees, and 315 degrees centered on the optical axis.
Then, each region sandwiched between adjacent diagonal lines is defined, namely, an X-direction right region 407, a Y-direction upper region 408, an X-direction left region 409, and a Y-direction lower region 410 (hereinafter referred to as the XR region 407, the YU region 408, the XL region 409, and the YD region 410, respectively).

Furthermore, the integrated values of the light intensities measured by the CCD pixels included in the thus defined regions XR, YU, XL, and YD are defined as _LXR , _LYU , _LXL , and _LYD , respectively.
Then, the XY intensity ratio in the effective light source distribution of the illumination light acquired by the two-dimensional CCD sensor 405 is calculated from the following equation (5).
XY intensity ratio = (L _XL + L _XR ) / (L _YU + L _YD + L _XL + L _XR ) ... (5)
In the effective light source distribution 404 of the illumination light having an annular shape, L _XR , L _YU , L _XL and L _YD are approximately the same value, and therefore the XY intensity ratio is close to 50%.

FIG. 4C shows a state where the two-dimensional CCD sensor 405 is irradiated with illumination light having a dipole shape in the X direction, that is, an effective light source distribution 412 (hereinafter referred to as X-dipole illumination).
FIG. 4D shows a state in which the two-dimensional CCD sensor 405 is irradiated with illumination light having a dipole shape in the Y direction, that is, an effective light source distribution 413 (hereinafter referred to as Y dipole illumination).

In such a case, the XY intensity ratio actually measured in the effective light source distribution 412 by the X dipole illumination becomes a value close to 100% according to the formula (5).
On the other hand, the XY intensity ratio actually measured in the effective light source distribution 413 by the Y dipole illumination is close to 0% according to the formula (5).

The information processing device 20 according to this embodiment quantitatively classifies the characteristics of the effective light source distribution of the illumination light based on the above-mentioned XY intensity ratio using the table shown in Table 2 below.

For example, when the XY intensity ratio is measured to be 55% in the effective light source distribution 404 of the illumination light having an annular shape, it is classified as "50% or more and less than 60%".
Moreover, when the XY intensity ratio is measured to be 98% in the effective light source distribution 412 of the X-dipole illumination, it is classified as "90% or more and 100% or less".
Furthermore, when the XY intensity ratio is measured to be 1% in the effective light source distribution 413 of the Y dipole illumination, it is classified as "0% or more and less than 10%".

Next, we will explain the relationship between the exposure conditions in the information processing device 20 according to this embodiment, the characteristics of the density distribution of the exposure energy in the projection optical system 114, and the imaging characteristics of the projection optical system 114, specifically the variation in magnification.

Figures 5(a) and (b) respectively show a schematic diagram of an effective light source distribution 501 of illumination light as an example of an exposure condition, and a circuit pattern 502 formed on a reticle 113.

As shown in FIG. 5A, an effective light source distribution 501 is an effective light source distribution of X-dipole illumination, and has a characteristic that the light intensity is strong in the X direction, that is, the XY intensity ratio is close to 100%.
Also, as shown in FIG. 5B, in the circuit pattern 502, the vertical L&S pattern is larger than the horizontal L&S pattern, i.e., the HV ratio is less than 50%, and the diffracted light has a directivity that scatters in the X direction.

Figure 5(c) shows how illumination light is irradiated onto the pupil plane 503 of the projection optical system 114.

Specifically, illumination light having an effective light source distribution 501 irradiates a zero-order light 504 that is symmetric in the X direction across a YZ cross section that includes the optical axis AX of the projection optical system 114 and is parallel to the Y direction.
The vicinity of the zero-order light 504 is irradiated with diffracted lights 505 and 506 that are generated by the illumination light being diffracted by the vertical L&S pattern included in the circuit pattern 502 .
Typically, the amount of light of the zero-order light 504 is greater than the amount of diffracted light 505 and 506 .

When exposure is continuously performed based on the density distribution of exposure energy in the projection optical system 114 formed from the effective light source distribution 501 and the circuit pattern 502, a lot of heat is absorbed in the areas irradiated with the zero-order light 504 and the diffracted light 505 and 506, causing the temperature to rise.
As a result, the lenses included in the projection optical system 114 undergo thermal deformation, that is, thermal expansion, along the X direction.

Figure 5(d) shows the chip area on wafer 115 where projection exposure is performed.

As described above, due to the thermal deformation of the lenses included in the projection optical system 114 along the X direction, as shown in FIG. 5D, projection exposure is performed in an area 508 where a magnification error occurs in the X direction with respect to a designed chip area 507.
That is, when the density of exposure energy is higher in the X direction than in the Y direction, the magnification fluctuation is also larger in the X direction than in the Y direction, resulting in a directional difference (anisotropy) in the magnification fluctuation.

Figures 6(a) and (b) respectively show a schematic diagram of an effective light source distribution 701 of illumination light as another example of exposure conditions and a circuit pattern 702 formed on a reticle 113.

As shown in FIG. 6A, an effective light source distribution 701 is an effective light source distribution of Y dipole illumination, and has the characteristic that the light intensity is strong in the Y direction, that is, the XY intensity ratio is close to 0%.
Also, as shown in FIG. 6B, in the circuit pattern 702, the horizontal L&S pattern is larger than the vertical L&S pattern, i.e., the HV ratio is greater than 50%, and the diffracted light has a directivity that scatters in the Y direction.

Figure 6(c) shows how illumination light is irradiated onto the pupil plane 503 of the projection optical system 114.

Specifically, illumination light having an effective light source distribution 701 irradiates a zero-order light 704 that is symmetric in the Y direction across an XZ cross section that includes the optical axis AX of the projection optical system 114 and is parallel to the X direction.
The vicinity of the zero-order light 704 is irradiated with diffracted lights 705 and 706 that are generated by the illumination light being diffracted by the horizontal L&S pattern included in the circuit pattern 702 .
Typically, the amount of light of the zero-order light 704 is greater than the amount of diffracted light 705 and 706 .

When exposure is continuously performed based on the density distribution of exposure energy in the projection optical system 114 formed from the effective light source distribution 701 and the circuit pattern 702, a lot of heat is absorbed in the areas irradiated with the zero-order light 704, the diffracted light 705 and 706, causing the temperature to rise.
As a result, the lenses included in the projection optical system 114 undergo thermal deformation, that is, thermal expansion, along the Y direction.

Figure 6(d) shows the chip area on wafer 115 that is projected and exposed.

As a result of the lens included in the projection optical system 114 being thermally deformed along the Y direction as described above, projection exposure is performed in an area 708 where a magnification error occurs in the Y direction with respect to the designed chip area 707, as shown in FIG. 6( d ).
That is, when the density of exposure energy is higher in the Y direction than in the X direction, the magnification fluctuation is also larger in the Y direction than in the X direction, resulting in a directional difference in the magnification fluctuation.

As described above, in the information processing apparatus 20 according to this embodiment, the XY intensity ratio is calculated and classified in the effective light source distribution formed by the illumination optical system 104 .
Furthermore, the L&S pitch and HV ratio of the circuit pattern 702 formed on the reticle 113 are calculated and classified.

Then, input data is created by quantitatively classifying the characteristics of the density distribution of exposure energy from the exposure conditions including the classified XY intensity ratio, L&S pitch, and HV ratio.
In other words, in the information processing device 20 according to this embodiment, input data is created by classifying the exposure conditions in the exposure device 1 with respect to anisotropy in cross sections parallel to the scanning direction and non-scanning direction.

Next, exposure is performed under the above exposure conditions in the exposure apparatus 1 to actually measure the change over time in the imaging characteristics (aberration) of the projection optical system 114, and correction coefficients for the prediction model are calculated to create training data.

Then, learning data consisting of input data and teacher data is repeatedly created, that is, the correlation between the feature amount of the density distribution of exposure energy and the correction coefficient is repeatedly learned, and a learning model is created.
This makes it possible to appropriately and efficiently predict the change in the imaging characteristics of the projection optical system 114 over time during exposure by obtaining output data from the created learning model indicating the correction coefficients of the predictive model corresponding to the actual exposure conditions.

Next, we will explain the processing that is actually performed by the information processing device 20 according to this embodiment.

Figure 7 shows a flowchart of the processing performed in the information processing device 20 according to this embodiment.

First, in step S1, a feature amount of the effective light source distribution formed by the illumination optical system 104 is calculated.
For example, consider the case where the X-dipole illumination shown in Fig. 5A is irradiated. In this case, the effective light source measurement device 135 measures a light intensity distribution equivalent to the effective light source distribution of the X-dipole illumination, and calculates that the light intensity is strong in the X direction (non-scanning direction), for example, the XY intensity ratio is 98%.

Next, in step S2, the feature amount of the circuit pattern formed on the reticle 113 is calculated.
For example, consider the case where the circuit pattern shown in FIG.
At this time, a reticle line width measuring instrument (not shown) measures the circuit pattern formed on the reticle 113, and calculates that the L&S pitch is, for example, 600 nm. In addition, it is calculated that the circuit pattern contains many vertical L&S patterns, for example, that the HV ratio is 85%.
This shows that the circuit pattern formed on the reticle 113 has directivity that generates diffracted light along the horizontal direction (X direction).

In step S3, input data is created by classifying the feature amount of the density distribution of the exposure energy in the projection optical system 114 from the feature amounts calculated in steps S1 and S2.
An example of a table used in creating the input data in step S3 is shown in Table 3 below.

From the table shown in Table 3, the XY intensity ratio of 98% calculated in step S1 is classified as "90% or more and 100% or less", and the L&S pitch of 600 nm calculated in step S2 is classified as "500 nm or more and less than 700 nm".
Moreover, the HV ratio of 85% calculated in step S2 is classified as "75% or more and 100% or less."

In this manner, the input data is created by classifying the feature amounts of the density distribution of the exposure energy in the projection optical system 114 .
The above steps S1 to S3 are referred to as an input data creation process 801.

Next, in step S4, a correction coefficient appropriate to the exposure conditions, i.e., corresponding to the input data created in step S3, is estimated. Specifically, it is determined whether the learning model already contains a correction coefficient.

If the learning model does not include a correction coefficient (No in step S4), the process proceeds to step S5, where a correction coefficient of a preset design condition, that is, an initial value of the correction coefficient, is selected.
On the other hand, if the learning model includes a correction coefficient (Yes in step S4), the process proceeds to step S6, where the correction coefficient having the highest correlation with the input data is selected (the correlation will be described later).
The above steps S4 to S6 are referred to as a correction coefficient estimation step 802.

FIG. 8A shows a prediction model 601 using the correction coefficients M _m , M ₀ , T _h and T _c of the design conditions selected in step S5.
In FIG. 8A, the magnification M is shown as the imaging characteristic of the projection optical system 114, that is, _Fm and _F0 in the expressions (1) and (2) are replaced with _Mm and _M0 , respectively.

As shown in FIG. 8A, the initial value of the magnification M, i.e., the initial magnification, is _M0 . When the exposure operation is started at time _t0 when the exposure light emitted from the laser light source 101 passes through the projection optical system 114, the magnification M changes over time.
Then, after the magnification M reaches the maximum fluctuation magnification _Mm at time _t1 , the magnification M hardly changes from the maximum fluctuation magnification _Mm .

Then, when the exposure operation is stopped at time _t2 , the magnification M changes so as to approach the initial magnification _M0 again as time t passes, and reaches the initial magnification _M0 at time _t3 .
It should be noted here that in the prediction model 601, no directional difference occurs with respect to the change over time in the magnification M of the projection optical system 114.

Next, in step S7, the change over time in the magnification M of the projection optical system 114 is actually measured during the exposure process of the product lot under the exposure conditions shown in Figures 5(a) and (b).

FIG. 8B shows an example of the change over time of the magnification M measured in step S7.
Specifically, the graph shows changes over time M _X1 to M _X10 of the magnification M measured at a predetermined X coordinate of the projection optical system 114, and changes over time M _Y1 to M _Y10 of the magnification M measured at a predetermined Y coordinate.

Next, in step S8, correction coefficients of the prediction model expressed by equations (1) and (2) for the data measured in step S7 are calculated.
Specifically, for the time changes M _X1 to M _X7 (exposure characteristics) of the magnification M in the X direction from time _t0 to _t1 , fitting is performed using a prediction model expressed by equation (1) to calculate the correction coefficients M _X and T _hX . Then, for the time changes M _X8 to M _X10 (non-exposure characteristics) of the magnification M in the X direction from time _t2 to _t3 , fitting is performed using a prediction model expressed by equation (2) to calculate the correction coefficients M ₀ and T _cX .

Similarly, for the time changes M _Y1 to M _Y7 of the magnification M in the Y direction from time _t0 to _t1 (exposure characteristics), fitting is performed using the prediction model expressed by equation (1) to calculate correction coefficients M _Y and T _hY . Then, for the time changes M _Y8 to M _Y10 of the magnification M in the Y direction from time _t2 to _t3 (non-exposure characteristics), fitting is performed using the prediction model expressed by equation (2) to calculate correction coefficients M ₀ and T _cY .
The prediction models thus obtained are shown as prediction models 602 and 603 in FIG. 8(b), respectively.

なおここで、最大変動倍率Ｍ_Ｘ及びＭ_ｙそれぞれについては、初期倍率Ｍ_０に対するずれの割合、すなわち（Ｍ_Ｘ－Ｍ_０）／Ｍ_０及び（Ｍ_Ｙ－Ｍ_０）／Ｍ_０で表している。 Moreover, examples of the correction coefficients M _X , M _Y , T _hX , T _hY , T _cX and T _cY calculated in step S8 above are shown in Table 4 below.

Here, the maximum fluctuation magnifications M _X and M _Y are expressed as the ratio of deviation to the initial magnification M ₀ , that is, (M _X −M ₀ )/M ₀ and (M _Y −M ₀ )/M ₀ , respectively.

In this manner, training data corresponding to the input data shown in Table 3 above can be obtained from the table shown in Table 4.
The above steps S7 and S8 are referred to as a teacher data creation process 803.

Next, in step S9, the degree of correlation between the input data, which is the characteristic amount of the density distribution of the exposure energy in the projection optical system 114, and the training data and output data, which is the correction coefficient, is calculated from the prediction error in the prediction model.

FIG. 8C is a schematic diagram showing how the correlation degree is calculated in step S9.
Specifically, FIG. 8C shows a plurality of measurement points obtained by measuring the change over time in the magnification M of the projection optical system 114 under specified exposure conditions, and a predictive model 604 using specified correction coefficients.

Here, in step S9, the value with the largest error from the prediction model 604 among the multiple measurement points acquired (the difference from the measurement point that is farthest from the prediction model) is taken as the prediction error D, and the prediction error D is recorded as the degree of correlation with the prediction model 604.
That is, in the above example, for the multiple measurement points measured in step S7, a prediction error D is calculated from each of the prediction model 601 selected in step S5 and the prediction models 602 and 603 obtained in step S8, and recorded as the degree of correlation.

For example, the correlation degree calculated in each of the prediction models 602 and 603, that is, the value of the prediction error D, is shown in Table 4.
Here, the degree of correlation, that is, the prediction error D, is expressed as the ratio of deviation of the magnification M _meas at the measurement point when calculating the prediction error D to the magnification M _calc calculated from the prediction model, that is, (M _meas -M _calc )/M _calc .

This makes it possible to perform quantitative learning from the input data corresponding to the exposure conditions exemplified in Table 3 and the teacher data and correlation degree exemplified in Table 4.
The above-described step S9 is referred to as a learning step 804.

Then, when performing exposure processing under similar exposure conditions from the next time onwards, the correction coefficients can be estimated based on the correlation data acquired up to this point.
In step S10, it is determined whether or not a product lot is to be repeatedly produced under the same exposure conditions.
If production is to be performed (Yes in step S10), the process returns to step S4, and if production is not to be performed (No in step S10), the process ends.

Returning to step S4, the correction coefficient estimation step 802 is performed again. That is, in step S4, the correction coefficients of the prediction model are estimated.
At this time, the learning model includes the correction coefficients learned in the previous step S9 (Yes in step S4).
Therefore, after proceeding to step S6, the correction coefficient with the smallest prediction error D, that is, the correction coefficient with the highest correlation, is selected from the correction coefficients stored in the learning model in step S6.

Then, the process proceeds to steps S7 to S10, whereby learning is repeated until the production of the product lot under similar exposure conditions is completed.
In this way, by repeatedly learning the time change of the magnification M of the projection optical system 114 for each exposure condition during the exposure process of a product lot and accumulating correction coefficients, the estimation accuracy of the correction coefficient can be improved even if the magnification M shows different time changes in the X direction and Y direction.

In the exposure apparatus 1, when correcting the magnification, distortion, astigmatism, spherical aberration, coma aberration, and wavefront aberration of the projection optical system 114 based on the estimated correction coefficients, the field lens 127 is moved according to time.
Furthermore, when correcting the focus of the projection optical system 114 based on the estimated correction coefficient, the position of the wafer stage 116 in the optical axis direction is changed according to time.

As described above, in the information processing device 20 according to this embodiment, a learning model is created using the exposure conditions in the exposure device as input data and the correction coefficients of a prediction model that predicts the time changes in the imaging characteristics of the projection optical system 114 as teacher data.
Furthermore, even if the number of exposure conditions increases, such as the number of types of originals, the change over time in the imaging characteristics of the projection optical system can be efficiently predicted by obtaining output data that provides correction coefficients for the predictive model from the created learning model.

In the above, the change in the magnification M over time has been described as a variation in the imaging characteristics of the projection optical system 114, but this is not limiting. For example, the information processing device 20 according to this embodiment can also be applied to focus, distortion, astigmatism, spherical aberration, coma aberration, wavefront aberration, and the like as the imaging characteristics of the projection optical system 114.
Although the above describes the characteristic quantities of the density distribution of exposure energy by X-dipole illumination shown in Fig. 5(a) to (d), the present invention is not limited to this. For example, the information processing device 20 according to the present embodiment can also be applied to the characteristic quantities of the density distribution of exposure energy by Y-dipole illumination shown in Fig. 6(a) to (d).

[Second embodiment]
FIG. 9 shows a block diagram of an exposure processing system 1000 provided with an information processing apparatus 220 according to the second embodiment.

As shown in FIG. 1, an information processing apparatus 20 according to the first embodiment is a so-called stand-alone type apparatus that is connected to a main controller 103 in an exposure apparatus 1 via an interface cable.
On the other hand, the information processing device 220 according to this embodiment is a so-called online type device that is connected to a plurality of exposure apparatuses including exposure apparatus 1001 and exposure apparatus 1002 via a network cable 1003 in an exposure processing system 1000, as shown in FIG.
As shown in FIG. 9, the information processing apparatus 220 according to this embodiment is also connected to a reticle pattern measuring instrument 1005 for measuring the circuit pattern of the reticle 113 via a network cable 1003 .

That is, the information processing device 220 according to this embodiment is a computer serving as information processing means, and can perform machine learning regarding the correction coefficients of a prediction model based on the learning data for each of a plurality of exposure apparatuses.
Specifically, input data is created from effective light source distribution data measured in each exposure tool and circuit pattern data measured by the reticle pattern measurement instrument 1005 .

Then, training data is created by calculating correction coefficients for a predictive model from the measurement values of the imaging characteristics of the projection optical system 114 obtained in each exposure apparatus, and correlation data between the input data and the training data is calculated and stored.
In this way, by recording the correlation data for each exposure condition learned for a plurality of exposure tools in the recording device 30 and creating a learning model, it is possible to further improve the estimation accuracy of the correction coefficients.

As described above, in the information processing device 220 according to this embodiment, a learning model is created using the exposure conditions in the exposure device as input data and the correction coefficients of a prediction model that predicts the time changes in the imaging characteristics of the projection optical system 114 as teacher data.
Even if the number of exposure conditions increases, such as an increase in the number of types of originals, it is possible to obtain output data that gives correction coefficients for the prediction model from the created learning model.
In addition, in the information processing device 220 according to this embodiment, a learning model can be created using learning data from multiple exposure apparatuses, thereby making it possible to more efficiently predict changes over time in the imaging characteristics of the projection optical system.

[Production method of the article]
Next, a method for manufacturing an article using an exposure apparatus equipped with an information processing apparatus according to either the first or second embodiment will be described.

The article is a semiconductor device, a display device, a color filter, an optical component, a MEMS, or the like.
For example, semiconductor devices are manufactured through a front-end process for creating circuit patterns on a wafer and a back-end process including processing steps for completing the circuit chips created in the front-end process into products.

The pre-processing step includes an exposure step of exposing a wafer coated with a photosensitive agent using an exposure apparatus equipped with an information processing apparatus according to either the first or second embodiment, and a development step of developing the photosensitive agent.
Then, the developed photosensitive pattern is used as a mask to carry out etching steps, ion implantation steps, etc., to form a circuit pattern on the wafer.

By repeating these steps of exposure, development, etching, etc., a circuit pattern consisting of multiple layers is formed on the wafer.
In a post-process, dicing is performed on the wafer on which the circuit pattern has been formed, and the chips are mounted, bonded, and inspected.

The display device is manufactured through a process of forming a transparent electrode. The process of forming the transparent electrode includes a process of applying a photosensitive agent onto a glass wafer on which a transparent conductive film has been deposited, and a process of exposing the glass wafer on which the photosensitive agent has been applied using an exposure device equipped with an information processing device according to either the first or second embodiment. The process of forming the transparent electrode also includes a process of developing the exposed photosensitive agent.

The method for manufacturing articles according to this embodiment makes it possible to manufacture articles of higher quality and with higher productivity than ever before.

Although the preferred embodiments have been described above, the present invention is not limited to these embodiments, and various modifications and changes are possible within the scope of the gist thereof.
Furthermore, although the above describes an information processing device according to this embodiment, the above-described information processing method, a program for implementing the method, and a computer-readable recording medium on which the program is recorded are also included within the scope of this embodiment.

1 Exposure apparatus 20 Information processing apparatus 114 Projection optical system

Claims (11)

1. An information processing apparatus that predicts a time change in an imaging characteristic of a projection optical system provided in an exposure apparatus by using a learning model acquired by machine learning, comprising:
An information processing device characterized in that the input data for the learning model is created by classifying at least one of the effective light source distribution in the exposure device and the pattern drawn on the original in terms of anisotropy in cross sections parallel to the scanning direction and non-scanning direction . 2 . The information processing apparatus according to claim 1 , wherein the teacher data for the learning model is created by calculating correction coefficients by measuring changes over time in the imaging characteristics of the projection optical system and fitting the changes over time with a prediction model. 3. The information processing apparatus according to claim 1 , wherein the imaging characteristics include at least one of magnification, focus, distortion, astigmatism, spherical aberration, coma aberration, and wavefront aberration. The information processing device includes:
creating the input data by classifying at least one of the effective light source distribution and the pattern with respect to the anisotropy;
creating training data by measuring a time change in the imaging characteristics of the projection optical system and calculating correction coefficients by fitting the measurement results to a prediction model;
The information processing device according to any one of claims 1 to 3, characterized in that the learning model is created by performing learning by calculating the correlation between the measured changes in imaging characteristics over time and the teacher data and the output data for estimating the correction coefficients. 5 . The information processing apparatus according to claim 4 , wherein the information processing apparatus calculates, as the degree of correlation, a prediction error of a measurement point that is farthest from the prediction model among the measurement points of the imaging characteristics. 6. The information processing apparatus according to claim 4 , wherein the information processing apparatus creates the learning model by performing the learning for a plurality of the exposure apparatuses. 7. The information processing apparatus according to claim 1, wherein the information processing apparatus uses the learning model to predict changes over time in imaging characteristics of the projection optical system provided in each of the multiple exposure apparatuses. 1. An exposure apparatus that exposes a substrate to light so as to transfer a pattern drawn on an original onto the substrate, comprising:
An exposure apparatus comprising the information processing apparatus according to claim 1 . exposing the substrate using the exposure apparatus according to claim 8 ;
developing the exposed substrate;
producing an article from the developed substrate;
A method for producing an article, comprising: An information processing method performed by an information processing device,
predicting a time change in an imaging characteristic of a projection optical system provided in the exposure apparatus using a learning model obtained by machine learning ;
An information processing method characterized in that the input data for the learning model is created by classifying at least one of the effective light source distribution in the exposure device and the pattern drawn on the original in terms of anisotropy in cross sections parallel to the scanning direction and non-scanning direction . A computer-readable recording medium having a program recorded thereon for causing a computer to perform information processing,
A step of predicting a time change in an imaging characteristic of a projection optical system provided in an exposure apparatus using a learning model obtained by machine learning by a computer ;
A computer-readable recording medium having a program recorded thereon, characterized in that input data for the learning model is created by classifying at least one of the effective light source distribution in the exposure apparatus and the pattern drawn on the original in terms of anisotropy in cross sections parallel to the scanning direction and non-scanning direction .

Images