A snapshot-style overlay error measurement method and system

Abstract

The present application belongs to the field of integrated circuit manufacturing online measurement, and discloses a snapshot overlay error measurement method and system, which comprises the following steps: measuring a measured object, obtaining two measurement spectra with positive and negative preset deviations respectively, coherently demodulating the measurement spectra, and shifting the specific frequency channel to the zero frequency channel; processing the spectrum data after frequency shifting by using a linear operator to obtain the coefficients corresponding to the positive and negative preset deviations; constructing a characteristic quantity according to the linear combination of the real part and the imaginary part of the coefficients, then calculating the corresponding characteristic quantity according to the coefficients corresponding to the positive and negative preset deviations; and calculating the overlay error according to the characteristic quantity based on the linear relationship between the characteristic quantity and the overlay error. The present application can solve the overlay error with multi-wavelength coupling without using traditional Fourier analysis and truncation operation, has high precision and robustness to noise, and can be used for data processing of snapshot overlay error measurement in multiple scenes.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of integrated circuit manufacturing online measurement, and more particularly, to a snapshot overlay error measurement method and system. BACKGROUND

[0002] Overlay refers to the alignment error between the current layer (the pattern retained on the photoresist after exposure and development) and the reference layer (the pattern already on the wafer) at the time of exposure, that is, the relative offset from the designed position of the current layer. With the rapid development of semiconductor manufacturing processes and technologies, integrated circuit (IC) components are facing the trend of miniaturization and three-dimensionality, requiring a transition from simple two-dimensional structures to complex three-dimensional structures, which often require the overlay error between the current layer and the previous layer in the key lithography layer to be less than 1 / 3 to 1 / 5 of the feature line width. Overlay error can greatly affect the yield of manufacturing. Overlay error, resolution, and yield are known as the three major performance indicators of a lithography machine, and their rapid measurement and accurate evaluation are critical to ensuring the optimization of lithography machine operating parameters and process yield control. Currently, the international semiconductor technology node has been reduced to 7nm, and in the future, the overlay error of the key lithography layer will be reduced to sub-nanometer level, which poses extremely stringent requirements on overlay error measurement technology.

[0003] Optical measurement technology has become the most commonly used measurement technology in production due to its speed and non-destructive advantage. Optical measurement technology has developed into two categories: image-based measurement method (IBO) and diffraction-based measurement method (DBO). IBO has a large error and cannot break through the optical resolution limit, and is facing severe challenges as the manufacturing technology node continues to decrease. DBO obtains the optical diffraction signal of the overlay mark and calculates the overlay error using certain mathematical methods, which can achieve high precision and break through the optical resolution limit, so it has become the widely used overlay error measurement technology.

[0004] Empirical diffraction-based overlay (eDBO) can convert the measurement of overlay error OV into a linear combination of the off-diagonal elements of the Mueller matrix corresponding to the zero-order diffraction light ΔM (ΔM can be M 13 +M 31 , M 23 +M 32 , M 41 -M 14 or M 42 -M 24) and the overlay error is solved by using the linear relationship between them, as shown in the following formula:

[0005] ΔM=k·OV

[0006] Wherein, K is the slope. On the basis of the above formula, preset the deviation of +D and -D on the two overlay marks respectively, so that the total overlay error in the two marks is (OV+D) and (OV-D) respectively, and the corresponding two non-diagonal element combinations ΔM + and ΔM - As shown in the following formula:

[0007] ΔM + =k·(OV+D)

[0008] ΔM - =k·(OV-D)

[0009] So the overlay error can be calculated as:

[0010]

[0011] Most snapshot overlay error measurement methods are to adjust the hardware configuration, modulate ΔM to a specific carrier channel, and then perform "Fourier transform, channel interception and inverse Fourier transform" operation to quickly measure ΔM in a wide spectral range, so as to obtain the overlay error. However, due to the use of traditional Fourier analysis method and channel interception operation, this kind of snapshot overlay error measurement method usually shows large error at both ends of the spectral range, and is very sensitive to random noise, which is a common problem that reduces the effective spectral range, affects the accuracy of overlay error measurement, and is easily affected by random error. In addition, after obtaining the overlay error under multiple wavelengths, the commonly used method is to take the average value of the overlay error values under all wavelengths as the final result. Such a multi-wavelength coupling solving method will be affected by a large error.

[0012] Therefore, there is an urgent need for a snapshot overlay error measurement method that can avoid traditional Fourier analysis, resist random noise interference, and realize new multi-wavelength coupling solving. SUMMARY

[0013] In view of the above defects or improvement needs of the prior art, the present application provides a snapshot overlay error measurement method and system, which aims to realize multi-wavelength coupling solving of overlay error, avoid errors introduced by traditional Fourier analysis and channel interception, and thus greatly improve the robustness of the measurement method to random error.

[0014] To achieve the above purpose, according to one aspect of the present application, a snapshot overlay error measurement method is provided, comprising the following steps:

[0015] S1, measuring the measured piece, respectively acquiring two measurement spectra with positive and negative preset deviations, coherently demodulating the measurement spectra, and shifting a specific frequency channel to a zero frequency channel;

[0016] S2, processing the frequency-shifted spectral data using a linear operator to obtain coefficients F + and F － corresponding to positive and negative preset deviations.

[0017] S3, constructing a feature quantity by linearly combining the real and imaginary parts of the coefficients, then calculating the corresponding feature quantities Spe + and Spe － according to the coefficients F + and F － . + -

[0018] S4, based on the linear relationship between the feature quantity and the overlay error, calculating the overlay error according to the feature quantities Spe + and Spe - .

[0019] As a further preferred, the linear operator is specifically:

[0020]

[0021] where f n is the frequency component of the linear operator, n=0, 1, 2, …; N is the signal length of the spectrum, i is the imaginary unit; f max is determined by the following formula:

[0022]

[0023] where [λ min , λ max ] is the spectral range, B max is the maximum birefringence of the wave plate in this spectral range, t is the thickness of the thickest multi-stage wave plate in the measurement device, and c is the number of frequency channels of the channel spectrum.

[0024] As a further preferred, the frequency-shifted spectral data is processed using a linear operator, specifically:

[0025]

[0026] wherein, is the spectral data, is the calculated coefficient.

[0027] As a further preferred, the linear operator is optimized using a multi-objective genetic algorithm or a deep learning algorithm, and the optimization parameter is the frequency component f n .

[0028] As a further optimization, the linear operator is optimized by using a slime mold optimization algorithm, and the optimization target is to minimize the product of the norm of the linear operator and the norm of its generalized inverse.

[0029] As a further optimization, in the spectral range [400nm, 800nm], the characteristic quantity Spe and the overlay error OV have a linear relationship that is resistant to noise interference: Spe = k·OV; where k is the slope.

[0030] As a further optimization, the overlay error OV is calculated as follows:

[0031]

[0032] where D is a preset deviation, i.e., the overlay marks with preset deviations +D and -D are used in step S1.

[0033] According to another aspect of the present application, a snapshot overlay error measurement system is provided, comprising a processor configured to execute the above snapshot overlay error measurement method.

[0034] Overall, compared with the prior art, the above technical solutions conceived by the present application mainly have the following technical advantages:

[0035] 1. The present application is based on the linear relationship between the characteristic quantity and the overlay error that is resistant to noise interference, and uses an optimizable linear operator to process the frequency-shifted spectrum, and then uses the calculated coefficient F to construct the characteristic quantity Spe, so as to calculate the overlay error. This method does not need to use traditional Fourier analysis and channel truncation operation, avoids the errors introduced by them, is more accurate, realizes multi-wavelength coupled solution of overlay error, and has stronger robustness to random noise in actual application due to the use of the constructed characteristic quantity for calculation.

[0036] 2. The overlay error is calculated using the optimized characteristic quantity Spe, which couples the overlay error values in the full spectral range, and the construction method of the optimized characteristic quantity is also optimized simultaneously. Compared with the arithmetic mean in the commonly used method, it is more stable and can more effectively adapt to various application scenarios and is not easily affected by noise.

[0037] 3. The frequency-shifted spectrum is processed using an optimizable linear operator, which can significantly reduce the calculation amount compared with the commonly used fast Fourier transform, reduce the data storage and processing overhead, and further speed up the measurement process. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 The flowchart of the snapshot overlay error measurement method of the embodiment of the present application;

[0039] Figure 2aThe linear relationship between the characteristic quantity Spe of the simple structure and the overlay error OV is calculated according to the rigorous coupled wave analysis (RCWA) for the embodiment of the application.

[0040] Figure 2b The linear relationship between the characteristic quantity Spe (F(0) real part) of the simple structure and the overlay error OV is calculated according to the rigorous coupled wave analysis (RCWA) for the embodiment of the application.

[0041] Figure 3 The measurement error absolute value of the overlay error solved by the channel is calculated according to the rigorous coupled wave analysis (RCWA) for the embodiment of the application after adding random errors. 31 +m 13 The measurement error absolute value of the overlay error solved by the channel is calculated according to the rigorous coupled wave analysis (RCWA) for the embodiment of the application after adding random errors.

[0042] Figure 4 The measurement error absolute value of the overlay error solved by the channel is calculated according to the rigorous coupled wave analysis (RCWA) for the embodiment of the application after adding random errors. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and do not limit the application. In addition, the technical features involved in each embodiment of the application described below can be combined with each other as long as they do not conflict with each other.

[0044] The snapshot overlay error measurement method provided by the embodiment of the application, as shown in Figure 1 , comprises the following steps:

[0045] S1, using a snapshot overlay error measurement device to measure a test piece, respectively acquiring two measurement spectra with positive and negative preset deviations, coherently demodulating the measurement spectra, and shifting the frequency of a specific channel to zero frequency channel.

[0046] In this embodiment, the measurement spectrum obtained based on the specific hardware configuration in the patent CN113219792A is used to illustrate the application, and the measurement spectrum obtained in other ways is also applicable in the application.

[0047] The analytical expression of the measurement spectrum is as follows:

[0048]

[0049] Wherein, C is the zero frequency channel, i is the imaginary unit, represents the frequency component of the specific channel; I out is the measurement spectrum, I out ∈RN×1 is a column vector with N elements, I in is the incident light intensity, m ij (i,j∈[1,4] are integers) are the Mueller matrix elements of the sample, is the base frequency component of the channel spectral channel. Extract the Mueller matrix element combination m 31 +m 13 When, the spectral signal is coherently demodulated, and the carrier signal is multiplied on both sides of the equation After arrangement, we get:

[0050]

[0051] In this way, the Mueller matrix element combination m 31 +m 13 is shifted to the zero frequency channel.

[0052] S2, the linear operator with optimized parameters is used to process the spectral data after frequency shift, and the coefficients F + and F － corresponding to the positive and negative preset deviations are obtained.

[0053] Specifically, the linear operator is specifically:

[0054]

[0055] Where, f n is the frequency component of the linear operator, n=0,1,2…; N is the signal length of the spectrum; f max is determined by the following expression:

[0056]

[0057] Where, [λ min ,λ max ] is the spectral range, B max is the maximum birefringence of the wave plate in this spectral range, t is the thickness of the thickest multi-stage wave plate in the measurement device, and c is the number of channel spectra.

[0058] The linear operator is used to process the spectral data after frequency shift, specifically:

[0059]

[0060] Where, I out is the spectral data, and F is the calculated coefficient.

[0061] Further, the linear operator can be optimized using multi-objective genetic algorithm, deep learning and other optimization methods, and the optimization parameter is the frequency component f n, so that the linear operator can better analyze the characteristics of the channel, and has stronger robustness to random noise in the spectral data, thereby improving the accuracy and robustness of the overlay error measurement.

[0062] In this embodiment, the linear operator optimization method used is the slime optimization algorithm, and the optimization objective is to minimize the product of the norm of the linear operator and the norm of its generalized inverse, as shown below:

[0063]

[0064] where M is the linear operator, is the generalized inverse of the linear operator, and after optimization, the parameter values of the linear operator are as follows:

[0065]

[0066] Using this linear operator to extract features from the coherent demodulated data, for the spectra corresponding to the two preset biases, the coefficients F + and F － can be obtained, which are complex numbers; it can also be seen that the operation amount is significantly reduced compared to the traditional FFT.

[0067] S3, according to the coefficients F + and F － , the corresponding characteristic quantities Spe + and Spe - are calculated.

[0068] Specifically, the calculated coefficients F are complex numbers, and the characteristic quantities Spe are constructed using themselves or linear combinations of real and imaginary parts; and in a wider spectral range, there is a linear relationship between the characteristic quantities Spe constructed by the method and the overlay error OV: Spe=k·OV. The characteristic quantities Spe obtained by different combination methods have different linear relationship characteristics, so the tolerance to various types of noise is also different. The specific linear combination method of the characteristic quantity can be optimized using an optimization algorithm to make it more resistant to random noise.

[0069] In this embodiment, the generalized inverse matrix of the linear operator is used to perform an inner product with the calculated coefficients, thereby constructing the characteristic quantity Spe. In the commonly used spectral range [400nm, 800nm], the linear relationship is verified: take the overlay error OV as a sequence with an interval of 0.5nm from -20nm to +20nm, and respectively according to the above process, generate the spectrum by rigorous coupled wave analysis (RCWA) simulation and process to obtain the characteristic quantity Spe, and the linear relationship between the two is as follows: Figure 2aAs shown, it can be seen that the linear relationship is established in the range of overlay error of-20nm to +20nm, which completely covers the common overlay error value. The feature quantity Spe is taken as the real part of the first value F(0) of the calculated coefficient, and the linear relationship is verified again, as shown in Figure 2b As shown. It can be seen that when the construction method of the feature quantity Spe is changed, the characteristics of the linear relationship between the feature quantity Spe and the overlay error OV will also change accordingly. In different application scenarios, the noise characteristics of the measurement process are also different. Thus, the construction method of the feature quantity Spe can be adjusted by various optimization methods so that it can be better applied to a specific scene, enhance the robustness to noise, and improve the accuracy of overlay error measurement.

[0070] S4, based on the anti-interference linear relationship between the feature quantity and the overlay error, the feature quantity Spe + and Spe - are calculated to obtain the overlay error.

[0071] Specifically, two overlay marks with +D and -D deviations (bias amount is ±5nm) are used for measurement and processing to obtain the feature quantities Spe + and Spe - , and then the corresponding overlay error OV is obtained by solving the above linear relationship:

[0072]

[0073] In this embodiment, the theoretical overlay error OV is set as a sequence with an interval of 0.4nm from-2nm to +2nm. The theoretical measurement spectrum is calculated by rigorous coupled wave analysis (RCWA) simulation. In order to reflect the robustness of the method to random error, 1% random noise of spectral intensity is added to the obtained theoretical measurement spectrum:

[0074] I out-error = I out + 0.01×I out ×random

[0075] Where, I out-error is the spectral data after adding noise, and random is a random number obeying Gaussian distribution in [-1, 1].

[0076] Using the above method, the feature quantity Spe is taken as the inner product of the generalized inverse matrix of the linear operator and the calculated coefficient, the absolute value of the measurement error of the overlay error, and the absolute value of the measurement error of the overlay error calculated by the traditional method under the same conditions are obtained, as shown in Figure 3As shown, it can be seen that the absolute value of the measurement error of the method is relatively stable and is basically within 0.1 nm, and the robustness to random noise is significantly improved. After further optimizing the construction mode of the feature quantity Spe, the measurement error of the method will also decrease.

[0077] Since the added random noise has randomness, the error value of a single time may be an extreme case of being very small, so 50 repeated tests are performed in this embodiment, and the absolute value distribution histogram of the measurement error of the overlay error obtained is as shown in Figure 4 As shown, it can be seen that the absolute value of the measurement error of the overlay error is basically controlled within 0.15 nm, and it can be seen that the robustness of the method to random noise is very high.

[0078] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present application, and is not used to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method of snapshot-style overlay error measurement, the method comprising: The method comprises the following steps: S1, measuring the measured piece to obtain two measurement spectra with positive and negative preset deviations respectively, coherently demodulating the measurement spectra, and shifting the specific frequency channel to a zero frequency channel; S2, processing the frequency-shifted spectral data with a linear operator to obtain coefficients F corresponding to the positive and negative preset deviations + and F － ; S3, construct the feature quantity by linear combination of the real part and the imaginary part of the coefficient, then according to the coefficient F + and F － , calculate the corresponding feature quantity Spe + and Spe - , respectively; S4, based on a linear relationship between the characteristic quantity and the overlay error, calculating the overlay error from the characteristic quantity Spe + and Spe - .

2. The snapshot-style overlay error measurement method of claim 1, wherein, The linear operator is specifically: where f n is the frequency component of the linear operator, n = 0, 1, 2,...; N is the signal length of the spectrum, and i is the imaginary unit; f max is determined by the following equation: where [λ min , λ max ] is the spectral range, B max is the maximum birefringence of the waveplate in this spectral range, t is the thickness of the thickest multi-order waveplate in the measurement device, and c is the number of channels of the channel spectrum.

3. The snapshot-style overlay error measurement method of claim 2, wherein, The linear operator is specifically: wherein is the spectral data, are the calculated coefficients.

4. The snapshot-style overlay error measurement method of claim 2, wherein, The linear operator is optimized by using a multi-objective genetic algorithm or a deep learning algorithm, and the optimization parameter is a frequency component f n .

5. The snapshot-style overlay error measurement method of claim 4, wherein, The linear operator is optimized by using the slime mold optimization algorithm, and the optimization target is to minimize the product of the norm of the linear operator and the norm of its generalized inverse.

6. The snapshot overlay error measurement method of claim 1, wherein, In the spectral range [400nm, 800nm], the characteristic quantity Spe and the overlay error OV have a linear relationship against noise interference: Spe=k·OV; wherein k is the slope.

7. The snapshot pinning error measurement method of any one of claims 1-6, wherein, The calculation method of the overlay error OV is as follows: Wherein, D is the preset deviation, that is, the overlay marks with preset deviations +D and-D are used in step S1.

8. A snapshot-style overlay error measurement system, characterized by, The method comprises a processor for executing the snapshot overlay error measurement method according to any one of claims 1-7.

Images