一种基于光谱成像的刻蚀终点检测方法及检测系统

Abstract

本申请提供一种基于光谱成像的刻蚀终点检测方法及检测系统，涉及半导体检测技术领域；刻蚀终点检测方法包括，将特征光照射至数字微镜器件的表面；控制数字微镜器件上的多个微镜按预设的二进制图案进行状态切换，将空间分布信息转换为随时间变化的时域测量序列；构建测量矩阵A，还原二维空间分布光谱图像X；获取特征谱线的强度时间序列及特征谱线在空间分布上的均匀性指数；根据强度时间序列与均匀性指数判定是否到达刻蚀终点。刻蚀终点检测系统包括沿光路依次设置的光学收集系统、数字微镜器件、单通道探测器以及光吸收阱。本发明的检测方法兼具高空间分辨率和高信噪比。

Description

Technical Field

[0001] This invention relates to the field of semiconductor inspection technology, and in particular to an etching endpoint detection method and system based on spectral imaging. Background Technology

[0002] As semiconductor device feature sizes continue to shrink to the nanometer scale, especially with the advent of three-dimensional device structures such as FinFET and GAA (gate all around), the requirements for precision and uniformity in dry etching processes are becoming increasingly stringent. Etching endpoint detection, as a core monitoring method for controlling critical dimensions, sidewall morphology, etching depth, and selectivity, directly impacts device performance, yield, and reliability with its accuracy and real-time performance.

[0003] In existing technologies, optical emission spectroscopy (OES) has become the mainstream endpoint detection method due to its non-invasiveness and ease of integration. This method determines whether the etching process is complete by monitoring the light intensity changes of specific active materials or products at characteristic wavelengths during the etching process. However, as advanced processes develop towards more complex multilayer stacked structures, higher aspect ratios, and larger wafer sizes, traditional OES technology has gradually exposed the following inherent limitations: First, it lacks spatial resolution. Traditional OES systems collect integrated spectral signals from the entire reaction chamber space, and cannot obtain etching status information at different locations on the wafer surface (such as the center and edges). In wafer processing of 300mm and larger sizes, the problem of process uniformity is particularly prominent. The lack of spatially resolved monitoring methods makes it difficult to reflect regional differences in etching rate in real time, leading to local over-etching or under-etching, affecting the uniformity of the entire wafer and product yield. Second, it suffers from low signal-to-noise ratio under weak signal conditions. In some etching processes (such as low aperture ratio etching, passivation layer etching, or the use of low-concentration reactive gases), the characteristic light intensity is weak, and the characteristic spectral signal is easily drowned out by background noise. Furthermore, as process steps advance to the atomic level, the changes in the signals to be detected become extremely subtle. The detection sensitivity and signal-to-noise ratio of traditional OES systems are insufficient to meet the requirements for high-precision endpoint determination, easily leading to endpoint misjudgment or delay. Thirdly, the response speed is limited. Traditional OES data acquisition and processing methods are usually based on the cumulative averaging of time-series spectra, resulting in low temporal resolution. In high-speed etching processes or applications requiring rapid feedback control, this response delay may lead to untimely endpoint detection, preventing true real-time process control and hindering fine-grained control of the process window.

[0004] To address these issues, the industry has attempted various improvement solutions, such as using multi-channel OES, combining mass spectrometry analysis, or laser interferometry. However, these solutions have not fundamentally achieved endpoint detection that integrates high spatial resolution, high signal-to-noise ratio, and fast response.

[0005] Therefore, it is necessary to provide a new etching endpoint detection method and system based on spectral imaging to solve the above-mentioned problems in the prior art. Summary of the Invention

[0006] The technical problem to be solved by this application is how to provide a method and system for detecting etching endpoints based on spectral imaging that has both high spatial resolution and high signal-to-noise ratio.

[0007] To address the aforementioned technical problems, according to embodiments of this application, an etching endpoint detection method based on spectral imaging is provided, comprising the following steps: acquiring characteristic light emitted by plasma within a semiconductor cavity; irradiating the characteristic light onto the surface of a digital micromirror device (DMM); controlling multiple micromirrors on the DMM to switch states according to a preset binary pattern, such that a micromirror in a first flip state reflects the characteristic light to a single-channel detector, and a micromirror in a second flip state reflects the light to a light absorption trap; sequentially switching the shape of the binary pattern according to a preset time sequence, such that the single-channel detector receives light signals from different spatial locations on the wafer surface at different times, converting the spatial distribution information into a time-domain measurement sequence that varies with time; constructing a measurement matrix A based on the time-domain measurement sequence and the binary pattern to reconstruct a two-dimensional spatial distribution spectral image X of the characteristic light; performing feature extraction on the two-dimensional spatial distribution spectral image X to obtain the intensity time series of the characteristic spectral lines and the uniformity index of the characteristic spectral lines in spatial distribution; and determining whether the etching endpoint has been reached based on the intensity time series and the uniformity index.

[0008] According to an embodiment of this application, the step of constructing a measurement matrix A based on a time-domain measurement sequence and a binary pattern to reconstruct a two-dimensional spatial distribution spectral image X of the characteristic light includes: converting the binary pattern into a binary matrix, wherein multiple binary matrices constitute the measurement matrix A; the rows of the binary matrix correspond to the time-domain measurement value Y, and the columns of the binary matrix correspond to the spatial pixel positions; and establishing a minimization problem model to obtain the objective function F(X).

[0009] ;

[0010] in, The L2 norm of the reconstruction error vector is given; TV(X) is the total variation regularization term; λ is the regularization parameter; the two-dimensional spatial distribution spectral image X is reconstructed based on the objective function F(X).

[0011] According to an embodiment of this application, the step of extracting features from the two-dimensional spatial distribution spectral image X to obtain the intensity time series of the feature spectral lines and the uniformity index of the feature spectral lines in spatial distribution includes: repeatedly performing the step of restoring the two-dimensional spatial distribution spectral image X of the feature light at preset time intervals to obtain the two-dimensional spatial distribution spectral image X corresponding to different times t, forming an image time series; obtaining pixel intensity data at the wavelength corresponding to the feature spectral lines based on the image time series; calculating the statistical average intensity of the wafer surface based on the pixel intensity data; and arranging the statistical average intensity at each time t in chronological order to generate the intensity time series.

[0012] According to an embodiment of this application, the step of extracting features from the two-dimensional spatial distribution spectral image X to obtain the intensity time series of the feature spectral lines and the uniformity index of the feature spectral lines in spatial distribution further includes: calculating the average value μ(t) and standard deviation σ(t) of the luminous intensity at each position on the wafer surface based on the pixel intensity data; calculating the coefficient of variation CV(t) based on the average value μ(t) and the standard deviation σ(t), where CV(t) = σ(t) / μ(t); calculating the uniformity index U(t) based on the coefficient of variation CV(t), where U(t) = 1 - CV(t); and arranging the uniformity index U(t) corresponding to each time t in chronological order to form the uniformity index time series.

[0013] According to an embodiment of this application, the step of determining whether the etching endpoint has been reached based on the intensity time series and the uniformity index includes: obtaining the intensity value at the current moment based on the intensity time series; setting an intensity threshold A, and when the intensity value is less than the intensity threshold A, outputting an intensity decrease criterion C1(t) as 1, otherwise outputting an intensity decrease criterion C1(t) as 0; calculating and obtaining a rate of change value based on the intensity time series; setting a rate of change threshold B, and when the absolute value of the rate of change value is less than the rate of change threshold B, outputting a rate of change criterion C1(t) as 0. If C2(t) is 1, otherwise the rate of change criterion C2(t) is 0; obtain the spatial uniformity value based on the uniformity index time series; set a spatial uniformity threshold C, and when the spatial uniformity value is greater than the spatial uniformity threshold C, output the spatial uniformity criterion C3(t) as 1, otherwise output the spatial uniformity criterion C3(t) as 0; when the sum of the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t) is greater than or equal to 2, it is determined that the etching endpoint has been reached.

[0014] According to an embodiment of this application, the step of calculating the rate of change value based on the intensity time series includes: smoothing the intensity time series using a sliding window, wherein the sliding window contains multiple sampling periods; normalizing the smoothed intensity time series to obtain normalized data; and calculating the first difference of the normalized data to obtain the rate of change value.

[0015] According to an embodiment of this application, determining that the etching endpoint has been reached when the sum of the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t) is greater than or equal to 2 includes: triggering continuous confirmation timing when the sum of the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t) is greater than or equal to 2; and determining that the etching endpoint has been reached when multiple consecutive sampling periods satisfy the condition that the sum is greater than or equal to 2.

[0016] According to an embodiment of this application, the step of collecting characteristic light emitted by plasma in a semiconductor cavity includes: collecting broadband light emitted by the plasma; and selecting a narrowband band corresponding to the characteristic spectral line from the broadband light using a filter assembly to form the characteristic light.

[0017] An etching endpoint detection system based on spectral imaging is provided for implementing the etching endpoint detection method described above. The etching endpoint detection system includes an optical collection system, a digital micromirror device, a single-channel detector, and a light absorption trap arranged sequentially along an optical path. The optical collection system is disposed in a semiconductor cavity to collect characteristic light within the semiconductor cavity. The digital micromirror device includes a micromirror array, and each micromirror in the micromirror array has a first flip state or a second flip state. In the first flip state, the micromirror receives and reflects the characteristic light transmitted through the optical collection system, forming a first reflected optical path. In the second flip state, the micromirror receives and reflects the characteristic light transmitted through the optical collection system, forming a second reflected optical path. The single-channel detector is located in the first reflected optical path to receive the characteristic light reflected by the micromirror in the first flip state. The light absorption trap is located in the second reflected optical path to receive the characteristic light reflected by the micromirror in the second flip state.

[0018] According to an embodiment of this application, the optical collection system includes a first lens group, a filter, and a second lens group arranged sequentially along the optical path; the first lens group is positioned opposite the observation window of the semiconductor cavity to collect broadband light emitted by the plasma inside the semiconductor cavity; the filter is used to filter the light so that the broadband light becomes a narrowband corresponding to the characteristic spectral lines; the second lens group is used to collimate the light beam passing through the filter into parallel light.

[0019] By adopting the above technical solution, characteristic light within the semiconductor cavity is collected to obtain broadband light containing characteristic spectral lines of specific elements. After passing through a filter and a second lens group, the characteristic light is irradiated onto the surface of the digital micromirror device. By switching the micromirror pattern, light signals from different spatial locations on the wafer surface are collected sequentially. The spatial distribution information is converted into a time-domain measurement sequence, and a two-dimensional spatial distribution spectral image X is reconstructed from it. The intensity time series of the characteristic spectral lines of CN free radicals, the intensity time series of the characteristic spectral lines of Si atoms, and their spatial uniformity index are extracted from the two-dimensional spatial distribution spectral image X. The etching endpoint is determined by the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t), thereby reducing noise interference caused by plasma fluctuations and improving detection efficiency. Attached Figure Description

[0020] Figure 1 This is a step diagram of an etching endpoint detection method according to an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram showing the distribution of various components in an etching endpoint detection device according to an embodiment of the present invention.

[0022] Figure label:

[0023] 100, Semiconductor chamber; 110, Observation window; 200, Optical collection system; 210, First lens group; 220, Filter; 230, Second lens group; 300, Digital micromirror device; 400, Single-channel detector; 500, Light absorption trap. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.

[0025] The following is combined Figures 1-2 The specific embodiments of the present invention will be further described in detail below.

[0026] Embodiments of the present invention provide an etching endpoint detection method and system based on spectral imaging. The etching endpoint detection system is used to detect whether the wafer has reached the etching endpoint during the etching process. The etching endpoint detection method is applied to the etching endpoint detection system to more accurately determine whether the wafer has reached the etching endpoint during the etching process. During the wafer etching process, the wafer is placed in a semiconductor chamber 100, which has an observation window 110.

[0027] The etching endpoint detection method and system of this application are applied to the determination of the etching endpoint during the etching of polysilicon gates. The material being etched is polysilicon; the mask material is an organic polymer containing C, H, and N.

[0028] The etching endpoint detection system includes an optical collection system 200, a digital micromirror device 300, a single-channel detector 400, and a light absorption trap 500 arranged sequentially along the optical path.

[0029] In some embodiments, an optical collection system 200 is disposed in a semiconductor chamber 100 to collect characteristic light within the semiconductor chamber 100. Specifically, within the semiconductor etching chamber, a radio frequency power supply excites process gas to form plasma. High-energy electrons in the plasma collide with etching reaction products (such as Si atoms) and mask pyrolysis products (such as CN radicals), exciting them to a high-energy state. When these excited-state particles in the high-energy state transition to a low-energy state, they release light with characteristic wavelengths of the element. The wavelengths of the spectral lines of these lights are unique; for example, the Si atom line is 288 nm, and the CN radical emission band is 388 nm. Other elements do not produce these two wavelengths. Therefore, in this application, the corresponding characteristic light is selected for detection to determine the etching endpoint.

[0030] In some embodiments, the digital micromirror device 300 includes a micromirror array, where each micromirror has a first flipped state or a second flipped state. A micromirror in the first flipped state receives and reflects characteristic light from the optical collection system 200, forming a first reflected light path; a micromirror in the second flipped state receives and reflects characteristic light from the optical collection system 200, forming a second reflected light path. Specifically, the first and second flipped states are symmetrically arranged about the normal of the micromirror. In this embodiment, the angle between the micromirror in the first and second flipped states and the normal of the micromirror is selected to be 12°, meaning that the micromirror can flip at ±12° angles to form different light paths. Since each micromirror in the micromirror array can be flipped, during the detection process, multiple micromirrors in the array are selected and flipped according to a preset pattern. By sequentially switching different patterns, the light signals emitted from different spatial positions on the wafer surface can be collected.

[0031] In some specific embodiments, a single-channel detector 400 is located in the first reflective optical path to receive characteristic light reflected by the micromirror in the first flipped state. The single-channel detector 400 is well-known to those skilled in the art and will not be described in detail here. An optical absorption trap 500 is located in the second reflective optical path to receive characteristic light reflected by the micromirror in the second flipped state. The single-channel detector 400 is used to receive the selected spatial location optical signal and convert it into an electrical signal; the optical absorption trap 500 is used to absorb the excluded spatial location optical signal to eliminate stray interference. Specifically, by controlling the micromirror to switch between two states, the single-channel detector sequentially acquires optical signals from different spatial locations, transforming the traditional parallel spatial acquisition of area array detection into serial temporal acquisition, thereby achieving high spatial resolution spectral imaging with a single high-sensitivity detector.

[0032] In some specific embodiments, the optical collection system 200 includes a first lens group 210, a filter 220, and a second lens group 230 arranged sequentially along the optical path direction; the first lens group 210 is positioned directly opposite the observation window 110 of the semiconductor chamber 100 to collect broadband light emitted by the plasma in the semiconductor chamber 100; the filter 220 is used to filter the light so that the broadband light becomes a narrow band corresponding to the characteristic spectral lines; the second lens group 230 is used to collimate the light beam after passing through the filter 220 into parallel light. Specifically, the first lens group 210 is positioned directly opposite the observation window 110 of the semiconductor chamber 100, allowing the characteristic light generated within the semiconductor chamber 100 to pass through the observation window 110 and illuminate the first lens group 210. The first lens group 210 employs an infinite conjugate distance design, meaning the object side is at infinity and the image side is at a finite distance. This means the first lens group 210 collects omnidirectional broadband light emitted through the observation window 110 and completes the initial imaging at its imaging surface, ensuring sufficient light flux enters the etching endpoint detection system. Simultaneously, the receiving surface of the digital micromirror device 300 is located at the imaging surface of the first lens group 210, meaning the light beam passing through the first lens group 210 is a beam that can be focused onto the digital micromirror device 300. Therefore, a second lens group 230 is also required. The second lens group 230 is used to correct the light beam into parallel light, thereby ensuring that the light beam illuminating the digital micromirror device 300 is parallel light. In addition, a filter 220 is disposed between the first lens group 210 and the second lens group 230. The filter 220 can filter out the narrow band corresponding to the target characteristic spectral line from the broadband light, thereby eliminating interference from other wavelengths and improving the signal-to-noise ratio. It is worth noting that since this embodiment is for determining the etching endpoint of polysilicon gate etching, the filter 220 can be selected to filter out the light emitted by the etched material and the mask material.

[0033] This application also discloses an etching endpoint detection method based on spectral imaging, which includes the following steps:

[0034] S1. Collect the characteristic light emitted by the plasma in the semiconductor chamber 100. Specifically, by setting an optical collection system 200 on the semiconductor chamber 100, the first lens group 210 in the optical collection system 200 can receive the characteristic light emitted through the observation window 110 of the semiconductor chamber 100, thereby realizing the collection of the characteristic light emitted by the plasma in the semiconductor chamber 100.

[0035] S2. The characteristic light is irradiated onto the surface of the digital micromirror device 300. Specifically, after the characteristic light is collected, the characteristic light collected by the first lens group 210 is filtered through the filter 220 to retain the required characteristic light. The retained characteristic light is then irradiated onto the second lens group 230, which is used to collimate the light so that the characteristic light becomes parallel light after passing through the second lens group 230. More specifically, in this embodiment, the retained characteristic light includes light with a center wavelength of 388nm from the CN radical emission band and light with a center wavelength of 288nm from the Si atomic lines; that is, the retained characteristic light is a characteristic spectral line. Among them, the light with a center wavelength of 388nm from the CN radical emission band originates from the pyrolysis products of the mask material and is used to monitor the photoresist consumption state; the light with a center wavelength of 288nm from the Si atomic lines originates from the atomic excitation of the etched material and is used to monitor the etching depth of the polysilicon layer.

[0036] S3. Multiple micromirrors on the digital micromirror device 300 are controlled to switch states according to a preset binary pattern. Micromirrors in the first flip state reflect characteristic light to the single-channel detector 400, while micromirrors in the second flip state reflect light to the light absorption trap 500. Specifically, the single-channel detector 400 is a single-channel photomultiplier tube. By controlling the micromirrors to be in the first or second flip state, the spatial parallel acquisition required by traditional area array detectors is converted into temporal serial acquisition. A single-channel photomultiplier tube replaces the traditional two-dimensional area array image sensor, such as a CCD or CMOS, thereby increasing detection sensitivity. More specifically, controlling the micromirror array to flip according to a preset binary pattern (e.g., a Hadamard matrix) selectively reflects characteristic light from specific spatial locations on the wafer surface to the single-channel detector 400, thus converting the optical signal into an electrical signal. Simultaneously, the light absorption trap 500 eliminates stray light from non-gated locations to ensure modulation contrast, thereby reducing the problem of low photon-to-electron conversion efficiency in the deep ultraviolet band caused by the strong absorption of deep ultraviolet light by the silicon substrate of traditional two-dimensional area array image sensors.

[0037] S4. The shape of the binary pattern is switched sequentially according to a preset timing sequence, so that the single-channel detector 400 receives light signals from different spatial positions on the wafer surface at different times, and converts the spatial distribution information into a time-domain measurement sequence that varies with time. Specifically, by switching the shape of the binary pattern sequentially according to a preset time interval, different binary patterns are loaded on the digital micromirror device 300, so that the micromirror array is in a dynamic combination of a first flip state and a second flip state, so that the characteristic light from different spatial positions on the wafer surface is selectively reflected to the single-channel detector 400 at different times. The single-channel detector 400 records the sum of light intensity at each time, forming a time-domain measurement sequence corresponding to the spatial position on the wafer surface, thereby converting the spatial distribution information into a time-domain measurement sequence that varies with time.

[0038] S5. Construct a measurement matrix A based on the time-domain measurement sequence and binary pattern to reconstruct the two-dimensional spatial distribution spectral image X of the characteristic light. Specifically, the two-dimensional spatial distribution spectral image X is obtained by converting the preset binary pattern into the row vector of the measurement matrix A, using the time-domain measurement sequence as the observation value, and establishing function values ​​for reconstruction (this step will be detailed later).

[0039] S6. Feature extraction is performed on the two-dimensional spatial distribution spectral image X to obtain the intensity time series of characteristic spectral lines and the uniformity index of the spatial distribution of characteristic spectral lines. That is, the intensity time series of characteristic spectral lines can monitor the changing trends of etching depth and reaction rate; the uniformity index of the spatial distribution of characteristic spectral lines can quantify the uniformity of wafer surface etching and detect spatial non-uniformity phenomena such as micro-loading effects. Therefore, during the judgment process, it is possible to simultaneously determine whether the etching endpoint has been reached and whether the wafer surface etching is uniform, thereby enabling the judgment of whether the wafer surface has uniformly reached the etching endpoint, thus reducing the risk of misjudgment.

[0040] S7. Determine whether the etching endpoint has been reached based on the intensity time series and uniformity index.

[0041] In some embodiments, a measurement matrix A is constructed based on a temporal measurement sequence and a binary pattern to reconstruct a two-dimensional spatial distribution spectral image X of the characteristic light. This includes converting the binary pattern into a binary matrix, with multiple binary matrices constituting the measurement matrix A. Each row of the binary matrix corresponds to a temporal measurement, and each column corresponds to a spatial pixel position. That is, each preset binary pattern is flattened into a row vector, i.e., a binary matrix, where "1" indicates that the light at that spatial pixel position is allowed to enter the single-channel detector 400, and "0" indicates that it is blocked to the light absorption trap 500. The row vectors of the patterns at multiple different times are stacked vertically to form a K×N dimensional measurement matrix A, where K is the number of measurements and N is the total number of spatial pixels. The temporal measurement value Y is the sum of the light intensities sequentially collected by the single-channel detector 400 at different times, while the spatial pixel position corresponds to multiple resolution units (i.e., each pixel point of the two-dimensional spatial distribution spectral image X) divided on the wafer surface. Therefore, the relationship Y=A·X can be obtained. Since the number of measurements K is much smaller than the total number of spatial pixels N, directly solving for X yields infinitely many solutions. Therefore, it is necessary to establish a minimization problem model to find the image that best matches the physical reality from among these infinitely many solutions, namely, the two-dimensional spatial distribution spectral image X. Specifically, by establishing a minimization problem model, the objective function F(X) is obtained.

[0042] ;

[0043] in, The L2 norm of the reconstructed error vector; TV(X) is the total variation regularization term; λ is the regularization parameter;

[0044] Reconstruct the two-dimensional spatial distribution spectral image X based on the objective function F(X).

[0045] In some specific embodiments, the objective function F(X) comprises two parts, one of which is a data fidelity term. This enables the theoretical value (A·X) of the two-dimensional spatial distribution spectral image X, encoded by the measurement matrix A, to match the time-domain measurement value Y. Another is the regularization term. TV(X) is defined as the sum of the gradient magnitudes between adjacent pixels in the image. By utilizing the spatial continuity of the plasma emission distribution to reduce the influence of noise, and by adjusting the weights of both through the regularization parameter λ, the X that minimizes F(X) can be solved, thus finding the image that best matches the physical reality from infinitely many solutions.

[0046] In some embodiments, feature extraction is performed on a two-dimensional spatial distribution spectral image X to obtain the intensity time series of characteristic spectral lines and the uniformity index of the spatial distribution of characteristic spectral lines. This includes repeatedly executing the step of restoring the two-dimensional spatial distribution spectral image X of the characteristic light at preset time intervals to obtain two-dimensional spatial distribution spectral images X corresponding to different times t, forming an image time series; obtaining pixel intensity data at the wavelengths corresponding to the characteristic spectral lines based on the image time series; calculating the statistical average intensity of the wafer surface based on the pixel intensity data; and arranging the statistical average intensity of each time t in chronological order to generate an intensity time series. Specifically, in this step, based on the image time series, pixel intensity data at the wavelengths corresponding to the light with a center wavelength of 388nm for the CN free radical emission band and the light with a center wavelength of 288nm for the Si atomic lines are obtained, the average value of the pixel intensity data of the wafer surface at each time t is calculated, and these average values ​​are arranged in chronological order to generate an intensity time series. The intensity time series can reflect the intensity changes of the characteristic spectral lines during the etching process.

[0047] In some embodiments, feature extraction is performed on the two-dimensional spatial distribution spectral image X to obtain the intensity time series of the feature spectral lines and the uniformity index of the feature spectral lines in spatial distribution. This also includes calculating the average value μ(t) and standard deviation σ(t) of the luminescence intensity at each position on the wafer surface based on the pixel intensity data; calculating the coefficient of variation CV(t) based on the average value μ(t) and standard deviation σ(t), where CV(t) = σ(t) / μ(t); calculating the uniformity index U(t) based on the coefficient of variation CV(t), where U(t) = 1 - CV(t); and arranging the uniformity index U(t) corresponding to each time t in chronological order to form a uniformity index time series. Specifically, by calculating the average value μ(t) and standard deviation σ(t) of the luminescence intensity at each position on the wafer surface, and further calculating the uniformity index U(t) based on the coefficient of variation CV(t), the spatial uniformity of the etching reaction on the wafer surface during the etching process is quantified. The closer U(t) is to 1, the more uniform the etching is. By arranging U(t) at each time point according to different time points t to form a uniformity index time series, a spatial consistency criterion is provided for the endpoint determination, ensuring that the entire wafer reaches the etching endpoint simultaneously during the etching process.

[0048] In some specific embodiments, feature extraction is performed on the two-dimensional spatial distribution spectral image X to obtain the intensity time series of the characteristic spectral lines corresponding to CN free radicals, the intensity time series of the characteristic spectral lines corresponding to Si atoms, and the uniformity index of the spatial distribution of the characteristic spectral lines corresponding to Si atoms.

[0049] In some embodiments, determining whether the etching endpoint has been reached based on the intensity time series and the uniformity index includes: obtaining the intensity value at the current moment based on the intensity time series; setting an intensity threshold A, and outputting an intensity decrease criterion C1(t) as 1 when the intensity value is less than the intensity threshold A, otherwise outputting an intensity decrease criterion C1(t) as 0; calculating and obtaining the rate of change value based on the intensity time series; setting a rate of change threshold B, and outputting a rate of change criterion C2(t) as 1 when the absolute value of the rate of change value is less than the rate of change threshold B, otherwise outputting a rate of change criterion C2(t) as 0; obtaining the spatial uniformity value based on the uniformity index time series; setting a spatial uniformity threshold C, and outputting a spatial uniformity criterion C3(t) as 1 when the spatial uniformity value is greater than the spatial uniformity threshold C, otherwise outputting a spatial uniformity criterion C3(t) as 0; determining that the etching endpoint has been reached when the sum of the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t) is greater than or equal to 2.

[0050] In some specific embodiments, the intensity value at the current moment is obtained based on the intensity time series of the characteristic spectral lines corresponding to CN radicals to determine whether the mask material has been completely consumed; the rate of change value is calculated based on the intensity time series of the characteristic spectral lines corresponding to Si atoms to determine the etching depth of the polycrystalline silicon layer; and the spatial uniformity value is obtained based on the uniformity index of the spatial distribution of the characteristic spectral lines corresponding to Si atoms to determine the etching uniformity of the polycrystalline silicon layer.

[0051] In some specific embodiments, intensity value, intensity change rate value, and spatial uniformity value are used together to determine whether the etching endpoint has been reached. First, the intensity value at the current moment is obtained based on the intensity time series of the characteristic spectral lines corresponding to the CN radical, and compared with a preset intensity threshold A. Intensity threshold A is 30%-40% of the luminescence intensity benchmark value in the initial stable stage of the etching process. This threshold value is set according to different needs during actual detection; in this embodiment, intensity threshold A is selected as 35% of the luminescence intensity benchmark value in the initial stable stage of the etching process. When the intensity value is less than intensity threshold A, the intensity decrease criterion C1(t) is output as 1; when the intensity value is greater than or equal to intensity threshold A, the intensity decrease criterion C1(t) is output as 0. Simultaneously, the change rate value is calculated based on the intensity time series of the characteristic spectral lines corresponding to Si atoms and compared with a preset change rate threshold B, where the change rate threshold B can be set to 0.005s. -1When the absolute value of the rate of change is less than the rate of change threshold B, the output rate of change criterion C2(t) is 1; otherwise, the output rate of change criterion C2(t) is 0. Finally, the spatial uniformity value is obtained based on the uniformity index of the characteristic spectral lines corresponding to Si atoms in spatial distribution, i.e., the uniformity index U(t) corresponding to time t, and a spatial uniformity threshold C is set, where the spatial uniformity threshold C is 0.8. When the spatial uniformity value is greater than the spatial uniformity threshold C, it indicates that the wafer surface etching has reached spatial uniformity, and the output spatial uniformity criterion C3(t) is 1; otherwise, the output spatial uniformity criterion C3(t) is 0. Finally, when the sum of C1(t), C2(t), and C3(t) is greater than or equal to 2 (i.e., at least two of them are satisfied), the etching endpoint is confirmed to have been reached. This multi-dimensional feature fusion effectively reduces the risk of misjudgment caused by process fluctuations in a single criterion.

[0052] In some embodiments, the rate of change value is calculated based on the intensity time series, including smoothing the intensity time series using a sliding window containing multiple sampling periods; normalizing the smoothed intensity time series to obtain normalized data; and calculating the first difference of the normalized data to obtain the rate of change value. Specifically, the rate of change value is calculated based on the intensity time series of the characteristic spectral lines of the CN radical. First, the intensity time series is smoothed using a sliding window containing multiple sampling periods to suppress inherent plasma fluctuations and measurement noise. Then, the smoothed data is normalized relative to the intensity value in the initial stable stage of the process to eliminate benchmark differences caused by different process conditions. Finally, the rate of change of intensity over time is obtained by calculating the first difference of the normalized data, which facilitates subsequent comparison with the rate of change threshold B. The normalization process and the calculation of the first difference of the normalized data are well known to those skilled in the art and will not be described in detail here.

[0053] In some embodiments, the etching endpoint is determined to have been reached when the sum of the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t) is greater than or equal to 2. This includes triggering continuous confirmation timing when the sum of the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t) is greater than or equal to 2; and determining that the etching endpoint has been reached when the sum is greater than or equal to 2 for multiple consecutive sampling periods. Specifically, when at least two of the three criteria—intensity decrease criterion C1(t), rate of change criterion C2(t), and spatial uniformity criterion C3(t)—are simultaneously met, the system does not immediately confirm the endpoint but triggers a continuous confirmation timing mechanism. Only when the primary triggering condition remains valid for multiple consecutive sampling periods is the etching endpoint finally determined. This filters out misjudgments caused by noise such as instantaneous plasma fluctuations, RF power jitter, or arc discharge, ensuring that the endpoint determination is based on the continuous characteristics of the etching reaction having reached a stable endpoint state in terms of depth, rate, and spatial uniformity, rather than instantaneous interference signals.

[0054] In some embodiments, the acquisition of characteristic light emitted by the plasma within the semiconductor chamber 100 includes collecting broadband light emitted by the plasma; selecting a narrow band corresponding to the characteristic spectral line from the broadband light using a filter assembly to form characteristic light. Notably, in this embodiment, the acquired characteristic light is broadband light generated by plasma radiation within the semiconductor chamber 100, which includes continuous background radiation and characteristic line spectra of atoms / molecules in the range of 200-800 nm. This characteristic light is generated due to energy level transitions of process gas ionization and etching reaction products under radio frequency excitation, rather than being actively induced by an external laser source. Subsequently, a filter 220 is used to filter out the narrow band corresponding to the characteristic spectral line of a specific element from the broadband light, filtering out interference from other wavelengths, thereby forming characteristic light containing only the target characteristic spectral line. This ensures that the subsequent detection process uses the radiation signal of the plasma within the semiconductor chamber 100, rather than light introduced from outside the semiconductor chamber 100.

[0055] The implementation principle of the etching endpoint detection method and system based on spectral imaging in this application is as follows: Characteristic light is collected from a semiconductor chamber 100 to obtain a wide-band light containing characteristic spectral lines of specific elements. After passing through a filter 220 and a second lens group 230, the characteristic light is irradiated onto the surface of a digital micromirror device 300. By switching the micromirror pattern, light signals from different spatial locations on the wafer surface are collected sequentially. The spatial distribution information is converted into a time-domain measurement sequence, and a two-dimensional spatial distribution spectral image X is reconstructed. The intensity time series of the characteristic spectral lines of CN free radicals, the intensity time series of the characteristic spectral lines of Si atoms, and their spatial uniformity index are extracted from the two-dimensional spatial distribution spectral image X. The etching endpoint is determined using the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t), thereby reducing noise interference caused by plasma fluctuations and improving detection efficiency.

[0056] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the invention as set forth in the claims. Furthermore, the invention described herein may have other embodiments and can be implemented or carried out in various ways.

Claims

1. A method for detecting etching endpoints based on spectral imaging, characterized in that, Includes the following steps: Collect the characteristic light emitted by the plasma inside the semiconductor cavity; The characteristic light is irradiated onto the surface of the digital micromirror device; Multiple micromirrors on a digital micromirror device are controlled to switch states according to a preset binary pattern, so that the micromirror in the first flip state reflects the feature light to a single-channel detector, and the micromirror in the second flip state reflects the light to a light absorption trap. The shape of the binary pattern is switched sequentially according to a preset timing sequence, so that the single-channel detector receives light signals from different spatial positions on the wafer surface at different times, and converts the spatial distribution information into a time-domain measurement sequence that varies with time. Based on the time-domain measurement sequence and the binary pattern, a measurement matrix A is constructed to reconstruct the two-dimensional spatial distribution spectral image X of the characteristic light; Feature extraction is performed on the two-dimensional spatial distribution spectral image X to obtain the intensity time series of the feature spectral lines and the uniformity index of the feature spectral lines in spatial distribution; The etching endpoint is determined based on the intensity time series and the uniformity index.

2. The etching endpoint detection method according to claim 1, characterized in that, The measurement matrix A is constructed based on the time-domain measurement sequence and binary pattern to reconstruct the two-dimensional spatial distribution spectral image X of the characteristic light. include, The binary pattern is converted into a binary matrix, and multiple binary matrices constitute the measurement matrix A; The rows of the binary matrix correspond to the time-domain measurement value Y, and the columns of the binary matrix correspond to the spatial pixel positions; Establish a minimization problem model to obtain the objective function F(X); in, The L2 norm of the reconstructed error vector; TV(X) is the total variation regularization term; λ is the regularization parameter; The two-dimensional spatial distribution spectral image X is reconstructed based on the objective function F(X).

3. The etching endpoint detection method according to claim 1, characterized in that, The step of extracting features from the two-dimensional spatial distribution spectral image X to obtain the intensity time series of the feature spectral lines and the uniformity index of the spatial distribution of the feature spectral lines includes, The step of reconstructing the two-dimensional spatial distribution spectral image X of the characteristic light is repeated at preset time intervals to obtain the two-dimensional spatial distribution spectral image X corresponding to different times t, forming an image time series; Based on the image time series, obtain the pixel intensity data at the wavelength corresponding to the feature spectral line; Calculate the statistical average intensity of the wafer surface based on the pixel intensity data; The intensity time series is generated by arranging the statistical average values ​​of the intensity at each time t in chronological order.

4. The etching endpoint detection method according to claim 3, characterized in that, The step of extracting features from the two-dimensional spatial distribution spectral image X to obtain the intensity time series of the feature spectral lines and the uniformity index of the spatial distribution of the feature spectral lines also includes, Calculate the average value μ(t) and standard deviation σ(t) of the luminous intensity at each position on the wafer surface based on the pixel intensity data; The coefficient of variation CV(t) is calculated based on the mean μ(t) and the standard deviation σ(t). CV(t) = σ(t) / μ(t) The uniformity index U(t) is calculated based on the coefficient of variation CV(t). U(t) = 1 - CV(t) The uniformity index U(t) corresponding to each time t is arranged in chronological order to form a uniformity index time series.

5. The etching endpoint detection method according to claim 4, characterized in that, The step of determining whether the etching endpoint has been reached based on the intensity time series and the uniformity index includes: The intensity value at the current moment is obtained based on the intensity time series; Set an intensity threshold A. When the intensity value is less than the intensity threshold A, output the intensity decrease criterion C1(t) as 1; otherwise, output the intensity decrease criterion C1(t) as 0. The rate of change value is calculated based on the intensity time series. Set a change rate threshold B. When the absolute value of the change rate is less than the change rate threshold B, output the change rate criterion C2(t) as 1; otherwise, output the change rate criterion C2(t) as 0. The spatial uniformity value is obtained based on the uniformity index time series; Set a spatial uniformity threshold C. When the spatial uniformity value is greater than the spatial uniformity threshold C, output the spatial uniformity criterion C3(t) as 1; otherwise, output the spatial uniformity criterion C3(t) as 0. When the sum of the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t) is greater than or equal to 2, the etching endpoint is determined to have been reached.

6. The etching endpoint detection method according to claim 5, characterized in that, The step of calculating the rate of change value based on the intensity time series includes: The intensity time series is smoothed using a sliding window, which contains multiple sampling periods; The smoothed intensity time series is normalized to obtain normalized data; The first difference of the normalized data is calculated to obtain the rate of change value.

7. The etching endpoint detection method according to claim 5, characterized in that, The etching endpoint is determined to have been reached when the sum of the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t) is greater than or equal to 2. When the sum of the intensity decrease criterion C1(t), the rate of change criterion C2(t), and the spatial uniformity criterion C3(t) is greater than or equal to 2, continuous confirmation timing is triggered; If the sum is greater than or equal to 2 in multiple consecutive sampling periods, the etching endpoint is determined to have been reached.

8. The etching endpoint detection method according to claim 1, characterized in that, The characteristic light emitted by the plasma within the semiconductor cavity is collected, including: Collect the broadband light emitted by the plasma; The characteristic light is formed by selecting the narrow band corresponding to the characteristic spectral line from the wide-band light using a filter component.

9. An etching endpoint detection system based on spectral imaging, characterized in that, For implementing the etching endpoint detection method according to any one of claims 1-8, the etching endpoint detection system includes an optical collection system, a digital micromirror device, a single-channel detector, and a light absorption trap arranged sequentially along the optical path; The optical collection system is located in the semiconductor chamber to collect characteristic light within the semiconductor chamber; The digital micromirror device includes a micromirror array, each micromirror in the micromirror array having a first flip state or a second flip state; in the first flip state, the micromirror receives and reflects the characteristic light via the optical collection system to form a first reflected light path; in the second flip state, the micromirror receives and reflects the characteristic light via the optical collection system to form a second reflected light path. The single-channel detector is located in the first reflected optical path to receive the characteristic light reflected by the micromirror in the first flipped state; The light absorption trap is located in the second reflected light path to receive the characteristic light reflected by the micromirror in the second flipped state.

10. The etching endpoint detection system according to claim 9, characterized in that, The optical collection system includes a first lens group, a filter, and a second lens group arranged sequentially along the optical path. The first lens group is positioned opposite the observation window of the semiconductor cavity to collect broadband light emitted by the plasma inside the semiconductor cavity. The filter is used to filter the light so that the broadband light becomes a narrowband corresponding to the characteristic spectral lines. The second lens group is used to collimate the light beam passing through the filter into parallel light.

Images