METHOD AND SYSTEM FOR OPTICAL INSPECTION OF A SUBSTRATE

Description

Domaine technique

[0001] The present invention relates to a method for inspecting a substrate, such as a wafer, for example for microelectronics, optics or optoelectronics. It also relates to a system for inspecting a substrate implementing such a method.

[0002] The field of the invention is more particularly, but in a non-limiting manner, that of optical detection by Doppler effect. État de la technique

[0003] Substrates, such as wafers (or « wafers » according to Anglo-Saxon terminology) for electronics, optics or optoelectronics, must be inspected during and after their manufacture in order to be able to detect and identify any defects present on their surfaces or in their volume. These defects, generally very small in size, can be crystal defects, scratches, roughness, etc.

[0004] Inspection is generally not only aimed at detecting the presence or absence of a defect or particle, but also at classifying these defects and providing qualitative information and / or statistical data on the defects, such as their location, size and / or nature, for example. This information may be representative of the quality of the substrate manufacturing process or of a production step in which the substrate is used.

[0005] Inspection systems have been developed to detect increasingly smaller defects and provide the above information.

[0006] Different substrate inspection techniques are known.

[0007] We know of techniques based on measuring the intensity of diffusion signals in the dark field, or dark field. (« dark field » according to English terminology), as described, for example, in document US 6,590,645 B1. Only the light scattered by a defect or a particle on the wafer is collected on a detector; the specular reflection is not acquired by the detector, which makes it possible to improve the detection contrast. A variation in the scattered intensity reveals the presence of a defect on the surface of the wafer.

[0008] Another technique is laser Doppler velocimetry (LDV, acronym for the Anglo-Saxon term " Laser Doppler Velocimetry » ) .

[0009] This technique is described, for example, in document WO 2009 / 112704 and is based on interferometric detection. The operating principle is to create a volume of interference fringes at the intersection of two light beams from the same light source, and to pass this measurement volume through the wafer to be inspected. The light beams and the detection elements are configured in such a way as to perform dark field detection. When a defect or a particle crosses the interference fringes, their presence is reflected on the detector by a measurement signal, for example an electrical signal, constituting a Doppler burst.A Doppler burst is a signal with a double frequency component: a low frequency component, forming the envelope of the signal and corresponding to the average light intensity diffused by the defect, and a high frequency component corresponding to the Doppler frequency containing information on the speed of movement of the defect and linked to the distance between the interference fringes.

[0010] A similar interferometric technique is described in WO 2016 / 050735.

[0011] Known techniques are based on intensity measurements. The quality of defect detection is then directly linked to the dynamics and signal-to-noise ratio of the detection system. However, the nature and size of defects present on a substrate vary greatly. Various difficulties can then arise.

[0012] Thus, certain types of defects can cause saturation of the detection signal and therefore be unquantifiable. In addition, spurious signals that do not correspond to the actual presence of a defect but to artifacts can be captured. Conversely, small defects or parts of defect structures may not be detected.

[0013] To overcome the numerous sources of parasitic measurements in order to optimize the signal-to-noise ratio and unambiguously extract signals enabling information to be obtained on the nature and characteristics of defects, the implementation of measurement systems based on dark field techniques requires significant resources, such as the multiplication of light sources and detectors or significant data processing. Exposé de l'invention

[0014] An aim of the present invention is to propose a method and a system for inspecting a substrate which make it possible to overcome these drawbacks.

[0015] An aim of the present invention is to propose a method and a system for inspecting a substrate allowing greater reliability of defect detection.

[0016] Another aim of the present invention is to propose a method and a system for inspecting a substrate making it possible to detect, discriminate and classify defects present on the substrate without complex or expensive installation of signal measurement and processing means.

[0017] Another object of the present invention is to provide a method and system for inspecting a substrate that allows defects to be mapped with current inspection devices with very little, or even without, modification of their hardware architecture.

[0018] These objectives are achieved at least in part with a method for inspecting a substrate according to claim 1.

[0019] According to the invention, the measurement volume is created at the intersection of two (or more) optical beams coming from the same source and which cross at a non-zero angle. These beams thus generate interference fringes in this measurement volume.

[0020] The measurement signal representative of the light scattered by the substrate may be, for example, a signal acquired by a photodetector (such as an avalanche photodiode). This signal corresponds to the light scattered by the substrate as it passes through the measurement volume in a volume or one or more solid angles. Preferably, this scattered light excludes specular reflection. Thus, in the presence of a purely reflective substrate that does not generate scattering, the measurement signal should be zero.

[0021] When a scattering source, such as a substrate defect (e.g. a hole or a particle) passes through the measurement volume, it generates scattered light whose intensity is modulated over time, due to the movement of the wafer, by the interference of the light beams. This intensity is thus modulated at a modulation frequency which depends on the arrangement of the light beams and the speed of movement of the substrate.

[0022] The measurement signal corresponding to the passage of a scattering source or a defect with the modulation frequency can be called a Doppler burst. The presence of the modulation frequency in the measurement signal constitutes a frequency signature, which makes it possible to discriminate between defects that pass through the measurement volume and defects that generate scattered light outside this measurement volume. Indeed, in the latter case, the defects do not pass through the interference fringes and therefore do not generate a modulation frequency.

[0023] Advantageously, the method according to the invention thus makes it possible to identify and discriminate defects at least by a frequency signature, which can only be emitted by defects crossing the measurement volume.

[0024] To this end, the method according to the invention implements a step of predicting or calculating an expected signal representative of the passage of a defect in the measurement volume. This calculation may relate to the expected modulation frequency. It may also relate to other expected parameters such as a signal shape or an envelope, which are translated in the spectral domain by a modulation spectrum around the expected modulation frequency.

[0025] The method according to the invention further comprises a step of determining characteristic values ​​representative of a frequency content of the measurement signal in a neighborhood around the expected modulation frequency. These characteristic values ​​are obtained by carrying out a comparison of the measurement signal with characteristics of expected signals representative of passages of defects in the measurement volume, essentially linked to the expected modulation frequency. A validated signal is thus formed from these characteristic values ​​which is much more representative than the raw measurement signal of the presence of defects in the measurement volume.

[0026] Thus, the detection of a defect is decorrelated from the light intensity backscattered by this defect. Even very small or very weakly scattering defects can be detected, in particular thanks to an improvement in the signal-to-noise ratio.

[0027] Furthermore, the method according to the invention makes it possible to distinguish a defect or a particle which generates a Doppler burst from a continuous scattering background. Indeed, the continuous background is due to a superposition of a plurality of scatterers which generate an intensity noise without a notable frequency component, since it corresponds to an incoherent superposition, or of random phases, of bursts.

[0028] The method also makes it possible to distinguish a signal coming from the measurement volume from a signal generated outside this volume, because the latter does not have a notable frequency component at the expected modulation frequency. This is particularly the case for signals coming from the surface opposite the inspected surface for transparent substrates.

[0029] The invention thus makes it possible to generate, for analysis purposes, a validated signal whose characteristics for the detection of defects (specificity, discrimination capacity, signal to noise ratio) are significantly improved compared to the intensity measurement signal.

[0030] The steps of calculating the expected parameters and determining characteristic values ​​to constitute the validated signal can be carried out during, before and / or after the acquisition of the measurement signal representative of the scattered light. These steps can thus be carried out in real time, and / or with saved measurement signals. They can be carried out by digital and / or analog data processing devices.

[0031] All the steps of the method according to the invention can be carried out for one or a plurality of measurement signals measured during a movement of the substrate carried out so as to travel with the measurement volume over all or part of a surface of the substrate. This surface can be in particular one of the two external surfaces of the substrate or a section or an interface in the volume of the substrate.

[0032] The method according to the invention comprises the following steps: determining a threshold value of the characteristic values; and comparing the validated signal with said threshold value.

[0033] Such a comparison of the validated signal with a threshold value makes it possible to select all or only certain types of defects, and / or to reject residual background noise.

[0034] According to one embodiment, the step of determining a threshold value can be carried out from a validated signal obtained with a test substrate of known characteristics.

[0035] In particular, this step of determining a threshold value can be carried out from a statistical study on a clean substrate free from scattering sources, and / or on a substrate on which scattering spheres of known characteristics have been deposited. The threshold value can be determined so as to be able to detect as many defects as possible during the inspection of a substrate, while avoiding false detections.

[0036] According to an advantageous embodiment, the step of determining the characteristic values ​​can comprise the following steps: modeling of the expected signal according to a model function, to produce a modeled signal; comparison of the modeled signal to the measurement signal, including the calculation of a distance in the sense of a Euclidean norm between the modeled signal and the measurement signal.

[0037] The modeling step can be carried out, for example, by modeling a Doppler burst (the expected signal) by a sinusoidal function modulated in amplitude by a Gaussian function, possibly with a continuous component. The characteristic values ​​can then be obtained by performing, for example, a likelihood calculation, as explained later, or a correlation calculation between the Doppler burst thus modeled and the acquired measurement signal. They are then representative of a measure of resemblance between the acquired measurement signal and an expected Doppler burst, or of a probability that the acquired measurement signal contains a Doppler burst.

[0038] This embodiment has the advantage of taking into account, in addition to the local noise of the measurement signal and the expected modulation frequency, also the shape of the envelope of the expected Doppler burst.

[0039] According to another embodiment, the step of determining the characteristic values ​​may comprise bandpass filtering of the measurement signal according to a bandwidth adapted to transmit only the frequency content of the measurement signal in a neighborhood around the expected modulation frequency.

[0040] In this case, the characteristic value can take into account, for example, the modulation amplitude of the filtered measuring signal, or the ratio between the modulation amplitude of the filtered measuring signal and a continuous value of the measuring signal before filtering.

[0041] According to another embodiment, the step of determining the characteristic values ​​may comprise the following steps: calculating a local Fourier transform of the measurement signal to obtain a local power spectral density; determining a characteristic value from the power spectral density at, or in a neighborhood including, the expected modulation frequency.

[0042] A local Fourier transform is a Fourier transform calculated on a portion or neighborhood of the measurement signal, for example in a sliding window.

[0043] Of course, the step of determining characteristic values ​​can also implement other signal processing methods, such as wavelet transforms, or time-frequency transforms.

[0044] Advantageously, the method according to the invention can further comprise a step of constructing an image of the substrate using the validated signal.

[0045] Indeed, by exploiting a plurality of characteristic values ​​determined for the entire substrate, it is possible to construct an image, or a map, of the substrate representing all of the defects retained from the validated signal.

[0046] According to a non-limiting exemplary embodiment, the image construction step may comprise a step of assigning intensity values ​​to characteristic values ​​corresponding to (or located at) positions on or in the substrate, which intensity values ​​correspond to pixels of the image.

[0047] Thus, the method according to the invention makes it possible to visualize the presence or absence of a defect on the substrate, the intensity values ​​attributed to the pixels being independent or distinct from a signal of amplitude of light backscattered by the defects. In other words, the method according to the invention makes it possible to obtain greater reliability in detecting the presence of defects, even for very small defects.

[0048] Depending on the implementation modes, the method according to the invention may comprise a step of determining a representative parameter deduced from the measurement signal and / or the validated signal.

[0049] This representative parameter can be deduced from the validated signal, and / or from the measurement signal for the validated Doppler bursts based on the validated signal. This representative parameter can be, for example, a likelihood ratio from the validated signal, or the visibility, the modulation amplitude of the burst, the maximum amplitude of the burst or an amplitude average from the measurement signal or from the validated signal if this information is present there.

[0050] According to embodiments, the method according to the invention may comprise an image construction step comprising an allocation of intensity values ​​to representative parameters corresponding to (or located at) positions on or in the substrate, which intensity values ​​correspond to pixels of the image.

[0051] Advantageously, the method according to the invention can further comprise a step of classifying the pixels of the image into logical objects to reconstruct the validated defects.

[0052] Indeed, the constructed image indicating only by pixel the presence or absence of defects, the stage of classifying the pixels into logical (binary) objects makes it possible to reconstruct the defects and thus to represent and classify them according to their size, their shape, their location, etc. The binary logical objects thus obtained can be processed and analyzed according to known statistical analysis or pattern recognition methods.

[0053] Logical object maps can thus be formed from the characteristic values. Several maps constructed from different types of characteristic values ​​representative of the same detected signal and obtained by applying different processing can be constructed and combined with each other to further improve fault discrimination and classification. Also, for example, several fault maps corresponding to different threshold values ​​of validated signals can be obtained for measurement validation and improvement of fault classification.

[0054] In addition, logical object maps formed from the characteristic values ​​can be used to perform statistical analyses and study the defect characteristics following a certain substrate manufacturing process. The corresponding steps of the manufacturing process can then be adapted based on these analyses.

[0055] Conventionally obtained scattering intensity maps can be combined with maps established using the method of the invention for analyzing defects in a substrate.

[0056] The method according to the invention also makes it possible to create and compare images corresponding to characteristic or visibility values ​​validated by different threshold levels. The defects reconstructed into logical objects can then appear or disappear entirely or partially since the images obtained can present different levels of contrast or dynamics. The chosen validation thresholds can then be confirmed or modified. Thus, the analysis and classification of defects can be improved.

[0057] Furthermore, the method according to the invention makes it possible to compare images, or maps, corresponding to characteristic values ​​and / or to intensity images obtained directly by the diffusion of light by the substrate. The intensity images depend on the illumination conditions of the substrate, such as the light intensity, the polarization of the light beams, their angle of incidence, and the characteristics of the defects. The combination of characteristic value maps validated by a detection threshold for the presence of the expected modulation frequency with directly measured intensity maps makes it possible, for example, to collect information such as the roughness of the substrate, and to inspect substrates transparent to the illumination wavelengths by distinguishing defects present on the surface or in the volume of the substrate and / or on both faces of the substrate.Comparing these maps also makes it possible to distinguish the nature of the defects, their structures or their sizes.

[0058] Advantageously, the method according to the invention can be used for the inspection of opaque or transparent substrates at the inspection wavelength. In the case of an opaque substrate, only the surface facing the measurement volume can be inspected.

[0059] In particular, the method according to the invention can be implemented for the inspection of a transparent or opaque wafer for electronics, optics or optoelectronics, and in particular for the representation of all the defects present on a surface of such a wafer.

[0060] According to another aspect of the invention, there is provided a system for inspecting a substrate according to claim 13.

[0061] The processing module can be integrated into the measurement module, or be external to the measurement module and connected to the measurement module wired or wirelessly.

[0062] According to embodiments, the system of the invention may comprise a device for rotating the substrate around an axis of rotation perpendicular to a main surface of said substrate, and a device for translating the interferometric device arranged to move the measurement volume in a radial direction relative to the axis of rotation.

[0063] It should be noted that in this case, if the substrate is rotated at constant speed, the expected modulation frequency depends on the radial position of the measurement volume relative to the axis of rotation. Description des figures et modes de réalisation

[0064] Other advantages and characteristics of the invention will appear on reading the detailed description of implementations and embodiments which are in no way limiting, and the appended figures in which: there Figure 1 is a schematic representation of a non-limiting exemplary embodiment of a system according to the invention; Figure 2 is a schematic representation of a non-limiting embodiment of a measuring method according to the invention; Figure 3 shows an example of a Doppler burst present in a measurement signal acquired during the inspection of a substrate; Figures 4a-4d represent an example of attribution of characteristic value according to a step of the method according to the invention; the Figures 5a-5d represent another example of attribution of characteristic value according to a step of the method according to the invention; Figure 6a-6d represent another example of attribution of characteristic value according to a step of the method according to the invention; the Figures 7a et 7b show examples of histograms for noise threshold determination; Figures 8a et 8b show non-limiting examples of inspection of a substrate with the method of the present invention; Figure 9 is a schematic representation of another non-limiting embodiment of the measuring method according to the invention; and the Figure 10 shows a non-limiting example of inspection of a substrate with the method of the present invention.

[0065] It is understood that the embodiments which will be described below are in no way limiting. In particular, it is possible to imagine variants of the invention comprising only a selection of characteristics described below isolated from the other characteristics described, if this selection of characteristics is sufficient to confer a technical advantage or to differentiate the invention compared to the state of the prior art. This selection includes at least one preferably functional characteristic without structural details, or with only a part of the structural details if this part alone is sufficient to confer a technical advantage or to differentiate the invention compared to the state of the prior art.

[0066] In particular, all the variants and embodiments described can be combined with each other if there is no technical obstacle to this combination.

[0067] In the figures, elements common to several figures retain the same reference.

[0068] In the embodiments presented, a substrate to be inspected according to the method of the present invention may be any wafer intended for use in the field of electronics, optics or optoelectronics. The wafer may have a circular circumference or any other shape consistent with the desired application of the wafer.

[0069] There Figure 1 schematically illustrates an example of an inspection system according to the invention.

[0070] The system 1 comprises a light source 20 and an interferometric device 30 coupled to the light source arranged opposite a surface S of the substrate 2 to be inspected.

[0071] The interferometric device 30 comprises a device intended to separate the light emitted by the light source 20 into two beams 4 and 5.

[0072] This interferometric device 30 can be produced with massive optical elements, for example a splitter cube, deflecting mirrors and focusing optics.

[0073] It can also be realized with a fiber optic coupler that directs the light injected into one optical fiber to two other optical fibers and focusing optics forming the two beams 4 and 5.

[0074] It can also be produced with a planar optical light guide whose input is coupled to the light source 20 and comprising two branches to divide the beam coming from the light source 20 into two incident beams 4 and 5.

[0075] In all cases, the incident beams 4, 5 are oriented relative to each other so as to intersect and form, at their intersection, a measurement volume comprising a plurality of parallel interference fringes, or at least oriented essentially in a direction determined by the directions of incidence of the beams 4, 5. The beams 4, 5 can be shaped by focusing optics. They can, in particular, be focused so that their intersection forming the measurement volume corresponds to the focusing point or « waist » (in English) of these beams, so as to minimize the size of the measurement volume. Different polarization states of light can be used.

[0076] The light source 20 is a laser having a coherence length suitable for creating the volume of interference fringes.

[0077] In the embodiments presented, the substrate 2 is rotated around an axis of rotation or symmetry X perpendicular to the surface S. In practice, the substrate 2 is placed and held on an appropriate support, of the type « chuck » (in English), not shown. The support is mobile and rotated, for example by a motor. It includes a sensor, for example of the angular encoder type, making it possible to know the angular position of the substrate as a function of time.

[0078] A detection module 50 of the system 1 makes it possible to collect and receive the light scattered by the substrate 2 and to generate a measurement signal from the collected light. The detection module 50 notably comprises a photodetector, for example of the photodiode, avalanche photodiode, CCD or CMOS type. The measurement signal reproduces the variation in the light intensity, corresponding to the interference fringes, of the collected light as a function of time.

[0079] System 1 as shown in the Figure 1 further comprises an optical fiber 40 arranged between the surface S of the substrate 2 and the detection module 50, so as to collect and guide the scattered light towards the detection module 50. Elements, not shown in the figure 1 , can be placed between the substrate 2 and the optical fiber 40 in order to increase the quantity of light collected by the fiber. These elements can be, in particular, lenses or concave mirrors. They can in particular be mirrors reproducing at least partially an elliptical shape with respectively the measurement volume and the end of the optical fiber 40 (or a collimation optic placed opposite this optical fiber 40) positioned at the foci of this ellipse. Thus, the optical fiber 40 can collect the light scattered on large portions of solid angles.

[0080] When a defect 3 or a particle passes through the measurement volume with the interference fringes, the measurement signal obtained by the detection element 50 comprises a Doppler burst with a frequency component at a modulation frequency modulated by an envelope.

[0081] An example of such a modulated burst is shown in the Figure 3 . The signal 11 can be expressed for example in the form of an electrical voltage (in Volts) at the output of the detection module 50 as a function of time, or an equivalent digitized signal.

[0082] Such a Doppler burst can be represented in the form of a signal 11 having a double frequency component: a low frequency component, forming the envelope of the signal, corresponding to the average light intensity diffused by the defect, and a high frequency component, at a modulation frequency, containing the information on the speed of the defect. The high frequency fD , also called Doppler frequency, is related to the velocity v of displacement of the defect in the direction perpendicular to the fringes and to the distance Δ between the interference fringes by the following relation: f D = v / Δ. The Figure 3 also illustrates the following parameters of a puff: the maxima of the envelopes I max and I min and the offset P of the signal.

[0083] The substrate 2 is installed on a rotating support whose angular speed is known and measured. In addition to the rotational movement, a radial movement system (not shown) which moves the substrate 2 and its support, or the optical assembly with the interferometric device 30 and the optical fiber 40 of the detection module 50, allows the volume of interference fringes to travel across the entire surface of the substrate, for example following a spiral movement. If the rotational speed of the substrate 2 is kept constant, which is the easiest to achieve, the speed with which a defect 3 crosses the measurement volume with the interference fringes depends on its radial position relative to the axis of symmetry X. In this case, according to the relationship presented above, the Doppler frequency f D decreases when the defects are closer to the center of the substrate 2. Of course, it is also possible to use a variable rotation speed. For example, a continuously decreasing rotation speed as the measurement volume moves away from the X axis of symmetry makes it possible to obtain a Doppler frequency f D constant.

[0084] The measurement volume, or volume of interference fringes, extends in a region of the substrate which encompasses the surface S. Preferably, the incident beams 4, 5 are arranged so that the dimension perpendicular to the surface S of the measurement volume is less than the thickness of the substrate 2. For example, for a substrate of approximately 1 mm thickness, the dimension of the measurement volume may be approximately 50 µm.

[0085] As explained previously, a defect present on the surface S in the measurement volume generates a measurement signal (the Doppler burst) with a different frequency content from a measurement signal generated by a defect present in the volume of the substrate or on the opposite surface, and which does not cross the measurement volume. Indeed, in this case the measurement signal does not include a modulation frequency.

[0086] The position of the measurement volume with the interference fringes relative to the substrate 2 is known as a function of time. Thus, it is possible to calculate the expected modulation frequency, corresponding to the Doppler frequency f D corresponding to the presence of a defect.

[0087] Finally, still in reference to the Figure 1 , the inspection system 1 according to the invention comprises a processing device 60, such as a microprocessor or a computer for example, connected to the interferometric device 30 and / or to the detection module 50 and configured to implement all the steps of a method for inspecting a substrate, such as for example steps 106-124 of the method 100 or steps 206-224 of the method 200 described below.

[0088] There Figure 2 is a schematic representation of a non-limiting embodiment of a method for inspecting a substrate according to the invention.

[0089] Process 100, shown in the Figure 2 , comprises a step 102 of acquiring a measurement signal corresponding to the light scattered by the region of the substrate passing through a measurement volume. The measurement signal reflects the intensity scattered by the substrate as a function of time. As detailed above, the measurement volume is created from two interfering light beams. The substrate rotates around its axis of symmetry (X) at a known angular velocity.

[0090] The method 100 further comprises a step 104 of determining an expected modulation frequency corresponding to a relative position of the measurement volume relative to the substrate. This expected modulation frequency is the Doppler frequency which is a function of the speed with which a defect crosses the measurement volume and the position on the substrate at which the defect is located. To carry out this step 104 of the method, it is therefore necessary to determine the relative position of the measurement volume relative to the substrate in step 106 and to deduce therefrom the speed of a defect crossing the measurement volume at this relative position. In a step 108, the modulation frequency corresponding to the relative position can then be calculated.

[0091] According to an embodiment of the method 100, the step 104 of determining the modulation frequency can be carried out in parallel or during the step 102 of acquiring the measurement signal, as indicated by the arrow 150 on the Figure 2 .

[0092] According to another embodiment, step 104 of determining the expected modulation frequency can be carried out after step 102 of acquiring the measurement signal is partially or completely completed, i.e. using saved measurement signals.

[0093] The method 100 according to the embodiment shown further comprises a step 114 of calculating a validated signal representative of the presence of defects 3, which comprises a determination of characteristic values ​​representative of a frequency content of the measurement signal in a neighborhood around the expected modulation frequency.

[0094] Once step 114 of calculating a validated signal has been carried out, the method 100 according to the embodiment shown in the Figure 1 continuous with step 116 in which a threshold value of the characteristic values ​​is determined, which threshold value is also called threshold. Of course, this step 116 of determining the threshold value may not be carried out at each iteration. Preferably, it is carried out in a prior phase, for example on test substrates, or during test measurements.

[0095] In a step 118, the validated signal is compared to the threshold value.

[0096] For example, we can consider that the validated signal is actually representative of a real fault when it is greater than the threshold value. We can then, for example, set all values ​​lower than the threshold value to zero or to a specific value to keep in the validated signal only the characteristic values ​​actually representative of a fault.

[0097] Examples of determining characteristic values ​​and determining the threshold value are given below.

[0098] In a first approach, an expected signal, corresponding to an ideal Doppler burst, is modeled, for example, as a sinusoidal function at an expected modulation frequency modulated in amplitude by a Gaussian function. The modeled signal also contains a continuous component, to which noise is added. The parameters of this ideal signal can be estimated, as illustrated in the Figure 3 . It is then possible to calculate a likelihood ratio between the ideal Doppler burst and a measured signal assumed to contain the expected modulation frequency. The more the measured signal resembles the ideal burst, the greater the likelihood ratio.

[0099] The likelihood ratio R can, for example, be defined as follows: R = exp − f mes . − f est . 2 σ ω 2 exp − f mes . 2 σ ω 2

[0100] In this equation, f mes . is the measured signal, f est . is the estimated ideal signal and σ ω is the representative standard deviation of residual noise between the estimated signal and the measured signal. ∥ ∥ represents the Euclidean norm.

[0101] The ratio can be simplified and expressed in logarithmic form: r = ln R = f mes . 2 − f mes . − f est . 2 σ ω 2

[0102] It appears that the report r increases when the noise decreases and / or when the estimated signal approaches the measured signal ( f est . = f mes. ) .

[0103] The likelihood ratio can be used as a characteristic value. A minimum likelihood ratio can be chosen to constitute the threshold value.

[0104] THE Figures 4a et 4b represent measurement signals 400, 402 acquired, for example, in step 102 of the method shown in Figure 2 , expressed as an electrical voltage (mV) as a function of time (ms). The 400 signal of the Figure 4a corresponds to a particle that crosses the measurement volume, and contains the expected modulation frequency. The signal 402 of the Figure 4b corresponds to a particle that generates diffusion but does not cross the measurement volume (for example a particle present on another face of the substrate 2). It therefore does not contain the expected modulation frequency. This is verified by calculating the likelihood ratio as detailed above.

[0105] The results of the calculation are shown on the Figures 4c et 4d respectively. When the measurement signal contains a Doppler burst, the likelihood ratio 404 increases and exceeds a threshold value 408 ( Figure 4c ). Otherwise, the likelihood ratio 406 is only noise and does not exceed the threshold value 408 ( Figure 4d ). We can therefore conclude that the measurement signal 402 of the Figure 4b is not due to an element passing through the measurement volume.

[0106] According to a second approach to determining characteristic values ​​of a measurement signal, a bandpass filtering of the measurement signal is carried out. The bandwidth of the filter is adapted to transmit only the part of the measurement signal around the expected modulation frequency. Other frequencies possibly present in the signal are therefore attenuated or rejected. The threshold value can then be chosen to retain only the validated signals with a modulation amplitude at the expected modulation frequency greater than a minimum value and to reject the other signals.

[0107] THE Figures 5a et 5b illustrate, respectively, measurement signals 500, 502, the first containing a Doppler burst, the second not containing the expected modulation frequency. The corresponding validated signals from the bandpass filtering 504, 506 are shown in the Figures 5c et 5d , respectively. The representative characteristic value may be, for example, the amplitude value of the modulation of the filtered signal 504, 506 or a visibility value representing the ratio of the modulation amplitude of the filtered signal to the continuous value of the unfiltered measurement signal. The threshold value 508 may be determined as a minimum value of the characteristic value accepted as being representative of an actual defect, as illustrated in the Figures 5c et 5d .

[0108] According to a third approach to determining characteristic values ​​of a measurement signal, spectral detection can be performed by locally performing a Fourier transform of the measurement signal, for example in a sliding time window. In the spectrum obtained, the peak corresponding to the expected modulation frequency can be chosen as the characteristic value representative of the real presence of a Doppler burst. As before, a threshold can be determined so that the amplitude value of this peak must be greater than a threshold value to be representative of the presence of a real defect. To take into account the amplitude modulation of the envelope, it is also possible to choose as a characteristic value a measurement of the energy or the power in a spectral band encompassing the expected modulation frequency.

[0109] An example of this approach is illustrated in the Figures 6a-6d . Measured signals 600, 602 ( Figures 6a et 6b ) as well as their respective spectra 604, 606 ( Figures 6c et 6d , in dB) are shown. In the example shown, the spectrum 604 of the signal 600 containing a Doppler burst has a peak 610 corresponding to the expected modulation frequency. The threshold value 608 is chosen to retain only the parts of the validated signal which contain a spectral component of amplitude greater than the threshold at the expected modulation frequency. In the spectrum 606 of the signal 602 not containing the Doppler burst, no peak apart from that of the continuous component (0 MHz) can be distinguished.

[0110] Other approaches to characteristic value assignment can also be used.

[0111] Subsequently, the choice of the threshold value is considered in more detail.

[0112] Generally, the threshold value depends on both the probability a priori whether a defect or particle is present on or in the substrate and the cost of false detection and non-detection, for the number of which a compromise can be found. The threshold value also depends on the frequency of the Doppler bursts, the signal-to-noise ratio of the detected Doppler bursts and the type of noise considered. The aim is to detect weak signals without the noise in the measurement signals leading to too many false detections. In addition, the threshold value can be fixed or adaptive depending on the preferred selection criterion. In order to determine a threshold value taking into account the constraints of the inspection system, a statistical study on substrates or calibration areas can be carried out. This study consists of observing a histogram of the characteristic values ​​obtained during an inspection of clean wafers (i.e. free from scattering sources) or with deposits of spheres with well-defined characteristics.The threshold value can be determined so that as many faults as possible can be detected, while avoiding false alarms as much as possible.

[0113] THE Figures 7a et 7b illustrate histograms of characteristic values ​​obtained by calculating likelihood ratios for two substrates.

[0114] For the Figure 7a , the substrate is assumed to be clean, or at least to contain very few particles, as substrates are never perfectly clean. The histogram illustrates the distribution of the calculated characteristic values, which essentially corresponds to the noise level of the inspection system.

[0115] For the Figure 7b , the substrate is polluted with particles. The histogram shows a very strong decrease in the occurrence of characteristic values ​​in the low levels corresponding to noise, and a flat area covering the entire histogram and corresponding to defect detections. We thus visualize, in addition to the false detections, the real detections corresponding to Doppler bursts caused by real defects.

[0116] Knowing the noise decay trend using the histogram of the Figure 7a corresponding to the clean substrate, it is possible to determine an optimal threshold value to both detect as many Doppler bursts as possible and avoid a maximum number of false detections. In the example illustrated, the threshold value can be established with a likelihood ratio between 20 and 50, for example. This threshold can be established, for example, by searching for the intersection with the abscissa axis of the histogram of a line representing the decrease in occurrences for low characteristic values ​​corresponding to the noise.

[0117] It can be observed that the threshold established with the clean substrate is applicable to the measurements carried out on the polluted substrate as illustrated in figure 7b .

[0118] The threshold value may also vary depending on the measurement volume parameters, substrate rotation speed, and defect density on the inspected substrates. When the signal acquisition system or method is changed significantly, it may be necessary to check and readjust the threshold value.

[0119] Of course, the detection threshold can be established in any other way, for example by analyzing false detections, and / or absences of detection on known samples.

[0120] The Doppler bursts whose respective characteristic values ​​are greater than the chosen threshold value are considered as bursts validated during step 116 of the method 100, i.e. indicating real presences of defects on the substrate.

[0121] Still referring to the Figure 2 , during a step 120 of the method 100 according to the invention, the characteristic values ​​associated with the validated puffs and therefore representative of real defects are associated with the positions on the substrate at which they were detected to form an image. To do this, each pixel of the image is associated with an intensity value corresponding to the characteristic value. For example, the value 1 can be associated with a characteristic value of a validated puff, and the value 0 can be associated with a non-validated puff, that is to say one whose characteristic value does not exceed the chosen threshold level.

[0122] THE Figures 8a et 8b illustrate step 120 of method 100.

[0123] There Figure 8a shows the measurement signal, represented in grayscale. The Figure 8b shows the validated signal after applying the threshold on this same substrate. The characteristic value used in the Figure 8b is the likelihood ratio. We clearly observe the discrimination capacity (detection of “real” defects) of the validated signal.

[0124] The method 100 according to the embodiment shown in the Figure 2 further comprises a step 122 of classifying the pixels of the image constructed with the characteristic values ​​into binary logical objects (called « blob » according to the Anglo-Saxon term " binary large object "). This step 122 allows the pixels to be grouped into defects. Known methods of image processing and statistical analysis can be used to carry out this step.

[0125] In a step 124, for each blob or defect, its physical parameters and characteristics are statistically analyzed. These parameters and characteristics may include, for example, area, shape, ellipticity, orientation, likelihood (if this characteristic value was used), etc. Shape recognition may also be performed.

[0126] There Figure 9 is a schematic representation of another non-limiting embodiment of the method for inspecting a substrate according to the invention.

[0127] Steps 202 to 218, 222 and 224 of the method 200, shown in the Figure 9 , are identical to steps 102 to 118, 122 and 124 of method 100, shown in the Figure 2 , and will therefore not be described in the following.

[0128] Once the Doppler bursts have been validated by comparing the validated signal or the respective characteristic values ​​with the threshold values ​​in step 218, in a step 220a, a representative parameter of the validated bursts is determined. This representative parameter can be deduced from the validated signal, and / or from the measurement signal for the validated Doppler bursts based on the validated signal. This representative parameter can be, for example, a likelihood ratio from the validated signal, or the visibility, the modulation amplitude of the burst, the maximum amplitude of the burst or an amplitude average from the measurement signal or from the validated signal if this information is present therein.

[0129] As an example, it is possible to obtain, from the modulation amplitude at the modulation frequency characteristic of the passage of a defect, the size of the defect crossing the measurement volume. This relationship is represented on the Figure 10 , with a circular defect 3 having a diameter d. Plus size d of defect 3 is large compared to the interfringe distance δ in the measurement volume, the smaller the ratio between the modulation amplitude and the maximum of the envelope of the measurement signal, corresponding to the contrast.

[0130] The modulation amplitude can be calculated in several ways. One method is described in W.M. Farmer, "Measurement of Particle Size, Number Density, and Velocity using a Laser Interferometer," Appl. Opt., 11, pp. 2603-2612, 1972. The modulation amplitude is reduced to a visibility parameter V defined as follows: V = I max − I min I max + I min − 2 P

[0131] The intensities I max and I min as well as the offset P are defined on the Figure 3 . Visibility V is then used to directly deduce the diameter of a particle.

[0132] Following step 220a, during a step 220b, an image is constructed by associating the representative parameter, such as visibility, of each validated puff with a position on the substrate, in a similar manner to step 120 of the method 100 described with reference to the Figure 2 .

[0133] According to another embodiment, the image construction step of the method according to the invention uses the representative envelope parameters of the detected bursts, originating from the validated signal. These representative parameters can be the maxima of the envelopes I m ax, I min and offset P, as illustrated on the Figure 3 . These parameters can be estimated during the characteristic value assignment step, as described previously for the likelihood ratio calculation. Thus, the amplitude A from the measurement signal can be deduced directly by calculating: = I max + I min / 2 .

[0134] Each parameter A can then be assigned image intensity values, to construct a map of validated defects.

[0135] Of course, the invention is not limited to the examples which have just been described and numerous adjustments can be made to these examples without departing from the scope of the invention.

Claims

1. Method (100, 200) for inspection of a substrate (2), the method (100, 200) comprising the following steps: - creating, based on at least two light beams (4, 5) originating from one and the same light source (20), a measurement volume at the intersection between said at least two light beams (4, 5), the measurement volume containing interference fringes and being positioned so as to extend over or into the substrate (2), said substrate (2) being in movement with respect to said measurement volume in a direction parallel to a main surface (S) of said substrate (2); - acquiring (102, 202) a measurement signal representative of the light scattered by the substrate (2), as a function of the location of the measurement volume on said substrate (2); - calculating (108, 208) at least one expected parameter, including at least one expected modulation frequency being an expected Doppler frequency, of an expected signal representative of the passage of a defect (3) of said substrate (2) through the measurement volume; - determining (114, 214) characteristic values representative of a frequency content of the measurement signal in a neighbourhood around the expected modulation frequency, so as to constitute a validated signal representative of the presence of defects (3); the step (114, 214) of determining the characteristic values comprising the following steps: - modelling the expected signal according to a model function, in order to produce a modelled signal; - comparing the modelled signal to the measurement signal, including calculating a distance within the meaning of a Euclidian norm between the modelled signal and the measurement signal; and - analyzing said validated signal in order to locate and / or identify defects (3).

2. Method (100, 200) according to claim 1, characterized in that it also comprises the steps of: - determining (116, 216) a threshold value of the characteristic values; and - comparing (118, 218) the validated signal to said threshold value.

3. Method (100, 200) according to claim 2, characterized in that the step of determining (116, 216) a threshold value is carried out based on a validated signal obtained with a test substrate (2) having known characteristics.

4. Method (100, 200) according to one of claims 1 to 3, characterized in that the step (114, 214) of determining characteristic values comprises pass-band filtering of the measurement signal using a pass-band suitable for transmitting only the frequency content of the measurement signal in a neighbourhood around the expected modulation frequency.

5. Method (100, 200) according to claim 4, characterized in that the characteristic value takes account of: - the modulation amplitude of the filtered measurement signal. - the ratio between the modulation amplitude of the filtered measurement signal and a continuous value of the measurement signal.

6. Method (100, 200) according to one of claims 1 to 3, characterized in that the step (114, 214) of determining characteristic values comprises the following steps: - calculating a local Fourier transform of the measurement signal in order to obtain a local power spectral density. - determining a characteristic value based on the power spectral density at, or within a neighbourhood comprising, the expected modulation frequency.

7. Method (100, 200) according to any one of the preceding claims, characterized in that it also comprises a step (120, 220) of constructing an image of the substrate (2) by using the validated signal.

8. Method (100, 200) according to claim 7, characterized in that the image construction step (120) comprises a step of assigning intensity values to characteristic values corresponding to positions on or in the substrate (2), said intensity values corresponding to pixels of the image9. Method (100, 200) according to any one of the preceding claims, characterized in that it comprises a step (220a) of determining a representative parameter deduced from the measurement signal and / or the validated signal.

10. Method (100, 200) according to claim 9, characterized in that it comprises an image construction step (220b) comprising assigning intensity values to representative parameters corresponding to positions on or in the substrate (2), said intensity values corresponding to pixels of the image.

11. Method (100, 200) according to one of claims 8 to 10, characterized in that it also comprises a step (122, 222) of classifying the pixels of the image as binary objects in order to reconstruct the validated defects.

12. Method (100, 200) according to any one of the preceding claims, characterized in that it is implemented for the inspection of a transparent or opaque wafer (2) for electronics, optics or optoelectronics.

13. System (1) for inspecting a substrate (2), the system (1) comprising: - an interferometric device (30) coupled to a light source (20) in order to create, based on at least two light beams (4, 5) originating from the light source (20), a measurement volume at the intersection between said at least two light beams (4, 5), the measurement volume containing interference fringes and being positioned so as to extend over or into the substrate (2); - a device for moving said substrate (2) relative to said measurement volume in a direction parallel to a main surface (S) of said substrate; - an optoelectronic device (50) for acquiring a measurement signal representative of the light scattered by the substrate (2), as a function of the location of the measurement volume on said substrate; the system also comprising a processing module (60) arranged for: - carrying out a calculation (108, 208) of at least one expected parameter, including at least one expected modulation frequency, of an expected signal representative of the passage of a defect (3) of said substrate (2) through the measurement volume; - carrying out a determination (114, 214) of characteristic values representative of a frequency content of the measurement signal in a neighbourhood around the expected modulation frequency, so as to constitute a validated signal representative of the presence of defects (3), wherein the processing module (60) is arranged for: - carrying out a modelling of the expected signal according to a model function, in order to produce a modelled signal; - carrying out a comparison of the modelled signal to the measurement signal, including calculating a distance within the meaning of a Euclidian norm between the modelled signal and the measurement signal; and - carrying out an analysis of said validated signal in order to locate and / or identify defects (3).

14. System (1) according to the preceding claim, characterized in that it comprises a device for rotating the substrate (2) about an axis of rotation (X) perpendicular to a main surface (S) of said substrate, and a device for moving the interferometric device (30) in translation, arranged in order to move the measurement volume in a radial direction relative to the axis of rotation (X).