TW202526320A - An acoustic subsurface inspection device and method - Google Patents

Abstract

The present application discloses an acoustic subsurface inspection device (1) an a corresponding method. The disclosure comprises performing a measurement session including at least two acoustic measurements, each acoustic measurement comprising inputting an acoustic wave at an input location at a surface (114) of a sample (11) and generating a respective sense signal (S32a, S32b) that is indicative for a measured reflections of the acoustic wave at a measurement location. A measurement session comprises at least two acoustic measurements that are performed with a different input location and/or a different measurement location. Subsequently output data indicative for subsurface features in the sample is generated based on a computation using the respective sense signals generated with the at least two acoustic measurements in the measurement session.

Description

Acoustic subsurface detection device and method

Field of the Invention This application relates to an acoustic subsurface detection device.

This application further relates to an acoustic subsurface detection method.

Background of the Invention Acoustic inspection devices and methods can be used in the lithography industry to enable subsurface inspection of finished and/or intermediate products to properly control process alignment and process parameters and, if necessary, remove defective products. According to Moore's law, critical dimensions of semiconductor devices are becoming increasingly smaller to achieve ever-increasing integration. There is also a trend toward fabricating devices with an increasing number of layers.

Existing acoustic detection methods operate according to a first, low-frequency-based approach and a second, high-frequency-based approach. In the first approach, the stiffness of a sample's surface is measured at relatively low acoustic frequencies (e.g., in the range of approximately 1 MHz to several hundred MHz). Consequently, subsurface features can be identified with sub-nanometer resolution at depths up to approximately 2 microns. In the second approach, elasticity and wave scattering are measured at relatively high acoustic frequencies, e.g., exceeding 1 GHz. This approach allows for deeper detection of samples than is possible with the first approach, but to date, the resolution achievable has been poor compared to that achieved with the first approach.

Summary of the Invention An object of the present disclosure is to provide an improved acoustic subsurface detection device to address the aforementioned shortcomings.

A further object of the present disclosure is to provide an improved acoustic subsurface detection method to address the above-mentioned shortcomings.

According to the first-mentioned object, an improved acoustic subsurface detection device includes a carrier for holding a sample to be detected, a signal generator, at least one AFM tip, a lateral positioning device, a distance control device and a signal processor.

As noted above, the device is particularly relevant for applications in the lithography industry, where the sample to be inspected is a semiconductor device, such as a (semi-finished) integrated circuit, such as a NAND memory. In other examples, the sample is a micro-optical device comprising a highly integrated optical circuit system. In yet another example, the sample is a MEMS device. In yet another example, the sample comprises a combination of two or more selected from electronic, optical, and mechanical components.

A signal generator is configured to provide a drive signal that drives at least one AFM tip to induce acoustic waves into the sample at an input location on the sample's surface. The electrical drive signal can then be converted into an acoustic signal for the at least one AFM tip by an actuator (such as an electrostatic or piezoelectric actuator). According to another approach, the acoustic signal is generated by a laser that directs a modulated laser beam toward the tip or a cantilever or diaphragm supporting the tip.

At least one AFM tip or at least one other AFM tip is (also) configured to receive reflections of acoustic waves in the sample at a receiving location on the surface of the sample. To this end, it is coupled to an acoustic sensor (e.g., a piezoelectric element) configured to generate a sensing signal indicative of the received reflected acoustic waves. In the case where a single AFM tip is configured to both sense and receive reflected acoustic waves, an actuator (e.g., a piezoelectric element) for converting electrical signals into acoustic signals can also serve as the sensor for converting acoustic signals into electrical signals. In the case where there is only a single AFM tip configured to sense and receive reflected acoustic waves, the actuator, the signal generator, is preferably configured to generate a pulse signal. This simplifies the signal processing operations to be performed on the sensing signal. In some examples, the signal processor generates a pulse signal having a pulse duration that is short compared to the time it takes for a reflection to reach a single AFM tip, so that a sensed signal resulting from the reflection can be easily distinguished from a sensed signal resulting from the acoustic input signal. In some of these examples, the signal processor generates pulses in a pulse train in which the time interval between subsequent pulses exceeds a response duration. The response duration is defined, for example, as the time interval in which the response amplitude has decreased to a value substantially less than its maximum value (e.g., less than 10 times the maximum value or less than 100 times the maximum value).

In cases where a single AFM tip is used to sense acoustic waves and receive reflected acoustic waves, it is easier to distinguish the sensed signals due to reflections from those due to the acoustic input signal. The signal processor can generate the drive signal as a continuous high-frequency signal or as a pulsed signal as described for the case where a single AFM tip is used.

The lateral positioning device is used to control the lateral position of at least one AFM tip relative to the surface of the sample. Typically, the lateral positioning device is configured to control the lateral position of the sample surface in two orthogonal directions, for example, by positioning the sample and/or at least one AFM tip.

The distance control device controls the distance of at least one of the AFM tips relative to the surface of the sample. In one example, the distance control device controls the position of a device head carrying the at least one AFM tip in a direction transverse to the sample surface. Alternatively or additionally, the distance control device controls the position of the AFM tip disposed in the device head in a direction transverse to the sample surface.

The signal processor is configured to generate output data related to subsurface features in the sample. The output data includes, for example, the location of the subsurface feature, the shape of the subsurface feature, the material properties of the subsurface feature, and the like.

The acoustic subsurface detection device is configured to perform a measurement session comprising at least two acoustic measurements performed using different input positions and/or different receiving positions. The signal processor is configured to combine information from respective sensed signals generated from the at least two acoustic measurements during the measurement session to calculate information about subsurface features in the sample.

Due to the fact that the signal processor combines information from the respective sensing signals resulting from at least two acoustic measurements applied during the measurement period, it can obtain information about sub-surface features in the sample that would otherwise not be available.

As noted above, in some examples, an AFM tip is used that is configured to sense acoustic waves and receive reflections of the acoustic waves. In these examples, at least two acoustic measurements are performed at different lateral positions of the AFM tip relative to the sample surface and at different time intervals. The first of the lateral positions serves as the input and receiving position in the first acoustic measurement. The second of the lateral positions serves as the input and receiving position in the second acoustic measurement.

In another embodiment, an acoustic subsurface detection apparatus includes a plurality of transmitting AFM tips configured to induce acoustic waves into a sample and at least one separate receiving AFM tip configured to receive reflected acoustic waves. In that embodiment, a signal processor is configured to generate output data based on a comparison of generated sensing signals resulting from reflections of acoustic waves from each of the plurality of transmitting AFM tips.

In another embodiment, the acoustic subsurface detection device includes a plurality of receiving AFM tips configured to receive reflected acoustic waves and at least one transmitting AFM tip, and the signal processor is configured to generate output data based on a comparison of generated sensing signals provided by each of the plurality of receiving AFM tips.

An AFM tip need not be statically configured as a transmitting AFM tip, a receiving AFM tip, or both. In some examples, an acoustic subsurface probing apparatus includes at least one AFM tip that is dynamically configured as a transmitting AFM tip, a receiving AFM tip, or both.

As noted above, the drive signal can be provided as a pulse train. During the same acoustic measurement period during an acoustic measurement session, that is, while the input and receiving positions remain unchanged, multiple sensed signals can be obtained for each pulse in the pulse train. Statistical analysis can be performed on the multiple sensed signals obtained, for example, to derive an estimate of measurement noise and/or to provide a de-noised sensed signal. Furthermore, the drive signal can be provided as a pulse train to facilitate lock detection.

In some embodiments, a signal processor of an acoustic subsurface detection device includes a peak detection unit and a calculation unit. The peak detection unit is configured to determine the respective delays at which peaks occur in respective sense signals obtained in respective ones of at least two acoustic measurements during a measurement period. The calculation unit is configured to calculate the position of a subsurface feature based on the difference between the respective delays. The difference between the respective delays is a monotonic function of the lateral position of the subsurface feature. The lateral position of the subsurface feature can then be calculated from the measured differences using an inverse function or an approximation thereof, such as an estimate using a lookup table or a polynomial approximation.

In the presence of more subsurface features, the peak detection unit can be configured to detect multiple peaks in the signal and correlate the delay differences between corresponding peaks. A peak appearing in the sensed signal of a first acoustic measurement can be paired with a peak appearing in the sensed signal of a second acoustic measurement, provided that the input and receiving locations in the first measurement are sufficiently close to the input and receiving locations in the second measurement.

Furthermore, measurement sessions with more than two acoustic measurements can be performed at mutually different input positions and/or mutually different receiving positions, for example to determine more than one coordinate of a subsurface feature.

In an alternative embodiment, the signal processor of the acoustic subsurface detection device includes a prediction module configured to generate a predicted detection signal based on a sample-based model and to generate output data based on a comparison of the predicted detection signal with the actual detection signal. In this embodiment, the comparison between the predicted detection signal and the actual detection signal indicates whether the properties of the sample deviate from their expected properties. If so, further measurements can be performed on the sample to identify the cause of the deviation.

In yet another embodiment of an acoustic subsurface detection device, a signal processor includes a trained neural network that receives individual sensor signals and is configured to apply neural network operations to the received sensor signals to estimate properties of subsurface features. The acquired sensor signals include convolutional information about subsurface features present in a sample. The neural network can be trained to reconstruct information about the subsurface features. Due to the fact that the sensor signals are dominated by local information, the neural network can be trained using training samples with a large number of training areas having mutually different subsurface patterns. The properties of the training samples determined at design time then serve as ground truth.

In a further embodiment, the signal processor of an acoustic subsurface detection device includes two or more of the signal analysis components described above to analyze the sensed signal in various ways. This application further relates to an improved acoustic subsurface detection method that holds a sample to be detected.

The method includes performing a measurement session comprising at least two acoustic measurements. Each acoustic measurement includes inputting an acoustic wave at an input location on a sample surface and generating a respective sensing signal that measures a reflection of the acoustic wave at a measurement location on the sample surface. The at least two acoustic measurements are performed using different input locations and/or different measurement locations.

Then, output data indicative of subsurface features in the sample is generated based on calculations using respective sensing signals generated by at least two acoustic measurements during the measurement period.

Embodiments of the acoustic subsurface inspection method further include an additional step prior to the step of holding the sample to be inspected.

Additional steps include: Providing a neural network; Providing at least one test sample; Performing a plurality of acoustic measurements, each of the plurality of acoustic measurements comprising: Inputting an acoustic input signal at an input location and obtaining a sensing signal indicative of an acoustic reflection signal at a sensing location, wherein the input location and/or sensing location of at least two of the acoustic measurements are located at different lateral positions relative to the sample surface.

The neural network is then trained to reconstruct information about subsurface features. Since sensory signals are dominated by local information, training can be performed using training samples from a large number of training regions with distinct subsurface patterns. The properties of the training samples, determined at design time, then serve as ground truth.

In an embodiment of the acoustic subsurface detection method, the calculation includes generating a difference signal indicating the difference between the respective sense signals. This serves as an intermediate calculation step that simplifies further calculation steps for generating output data.

In one embodiment, an acoustic subsurface inspection device includes a prediction module configured to generate a predicted detection signal based on a sample model and to generate output data based on a comparison of the predicted detection signal with the actual detection signal. In this case, it is not necessary to reconstruct subsurface features from the detection signal. Instead, the presence of a subsurface defect can be determined based on the comparison.

In an embodiment of an acoustic subsurface detection device, at least one AFM tip configured to detect reflections is located at a fixed distance from at least one AFM tip configured to provide an acoustic input signal. This embodiment is advantageous in that it has a mechanically simple construction and the relative position can be calibrated very accurately.

In another embodiment of the acoustic subsurface detection device, at least one AFM tip configured to detect reflections is at a configurable distance relative to at least one AFM tip configured to provide an acoustic input signal. This allows for additional measurements to be performed.

In some examples, the relative lateral positioning between the transmitting and receiving AFM tips is not fixed but adjustable. This adjustability allows, for example, the receiving AFM tip to be positioned at a plurality of selectable/adjustable lateral offsets relative to the transmitting AFM tip. By selecting different lateral separations, measurement conditions can be optimized to suit the specific characteristics of the sample being inspected, its secondary surface structure, and the desired measurement resolution and accuracy.

Optionally, the configurable distance is adjusted based on the (e.g. expected) buried structure to be imaged and/or parameters derived therefrom. Having a configurable distance allows optimization to be performed depending on the properties of the buried structure to be imaged and/or parameters derived therefrom.

By adjusting the transmitter-receiver spacing, the path length of the acoustic wave traveling to and from subsurface features can be modified, which in turn affects the delay time and the characteristics of the detected signal. This flexibility facilitates fine-tuning the measurement to increase signal contrast, reduce measurement uncertainty, and achieve an improved signal-to-noise ratio. For example, by increasing the lateral distance between the transmitting and receiving AFM tips, certain interference patterns in the sensed signal can be mitigated, or previously indistinguishable subsurface features can now produce distinguishable peaks in the reflected signal. Conversely, by reducing the lateral distance, it is possible to increase the measurement sensitivity of specific local subsurface structures. Therefore, having a configurable distance offers a metrological advantage: the operator or an automated control system can select or tune the AFM tip position to best suit the material and configuration of the sample at hand.

Embodiments of an acoustic subsurface sensing device having more than one AFM tip may include a respective actuator for individually positioning each AFM tip in a direction toward the surface of the sample.

Advantageously, effectively controlled individual coupling levels can be ensured. By providing each AFM tip with a separate actuator, the vertical positioning of the tip relative to the sample surface can be finely tuned, thereby allowing precise control of the contact force and, therefore, the acoustic coupling at each tip position. The acoustic coupling between the AFM tip and the sample surface, influenced by the applied force, directly impacts the signal-to-noise ratio (SNR) of the measured reflected signal.

In some cases, ensuring perfect parallel alignment of the tip and sample substrate can be challenging, and small angular misalignments (tilts) may exist. Consequently, a fixed vertical position of the tip relative to the sample can cause the tip-to-substrate distance to vary across multiple AFM tips, as each tip can engage the substrate surface at a slightly different angle or vertical offset. These variations can lead to differences in the contact force applied to the substrate by individual AFM tips, resulting in uneven acoustic coupling conditions that can adversely affect measurement quality. Advantageously, by employing a separate actuator for each AFM tip, the vertical displacement and contact force of each tip can be individually controlled, even in the presence of tilt or other spatial variations in the substrate surface. For example, an AFM tip acting as an acoustic source can be positioned at a certain contact force level to achieve improved acoustic wave transmission into the sample, while the AFM tip designated as a receiver can be independently adjusted to ensure the appropriate force level for optimal detection and coupling efficiency. In some examples, this individual control can be further combined with an embodiment in which one of the AFM tips configured to receive reflections is placed at a configurable lateral distance from the source AFM tip. By doing so, the force applied by the sensing tips can be tuned to provide improved acoustic coupling (proper coupling of each tip), thereby enhancing the SNR of the reflected signal and improving the fidelity and reliability of subsurface feature detection.

In one embodiment, the head of the acoustic subsurface detection device has one or more AFM tips removably mounted therein.

The present disclosure further provides a system that includes, in addition to the improved acoustic subsurface detection device, a calibration sample and a calibration unit. The calibration unit is configured to cause the acoustic subsurface detection device to generate output data indicative of subsurface features in the calibration sample and to calibrate the acoustic subsurface detection device based on a comparison of the output data with output data expected for the calibration sample.

Detailed Description of Preferred Embodiments FIG1 schematically illustrates an embodiment of an improved acoustic subsurface detection device 1. In the embodiment of FIG1 , the acoustic subsurface detection device 1 includes: a carrier 10 for holding a sample 11 to be detected; a signal generator 20; at least one AFM tip 31 configured to sense an acoustic wave W; at least one AFM tip 32, 32a, 32b configured to receive reflections _Rca and _Rda of the acoustic wave; a lateral positioning device 16; a distance control device; and a signal processor 50.

When driven by the signal generator 20, at least one AFM tip 31 is configured to induce acoustic waves W into the sample at input positions _pa , _pd , _pe on the surface 114 of the sample.

At least one AFM tip 32, 32a, 32b is configured to receive reflections _Rca , _Rda of acoustic waves in the sample at receiving positions _pb , _pc , _pf on the surface (114) of the sample and is coupled to an acoustic sensor configured to generate sense signals _S32a , _S32b indicative of the received reflected acoustic waves.

The lateral positioning device 16 is configured to control the lateral position of at least one AFM tip 31 , 32 relative to the surface of the sample 11 .

A distance control device 51 is provided for controlling the distance of at least one of the AFM tips relative to the surface of the sample 11 .

The signal processor 50 is configured to generate output data S _im relating to the sub-surface features 113 c , 113 d in the sample.

The acoustic subsurface detection device is configured to perform a measurement session having at least two acoustic measurements performed using mutually different input positions and/or mutually different receiving positions.

The signal processor 50 is configured to combine information from respective sense signals generated from at least two acoustic measurements taken during a measurement period to calculate information about sub-surface features 113c, 113d in the sample.

FIG2 schematically shows the surface 114 of a sample 11 and illustrates the operation of two embodiments of an acoustic subsurface detection apparatus. The upper portion of FIG2 illustrates the operation of the first embodiment, in which an AFM tip 31 driven by a signal generator 20 induces an acoustic wave W into the sample at an input position p _a on the surface 114 of the sample 11. At least one AFM tip receives reflections _Rab and _Rac of the acoustic wave in the sample at respective receiving positions p _b and _pc on the surface 114 of the sample. In one example, a single AFM tip is then positioned at receiving positions p _b and _pc . In another example, separate AFM tips are positioned at receiving positions P _b and _pc to simultaneously receive reflections _Rab and _Rac of the acoustic wave.

The lower portion of FIG. 2 illustrates the operation of a second embodiment, in which a single AFM tip positioned at a receiving position _pf receives reflections _Rdf and _Ref of acoustic waves sensed in a sample at respective positions _pd and _pe on the surface 114 of the sample 11. In one example, the single AFM tip is then positioned at positions _pd and _pe to sense the acoustic waves. In another example, separate AFM tips are positioned at positions _Pd and _pe to simultaneously sense the acoustic waves in the sample 11.

In each of these examples of these embodiments, respective sensing signals are obtained, each generated using at least two acoustic measurements during a measurement period. In the first example of the embodiment, respective sensing signals are obtained for reflections _Rab and _Rac . In the second example of the embodiment, respective sensing signals are obtained for reflections _Rdf and _Ref . Based on a comparison of the generated sensing signals, a signal processor generates output data.

3A and 3B illustrate the operation of an embodiment of the improved acoustic subsurface detection device in more detail. FIG3A shows an exemplary sample 11 having a lower layer 111 (e.g., a substrate), a matrix 112, and a series of features, the two closest to the transmission tip 31 of which are designated 113c and 113d.

In this embodiment, the improved acoustic subsurface detection device includes a sensing head 30 having an AFM tip 31 configured to induce an acoustic wave W into the sample 11 at an input location on the surface 114 of the sample 11. The sensing head 30 also includes a pair of AFM tips 32 _a and 32 _b symmetrically arranged with respect to the AFM tip 31 at a distance of 5 nm from each other and configured to receive reflections of the acoustic wave W at respective receiving locations. In one example, the AFM tips 31, 32 _a , and 32 _b are statically configured to perform these functions, i.e., their functions are fixed-wired. For example, AFM tip 31 is exclusively connected to signal generator 20 to receive drive signal _S31 , and AFM tips _32a and _32b are exclusively connected to signal processor 50 to provide their respective sense signals _S32a and _S32b . In this case, AFM tip 31, on the one hand, and AFM tips _32a and _32b, on the other hand, may have dedicated configurations to optimally perform their respective fixed-wire functions. In another example, AFM tips 31, _32a , and _32b are dynamically configured to perform these functions. For example, each of these AFM tips is mechanically coupled to a respective piezoelectric element. In this case, the AFM tip 31 is configured to induce the acoustic wave W into the sample 11 because its piezoelectric element is dynamically configured to receive the drive signal S ₃₁ from the signal generator 20, and the piezoelectric elements of the AFM tips 31, _32a , _32b are dynamically configured to provide their respective sense signals S _32a , S _32b to the signal generator 20.

For simplicity, the acoustic waves emitted by the transmission tip 31 and reflected by the features 113c, 113d are schematically shown in the form of arrows.

The drive signal _S31 and the sense signals _S32a , _S32b are typically electrical signals, but alternatively, one or more of these signals may be optical signals.

As shown in FIG3A , AFM tips 32 _a and 32 _b receive reflections R _ca and R _da, respectively, from the two closest subsurface features 113 c and 113 d in sample 11 and generate a sense signal S _{32 a} , as shown in FIG3B . In practice, it can be assumed that reflections from other subsurface features farther from these two subsurface features 113 c and 113 d are negligible. Similarly, AFM tip 32 _b receives reflections R _cb and R _db, respectively, from the two closest subsurface features 113 c and 113 d in sample 11 and generates a sense signal S _{32 b} , as shown in FIG3B .

FIG3B shows that sense signals _S32a and _S32b each have a main peak near the origin of the graph, attributed to the initial acoustic excitation caused by the wave propagating on sample surface 114. Sense signal _S32a has a first peak and a second peak following the main peak after time delays _tca and _tda , respectively. These time delays _tca and _tda indicate the path lengths from AFM tip 31 through subsurface feature 113c to AFM tip 32a, and from AFM tip 31 through subsurface feature 113d to AFM tip 32a, respectively. Similarly, sense signal _S32b has a first peak and a second peak following the main peak after time delays _tcb and _tdb , respectively. These time delays _tcb and _tdb indicate the path lengths from the AFM tip 31 through the subsurface feature 113c to the AFM tip 32b, and from the AFM tip 31 through the subsurface feature 113d to the AFM tip 32b, respectively.

In this example, a single pulse, W, is used to induce an acoustic wave. As shown in Figure 3B, the pulse has a short duration compared to the time it takes the sound to travel the distance to the subsurface feature to be detected. Consequently, in response, in this case, subsequent peaks following the initial peak can be easily distinguished from the initial peak. By way of example, the speed of sound in the substrate is approximately 2000 m/s. When measured within a range of 1 micron, the travel time is approximately 0.5 ns, and the pulse width is at most one-tenth of the travel time, or no more than 0.05 ns. In practice, the measurement pulses can be sensed as part of a pulse train, with the constraint that the time interval between subsequent pulses exceeds the time interval for measuring the reflection of the acoustic wave. While providing the acoustic wave in response to a pulse train such as drive signal S ₃₁ , the AFM tip 31, 32a, 32b can be maintained in a fixed position, allowing multiple measurement sessions with identical conditions to be performed, thereby enabling statistical operations to be performed, such as calculating a more accurate position estimate of a subsurface feature or calculating a measure of uncertainty in the position estimate.

In the following discussion, the coordinates (x, z) of tips 32a, 31, and 32b, and the coordinates of features 113c and 113d, are expressed relative to the coordinates of transmitting tip 31. Thus, the relative coordinates of the transmitting tip are (0, 0). It is further assumed that receiving tips 32a, 32b are equidistant from transmitting tip 31 at the same level z. Thus, the receiving tips have coordinates (-dx, 0) and (dx, 0), respectively. It is further assumed that the feature has a z-coordinate dz and relative positions -xc and +xd along the x-axis relative to transmitting tip 31. Thus, the coordinates of features 113c and 113d are (-xc, dz) and (xd, dz), respectively. In the example shown, first receive tip 32a generates a sense signal S _32a , which indicates the superposition of acoustic waves reflected from features in sample 11, specifically reflections R _ca and R _da received from features 113c and 113d closest to the tip, respectively. Similarly, second receive tip 32b generates a sense signal S _32b , which indicates the superposition of acoustic waves reflected from features in sample 11, specifically reflections R _cb and R _db received from features 113c and 113d closest to the tip, respectively. The reflections are received with different delays depending on the relative position of the feature to the tip.

The delay in receiving the reflection is determined by the path lengths and the speed of sound _v in matrix 112. The relative time position of the peak, _tca (i.e., the time interval between the emission of the acoustic pulse by the transmitting tip 31 and the appearance of the peak in the detected signal _S32a caused by the reflection of feature 113c), is approximated by: (1a)

Similarly, for the peak in detected signal S _32a caused by reflection R _da from feature 113d, the relative time position t _da is: (1b)

The time position _tcb of the peak in the detected signal _S32b caused by the reflection _Rcb from the feature 113c is: (2a)

Furthermore, the reflection R _db from feature 113 d causes a peak in the detected signal S _{32 b} at time position t _db : (2b)

The distance between the peaks in the detected signal S _32a is:

The distance between the peaks in the detected signal S _32b is:

Therefore, the distance difference is:

In the case of a single measurement, it can be determined that two features exist in the vicinity of the head 30, but their positions cannot be determined.

Due to the fact that the respective sensing signal is generated using at least two acoustic measurements during the measurement period, information about the subsurface features 113c, 113d in the sample 11 can be calculated based on the measured delay times _tca , _tcb , as shown below. (3)

Figure 3C shows the relationship between the position _xc of the feature relative to the position of the transmitter tip and the observed difference in delay Δc _,ab in arbitrary units. As is apparent from this, the observed difference is a monotonically increasing function of the distance, which allows the position of the feature to be determined.

FIG3D illustrates in greater detail the signal processor 50 used to perform the calculations discussed above. The exemplary signal processor of FIG3D includes a first delay time calculation module 55a that calculates the delay time _tca specified in Equation 1a. A second delay time calculation module 55b calculates the delay time _tcb specified in Equation 2a. A subtraction element 56 calculates the difference Δc _,ab between these delay times. A feature position calculation module 57 calculates the feature position _xc according to Equation 3.

In another approach, illustrated in Figures 4A and 4B , the apparatus includes a detection head 30 having a single transmitting tip 31 and a single receiving tip 32a, and a measurement session includes measurements at two or more different locations. For example, a first measurement is performed at a reference location, and a second measurement is performed at a location that is offset from the reference location by an amount in the x-axis direction. As a non-limiting example, assume the offset is dx.

As described above, for the first measurement, it is assumed that the position of the transmitting tip is at the origin and the position of the receiving tip 32a is at position -dx relative to the position of the transmitting tip 31. In that case, the observed delay, denoted as t _da1 , is:

For the second measurement, with the transmission tip at position dx, the observed delay, denoted as t _da2 , is: Therefore, the delay time difference Δ _c,12 can be calculated as

The location of the feature can then be determined using the same calculations described in the examples of Figures 3A, 3B, and 3C.

In a further embodiment, as illustrated in Figures 5A and 5B , an acoustic subsurface detection device includes an AFM tip 31 that can be dynamically configured to function as an AFM tip for sensing acoustic waves and as an AFM tip for receiving reflections. In operation, a signal generator 20 drives the AFM tip 31 using a drive signal S _{31_i1} , which causes the AFM tip 31 to sense an acoustic wave W1 into the sample at a first input position on the surface 114 of the sample 11. The signal processor 50 processes a sense signal S _{31_o1} generated by the AFM tip in response to the received reflection of the sensed acoustic wave. Without loss of generality, the position is assumed to have coordinates (x1 = 0, z1 = 0). The AFM tip 31 is also configured to receive reflections R _c1 , R _d1 of the acoustic wave in the sample at that location and generate a sense signal S _{31 — o1} indicative of the received reflected acoustic wave to be processed by the signal processor 50 .

The reflection of the acoustic signal W1 at the feature 113c causes a peak in the sense signal _{S31_o1} at a time interval _tc1 after the peak in the drive signal, where the time interval is specified by:

Similarly, the time positions of peaks caused by reflections from other features (e.g., feature 113d) can be calculated.

As shown in FIG5B , the head 30 is shifted by a distance dx in the positive direction of the x-axis and the measurement described above is repeated, thereby generating a second sensing signal S _{31_o2} . Now the time interval tc2 is measured, which is determined by the following formula:

The difference in the length of these time intervals, Δ _c,12 , is:

This is again a monotonically related function of the position _xc of feature 113c, allowing the position _xc to be calculated.

In summary, the acoustic subsurface detection device is configured to perform an acoustic subsurface detection method. When performing the method, the sample 11 to be detected is fixed on, for example, an xy stage, and the lateral position of the sample can be dynamically controlled using the xy stage. The method includes performing a measurement period including at least two acoustic measurements. In each acoustic measurement, an acoustic wave is input at an input position, and a respective sensing signal is generated indicating the measured reflection of the acoustic wave at the measurement position. At least two acoustic measurements in the measurement period are performed using different input positions and/or using different measurement positions. The method also includes using the respective sensing signals generated using at least two acoustic measurements in the measurement period to generate output data indicating subsurface features in the sample based on calculations.

As discussed above, the at least two acoustic measurements during a measurement session can be performed in various ways. For example, as discussed with reference to FIG2 , a measurement session can be performed using a single transmitting tip and multiple receiving tips, or using multiple transmitting tips and a single receiving tip. Furthermore, a measurement session can be performed using multiple transmitting tips and multiple receiving tips, or using a single transmitting tip and a single receiving tip as shown in FIG4A and FIG4B , or using a single tip configured to sense acoustic waves and receive their reflections as shown in FIG5A and FIG5B .

As discussed further, the functionality of the tip can be hard-wired or dynamically configurable.

Figures 6A to 6D illustrate an embodiment of an acoustic subsurface detection method improved for the purpose of overlay detection. Figure 6A shows two detection signals from two subsurface features at a depth of 2.5 microns below the sample surface and separated by 5 nm from each other. In this graph, the vertical axis indicates the signal intensity, and the horizontal axis indicates the distance from the subsurface feature to the detector calculated based on the time delay of the response to the sensed acoustic wave. Figure 6B shows an enlarged portion. Figure 6C shows a difference signal calculated based on the difference between the two detection signals. Figure 6D shows the difference signal obtained for various lateral offsets of the subsurface features. As is apparent from Figure 6D, the difference signal clearly indicates the offset between the subsurface features.

In an embodiment, the signal processor 50 configured to calculate information about subsurface features in a sample comprises a trained neural network 50 _NN . For this purpose, the trained neural network 50 _NN can be obtained by applying the following procedure to the network to be trained.

At least one training sample 11T is provided, wherein the at least one training sample comprises a variety of mutually different subsurface structures distributed laterally within the sample. The following sequence of training steps is repeatedly performed using the at least one training sample 11T.

Respective sensing data is obtained during a measurement period, wherein an input position and a measurement position are within a spatial range around position x. The measurement period includes at least two acoustic measurements, each acoustic measurement comprising inputting an acoustic wave at the input position and generating a respective sensing signal indicative of a measured reflection of the acoustic wave at the measurement position, wherein the at least two acoustic measurements are performed using different input positions and/or different measurement positions within the spatial range.

At the input of the neural network 50 _NN to be trained, a respective sensor signal ST(x) is provided, for example generated by the head 30 using at least two acoustic measurements during a measurement period.

In response thereto, the neural network 50 _NN to be trained applies a neural network operation to the respective sensor signal and provides output data F(x). The output data f(x) indicates, for example, the estimated offset of the subsurface feature in the training sample 11T in the region x.

In addition, the ground truth data GT(x) is obtained from the subsurface specifications of at least one training sample in the spatial range x.

Here, the loss L(x) is calculated by the loss calculation module 70 based on the comparison between the output data F(x) and the ground truth data GT(x).

The neural network parameters of the neural network 50 _NN to be trained are then updated by back propagating the calculated loss L(x).

After the update, a new spatial range is selected that is different from the previous spatial range and the next sequence of training steps is executed.

It should be noted that in yet another embodiment of the acoustic subsurface detection device, the signal processor includes a prediction module configured to generate a predicted detection signal based on a sample-based model and to generate output data based on a comparison of the predicted detection signal with the actual detection signal. In this embodiment, subsurface properties need not be reconstructed from the detection signal. Instead, the obtained detection signal is compared to the expected detection signal, and deviations indicate sample defects.

In a technical solution, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single component or other unit may fulfill the functions of several items described in the technical solution. The mere fact that certain measures are described in different technical solutions does not indicate that a combination of these measures cannot be used to advantage. Any reference signs in the technical solution should not be construed as limiting the scope.

1: Acoustic subsurface detection device 10: Carrier 11: Sample 11T: Training sample 16: Lateral positioning device 20: Signal generator 30: Head 31, 32, 32a, 32b: Tip 50: Signal processor _50NN : Trained neural network 51: Distance control device 55a: First delay time calculation module 55b: Second delay time calculation module 56: Subtraction element 57: Feature position calculation module 70: Loss calculation module 111: Lower layer 112: Matrix 113c, 113d: Subsurface feature 114: Surface F(x): Output data GT(x): Ground truth data L(x): Losses p _a , p _d , p _e : Input positions p _b , p _c ,p _f : Received position Rab ,R _ac ,R _c1 , _{R ca} ,R _cb ,R _d1 ,R _da ,R _db ,R _df ,R _ef : Reflection S ₃₁ ,S _{31_i1} : Drive signal S _{31_o1} ,S _32a ,S _32b ,ST(x): Sense signal S _{31_o2} : Second sense signal _Sim : Output data t _ca ,t _cb ,t _da ,t _db : Time delay W,W1 : Sound wave x _c : Position Δ _c,ab : Difference

These and other aspects of the present disclosure are illustrated in greater detail in the accompanying drawings. Among them: FIG. 1 schematically illustrates an embodiment of an improved acoustic subsurface detection device; FIG. 2 illustrates various measurement options; FIG. 3A through FIG. 3D illustrate the operation of an embodiment of the improved acoustic subsurface detection device 1 in greater detail; FIG. 4A and FIG. 4B illustrate another embodiment of the improved acoustic subsurface detection device; FIG. 5A and FIG. 5B illustrate yet another embodiment of the improved acoustic subsurface detection device; FIG. 6A through FIG. 6D illustrate an application for stacked pair detection; FIG. 7 illustrates a method for training a neural network for use in an embodiment of the improved acoustic subsurface detection device or in corresponding embodiments of the method.

11: Sample

30: Head

31, 32a, 32b: Tip

111: Lower layer

112: Matrix

113c, 113d: Subsurface features

114: Surface

R _ca ,R _cb ,R _da ,R _db : reflection

S ₃₁ : Drive signal

S _32a , S _32b : Sensing signal

W: Sound waves

x,z:component

Claims (17)

An acoustic subsurface detection device (1) comprises: a carrier (10) for holding a sample (11) to be detected; a signal generator (20); at least one AFM tip (31) configured to be driven by the signal generator and configured to induce an acoustic wave (W) into the sample at an input position ( _pa , _pd , _pe ) on a surface (114) of the sample; at least one AFM tip (32, 32a, 32b) configured to receive a reflection ( _Rca , Rda) of the acoustic wave in the sample at a receiving position ( _pb , _pc , _pf ) on the surface (114) of the sample and coupled to a sensing signal ( _S32a , _S32b ) configured to generate a sensing signal indicative of the received reflected acoustic wave. _32b ); a lateral positioning device (16) for controlling a lateral position of at least one AFM tip (31, 32) relative to the surface of the sample; a distance control device (51) for controlling a distance of at least one of the AFM tips relative to the surface of the sample; a signal processor (50) for generating output data ( _Sim ) related to subsurface features (113c, 113d) in the sample; wherein the acoustic subsurface detection device is configured to perform a measurement session having at least two acoustic measurements performed using different input positions and/or different receiving positions; And wherein the signal processor (50) is configured to combine information from the respective sense signals generated using the at least two acoustic measurements during the measurement period to calculate information about the sub-surface feature (113c, 113d) in the sample. An acoustic subsurface detection device as claimed in claim 1, comprising a plurality of transmitting AFM tips configured to sense an acoustic wave into the sample and at least one AFM tip configured to receive a reflected acoustic wave, wherein the signal processor is configured to generate output data based on a comparison of one of the generated sensing signals generated by reflections of the acoustic waves from each of the plurality of transmitting AFM tips. An acoustic subsurface detection device as claimed in claim 1 or 2, comprising a plurality of receiving AFM tips configured to receive a reflected acoustic wave and at least one transmitting AFM tip configured to sense an acoustic wave into the sample, wherein the signal processor is configured to generate output data based on a comparison of one of the generated sensing signals provided by each of the plurality of receiving AFM tips. An acoustic subsurface detection device as claimed in claim 1 or 2, which is configured to perform the at least two acoustic measurements at different time intervals, wherein at least one of an input position and/or a receiving position is changed after performing a first one of the at least two acoustic measurements and before performing a second one of the at least two acoustic measurements. An acoustic subsurface detection device as claimed in any of the preceding claims, wherein the signal processor comprises a peak detection unit (55a, 55b) and a calculation unit (57), the peak detection unit being used to determine a respective delay ( _tca , _tcb ) at which a peak appears in respective sensing signals ( _S32a , _S32b ) obtained in respective ones of the at least two acoustic measurements during the measurement period, and the calculation unit being used to calculate a position ( _xc ) of the subsurface feature (113c, 113d) based on a difference between the respective delays. In the acoustic subsurface detection device of any one of claims 1 to 4, the signal processor includes a prediction module configured to generate predicted detection signals based on a model of the sample and to generate output data based on a comparison of the predicted detection signals with one of the actual detection signals. An acoustic subsurface detection device as claimed in any one of claims 1 to 4, wherein the signal processor (50) comprises a trained neural network ( _50NN ) which receives the respective sensing signals and is configured to apply neural network operations to the received sensing signals to estimate properties of the subsurface feature. The acoustic subsurface detection device of claim 1, wherein the at least one AFM tip configured to detect a reflection is at a configurable distance relative to the at least one AFM tip configured to provide an acoustic input signal. An acoustic subsurface detection device as claimed in claim 1, comprising a head carrying at least one AFM tip configured to supply an acoustic input signal and at least two AFM tips configured to detect a reflection, the head comprising a respective actuator for individually positioning the AFM tips in a direction toward a surface of the sample. The acoustic subsurface detection device of claim 1 comprises at least three AFM tips, wherein the at least three AFM tips can be dynamically configured to serve as an AFM tip for supplying an acoustic input signal or an AFM tip for detecting a reflection. An acoustic subsurface detection device as claimed in claim 1, having one or more of said AFM tips removably mounted therein. An acoustic subsurface detection device as claimed in claim 1, which is configured to operate with at least two AFM tips to detect a reflection at a fixed lateral position relative to the sample while the at least one AFM tip for supplying an acoustic input signal is positioned at mutually different lateral positions relative to the sample, and the signal processor is configured to generate the output data based on the detection signals obtained for the mutually different positions. A system comprising, in addition to an acoustic subsurface detection device as claimed in any of the preceding claims, a calibration sample and a calibration unit, wherein the calibration unit is configured to cause the acoustic subsurface detection device to generate output data indicative of subsurface features in the sample and to calibrate the acoustic subsurface detection device based on a comparison of the output data with one of the output data expected for the calibration sample. An acoustic subsurface detection method comprises: holding a sample (11) to be detected, the sample having a sample surface; performing a measurement period including at least two acoustic measurements, each acoustic measurement comprising inputting an acoustic wave at an input position ( _pa , _pd , _pe ) and generating a respective sensing signal indicating reflection of the acoustic wave measured at a measurement position ( _pb , _pc , _pf ), wherein the at least two acoustic measurements are performed using a different input position ( _pa , _pd , _pe ) and/or a different measurement position; and generating output data indicating subsurface features in the sample based on a calculation using the respective sensing signals generated using the at least two acoustic measurements in the measurement period. The acoustic subsurface detection method of claim 14, wherein the calculation includes generating a difference signal indicating a difference between the respective sensing signals as an intermediate calculation step for the calculation of the output data. The acoustic subsurface detection method of claim 14, wherein the calculating comprises: providing, at an input of a trained neural network, the respective sensing signals generated using the at least two acoustic measurements during the measurement period; wherein the neural network applies a neural network operation to the respective sensing signals; and obtaining, at an output of the trained neural network, output data indicative of a subsurface feature in the sample. The acoustic subsurface detection method of claim 16, comprising: providing a neural network to be trained; preparing at least one training sample, wherein the at least one training sample comprises a variety of mutually different subsurface structures distributed laterally in the sample; repeating the following training steps: applying the steps of claim 10 to obtain respective sensing data during a measurement period, wherein the input positions and the measurement positions are within a spatial range; providing the respective sensing signals generated by the at least two acoustic measurements during the measurement period at an input of the neural network to be trained; wherein the neural network applies neural network operations to the respective sensing signals; Obtaining output data at an output of the neural network to be trained; Obtaining ground truth data, the ground truth data obtained from a surface specification of the at least one training example within the spatial range; Calculating a loss based on a comparison between the output data and the ground truth data; Updating neural network parameters of the neural network to be trained by backpropagating the calculated loss; Selecting a new spatial range that is different from the previous spatial range.