Some content of this application is unavailable at the moment.
If this situation persist, please contact us atFeedback&Contact
Note: Text based on automatic Optical Character Recognition processes. Please use the PDF version for legal matters

TITLE: Determining quantitative three-dimensional surface topography from two-dimensional microscopy images


The invention relates to a method and a computer program for determining quantitative three-dimensional surface topography data from data representing images of a surface obtained by a microscope.

Upon observing a sample surface using the appropriate detection conditions, contrasting variations in tint and/or color in images are caused by the topography of the sample, also referred to as 'topographic contrast'.

However, the extraction of topographical information from two-dimensional images with such topographic contrast is rather complex. Zhu et al., Exp. Mech. 51 (2011) 97 describes techniques for obtaining quantitative three-dimensional surface topography data by scanning electron microscopy (SEM) or from SEM image data. To obtain data of meaningful accuracy regarding the shape of the scanned surface in a Z-direction perpendicular to the (mutually perpendicular) X and Y dimensions of that surface, a calibration of the microscope stage movement for precisely determining movements of the scanned surface and dimensions of the chamber and of the holder of the sample of which the surface is scanned, is required. SEM and other

microscopes suitable for calibration are rare and very costly and calibration is time consuming. Moreover, such calibration may introduce several additional errors in the final results (Li et al., Opt. Lasers Eng. 50 (2012) 971).

Cop et al., Proceedings of the 10th International Conference on Pattern Recognition, Volume 1, 733 - 737, 1990 discloses a multi-resolution approach to the 3D-reconstruction from electron microscope tilt series with a limited number of projections where the tilt angles are assumed to be known. It solves the alignment problem from the image data without additional gold particles in a coarse-to-fine strategy, but quantitative results of the 3D reconstruction are not demonstrated.


It is an object of the present invention to provide a solution that allows obtaining quantitative three-dimensional surface topography data from two-dimensional images obtained by a microscope without requiring a priori information about the imaging equipment and in particular without requiring a calibration of the microscope stage movement determining movements of the scanned surface and of the dimensions of the chamber to obtain three-dimensional surface topography data of meaningful accuracy. According to the invention, this object is achieved by providing a method according to claim 1.

This method allows obtaining three-dimensional surface topography data from a series of at least three images acquired from the same sample surface positioned at different angles in the microscope. By solving equations of displacements of corresponding points in two-dimensional image planes between one image and at least two other images, all acquired from the same sample surface, but acquired from that surface while it is held at different orientations relative to the optical axis, quantitative three-dimensional surface topography data can be calculated.

Corresponding points may for instance be data points each representing a corresponding pixel or facet (consisting of a cluster of pixels), in particular properties of the respective pixel or facet and parameters defining the location of the respective pixel or facet, for instance in terms of the location of the center or of a predefined corner thereof. For a particularly precise

identification of the topographical features, it is preferred that variations in tint and or color in the image result from the surface topography of the sample.

The only physical requirement to obtain the 3D representation is the acquisition of at least three images of the surface region under study at

different tilt angles, without further need to know details of the tilting parameters. This method can be applied to sets of images acquired or provided without knowledge of the imaging and/or rotation conditions. The method according to the invention also allows finding the precise location and orientation of the sample with respect to the optical system. In general, this technique may be used to improve other techniques present in the same measurement systems which are influenced by surface topography (for instance EDS measurements in SEM systems).

The invention can also be embodied in a method according to claim 7 that also includes capturing of the images by microscopy and in a computer program product according to claim 8 which when executed on a computer, determines quantitative three-dimensional surface topography data from data representing one image and at least two other images, all acquired from the same sample surface, but acquired from that surface while it is held at different orientations relative to the optical axis.

Particular elaborations and embodiments of the invention are set forth in the dependent claims.

Further features, effects and details of the invention appear from the detailed description and the drawings.


Figs. la)-lc) are schematic illustrations of the influence of the height (level above a tilted plane) (Az) of a point P on the observed displacement of that point from a position on an untilted sample surface when the sample surface is tilted over a tilt angle Θ for:

a) a point in the sample plane (z = 0),

b) a point above the sample plane (z > 0), and

c) a point below the sample plane (z < 0);

Fig. 2 is a diagram representing the physical, optical and computational relationship between situations before and after the movement, in which examples of corresponding rotation, translation and observations are shown on the right side;

Fig. 3 is an illustration showing nominal orientations of tilting axes in three examples:

Case I: tilt about a sample stage axis only, imaging with x-axis parallel to the tilting axis,

Case II: tilt about the sample stage axis only, imaging with x-axis rotated over 35° relative to the tilting axis, and

Case III: tilt about the sample stage axis and about an axis perpendicular thereto, imaging with x-axis parallel to the sample stage axis;

Fig. 4a) is a map of displacements in y direction (<5y) obtained from digital image correlation (DIC) applied to images of Case II captured before and after tilting over an angle Qnom = 4.0°;

Fig. 4b) is a map of the result of fitting of the displacements to eq. 16b (see below) for calculation of Q4, Q5, and terms;

Figs. 5a), 5b) and 5c) are graphs showing the fitting of displacements in x and y direction to the values of {Q} (eqs. 16) for the three examples Case I, Case II and Case III, all the displacement data have been normalized for obtaining zero averages, each set of data is identified by the nominal angle of tilting;

Figs. 6a) 6b) and 6c) are plots of displacements Ay vs. Ax for the three examples Case I, Case II and Case III (see eq. 18), each set of data is again identified by the nominal angle of tilt, the alternative (k, ΐ) coordinate system being depicted for each Case;

Figs. 7a), 7b) and 7c) are graphs illustrating methods for finding a unique solution for the tilting in which the upper graphs illustrate method 1 (py vs. px plots) and the bottom graphs illustrate method 2 (pu vs. 6AC);

Fig. 8a), 8b), 8c) are graphs showing the outcome of comparisons between both methods of solution depicted in Figs. 7a), 7b) and 7c);

Fig. 9a)-9f) are 3D images obtained from Case I (all the scale bars representing 5 μηι) in which:

Fig. 9a) is an SEM reference picture,

Fig. 9b) is a reconstruction image obtained using δχ data from a set acquired at Onom = 0.5°,

Fig. 9c) is a reconstruction image obtained using <5y data from a set acquired at Onom = 0.5°,

Fig. 9d) is a reconstruction image obtained using <5y data from a set acquired at Onom = 2.0°,

Fig. 9e) is a reconstruction image obtained using <5y data from a set acquired at Onom = 2.0° and with smaller facet size in DIC,

Fig. 9f) is a graph with a set of histograms of reconstructions performed at different angles and facet sizes;

Figs. 10a)-10d) are 3D reconstructions from Case II, in which:

Fig. 10a) is a reconstruction using &c data from a set acquired at

Fig. 10b) is a reconstruction using d data from a set acquired at

Fig. 10c) is a reconstruction using 5k data from a set acquired at

Fig. lOd) is a graph containing histograms of reconstructions using data from different axes;

Fig. 11 is a flow chart of an architecture of an example of an algorithm for obtaining 3D maps from SEM images;

Fig. 12 is a flow chart of an example of an algorithm for correction of perspective deviations;

Fig. 13a) is a SEM image of a sample;

Fig. 13b) is a 3D reconstruction of the sample shown in Fig. 13a); and Fig. 13c) is a perspective correction applied to obtain the 3D

reconstruction of Fig. 13b).


Figs. la)-lc) illustrate basic phenomena occurring when tilting a sample on the displacement (δ) of the points at different distances from a plane in images acquired from the sample surface at different angles. The magnitude of the displacement (δ) of each point between its position in one image and its position in the other image acquired from the sample surface held at a different angle depends on the value of the z coordinates of the points (cf.

situations a, b and c in Fig. 1). As a consequence, from the displacements (<5) between the positions of corresponding points in the images before and after tilting, the z coordinates of these points can be determined. Determination of these displacements of the positions of corresponding points between different images can for instance be carried out by conventional 2D digital image correlation (DIC) (see e.g. B. Pan et al., Meas. Sci. Technol. 20 (2009) 062001) between the images before and after the tilting. In the following, it is described how data describing the three-dimensional profile can be obtained from the DIC data.

Mathematical definitions

An image can be considered as a matrix of data in which each pixel is characterized by its position (x and y coordinates) and a grayscale intensity. For a rigid body, the modification of coordinates before and after the tilting (situations labeled A and B, respectively) can be described as a combination of rotations and translations (see Figure 2). Therefore, a point i with starting ' ^ί ί is transformed to final coordinates

where [T] is the translation matrix:

and [R] is the rotation matrix:

Θ is the angle of rotation around the vector a with components ax, ay, az. This vector is unitary, so:

Both movements are shown in Fig. 2. Without loss of generality, the z coordinate of every point i can be expressed as a combination of a plane with components px, py and pz and the height of every point relative to this plane

Since we are working in laboratory coordinates, Azi vertical shifts are only perpendicular to the plane if px = py = 0 (horizontal). This plane can be characterized by a vector located at the origin normal to the plane with

Combining eqs. 1-3 and 5, the following expressions are found:

.real _rs»i , n r«ai , I "wf , „ . real' , _ , A (6a)

Both expressions combine the rotation and the translation of the sample before and after the movement (situations A and B, respectively).

The real coordinates of the specimen are to be determined from images taken before and after the movement (see Fig. 2). Therefore, in addition to the pure physical motion of the rigid body, there is also influence of the optics in the observed displacements. This effect is related to the distance between the sample and the optics along the optical axis, the so-called working distance (WD). As shown in Fig. 2, all points measured out-of-plane away from the center of the image (xo, yo) show a difference between the observed coordinates (xi, yi), and the real coordinates (xi, yi)real:

The magnitude of this distortion does not have a substantial influence in the measurements in SEM, since the working distance is much larger than the values of Az. In general, it is confined to the sub-pixel regime and negligibly small. However, since the present method is based on a comparison of images acquired before and after tilting the sample by small amounts, the

displacements are also in the sub-pixel regime, so the optical distortion may become relevant, in particular when comparing lower magnification images (i.e. large image size) before and after tilting. In such cases, the points close to the border of the image are strongly shifted, which causes important deviations. This effect can be corrected, as will be described with reference to Fig. 12. However, in the first examples only high magnification images are processed, where this effect is negligible and no correction for optical distortion is applied. An example of the application of such a correction for optical distortion will be described with reference to Figs.l3a)-13c).

Another optical effect that may be considered is a consequence of a translation tz during the movement of the sample. During the movement, the working distance can decrease or increase. As a result, the final image may be slightly expanded or contracted, respectively, by a factor ε in comparison to the initial image. In typical measurements this factor is close to 1, and its value can be calculated by evaluation of the change of the observed coordinates after a displacement in z direction only:

= (8a)

WD3 = WDA - tz (8c) In that case eq. 7a can be re-written, with help of eq. 8, as:

The combination of eq. 9b and eq. 9a leads to:

Since a factor ε can be defined as:

we can conclude that, by use of eq. 10 in eq. 11:

Therefore, the optical effect of the translation tz during the movement of the sample can be considered to be a uniform effect, while Azi « WD and tz « WD. For instance, at a working distance of 9 mm with an error of 0.1% (i.e. 9 pm), an uncertainty of 30 nm will be obtained in an image of 30x30 microns. If the image resolution is 1000 xlOOO pixels, then the pixel size is 30 nm, which is close to the uncertainty range. Actually, this value is quite large, since DIC displacements are calculated at sub pixel regime.

Application of the technique

In the following, application of the invention is explained with reference to three examples which illustrate different steps of the method according to the invention. Although these examples include SEM as image acquisition technique and DIC as displacement quantification technique, the present invention is not limited to methods including SEM or DIC, or its combination.

In the present examples, image acquisition was carried out with a Lyra (Tescan) FEG-SEM microscope operating at 10 kV and a working distance of 9 mm. The images (30x30 m2 and resolution of 768x768 pixels) were recorded in integration mode in order to minimize the presence of nonrandom noise. The sample was a steel sample with zirconia particles dispersed on its surface. Fig. 3 shows how the tilting axes were oriented in three "Cases" forming the present examples. In Cases I and II, the sample was tilted around a rotation axis of the sample stage only.

In Case I the images were taken with the x axis of the image essentially parallel (within experimental constraints) to the tilting axis. In Case II, the images were scanned in a direction at an angle of about 35° relative to the tilting axis (see Fig. 3). In Case III the sample was not tilted to different degrees about the same axis, but first tilted about an axis substantially in x direction and subsequently tilted about a second tilting axis substantially perpendicular to the x axis. In all cases, images have been captured with the sample tilted to several nominal tilt angles (θηοτη) between 0.5 and 12°). In each case, the images obtained after tilting were compared by DIC with the initial image before movement. The DIC process was carried out with Aramis 5.3 software, using facet sizes of 21 and 9 pixels for each (data) point and overlapping of 11 and 5 pixels, respectively. The DIC process was performed without activating any additional correction option in the software. The 3D reconstructions have been plotted with the WSxM software (I. Horcas et al., Rev. Sci. Instrum. 78 (2007) 013705).

In the following, first, the method for fitting the experimental data is presented. Next, the calculation of parameters for the three-dimensional reconstruction and two methods for finding a unique solution are described.

Fitting of experimental data

The influence of the optics considered in this example is a factor ε that multiplies the real coordinates after the movement in order to reach the observed ones. In that regard, the influence of the optical distortion can be introduced in eq. 6 by multiplication of one side of both expressions by ε (ε=1 in the other side). Therefore, eq. 6 can be re- written in the following form:

>ff,j = 5 A j + ft I' + ? 6 + *) where,

ft = ί(ΛΧΓ + Λ„Α ) (14a)

¾ = Β, + Λ^, ) (14e)

¾ = +¼) C14±)

The six terms {q} represent the rotation, translation and Optical expansion/contraction' of a flat plane without roughness (i.e. without z topography). All the random topographical information is stored in Az

Therefore, we can find the values of {q} by two fittings of {xi,e} and {yi,B} against their corresponding {xi,A, JIA) by using eq. 13. Nevertheless, the experimental data consists of displacements of the points (facets) in x and y directions obtained from the DIC procedure (S and <5y, respectively). These parameters can be calculated as:

&iJB = xi —xiTA (1 5a)

Therefore, eq. 13 can be re-formulated in 'DIC terms' instead of 'rigid

¾n = QA,A + Q5yirJ +Q6 + dtJtAzi (16b) where Qj = qh except that Qi = qi - 1 and Qs = qs ~ 1.

For each set of DIC data corresponding to a particular tilt, the values of {(¾ are calculated by fitting to eq. 16. Fig. 4a) shows an illustrative example of a map of DIC displacements in y direction from Case II and Fig. 4b) shows a

fitting performed using eq. 16b. In Fig. 4a), the displacements of the points appear distributed in a diagonal manner, in a direction perpendicular to the axis of tilt. However, as illustrated in Fig. 4b), the deviation of the

displacement of each point from the displacement of a corresponding point on the average plane described by the set of Q's depends on ε, on a rotation term and on Azu The quality of the fittings can be observed in Figs. 5a) to 5c). All plots show a slope of 1 and ordinate at the origin in zero. Larger ranges of variation of δχ and dy were obtained when larger tilt angles were used.

However, differences can be seen between plots of δχ and dy from the same Case. In Case I, the values of δχ are less scattered than the values of Sy of which dispersion increases with the tilt angle (the graphs of S and <5y are not of the same scale). Since in Case I, the tilt axis was substantially parallel to the x axis, all the topographical information has been transferred to the y axis only, which explains the difference of behavior. In fact, for <5y, the deviation of each point from the diagonal line is directly correlated to the value of its z coordinate. In the case of δχ, the deviations are mostly a consequence of noise caused by DIC and image processes. A different situation can be observed in Case II (cf. Fig. 4b). In this Case, the increase in dispersion with the tilt angle occurred in both plots. This is a consequence of tilting about an axis at 35° to the x-direction (see Fig. 3). Therefore, topographical information is expressed in deviation of each point from the diagonal line for both S and <5y. In Case III, the tilts about mutually perpendicular axes lead to different results. The tilt about the x-axis causes most of the dispersion and variation due to differences in z-coordinate values to be expressed in deviation of each point from the diagonal line for both <5y, while just noise appears in the direction of the x axis, as in Case I. The tilt about the y axis shows the opposite behavior, with most of the dispersion and variation due to differences in ^-coordinate values to be expressed in deviation of each point from the diagonal line for both δχ and only noise in the direction of the y axis.

The differences between the fittings to data that would have represented displacements of points on a perfectly plane surface tilted about the same angle and the real data is of particular interest, since the desired

topographical information can be derived from these differences. After eq. 16, we define:

Δ¾« = - ¾¾¾J + Qif A + 6i) = ¾» - ½i¾j + ¾J j + %}= ^Azf (17a)

The ratio is of interest as well, since:

where #> is the angle formed in the representation of Ady versus A&c. These plots are depicted in in Fig. 6a)-6c) for all Cases. Fig. 6a)-6c) show the differences between the corresponding data in Figs. 5a)-5c) and the diagonals (y=x). For Case I, the data form an almost vertical line, indicating that the topographical information is substantially stored in the y direction only, in agreement with tilting about an axis substantially in the x direction. In Case II, the data are dispersed substantially along a line deviating from the vertical, indicating that topographical information is stored not only in the y direction, but also in the x direction. In both Cases, the angle > is depicted (cf. eq. 18). In Case I, φ= 89,7°, i.e. almost vertical. In Case II, φ = 122,7°, i.e.

almost 35° deviated from the vertical, which is in agreement with the angle between the scanning direction and the tilt axis. In Case III, the differences found when tilting about the x axis are dispersed in substantially the y

direction (angle 88,7°) and the differences found when tilting about the y axis are dispersed in substantially the x direction (angle - 0,3°). Thus, the sample can be tilted about any angle, the values of φ angle automatically indicating in which direction or directions the differences representing the topographical information is to be found (eq. 18).

Calculation of parameters

For the calculation of the Azi data, the {q} or {(¾ sets obtained previously for evaluating the seven unknown parameters are used:

- four from rotation (<¾, ay, a∑ and Θ),

- two from the plane (p% and py) and

- one from the optical expansion/contraction (ε).

To reach that objective, we have six equations: (equations 4, 14a, 14b, 14d, 14e and 18). Therefore, one degree of freedom is left. (Equations 14c and 14d will not be used, since they include translations tx and ty which are also not known.) By combination of the six equations mentioned above, the following expressions can be found:

The angle Θ can be obtained from eq. 19 if az is known, as:

If £ equals 1, the three expressions in eq. 19 are not independent and the third equation does not provide any additional information. However, in any realistic setup this is not the case, and the value of ε can be calculated from parameters obtained from fitting after combining eqs. 19:

e = I q 5 cos φ— J sin φ)~ + ( 4 cosp— ql sin^)" (21)

At this moment, although one of the seven unknown parameters listed before has been calculated, there is still one degree of freedom in our system Therefore, a unique solution cannot be reached from two images alone, and

just a 'family of solutions' in terms of the ratio ax/ay can be obtained. In other words, we will get sets of permitted combinations of {<¾, ay, a∑, Θ, p , py}. If one of these parameters would be known, the others would be fixed immediately. The ratio a /ay can be calculated from any pair combinations from eq. 19:

a j, |f 4 cos — ( J— ε )sin fj]

Two solutions are obtained from each equation, depending on the sign of the square root, and therefore in total six solutions will be found. However, the correct one is the only one that is repeated three times, while the other ones are mutually different. After obtaining ε and ax/ay, there is one last degree of freedom. Values of Oz, Px, py and Θ can for example be calculated from eqs. 4, 14 and 20 after assuming a value for a or ay. The signs of az and Θ are linked by eq. 20.

Finding a unique solution

In practice, the sets of points within a 'family of solutions' form

corresponding 3D maps with different z scales. This degree of freedom can be eliminated by including a third SEM image C and (in addition to the DIC set AB) a DIC set AC representing the displacements of corresponding points between image A and a third image C. As depicted in Table 1, three

parameters are dependent on the starting situation A only: the plane parameters p and py and the sample topography Azi (see eq. 5). Therefore, by including another DIC set AC, we introduce only one more degree of freedom

similar to the one in the AB set, but we also introduce three additional constraints. As a consequence, we can eliminate all unknown parameters from the evaluations of the DIC sets AB and AC. In fact, a redundancy is obtained that allows two methods to reach a unique solution.

Thus, the actual rotation parameters can be calculated from the observed displacements by requiring that for at least two types of physical parameters of the sample, such as orientation of the sample plane in the first image and distances of corresponding points to the sample plane, the value resulting from the first set of displacements is identical to the value resulting from the second set of displacements. Accordingly, accurate calibration and control of the rotation is not required, and the need of complex calibrations has been circumvented. One method (hereinafter Method 1) uses only the plane parameters (pX,AB = p ,AC and pY,AB = Py,Ac) as physical parameters of the sample. In another method 2 (hereinafter Method 2), the topographical information (ΛΖΙ,ΑΒ = AZI,AC) is also used as a physical parameter of the sample together with a combination of the plane parameters.

Method 1

From corresponding 3D maps with different z scales, a unique solution for both tilts can be found by determining a solution for which the values of px and py agree for different tilts (i.e. a crossing point of plots of py vs. p for different tilts; in other words P ,AB= P%,AC, and pY,AB=py,Ac). Such

representations are depicted in the top graphs of Figs 7a)-7c). In each Case, the graph shows two crossing points, which actually represent the same solution for positive or negative reconstructions of the 3D topography. It can be seen that in Cases I and II the lines are quite close to each other and intersect at a very small angle. This entails a relatively large uncertainty in the determination of the crossing point. In contrast, the crossing of the lines of each pair is at a much larger angle in Case III, so that a much more accurate determination of the crossing point is obtained, assuming that the lines as such are of similar accuracy. This is a consequence of the number of axes of rotation about which the sample has been tilted. In Cases I and II, just one axis was used (i.e. identical values of φ ϊη Fig. 6 for different tilts), which leads to information regarding displacement in directions perpendicular to the single tilt axis only. Rotating the imaging conditions (i.e. Case II vs. Case I) does not make a big difference in this respect (cf. Case II in Figure 7), since also in this situation information is derived only from different amounts of displacement in a single common direction. Nevertheless, since typical sample holders for SEM apparatus only allow tilting about one axis of rotation, it is relevant for practice that also from images scanned from a sample in a starting position and from that sample tilted to different degrees about the same axis, a quantitative representation of the surface topography of practically

meaningful accuracy can obtained.

Method 2

While method 1 is based on information about the sample plane (py and Px) only, in method 2 one plane parameter (/½) and the topographical

information (Δζΐ) are used (see Table 1). Furthermore, while in method 1, the solution is found by comparison of independent sets of data from two tilts (AB vs. AC), in method 2 the information from both tilts is mixed. If two tilts are considered, we will find two values of φ that can be labeled as ψΑΒ and q>AC. The average of these angles can be defined as:


For the further analysis, a new set of coordinates (k, I, z) replacing (x, y, z) is defined as:

k = xc os ipjjc + y sin ¾JC (24a)

— xsni ψΛΒ + }'∞· %φ^1 (24b)

These new axes of coordinates are depicted in Fig. 6. The rotation vector can be expressed in the new set of coordinates with the use of eq. 24, i.e. (<¾, ai, a∑) instead of (<¾, ay, a∑). In a similar way, we can calculate the components of the plane in this new coordinate system, as:

Pk = Px cos φΜ€ + py sua <pMC (25a)

Pi = -Px sm <PMC + Py cosf^e (25'b)

In case of tilts around the same tilt axis or around similar tilt axes (Cases I and II), this change of coordinate system allows concentrating the topographical information in the direction of one of the axes, e.g. the k axis, and minimizing all information other than noise in the direction of the other axis, e.g. the I axis, as shown in Figure 6b). Thus, for each tilt we can get one expression equivalent to eq. 17 in the k direction by using eq. 24a (and in the I direction using eq. 24b), as:

Ai* J = Δ¾« eos¾jrc + A¾ «™if½jc C26a)

A* € = A*f c cos 9ABC + A¾JC s"19JBC (26B)

The overall deviations caused by the surface topography in the z direction can be expressed as the root-mean- square (RMS) deviation in the k direction, as:

where N is the number of points evaluated (number of facets), and A<¾LB is the average of the N A5ki,AB values. In this case, the average equals zero, as an ultimate consequence of the fitting procedure when calculating {Q}, which subtracts a fitted plane from the original data (cf. eqs. 16 and Fig. 5).

As illustrated in eq. 18, the topographical information is stored in the ratio of two rotation terms. In this case, the new coordinate system has been defined in such a way that most of the topographical information is in the k direction. In this, the theoretical requirement that ΛΖΙ,ΑΒ = AZI,AC (see Table 1) is converted into a practical relationship between the RMS of local deviation data from two different tilts:

ims( &M } _ R^m _ ia {l - em 0)+ ai ^0}M

RMS(ASkAC ) Rk AC {afcar (l - cos ^)+ a, sin d?) iC

This relationship between ΘΑΒ and 6AC is in addition to the relationship pk,AB = pk,AC A practical way to find a unique solution is the representation of a double-plot of/½ vs. 6AC. TO reach it, we calculate ε and ay/ax as before (see eqs. 21 and 22) and find functions ΡΗ(ΘΑΒ) and pk(0Ac) which depend only on the angle and known parameters. This is done through the subsequent use of the following equations: pu (eq. 25a), p and py (eq. 14), rotation terms (eq. 3), a∑ (eq. 19) and ay (eq. 4).

Finally, the relation between ΘΑΒ and 6AC (eq. 28) is used to define a function ΘΑΒ(ΘΑΟ), and to reach two independent relations that can be labeled as PII(QAC) and
These functions are shown in the bottom graphs of Figs 7a)- 7c).

By including the determination of the direction in which the largest variation in differences between the displacements and the fitting of

displacements is found in the fitting of displacements to fitting planes, the accuracy of calculated three-dimensional surface topography data is improved. More specifically, it can be seen that much less sharp crossing angles are obtained, particularly for Case II. This procedure is more accurate because it is based only on one of the parameters of the plane (/½), which is the one that contains the information when tilting using only one rotation axis.

Nevertheless, we can see that this method can be also used in Case III, where tilts were performed around two tilting axes oriented in substantially different directions. This is because, although in this situation not all the topographical information is expressed in the direction of the k axis (cf. Figure 6c)), there is still variation caused by topography in the direction of the k axis to achieve an accurate solution.

Both methods 1 and 2 are based on the insight that three connections link the two tilts from a starting position, as summarized in Table 1. By using the equations in a way similar to how it was described in the previous paragraph, it is possible to re-plot each curve in Figs. 7a)- 7c) as relations between both tilt angles 6AC and ΘΑΒ, as included in Fig. 8. Although this type of plot is illustrative for comparison purposes, it is inconvenient for

computation due to the appearance of many inconsistent solutions. Four lines appear in each plot. This is because pk, p and py are not independent (cf. eq. 28a). For Case I, the curves from pk and py are overlapping, since the axis k is very close to axis y (cf. Fig. 6a)). In addition, the information from p is very similar to that obtained from Azi (eq. 28). In the other two Cases, pu is between Px and py, while data from Azi appears different. Nevertheless, in all of the cases, the lines cross basically in the same point, and similar results are obtained from both methods (cf. Table 2).

Examples of 3D reconstructions of topographical information

In this section we will show examples of 3D reconstructions performed after the data treatment explained before. Some 3D mappings of Case I are shown in Figs. 9a)-9f. The original SEM picture is included in Fig. 9a) as a reference for comparison. Figure 9b shows the reconstruction obtained from the image tilted at θηοτη = 0.5° from the initial, using S data. It can be seen that this image does not correspond to the actual topography of the sample, and it is composed only by noise. This can be expected, considering that the topographical information is only included in the y axis (cf. Figure 6a)). In contrast, in the pictures 9c)- 9e) (based on Sy data) the observed features correspond very well to the image obtained by SEM. The image taken at θηοτη = 0.5° (Figure 9c)) appears noisy when compared with that obtained from θηοτη = 2.0° (Figure 9d)), which indicates that 0.5° was a too small tilt to get a good signal-to-noise ratio for the present sample in view of its roughness. A larger tilt angle is preferred for samples with a small roughness in order to improve the quality of the 3D reconstruction as is illustrated by Figs 9d) and 9e) which show the reconstructions from Onom = 2.0°. In Fig. 9e) smaller facet sizes were used during DIC than in Fig. 9d). For this scanned sample, this resulted in a better lateral resolution and a better agreement with the features appearing in the basic SEM image (Fig. 9a)). The additional detail shown in Fig. 9e) matches the qualitative information appearing from the basic SEM image (Fig. 9a)) very well and remains consistent with the more smoothed image of Fig. 9d). Therefore, use of the lowest facet size that allows DIC recognition is advantageous.

Figure 9f) shows histograms of the images presented in Figure 9c) to 9e). In addition, a histogram of an image obtained after Onom = 4.0° is also included. It can be seen that the histograms are very similar, in agreement with reconstructions of the same area. The histogram of the image obtained after Onom = 2.0° with small facet size appears is wider and lower than the others. This is because the improved lateral resolution leads to a larger set of z data, which increase the number of facets for which a z- value differing from 0 is found.

Figs. 10a)-10c) show 3D plots of the reconstructions obtained from Case II at Onom = 2.0°. In this case, different profiles obtained from x, y and k axes are shown. As opposite to Figure 9b) and 9c), the same features are identified from images obtained using δχ, or < y data (Figs. 10a) and b), since there is topographical information in both. In fact, this is the reason to define the new pair of working axis (k, I) to concentrate all the topographical information in one. The result using k axis is shown in Fig. 10c), which shows a profile matching the profiles shown in Figs. 10a) and 10b). Thus, consistency between methods 1 and 2 is verified. The histograms in Figure lOd) shown that also the z values determined in accordance with methods 1 and 2 are very similar.

In this particular Case, £=1.000036634. With such a small value, it would be tempting to approximate £ to 1. However, it is preferred not to apply this approximation since then the consistency between images suffers resulting in alterations of the z values by factors of 0.7 and 1.7. Therefore, preferably even very small variations in the working distance are taken into account during processing of image data in order to obtain accurate 3D images.

A schematic representation of an example of an algorithm for

performing a method according to the invention for obtaining a 3D

reconstruction of quantitative three-dimensional surface topography data of a sample is shown in Fig. 11. In steps 4 and 5, data representing three different images 1, 2, 3 (named A, B, C) scanned from the same sample surface held at different tilts are compared pairwise, resulting in two sets of DIC data 6, 7. In steps 8 and 9 these sets DIC data 6, 7 are each fitted to two fitting planes, resulting in data sets 10, 11 representing fitting parameters {q}i-6 and φ as well as the data δΔχ and 6Ay. From the fitting parameters 10, 11, the

expansion parameter ε is calculated in steps 12 and 13 and the ratio a /ay is quantified, leaving one degree of freedom in each set of fitted data 14, 15. In step 16, by combining the parameters 14, 15 obtained by fitting of the two sets of DIC data 6, 7 and compensating for expansion and/or contraction, unique solutions 17, 18 for each of the rotations AB and AC are obtained. From the known rotations, two 3D reconstructions 19, 20 are calculated (one per DIC). In other words, the uncertainty about the z scale that remains when

considering only one DIC set 6 or 7 is eliminated by adding a third image, so that the need of complex calibration procedures for determining the precise amount and axis or axes of rotation is avoided.

A method to correct the perspective distortions is described with reference to a flow chart of an example of an algorithm for performing this method, shown in Fig. 12. As indicated with reference to Fig. 2, the measured displacements as for instance obtained from the DICs result from two effects: the movement of the sample and the manner of observation of the sample. The perspective distortion can be corrected to obtain the real coordinates of the sample with the help of eq. 7. The process starts with the [x,y,z] data 19 or 20 obtained from the method shown in Fig. 11, which corresponds to an

observation of the sample in an orientation as in situation A (for instance, an average of both 3D profiles 19, 20 is used). Then, in step 21, eq. 7 is used to calculate a first iteration of the real coordinates 22 connected to the observed coordinates in situation A.

The further steps are for checking the accuracy of the iteration and initiating a further iteration if the accuracy does not meet a predetermined level. Steps 23, 24 consist of calculation of the real coordinates 25, 26 in the tilted positions B and C (images 2 and 3 in Fig. 11) corresponding with the first iteration 22, using the movement parameters 17, 18 (see Fig. 11) obtained before. Next, in steps 27-29, the perspective distortion for the three different situations is calculated, resulting in new sets 30-32 of coordinates for the monitored points for the situations A, B and C. In steps 33, 34, new sets of displacement data 35, 36 are determined from these coordinates by

subtraction. The new sets of displacement data can be regarded as a 'new DIC. In decision step 37, if this 'new DIC matches the observed DIC to a sufficient extent, the calculated set of the real coordinates 22 is accepted as being correct. If not, the 3D algorithm (i.e. starting with steps 8 and 9) shown in Fig. 11 is employed on the new [x,y, dx,dy] values to obtain new sets of [x,y,z] according to steps 19 and 20. This cycle continues until a predetermined extent of agreement between the 'new DICs' and the DIC obtained directly from the images has been achieved.

An illustration of the effect of the correction for perspective distortions is depicted in Figs. 13a)-13c) in which Fig. 13a) shows an image with a larger field of view than the image on which the previous examples of application were based. Its 3D reconstruction with correction for perspective distortion is shown in Fig. 13b) and the perspective correction used to reach that

reconstruction is depicted in Fig. 13c). The 'U' shape of the perspective correction in the y axis is a consequence of the tilting around a rotation axis close to the x axis. It can be seen that, although the perspective correction is smaller than the topography of the sample, its magnitude is important in certain situations.

While, for an easier understanding, the invention is described in the detailed description with reference to examples of its application, the scope of the invention as defined by the appending claims is not limited to these examples. As illustrated in Figure 11, the acquisition of a third image at a different tilting angle provides the information on the basis of which tilting angles and axis can be derived from the images only. However, the accuracy can be improved by inclusion of more images acquired at other angles.

Which angle provides the best results in terms of resolution and accuracy of determined z- values depends strongly on the roughness of the sample, smaller angles being more suitable for acquiring surface topology data from rougher samples. While larger angles allow an improved identification of small roughness, angles should not be too large in order to avoid problems with facet recognition during the DIC process. For instance, for the type of sample used in this example, an approximate tilt range would be between 1 and 15 degrees. No previous information about the SEM or other microscopic equipment is needed, and the 3D map can be just reconstructed from these data by software (DIC and calculation of rotation parameters) in period of time of typical user operation. The images can be acquired in various manners, for a useful determination of surface topography, variations in tint and or color in the images should allow corresponding points in the images to be identified with a sufficiently fine distribution to recognize topographical features, such as peaks, ridges and valleys, with a resolution that is sufficient for the desired level of detail at which the surface topography is to be determined.

Several features have been described as part of the same or separate embodiments. However, it will be appreciated that the scope of the invention also includes embodiments having combinations of all or some of these

features other than the specific combinations of features embodied in the examples.

Tables ;

Table 1. Tilt, linages, DIC processes, and criteria to find a unique solution.

Table 2. Angles obtained from both calculations methods for each Case under study.

ΘΑΒ (dee) ΘΑΟ (deg)


Nominal Method 1 Method 2 Nominal Method 1 Method 2

Case I 4.0 4.832 4.754 2.0 2.443 2.341

Case II 4.0 3.973 3.900 2.0 2.037 1.973

Case III 12.0 11.724 1 1.997 ca. -13.5 -13.764 -13.281