BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to an apparatus and method for measuring a thin film or layer thickness and, more particularly, to an electro-optical system which measures the thickness of an outer silicon layer of a silicon/silicon dioxide/silicon(Si/SiO.sub.2 /Si) structured semiconductor wafer.

2. Description of the Prior Art

In one particular application wherein the present invention is especially practical, a silicon-on-insulator (SOI) semiconductor wafer typically includes a Si/SiO.sub.2 /Si sandwich structure fabricated by growing a silicon dioxide film on onesurface of each of two silicon wafers and bonding the two silicon dioxide film surfaces together at high temperature. It will be understood that other materials such as, for example, silicon nitride, may be used for the insulator material and that othermaterials may be used for the wafer material. In such an application, one of the two outer silicon surfaces of the sandwich structure is mechanically ground and polished to an average thickness of several microns. This mechanical process unfortunatelyresults in large spatial variations in the thickness of this outer silicon layer over the surface of the wafer. To reduce these spatial variations, a thickness error map that indicates thickness non-uniformities of this outer silicon layer over theentire wafer surface, is required, for example, to initialize a further micropolishing process.

A sequence of measuring the spatial variations in the thickness of the outer silicon layer followed by thinning and smoothing this surface by micropolishing may need to be performed several times before the entire outer silicon layer achieves thedesired thickness. In order to reduce costs and increase production, a measurement of at least 400 points on a wafer surface in 60 seconds is desirable.

Current commercial instruments, however, typically provide film thickness measurements at only a single point on a surface. These instruments use a focused lens or a fiber bundle to locally illuminate the film surface with a beam ofmonochromatic light, and a grating or prism spectrograph to measure the surface spectral reflectance at each point. In all cases, this surface spectral reflectance data must be numerically corrected due to variations in the angle of incidence caused bythe illuminating beam f-number.

These commercial instruments may be extended to cover an entire wafer surface by moving either the measuring instrument or the wafer in a controlled manner. However, the time required for these instruments to determine the thin film layerthickness at a single point is on the order of several minutes and characterizing an entire film surface of at least 400 measurement points far exceeds the time desired for efficient wafer production.

SUMMARY OF THE INVENTION

The present invention contemplates an electro-optical imaging system for efficiently determining a thin film layer thickness of, for example, a wafer over a full aperture. Non-uniformities in this layer thickness are obtained by measuring thereflectance characteristics for a full aperture of a wafer surface and comparing this measured reflectance data to reference reflectance data by using numerical iteration or by using a calibration wafer having known layer thicknesses.

To efficiently measure the reflectance characteristics of a wafer layer, a filtered white light source is used to produce a sequence of collimated monochromatic light beams at several different wavelengths. These collimated monochromatic beamsare individually projected onto the entire surface of the wafer, and coherent interactions occur between this light as it is reflected from the physical boundaries in the wafer structure. As a result of these interactions an interference fringe patternis formed on the surface of the wafer for each projected beam and, consequently, for each wavelength. A reflected image of each fringe pattern is projected onto a detector array of, for example, a charge coupled device (CCD) camera, where the fullaperture of this image is then captured. The fringe pattern image is captured by digitizing pixels in the CCD camera detector array corresponding to the image present. A reflectance map of the entire wafer surface is generated from this captured fringepattern image. Several reflectance maps are generated from each measured wafer to eliminate thickness ambiguities which may result from outer layers having phase thicknesses greater than 2.pi..

The reference reflectance data for a wafer may be obtained by a theoretical calculation or through the use of a calibration wafer. The theoretical method consists of numerically computing reference reflectance characteristics based on assumedvalues of the intrinsic optical properties of the wafer materials. Alternatively, a calibration wafer, having a known thickness profile, may be constructed from the same batch of materials used to construct the wafer to be measured. By subjecting thiscalibration wafer to the measuring method of the present invention, reference reflectance data is obtained for the known wafer.

The comparison between the measured reflectance data and the reference reflectance data can then be performed by a computer. Upon performing this comparison, the computer can provide a mapping of layer thickness or a mapping of layer thicknessnon-uniformities over a full aperture of the wafer.

A primary objective of the present invention is to provide an economical means for measuring the outer layer thickness of a wafer surface.

Another objective of the present invention is to provide an efficient means for measuring layer thicknesses.

Another objective of the present invention is to provide a highly accurate means for measuring layer thicknesses.

Another objective of the present invention is to provide a means for measuring wafer layer thicknesses without mechanically scanning the wafer surface.

Another objective of the present invention is to provide a means for measuring layer thicknesses of thin film structures in general, provided that the structure materials are known and a sufficient number of wavelengths are used.

A still further objective of the present invention is to provide an economical means for measuring the outer silicon layer thickness of a SOI semiconductor wafer surface.

Another objective of the present invention is to provide an efficient means for measuring outer silicon layer thicknesses.

Another objective of the present invention is to provide a highly accurate means for measuring outer silicon layer thicknesses.

Another objective of the present invention is to provide a means for measuring SOI semiconductor wafer outer silicon layer thicknesses without mechanically scanning the wafer surface.

DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic representation of a wafer layer thickness measuring instrument.

FIG. 2 is a cross-sectional view of a SOI semiconductor wafer.

FIG. 3 is a plan view of a CCD camera detector array showing a wafer image outline and several reference surface image outlines.

FIG. 4 is a plan view of a calibration wafer having a stepped outer layer surface.

DETAILED DESCRIPTION OF THE PRESENT INVENTION

An electro-optical system for measuring a layer thickness of a wafer 24 is shown in FIG. 1. For the purposes of this description, the measurement of an outer silicon layer of a SOI semiconductor wafer 24 is described.

The present invention provides a white light source comprised of a circular aperture 14 illuminated by a halogen lamp 10 and a condensing lens 12. Light passing through aperture 14 impinges on a collimator lens 16 to form a beam 15 of collimatedlight. The size of the aperture 14 determines the field angles in the collimated light sections of the optical system and the orientation is chosen to allow an aperture image to be projected onto the SOI wafer 24. It should be noted that the condensinglens 12 may be replaced by a fiber optic light guide.

The white light source is spectrally filtered by a series of narrow band filters 17, nominally of 30 to 50 .ANG. half bandwidth, placed in the collimated beam 15. The series of filters 17 are placed around the periphery of a rotating filterwheel assembly 18, whereby a corresponding series of collimated monochromatic light beams 19 are produced. The wavelengths of these collimated monochromatic light beams 19 range from 550 nm to 950 nm. Locating the filter wheel assembly 18 in acollimated light section 15 minimizes the spectral broadening of the filtered beam 19 caused by the field angle defined by the size of the aperture 14. A pair of electronic signals 32 are generated by the filter wheel assembly 18 to serve as a timingreference 33 for a digitizing circuit 34. One of these signals indicates the beginning of a filter wheel revolution, whereas the other signal indicates the beginning of each filter period.

A second collimator lens 20 forms a monochromatic image of the aperture 14 about a point 21 in a focal plane of a third collimator lens 22. This third collimator lens 22 produces a collimated beam 23 which illuminates the full aperture of the100 millimeter diameter SOI wafer 24. Also, an extension of this wafer illumination technique to wafers of 150 millimeters or 200 millimeters in diameter requires that the size of the third collimator lens 22 match the wafer size. It should be notedthat a monochromator can replace the halogen lamp 10, the condensing lens 12, the first two collimator lenses 16, 20, and the narrow band filter wheel 18, provided that the slewing rate of the monochromator between different wavelengths is sufficientlyhigh, up to twenty different wavelengths in less than one second.

Referring to FIG. 2, a cross-sectional view of a SOI semiconductor wafer 24 is shown. This wafer 24 is constructed in a sandwich structure consisting of a mechanically polished outer silicon layer 40, an internal silicon dioxide (SiO.sub.2)film42, and a silicon wafer substrate 44. This sandwich structure creates three interfaces 46, 48, 50 from which light, incident upon the outer silicon layer 40, may be reflected. The reflectance characteristics of these interfaces 46, 48, 50 are basedupon the intrinsic optical and physical properties of the semiconductor materials in each layer 40, 42, 44 of the SOI wafer 24. These properties consist of the absorption coefficient, .alpha., the index of refraction, n, and the thickness, t, of thematerial layers 40, 42, 44. For an SOI wafer, it is assumed that the absorption coefficient, .alpha., of the SiO.sub.2 layer 42 is zero. However, in general it is permissible that the absorption coefficient be non-zero, provided that it is known.

When the surface of the SOI wafer 46 is illuminated with collimated monochromatic light from beam 23, a series of coherent interactions occur as this light is reflected between the three material interfaces 46, 48, 50 of the SOI structure 24. These interactions produce a wavelength dependent interference pattern that is visible upon the surface of the wafer. The reflectance at any point on the wafer is determined by the multiple reflections between the three surfaces and by the magnitudes oftheir physical properties, n.sub.1,.alpha..sub.1,t.sub.1, and n.sub.2,.alpha..sub.2,t.sub.2 as well as the properties of the substrate n.sub.8,.alpha..sub.8. In the unique case of an SOI wafer structure, the substrate indices are identical to those ofthe outer film indices (n.sub.8 =n.sub.2,.alpha..sub.8 =.alpha..sub.2) since both are fabricated from single crystal silicon. The wafer reflectance at any wavelength can be calculated explicitly as a function of the outer film thickness if all the otherparameters are known, however, the reverse problem of computing the thickness from a single measured reflectance is ambiguous. This ambiguity is created by the fact that as the outer film thickness is increased, the measured reflectance cycles betweenmaximum and minimum values as the phase thickness (n.sub.2 t.sub.2) increases by .pi./4. This multi-valued problem clearly makes the computation of the value of t.sub.2 from a single reflectance measurement impossible. The use of multiple wavelengthmeasurements can in principle overcome the multiple value problem but the wavelength dependent behavior of the material properties must be very accurately known otherwise large errors occur in the thickness computations.

An alternate approach is a statistical one where measured reflectance data at several wavelengths is compared on a least squares best fit basis with a library of computed spectral data at the same wavelengths. In the case of an SOI wafer, thelibrary of spectra is computed for all values of the outer film thickness and the selection is made by choosing that outer film thickness which minimizes the least squares fit.

Referring back to FIG. 1, a collimated light image of the interference fringe pattern is reflected off the surface of the SOI wafer 24 and returned through the third collimator lens 22. This third collimator lens 22 projects a condensed image ofthe reflected fringe pattern upon an off-axis mirror 26. This mirror 26 is positioned at a point 25 in the focal plane of the third collimator lens 22, alongside the position of the aperture image at focal point 21. The separation of these two focalpoints 21, 25 may be controlled with a slight lateral shift in the optical axis of the third collimator lens 22 with respect to the optical axis of the condensing lens 12 and the first two collimator lenses 16, 20. Equivalently, the wafer 24 may betilted through a small angle, less than one degree, to achieve this same effect. This image separation scheme avoids the use of a beamsplitter which contains metallic coatings with attendant optical losses.

The off-axis mirror 26 is used to redirect the reflected fringe pattern image from the wafer 24 to a final collimator lens 28. This final collimator lens 28 projects a collimated beam 29 containing an image of the fringe pattern onto a CCDcamera detector array 31. It should be noted that the filter wheel assembly 18 may also be placed in this collimated beam 29 provided that the field angle, which is approximately fifteen times larger than the field angle in the collimated beam 23illuminating the wafer 24, can be tolerated by the narrow band filters.

An alternate method of providing the reflected fringe pattern image to the CCD camera detector array 31 is shown in a dashed line block 61 in FIG. 1. An on-axis beamsplitter 60 is placed in the collimated light beam section 19 where the filterwheel assembly 18 is positioned. The beamsplitter 60 receives a collimated fringe pattern image from the second collimator lens 20 and reflects a portion 62 of this collimated beam to a final collimator lens 64. This final collimator lens 64 convergesthe fringe pattern image onto the CCD camera detector array 31. Although this alternate method results in optical losses which are inherent in beamsplitters, it does not require an image separation scheme which can introduce field angle errors in thecollimated light beam 23 reflected from an off-axis SOI wafer 24. As with the previous method, the filter wheel assembly 18 may be placed in the collimated beam 62 reflected by the beamsplitter 60, provided that the field angle can be tolerated by thenarrow band filters 17.

The determination of the method used to provide the reflected fringe pattern image to the CCD camera 30 is critically dependent upon tile optical performance of the third collimator lens 22. When using the off-axis mirror method, the opticaldesign of the third collimator lens 22 must possess an optimal off-axis performance quality and provide a minimal radial color distortion effect. Optimal off-axis performance minimizes the distortion effects associated with field angles that are createdwhen the collimated light beam 23 is reflected from an off-axis non-uniform surface of a SOI wafer 24. Also, the need for a consistent fringe pattern image size at the CCD camera detector array 31 requires radial color distortion correction over thewavelength region of the incident monochromatic light. When the on-axis beamsplitter method is used, however, only the radial color distortion correction requirement applies since the field angles produced in the collimated light beam 23 reflected froman on-axis SOI wafer surface 46 are negligible. Therefore, if the returned fringe pattern image is distorted due sub-optimal off-axis performance by the third collimator lens 22, then the off-axis mirror method is unsuitable and an on-axis beamsplitter60 must be used.

Referring to FIG. 3, the CCD camera detector array 31 is shown with an image of a scaled SOI wafer outline 52, a pair of reference alignment images 54, and a pair of reference reflecting images 56, projected upon its surface. These referenceimages are formed by placing reference alignment marks and reference reflecting surfaces along the same plane as the surface of the SOI wafer 24. When illuminated with a collimated light beam 23 from the third collimator lens 22, these referencesprovide reflections from their surfaces. Similar to the SOI wafer fringe pattern, images of these reflected references are returned through the third collimator lens 22 and are eventually projected upon the CCD camera detector array 31. The referencealignment marks provide aid in wafer alignment, whereas the reference reflecting surfaces serve to normalize the CCD signals so that actual wafer reflectances can be calculated.

Referring back to FIG. 1, the collimated beam 29 formed by the final collimator lens 28 contains an image of the reflected fringe pattern. This image is projected upon the CCD camera detector array 31 and captured by the CCD camera 30. Areflectance map is generated by digitizing the CCD pixel signals corresponding to the projected fringe pattern image with a digitizing circuit 34. This raw reflectance data may be normalized to eliminate variations in pixel sensitivity and may bereduced in size by averaging signals over blocks of pixels to match the spatial limitations of the subsequent chemical micropolishing process. In determining the thickness, t.sub.2, of the outer silicon layer of the SOI wafer 24, either a numericalcomputation method or a SOI calibration wafer may be used. Both of these methods require the use of a computer 36.

The numerical method of determining outer silicon layer thickness, t.sub.2, consists of assuming values for the thin film constants n.sub.1, .alpha..sub.1, t.sub.1, n.sub.2, .alpha..sub.2, n.sub.3, and .alpha..sub.3 and calculating spectralreflectances for a set of wavelengths corresponding to the monochromatic light produced by the filtered white light source. This calculation is done for a number of different outer layer thicknesses, t.sub.2, and provided that the initial thin filmconstant assumptions are correct should only need to be computed once. This calculation provides sets of reflectance values, R.sub.c (.lambda..sub.1, .lambda..sub.2, . . . .lambda..sub.n,t.sub.2) for thicknesses ranging from the thinnest t the thickestestimated value of the outer silicon layer. These computed spectral reflectances are then compared with the measured reflectance data, R.sub.m (.lambda..sub.1, .lambda..sub.2, . . . .lambda..sub.n,t.sub.2), at specific points on the wafer using a rootmean square (rms) merit function of the form ##EQU1## This merit function is evaluated for different values of t.sub.2 until a minimum or best match is found, which in turn indicates the most likely thickness. It is, of course, apparent that otherpattern matching merit functions can be used, if desired.

Unknown variations in any of the assumed thin film constants may cause errors to propagate through the computation process as outer layer thickness errors. Such first order error sources include the lack of knowledge of the SiO.sub.2 inner filmthickness, t.sub.1, over the wafer aperture and the dispersive effects of the silicon index of refraction, n.sub.1. If the value of the merit function is too large, indicating a poor match, then new computed spectral reflectances will have to begenerated for a closer set of t.sub.2 thicknesses, iterated with the absorption coefficients .alpha..sub.2, .alpha..sub.3, and the indices of refraction n.sub.2, n.sub.3, of the outer silicon layer 40 and the silicon substrate 44, respectively, or theindex of refraction, n.sub.1, and the thickness, t.sub.1, of the SiO.sub.2 layer 42.

The second method of determining outer silicon layer thickness is to generate a set of spectral reflectances from a SOI calibration wafer 58 containing a stepped outer surface, as shown in FIG. 4, where each square area 59 of the wafer has adifferent known outer silicon layer thickness. It is preferred that the wafer have at least 500 reference squares to cover the likely range of outer layer thicknesses. This wafer 58 can be calibrated for outer silicon layer thicknesses down to zerosilicon thickness by using a stylus profilometer. The calibration wafer 58 is manufactured from the same batch of materials and is subject to the same manufacturing conditions as the wafer 24 under measurement. Consequently, the intrinsic opticalproperties of the calibration wafer such as the absorption coefficient and the index of refraction, and also the dispersive behavior should match those of the wafer 24 being tested.

The spectral reflectances of the calibration wafer 58 are obtained by subjecting this wafer to the measuring method of the present invention. These spectral reflectances are stored in the computer 36 and are used as a reference for comparisonwith the spectral reflectances of a measured SOI wafer 24. The calibration wafer 58 has reference reflectances for at least 500 different outer silicon layer thicknesses and the reference reflectance that most closely matches the reflectance at anypoint on the surface of the measured SOI wafer 24 indicates the outer silicon layer thickness at that point.

Although both of these methods meet the objective of providing a 400 point measurement of outer silicon layer thickness in less than 60 seconds, the calibration wafer method is potentially more accurate than the numerical approach due to thesimilarity of the intrinsic optical properties between the calibration wafer 58 and the SOI wafer 24 to be measured. However, the calibration wafer method would require on-line production capability to generate a new calibration wafer 58 if the basicwafer process is significantly altered.

* * * * *