Coherence radar - new modifications of white-light interferometry for large object shape acquisition G. Ammon, P. Andretzky, S. Blossey, G. Bohn, P.Ettl, H. P. Habermeier, B. Harand, G. Häusler Chair for Optics, University of Erlangen, Staudtstraße 7/B2, D-91058 Erlangen tel: +49 9131 858382, fax: +49 9131 13508 email: haeusler@physik.uni-erlangen.de ABSTRACT The coherence radar is a 3D-sensor based on interferometry with a broad bandwidth light source. This sensor principle combines the advantages of absolut measurements with a low measurement uncertainty limited only by the surface roughness of the object under test. We introduce new sensor modifications which enable us to measure large objects (>100mm x 100mm), to measure the difference between an object under test and a master object and to measure strongly curved polished surfaces (e.g. aspheric lenses). We demonstrate a modification which reduces the measurement time considerably. We discuss new measurements for different applications as turbine blades, silicon, solar cells and circuit boards. 1. Introduction White light interferometry is a well known measuring principle [1]. Its usefulness was first demonstrated in microscopic setups [2-3] for measuring smooth object surfaces. The coherence radar [4] is able to measure rough object surfaces, as well. We introduce new powerful modifications of the coherence radar. The new modifications allow us to use the coherence radar as a tool for measuring large objects (>100mm x 100mm) with 1µm accuracy and telecentric imaging the difference between an object under test and a master object strongly curved polished surfaces layered surfaces and translucent surfaces like ceramics and skin coaxially, with extremely small aperture, without sacrificing longitudinal accuracy with improved speed of up tp 100µm/s in depth polished surfaces with distance uncertainty of less than 10nm 2. Measurements with the coherence radar The following results are some examples of measurements recently made with the coherence radar. Fig. 1 shows a 3D plot of a small turbine rotor blade. In additional measurements we could measure the edges of the blade.
Fig. 1: Part of a turbine rotor blade A scratch in a metal surface made by a tool is shown in fig. 2. The coherence radar can measure layered objects as well. Fig. 2: Inverted 3D plot of a scratch (height: 100µm) For Example, bubbles (missing glue) between the silicon surface and the protection glass of a solar cell can be measured, as displayed in fig. 3.
Fig. 3: Silicon surface of a solar cell, intensity encoded height map, and cross section Fig. 4 shows copper structures on a ceramic layer (volume scatterer!) [5]. Fig. 4: Copper structures on a ceramic layer 3. Interferometry with optical rough reference Acquisition of surface topology with optical sensors will be more and more important. We introduce a modification of the coherence radar that can measure large surfaces. This modification of our sensor uses an optically rough surface (ground glass) instead of the reference mirror [6]. Therefore our reference signal is mainly based on scattered light. This enables us to work with a divergent light beam and receive even a signal from object points far beyond the lateral limits of the beamsplitter. We can measure objects which are larger than the beam splitter dimensions, as will be shown in section 4. Fig. 5 shows the optical setup that is used for the measurement of objects much larger than the beamsplitter cube.
Fig. 5: Experimental setup with divergent beam It should be mentioned that although the object is larger than the optical elements, the sensor displays perfect telecentric imaging. Fig. 6: Intensity encoded height map of a turbine rotor blade In our experiments the beamsplitter has the size of 50 mm x 50 mm. Our rough reference has the size of 100 mm x 100 mm with a roughness σz of 2.5µm. By now the maximum measurable object size is 15 cm x 15 cm which is limited only by the available light power. The object under test was a turbine rotor blade (80 mm x 145 mm). An intensity encoded height map of the lower part of the blade can be seen in figure 6. To the left and right side of the blade there is some noise from the background of the picture. Fig. 7: Profile through the height map of figure 6 Fig. 7 displays a profile through the height map of the rotor blade. Here you can see the dimensions of
the object and its depth. Instead of the rough reference plane it is possible to use a master object. The reference plane will then be deformed to match the topology of the master object. The comparison of two identical objects results in a planar height map. Deviations of the test object from the master object are visible as deviations from the ideal plane. Master-test measurements are much faster because only the differences have to be scanned in height, not the total object. 4. White light interferometry of strongly curved polished surfaces With a classical interferometer strongly curved polished surfaces cannot be measured. The interference fringes get too narrow to be detectable and the reflected light misses the observing aperture. We overcome these limitations by using a white light interferometer with an optically rough reference plane and a divergent light beam. With the optical setup shown below (fig. 8) it is possible to measure objects, for example concave curved lenses, and aspheric lenses, without proper reference mirrors. Fig. 8: Optical setup with rough reference By using a rough reference plane we do not get interference fringes but a speckle pattern that we can detect with our coherence radar. With this setup we are able to measure strongly curved surfaces that usually cause narrow fringes which are not detectable in an interferometer. With this device a polished lens surface has been measured. Fig. 9 shows an intensity encoded height map of the measured lens (r = 1,25cm). Fig. 9: Intensity encoded height map (the brighter the deeper) A profile through the 3D map of fig. 9 is shown in fig. 10.
Fig. 10: Profile along the line shown in figure 9 Another application of this system is the detection of faults in non-planar surfaces. One example is the search for hair line cracks in silicon wavers. 5. Dynamical coherence radar The coherence radar is a scanning white-light-interferometer (SWLI). As the object is shifted through the reference plane or vice versa, each pixel of the observing camera watches a white-light correlogram as shown in fig. 11. Fig. 11: White light correlogram If the scanning speed v is low, the integration time T of the camera may be neglected. At higher speeds, however, each pixel integrates over large parts of a modulation period or even over several periods. This weakens the detected signal and the evaluation of the correlograms is no longer possible at higher speeds. Fig. 12 illustrates this problem by showing the Fourier transform of a correlogram and the envelope of the transfer function HCCD of a camera pixel as a function of vt (dashed lines). As vt, i.e. the distance traveled per video-cycle, is increasing, the detected signal is decreasing. Even if a high-speed camera is used (short T), the maximum speed of measurement cannot exceed 10 µm/s. Fig. 12: Correlogram in the Fourier domain There has not been a solution to this problem so far, i.e. all SWLIs scan quasistatic or very slow in order to minimize the influence of the integration time T. We are introducing a new method for evaluating the correlogram much faster [7]: The method is based on the fact that we are not interested in the modulation of the correlogram but in the position of the maximum of the envelope, since the modulation itself does not carry any information, if rough objects are measured! Hence, the correlogram I(z) is an amplitude-modulated (AM) signal with a DC offset. In communication theory many demodulating techniques for AM-signals are known, one of which reads as follows: An AM-signal only has to be modulated by a carrier of the same frequency in order to get the desired envelope. This can easily be applied to the coherence radar: If the light-source is modulated at the correlogram frequency, the output function of the light-source can be displayed in the Fourier domain as shown in fig. 13. Both spectra (figs. 12/13) must be convolved in order to obtain the spectrum of the signal which hits the CCD-target. Fig. 14 shows the result: The Fourier transform of the envelope is not influenced by HCCD, whereas the disturbing parts at higher frequencies are suppressed. Thus only the desired envelope is extracted.
Fig. 13: Output of the light source Additionally, the piezo commonly used in SWLIs, can now be replaced by a motor-driven linear stage, which helps to largely extend the longitudinal range of measurement. Fig. 14: Demodulated signal in the Fourier domain The new method allows speeds of measurement of several 100 µm/s at an accuracy δz which is proportional to the speed: δz = γ vt, where γ [ ; [ 0 1 only depends on the postprocessing. A speed of 50 µm/s at δz = 2 µm has already been achieved and a speed of 500 µm/s at δz = 0. 4 µm appears to be possible by using a high-speed camera and more sophisticated postprocessing techniques. 6. Conclusions The use of a rough reference plane instead of a mirror enables the measurement of objects larger than the optical elements and curved polished surfaces. The direct measurement of deviations between object under test and master object is possible, as well. In connection with the enhanced measurement speed the coherence radar is a widely useful tool for inspection of technical surfaces. The measurement uncertainty is only limited by the the roughness of the object - typically a few micrometers. This project is funded by BMBF (13N6667). References [1] A. A. Michelson, Determination experimentale de la valeur du mètre en longueurs d ondes lumineuses, Trav. Mem. Bur. Int. Poids Mes. 11, 1-42 (1895) [2] B. S. Lee, T. C. Strand, Profilometry with a coherence scanning microscope, Appl. Opt. 29, pp. 3785-3788 (1990) [3] G. S. Kino, S. S. C. Chim, Mirau correlation microscope, Appl. Opt. 29, p. 14 (1990) [4] Th. Dresel, G. Häusler, H. Venzke, Appl. Opt. 31, pp. 919, March 1992 [5] B. Harand, Diploma thesis, University of Erlangen-Nürnberg (1996) [6] P. Ettl, Diploma thesis, University of Erlangen-Nürnberg (1995) [7] S. Blossey, Dissertation thesis, University of Erlangen-Nürnberg (1996)