In this investigation, we propose a technique to obtain not only the dimensional surface profile but also tilt information of the rough dielectric surface having a few microns root-mean-square roughness. This technique is based on low coherence scanning interferometry (LCSI) using a compound light source by combining a superluminescent light-emitting diode with ytterbium-doped fiber amplifier. Tilt angle and direction of the measured surface is extracted by the principal component analysis (PCA) from the measurement surface data and the centroid peak detection algorithm. To verify the performance of the proposed tilt measurement method, standard angle gauge block and certified step height sample were used as specimens. LCSI tilt measurement was about 3 times superior to the conventional auto-collimator in terms of the measurement precision in the practical camera module manufacturing process of smartphones. The proposed method was also discussed the dynamic tilt evaluation for the moving object.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Non-contact angular measurements are important in various fields such as three-dimensional sensing in camera modules, vision ADAS (advanced driver assistance system) of cars, mirror monitoring in telescopes, optical alignment in lithography, etc. Two typical methods used in angle measurements are auto-collimator technique and laser interferometry. The auto-collimator is a sensitive angle measuring device and is thus traditionally used for the precision adjustment of optical or machine components. Due to the infinity adjustment of the collimated beam the measured angles are independent from the distance to the object under test. Michelson interferometry is another established method for an angle measurement . By the variations of optical path difference (OPD) between reference flat surface and specimen interference fringes change. In the present there are two main categories in Michelson interferometry. One is the homodyne interferometer using single frequency . Another is the heterodyne interferometer using two different frequencies for more accurate measurements with the phase information carried on an AC signal . The main drawback in the above two accredited angle measurements is that the measurement is confined to the reflective specimen surfaces. If the object under test is a rough surface, auto-collimator and laser interferometry cannot directly measure the angle. In addition, if we put or attach mirror-like objects such as corner cubes or flat mirrors on the rough surface to measure angles of the surface, there may be additional tilts unintentionally.
Recently, rough surfaces have been measured using low temporal coherence light sources such as white-light sources, light-emitting diodes (LED), superluminescent light-emitting diodes (SLD) and supercontinuum light sources by means of scanning interferometry [4–6]. This technique removes the well-known 2 π-phase ambiguity problem in spatial phase unwrapping processes at phase shifting interferometry (PSI) having monochromatic or quasi-monochromatic light sources, namely long coherence length . However, angle measurements for the entire surface of the rough dielectric object have not been yet reported.
In this paper, we propose a tilt metrology method based on low coherence scanning interferometry for the dielectric rough surface. This method utilizes a principal component analysis (PCA) algorithm which is from statistics for reducing data set to get tilts from the measured three-dimensional surface data . Measured tilts with the proposed method for the traceable angle gage standards and practical products such as smartphone camera modules are compared with those by the well-established auto-collimator technique. This proposed method is found to be reliable and especially useful to the rough surface where auto-collimators cannot be directly applied to. For reference, in this paper tilt expresses the small angle ranges which are smaller than 1 degree .
2. Low coherence scanning interferometer for a tilt metrology
The assembly completeness is important in an industrial field. In general, auto-collimators are used to measure angular displacements, flatness, and parallelism as a noncontact method. Figure 1(a) shows the usage of auto-collimator for the specular surface. We can know the degree of tilt of the target surface from the displacement of the spot on the 2-dimensional detector inside the auto-collimator reflected from the incoming collimated beam. The target is usually a mirror-like surface such as glasses and polished metals. The process control for the tilt of components inside module is important because the state of the optical components affects the optical axis of the lens assembly so that the whole optical performance of camera module in electronic communication devices such as smartphones is changed easily. Most surfaces except the camera sensor and lens in the optical module are diffusive and require additional specular surfaces to measure tilt angle of the seat surface on which the lens assembly will be seated in the next process as shown in Fig. 1(b). But the measured tilts using the conventional autocollimator include the manufacturing, the assembly, and the handling of tolerances of the dummy mirror used as the specular surface representing the seat surface so that the reliability of the measurement data is lowered. Errors in the tilt evaluation process should be removed for precise measurement.
Figure 1(b) shows a concrete example with the auto-collimator for the tilt evaluation. The average roughness and the external diameter of the seat surface are about 2 µm and 6 mm respectively so that it is difficult to keep a certain level of specular reflectance in a large injection molded surface such as the actuator carrier including a seat surface. The specular reflectance R is dependent on the surface roughness, the angle of incidence, and the wavelength. Theoretically R is expressed as R = R0 exp[-(4πσa cos θ/λ)2] where R0 is the reflectance for the perfectly smooth surface of the same material, σa is the root-mean-square (RMS) surface roughness, θ is the angle of incidence, and λ is the wavelength of the incident light source [10,11]. For the seat surface, surface roughness is about 3 times bigger than the wavelength of the incoming light so the diffusive light scatterings on the surface degrade the specular reflectance severely. The approach for the coherence of the light source is needed in consideration of the characteristics of the target surface to keep interferometric visibility adequately.
The collimation of incident light is more important because the seat surface of the camera module is in vertically inward and is composed of separated 4 pieces in a large area. Therefore, a spatially coherent light source is needed to overcome the geometrical constrained conditions of the measurement surface. In particular, the components of a camera module such as a carrier and a barrel are made through an injection process with dielectric material and make surface treatment to lower the reflectance in visible wavelength range to reduce the possibility for visible straylight reaching the surface of an image sensor chip inside a camera module. Therefore, near-infrared seed SLD with longer wavelength than visible light is amplified to compensate for low reflectivity due to surface roughness and low quantum efficiency of a few percent of silicon-based complementary metal-oxide-semiconductor (CMOS) camera (Basler Model#: acA2040-180kmNIR) in about 1050 nm wavelength.
Figure 2(a) shows the optical layout configured in this study to measure tilt for the rough dielectric surface by low coherence scanning interferometry. Polarization-Maintaining Superluminescent light-emitting diodes (SLD) offering a 46.8 nm spectral bandwidth at about a 1,040 nm center wavelength, a 4.6 mW output, and 25 °C mounting temperature is used as a seed light [12,13]. An ytterbium-doped fiber amplifier is connected in series to amplify the output power to 21.5 mW of which the output beam is made linearly polarized as shown in Fig. 2(b). The linear polarizer placed at the end of the amplification stage is to suppress the amplification of other polarized state light caused by spontaneous emission during the amplification process and to amplify the linearly polarized light same as the PM-SLD. As a result, it produces light with a high S/N ratio. Through the above configuration of light source, the gain narrowing effect of the fiber amplifier [14,15] reduces the light before amplification such as in Fig. 2(c) to the spectral bandwidth to 42.0 nm about a center wavelength of 1,052.6 nm as shown in Fig. 2(d). The merit of adopting the fiber amplifier is to enhance the spatial coherence through a single-mode fiber exit of about 6 µm core diameter so that the output beam can be extended to a large diameter without loss of the optical power. Further, the configured near-infrared light source can be detected with reasonable sensitivity by ordinary digital cameras of silicon multi-mega pixels used for industrial machine vision in the wavelength range of 400 ∼ 1,100 nm . Figure 2(e) shows a typical pattern of coherence correlogram obtained using the near-infrared light source by scanning the reference mirror with 50 nm steps on the Fig. 2(a) optical layout for a mirror surface. With a target surface being made of the same material as the reference mirror, the correlogram contrast is measured to be as high as 0.9. The temporal coherence length Lc is measured to be 11.6 µm, which similar to the analytical value estimated from Lc = 0.44 λ2/Δλ where the spectral bandwidth Δλ and center wavelength measured in Fig. 2(d) [17,18,19]. For successful implementation of coherence scanning interferometry, the temporal coherence length is required to be at least four times larger than the standard deviation of the surface height variations . This implies that the light source can apply to the surfaces with RMS roughness as large as 2.9 µm in the field of optical metrology under the assumption that the speckle effect occurs only due to the surface roughness. In the case of broadband light source, i.e., short coherence length and very rough surface, there is the effect of wavelength-dependent speckle decorrelation so that the contrast of correlograms is lowered by the incoherent superposition of speckles of different wavelengths .
For implementation of coherence scanning interferometry for the tilt measurement, the reference mirror of the interferometer is moved using a PZT micro-actuator along the optical axis in a scanning mode. Resulting coherence correlograms are captured with a digital camera of CMOS type that enhanced quantum efficiency in near-IR. The image plane of the camera is comprised of 2,048 X 2,048 pixels with a 5.5 µm pixel. A commercial macro lens is used as the imaging lens for a large field-of-view up to 14 mm diameter.
The specific measurement conditions are summarized in Table 1. Compared to conventional visible light source, the complex near-infrared source constructed in this study suffers a loss in camera sensitivity but offers advantage of less sensitive to the roughness of the surface.
In order to verify the performance of the low coherence scanning interferometer (LCSI) made in this study as a surface profile meter, a traceable step height standard (VLSI Standards Model#: SHS-9400QC) was used for the measurements of 3D surface and step heights. Figure 3(a) is a photograph of the standard specimen having 940 nm step height etched into Quartz and coated with Chromium.
The measured spot is the rectangular area in the middle of Fig. 3(a) and the measured surface profile is shown in Fig. 3(b) when the reference mirror in Fig. 2(a) was scanned with 500 nm step size. The localized correlogram can be obtained such as in Fig. 2(e) when the optical path difference (OPD) between the optical path length of reference (OPL1) and measurement arms (OPL2) becomes near zero by adjusting one of two path lengths. The position of the correlogram peak is determined with the centroid algorithm, which allows the scan step to be selected as large as 500 nm to minimize the total sampling and computation time . This method in LCSI has been attractive for measuring 3D surface profile and large step height of a sample because it can avoid the well-known 2π ambiguity problem. Figure 3(c) shows repeatability of 89 nm for the step height and the height is estimated to be 931 nm for the 940 nm step height block. There is 9 nm offset due to the accumulated scanning stage error and the low correlogram contrast by the background intensity, which affects the measurement uncertainty of coherence scanning interferometry .
3.1 Tilt measurements
Standard angle gauge blocks (Starrett-Webber Angle Gage Block of a calibration grade) of 1 and 5 min were prepared to confirm the accuracy of the tilt measured by LCSI. The angle gauge blocks are placed on the commercial flat mirror of λ/10 flatness at 633 nm wavelength as shown in Fig. 4(a). The surface of the angle block on the flat mirror is measured first and then the surface of the flat mirror is measured after removing the angle block as shown in Fig. 4(b). Figure 4(c) is the surface profile for the top surface of the angle gauge block configured such as Fig. 4(a) and Fig. 4(d) is for the flat gold mirror of Fig. 4(b). In our case of LCSI, principal component analysis (PCA) is applied to reduce the dimension of the measurement raw data which contains x, y coordinates and z position data into 2-dimensional plane and to get the normal vector of the plane. PCA finds linearly uncorrelated variables called principal components by orthogonally transforming the original basis in which the measurement data is into a more meaningful basis. So, the method extracts the angle between the normal vectors representing surfaces of the flat gold mirror and the angle gauge block.
Table 2 shows the results for 20 times consecutive measurements for the 1 min and the 5 min tilted block with the LCSI. There are around 0.5’ tilt differences between standard angle gage block and LCSI measurement result for the nominal value. The deviations of accuracy in each angle gage block is estimated due to the straightness deviation of the scan axis and the axial displacement of a measurement point between Fig. 4(a) and Fig. 4(b). The thickness of the angle gauge blocks used in Fig. 4(a) is around 10 mm. The distribution of measured angle is due to errors in the process of placing the angle gauge block on the flat gold mirror for every measurement.
3.2 Practical applications for tilt measurements
Figure 5 shows the two typical applications for the tilt measurement.
Figure 5(a) is the cross section of camera module before assembling the barrel into the carrier. Tilt measurements are to evaluate the parallelism and the squareness of the seat surface according to the base plate and the housing. The base plate is the reference surface for tilts as shown in Fig. 5(a). Figure 5(b) shows another tilt measurement surface to know the assembly matching degree of the barrel from measuring the barrel top surface after putting together all optical parts. The RMS roughness of the seat surface and the barrel top surface is a few microns, so it is important to maintain the contrast of interference signals high and stable over the whole measurement area.
The shape data of Fig. 6(c) and Fig. 6(e) are obtained by moving the reference mirror on Z axis after aligning the interference between the base plate and the reference mirror as a null fringe. Figure 6(a) and (d) show the shape and position of the seat and the barrel top surface of a typical camera module, although it differs depending on the product model. Figure 6(b) shows the measured surface profile of the sample mounting surface namely the base plate in Fig. 5, which is the reference surface for tilt measurement. Tilts are measured relative to the reference surface. The seat surface consists of four independent surfaces separated from each other and the surface height distributions of each separated seat surface are relatively different as shown in Fig. 6(c). Figure 6(e) shows the surface profile of the barrel top of the lens assembly holder. Figure 6(f) is the tilt graph showing the tilt angle and direction of the barrel top surface using the surface height measurement data of Fig. 6(b) and 6(e). The red solid circle of the tilt graph is the criterion of the good or faulty product for the tilt measurement in the manufacturing. The yellow dots in Fig. 6(f) represent the results of five repeated tilt measurements are superimposed and the detailed measurement data are shown in Table 3. The position and the tilt value of the yellow dots correspond to the reflected displacement for the dummy mirror on each target surface with the auto-collimator (PSI Model#: LBF-CH1) having 200 mm working distance and 100 mm focal length.
Table 3 shows how the tilt information is specifically extracted from the surface profile data and the tilt deviation of the repeat measurement. The data (a, b, c) in the left column of the table are the coefficients in each axis of the normal vector representing the reference surface by PCA and the coefficients of the right column are for the barrel top surface. It is possible to know in which quadrant the tilt is located from the tilt sign in the x and y axis. The deviation for the 5 repeated measurements is 0.05’ as shown in Table 3.
In order to confirm the measurement dispersion and data correlation of the auto-collimator, which is a conventional tilt measurement method, and LCSI proposed in this study, 40 actual products were prepared. The products were divided into four groups by combining a housing named as ‘H’ and a carrier of actuators inside camera modules named as ‘C’. The results of the above two measurements (Auto-collimator and LCSI) for four sample combinations are shown in Fig. 7(a) - (d). In Fig. 7, there are four kinds of a housing (H) and a carrier (C). For example, HACA indicates the assembly combination of an A type housing and an A type carrier. The tendency for the tilt direction is the same for both methods.
However, it can be seen from Fig. 7 that the tilt measurement deviation of auto-collimator is larger than that of LCSI. LCSI is more useful than auto-collimator for determining the tilt status for a combination of module parts. For comparison of measurement repeatability for both tilt measurement methods, the above 40 samples are measured 3 times in succession and the distributions of the maximum-minimum deviation for each tilt method are shown in Fig. 7(e). For the auto-collimator, the mean of the maximum-minimum deviation is 1.58 min and the standard deviation is 1.21 min. On the other hand, LCSI shows a deviation of 0.58 min and a standard deviation of 0.39 min. The practical application of this study is to determine whether the design and manufacturing results of metallic injection molds that make a carrier, a housing, and a barrel inside smartphone camera modules are suitable for mass production by measuring the tilting of the seat surface and the barrel top surface. Because mold making is expensive, it is important to reduce the number of mold modifications in actual product development and manufacturing.
Therefore, there is a need for a measurement method having excellent repeatability rather than accuracy of tilt measurement. As a result, LCSI is superior to the auto-collimator method in terms of tilt measurement precision in the camera module assembly process. The main reason for the difference in the measurement deviation between LCSI and auto-collimator is the difference in the measurement methods. In the case of auto-collimator, an additional dummy mirror is used for the tilt measurement. So, the machining error for making a dummy mirror and the assembling error for putting the dummy mirror in the seat surface as shown in Fig. 1(b) act as measurement noise. On the other hand, LCSI does not require a dummy mirror and can be directly applied to the diffusive surface such as a rough dielectric surface to obtain tilt information on the measured surface. Also, the difference in the number of measurement data processed in each method is the additional cause for measurement deviations. One point is used to obtain the tilt in auto-collimator but hundreds of thousands to millions of data are used to get the tilt in LCSI. The more data, the smaller random error.
Figure 8 shows that dynamic tilt evaluation is possible with LCSI. The yellow dots are the measured tilts while the lens assembly shifts from the static to the maximum zooming state in an optical axis direction. The total axial shift of the barrel top surface is 184.1 µm and each shift interval is 13.15 µm. However, it is difficult to evaluate the tilt dynamically with the auto-collimator due to the dummy mirror on the barrel top surface.
We devised a compound light source by combining a superluminescent light-emitting diode with an ytterbium-doped fiber amplifier to offer proper coherence length and optical power for the dielectric surface having a few µm roughness. The light source was used to illuminate the target surface with 21.5 mW optical power with high spatial coherence through a 6 µm core-diameter single-mode fiber. With the spectral range of the light source being centered at a 1,052.6 nm with a 42.0 nm bandwidth, the low coherence scanning interferometry (LCSI) was performed with a 11.6 µm coherence length to measure tilts and 3-dimensional surface profiles. Principal component analysis usually used in a machine learning field was applied to the measured surface data to obtain tilt information in our study. The accuracy for the tilt and the step height were confirmed with the traceable step height standard and angle gauge block. With the actual products having rough dielectric surface, we have demonstrated that LCSI, a non-contact direct optical method, is more advantageous for the precision of tilt measurements and for the evaluation of whole surface profile than the auto-collimator method which needs a mirror-like surface to use specular reflections. The tilt measurement deviation by LCSI method for the barrel top surface of the compact camera module was 3 times superior to the auto-collimator method. The proposed interferometric tilt measurement method may applicable to the assembly processes having various rough surfaces and the monitoring the dynamic tilt trajectory.
Central R&D Center of Samsung Electro-Mechanics Co. Ltd, Samsung (MOM17-ZZ01SC).
The measurements of Fig. 7 were conducted by the camera module department of Samsung Electro-Mechanics Co. Ltd. through the system made in this study.
Chang-Yun Lee: Samsung Electro-Mechanics Co. Ltd. (E, F), Joonho You: Nexensor Inc. (E), Yunseok Kim: Lasernics Co. Ltd. (E).
1. M. Ikram and G. Hussain, “Michelson interferometer for precision angle measurement,” Appl. Opt. 38(1), 113–120 (1999). [CrossRef]
2. P. R. Yoder, Opto-Mechanical Systems Design, 3rd ed. (CRC, 2006).
3. D. Malacara, Optical Shop Test, 3rd ed. (John Wiley & Sons, 2014).
4. L. Deck and P. de Groot, “High-speed noncontact profiler based on scanning white-light interferometry,” Appl. Opt. 33(31), 7334–7338 (1994). [CrossRef]
5. S. -W. Kim and G. -H. Kim, “Thickness-profile measurement of transparent thin-film layers by white-light scanning interferometry,” Appl. Opt. 38(28), 5968–5973 (1999). [CrossRef]
6. I. Shvrin, L. Lipiäinen, K. Kokkonen, S. Novotny, M. Kaivola, and H. Ludvigsen, “Stroboscopic white-light interferometry of vibrating microstructures,” Opt. Express 21(14), 16901–16907 (2013). [CrossRef]
7. T. Dresel, G. Häusler, and H. Venzke, “Three-dimensional sensing of rough surfaces by coherence radar,” Appl. Opt. 31(7), 919–925 (1992). [CrossRef]
8. J.-A. Owen, “Principal Component Analysis: Data Reduction and Simplification,” McNair Scholars Res. J. 1(2), 1–23 (2014).
9. H. Gross, B. Dörband, and H. Müller, Handbook of Optical Systems: Metrology of Optical Components and Systems, (Wiley, 2015), Chap. 50.
10. H. E. Bennett and J. O. Porteus, “Relation between surface roughness and specular reflectance at normal incidence,” J. Opt. Soc. Am. 51(2), 123–129 (1961). [CrossRef]
11. H. E. Bennett, “Specular reflectance of aluminized ground glass and the height distribution of surface irregularities,” J. Opt. Soc. Am. 53(12), 1389–1394 (1963). [CrossRef]
12. O. Sasaki, Y. Ikeada, and T. Suzuki, “Superluminescent diode interferometer using sinusoidal phase modulation for step-profile measurement,” Appl. Opt. 37(22), 5126–5131 (1998). [CrossRef]
13. C. -Y. Lee, S. -W. Hyun, Y. -J. Kim, and S. -W. Kim, “Optical inspection of smartphone camera modules by near-infrared low-coherence interferometrry,” Opt. Eng. 55(9), 091404 (2016). [CrossRef]
14. R. Paschotta, J. Nilsson, A. C. Tropper, and D. C. Hanna, “Ytterbium-doped fiber amplifiers,” IEEE J. Quantum Electron. 33(7), 1049–1056 (1997). [CrossRef]
15. Y. Chiba, H. Takada, K. Torizuka, and K. Misawa, “65-fs Yb-doped fiber laser system with gain-narrowing compensation,” Opt. Express 23(5), 6809–6814 (2015). [CrossRef]
16. P. Magnan, “Detection of visible photons in CCD and CMOS: A comparative view,” Nucl. Instrum. Methods Phys. Res., Sect. A 504(1-3), 199–212 (2003). [CrossRef]
17. T. H. Ko, D. C. Adler, J. G. Fujimoto, D. Mamedov, V. Prokhorov, V. Shidlovski, and S. Yakubovich, “Ultrahigh resolution optical coherence tomography imaging with a broadband superluminescent diode light source,” Opt. Express 12(10), 2112–2119 (2004). [CrossRef]
18. C. Akcay, P. Parrein, and J. P. Rolland, “Estimation of longitudinal resolution in optical coherence imaging,” Appl. Opt. 41(25), 5256–5262 (2002). [CrossRef]
19. A. F. Fercher, W. Drexler, C. K. Hitzenberger, and T. Lasser, “Optical coherence tomography – principles and application,” Rep. Prog. Phys. 66(2), 239–303 (2003). [CrossRef]
20. P. Pavliček and J. Soubusta, “Theoretical measurement uncertainty of white-light interferometry on rough surfaces,” Appl. Opt. 42(10), 1809–1813 (2003). [CrossRef]
21. T. Dresel, G. Häusler, and H. Venzke, “Three-dimensional sensing of rough surfaces by coherence radar,” Appl. Opt. 31(7), 919–925 (1992). [CrossRef]
22. C. Ai and E. L. Novak, “Centroid approach for estimating modulation peak in broad-bandwidth interferometry,” U.S. patent 5,633,715 (filed 19 May 1996; issued 27 May 1997).
23. P. Pavliček and O. Hŷbl, “White-light interferometry on rough surface measurement uncertainty caused by surface roughness,” Appl. Opt. 47(16), 2941–2949 (2008). [CrossRef]