Optical 3D inspection plays an important role in quality assurance of microelectronic products produced in large quantity. Compared to competing methods, optical inspection has many advantages. It is relatively compact and fast, and requires neither contact nor vacuum. Currently, a variety of 3D methods are used, including optical triangulation, confocal microscopy, and moiré and white-light interferometry. These methods have characteristic merits and drawbacks. Development of each is steadily progressing to meet ever-growing, diverse industrial requirements. Nonetheless, increasingly complex microelectronic products are difficult to inspect. Consider, for example, products with metallic bumps covering micropatterned surfaces for high-density packaging of semiconductor chips. A large quantity of bumps with multiple features, including material, size, and roughness, is placed on a substrate coated with a dielectric thin film that protects microelectric circuits underneath. A bump's surface is usually rough with a large sag, and its reflectance is either too low or too high when compared with the base material. All of these challenges make optical inspection extremely difficult.
Figure 1. Large field-of-view measurement of microbumps on a printed circuit-board substrate. The inset is a close-up of ellipsoidal solder bumps.
White-light interferometry with wide-spectrum sources has long been used to measure mirror-like surfaces with extremely fine resolutions in the subnanometer range. Recently, in response to industrial demands, researchers have focused on extending low-coherence interferometry to measure rough surfaces. To further this effort, researchers in our precision metrology group at the Korea Advanced Institute of Science and Technology have inspected complex 3D patterns of large metallic bumps fabricated on a film-coated substrate. Specifically, we optimally configured an optical hardware design to inspect large rough objects up to a few hundred micrometers in height. We expanded the lateral field of view of a single measurement up to a few tens of millimeters. In addition, we analyzed nonlinear spectral variation of interference signals for thin films to enable inspection of the metallic bumps.
White-light interferometry can be implemented in two ways: by z-scanning or dispersive k-scanning. Z-scanning uses a mechanical scanner to move a target surface or reference mirror to produce an interferogram as a function of the optical path difference z. The interferogram is then Fourier-transformed numerically, resulting in a spectrally resolved signal that gives the phase distribution of interference in the domain of wave number k. This method is theoretically sound, but is susceptible in practice to vibration that distorts the resulting interferogram with subsequent measurement error. Alternatively, dispersive k-scanning involves a grating or prism spectrometer, which enables capture of the spectrally resolved interference signal directly as a function of k. This dispersive method requires no mechanical scanning. Thus, the measurement process can proceed fast enough to avoid disturbance by vibration. In addition, dispersive k-scanning can incorporate readily the principles of thin-film measurement based on nonlinear phase analysis1,2 or reflectometry.3
Figure 2. Measurement of metallic bumps on a silicon dioxide (SiO2)-coated wafer. The inset shows a photographic view of the same target of bumped wafer.
Figure 3. 3D thickness profile measurement of a SiO2 film layer deposited over a gold (Au)-patterned Si wafer.
To speed inspection, we adopted the Twyman–Green configuration in the interferometer design and used a macro lens to image the target surface with a large lateral field of view up to 30 × 30mm. This also enables separation of the illumination and imaging optics, which is not possible in the long-used designs of Mirau or Linnik for microsurface metrology. We designed the illumination optics to provide high collimation so that the target object can be uniformly illuminated. This allows surface profile measurement over an extensive height range of more than 100μm, significantly improving inspection speed. The interference signal produced between the target surface and the flat reference mirror is monitored using a 4MB CMOS camera.
We demonstrate the overall performance of our method with results from actual measurements. Figure 1 shows a 3D array of solder bumps fabricated on a polymer substrate. The shape of a bump is ellipsoidal, the diameter at the bottom is about 100μm, and the height is about 35μm. The top surface is relatively rough because it is made of microsized solder balls. The light source is a green LED with a spectral range of 60nm around the central wavelength of 530nm. All the balls within the field of view are captured with a lateral resolution of 7μm. The total number of pixels used is about 4 million. We conducted repeatability tests by monitoring five sample balls over 15 consecutive measurements. The standard deviation of the measured height values ranged between 200 and 300nm.
Figure 4. Consecutive measurements of a thin film during the chemical-mechanical-polishing process. (a) 3D tomographical thickness profile of a SiO2 film layer deposited on a check-patterned Si substrate before CMP. Note the A to A’ line at the top, parallel to the x-axis. (b) Variation of the top surface profile of the A to A’ line for five time lapses during CMP: (1) 0s, (2) 15s, (3) 30s, (4) 40s, and (5) 50s. (c) Cross-sectional top surface profile and (d) cross-sectional film thickness profile of the A to A’ line for five time lapses during CMP.
Figure 2 shows another 3D array. Here, the image is of gold bumps fabricated on a silicon wafer. The field of view is 5 × 4mm with a corresponding lateral resolution of 2.3μm. The wafer is coated with a silicon dioxide thin-film layer. Each bump is about 15μm in height and about 14μm in width. The light source used is an LED with a wide spectral range of 450–730nm. The coplanarity of all the bumps is determined by gauging bump height above the thin film of the silicon wafer. Two more examples are presented in Figures 3 and 4, in which the emphasis is on measurement of thin films with nanometer resolutions in thickness.
To conclude, the optical hardware we designed allows us to measure large rough objects with accuracy in the submicrometer range. Furthermore, we accelerated inspection speed by extending the lateral field of view of a single measurement and gauged thin films by analyzing nonlinear spectral variation of interference signals in combination with reflectometry. Our enhancements of white-light interferometry enable high-speed 3D inspection of microelectronic products with composite features of metallic bumps and thin films.
Further improvements, especially regarding measuring speed and accuracy, are under way to meet stringent industrial requirements. This includes incorporating spectrally resolved confocal microscopy into our existing techniques. So doing will eventually enable real-time 3D metrology without any mechanical scanning.
Billionth Uncertainty Precision Engineering Research Group
Department of Mechanical Engineering, Korea
Advanced Institute of Science and Technology (KAIST)
Daejeon, South Korea
Seung-Woo Kim received his PhD in precision machine system design from Cranfield University, United Kingdom. Since 1985, he has been a professor in the Department of Mechanical Engineering at KAIST. He heads the Precision Engineering and Metrology postgraduate research group and the KAIST Institute of Optical Science and Technology. His research interests include precision machine design, optical metrology, and mechatronics. He is a member of SPIE, the Optical Society of America, and the American Society for Precision Engineering.