Inspection system classifies wafer defects

By Lawrence J. Curran,Contributing Editor

Integrated laser-sensor and linescan-camera-based inspection system combines 2-D and 3-D imaging to find faults in semiconductor wafers.

An inspection system that combines two-dimensional (2-D) and three-dimensional (3-D) measurements of semiconductor bumped wafers is strongly contributing to the operations of the Interconnection Systems Laboratory at the Motorola Semiconductor Products Sector (Tempe, AZ). The WS-1000 vision inspection system from the Electronics Division of Robotic Vision Systems Inc. (RSVI; Hauppauge, NY) provides precise measurements in the final inspection of bumped wafers used to produce chip-scale and flip-chip-packaged integrated circuits (ICs; see Fig. 1).

FIGURE 1. The WS-1000 vision system from Robotic Vision Systems Inc. provides precise measurements in the final inspection of bumped semiconductor wafers used to produce chip-scale and flip-chip-packaged integrated circuits. Patented 3-D laser technology is used to obtain surface metrology data for determining accurate, noncontact measurements of bump height and coplanarity. The addition of 2-D vision technology enables the system to identify defects such as bumps that are too large, too small, or damaged.

Using patented 3-D laser technology to obtain surface- metrology data provides accurate, noncontact measurements of bump height and coplanarity, says Reza Asgari, product manager for wafer-scanner products at the RVSI Electronics Division. True 3-D measurements for height-related data can enhance wafer yield and throughput by identifying coplanarity issues that might result in board-level failures. A straightforward 2-D inspection alone cannot provide this height-related measurement, according to Asgari.

The addition of 2-D vision allows the inspection system to identify wafer bumps that are too large, too small, or damaged. Other defects that can be detected include bridging and missing bumps. Motorola has been using this bumped-wafer inspection system on 6- and 8-in.-diameter wafers for about a year (see Fig. 2).

The wafers carry fab-produced high-end chips or die. One measurement challenge is that there could be as many as 2000 bumps per die, packed in at pitches as tight as 150 µm, according to Vasile Romega-Thompson, Motorola senior staff engineer. Since its installation, the vision inspection system has been modified with an upgraded laser lens and more advanced software to boost throughput and refine the system's ability to classify defects such as missing or bridged bumps.

Since those improvements, Romega-Thompson says, the vision inspection system "has allowed us to obtain all the bump statistics we require in a timely manner. It also gives us a coplanarity number for each die, which is handy for improving assembly yields, and saves money."

Two vision-system elements provide 3-D and 2-D views of the wafers. An RVSI Electronics infrared laser is linked to a PowerPC-based multiprocessor. The 2-D images are captured by a time-delay-and-integration linescan camera from Dalsa Inc. (Waterloo, ON, Canada), transferred by a digital camera link to an RVSI Acuity CiMatrix (Canton, MA) frame grabber, and delivered to an image-processing computer.

FIGURE 2. During WS-1000 system operation, a robotic arm picks off wafers from two cassettes and delivers them to an air-bearing x-yaxis motion stage. A prealigner device on the robotic arm orients and places the wafer on a vacuum chuck. Two-dimensional images are captured by a vertically mounted Dalsa TDI linescan camera and an RVSI Acuity CiMatrix frame grabber. Three-dimensional images are derived by a laser-based sensor mounted at an oblique angle.

"Both the 2-D camera and the 3-D laser sensor are mounted above the wafer chuck on a z-axis stage to accommodate different wafer thicknesses and bump heights," Schaefer explains. The 2-D camera uses both brightfield coaxial and darkfield red-light-emitting diode ring lighting to illuminate different aspects of the wafer being inspected.

This special lighting is provided by the Northeast Robotics Division of RVSI (Weare, NH) and helps to illuminate bump faults and some wafer-surface defects. However, brightfield illumination is best suited to detecting most wafer-surface defects. It comes from a coaxial lighting device for illuminating the 2-D images captured by the TDI linescan camera. This coaxial lighting is collimated illumination for the 2-D camera and appears to be aimed straight down on a wafer as if emanating from the camera positioned directly above the camera. The return image follows the same axis back to the camera and the frame grabber. In contrast, darkfield lighting is essentially side lighting for illuminating 3-D geometries—those particles that rise up from the surface of the wafer—and especially deposited bumps (see Fig. 4).

FIGURE 3. To start wafer inspection, a robotic arm selects a wafer from a storage cassette (bottom left) and places the wafer on a vacuum chuck (center). The 2-D camera (top right) in mounted in a vertical position and the 3-D infrared laser-based sensor is positioned at an oblique angle (top center). The laser sensor and the 2-D camera are linked to separate image processors. Two-dimensional imaging employs both brightfield coaxial and darkfield red LED ring lighting for illumination. In contrast, darkfield lighting provides side illumination for highlighting 3-D geometries.

A Pentium-II-based processor running Windows NT serves as the system host computer. "The NT-based host computer coordinates the activities of the x-yaxis stage, 2-D image processor, 3-D data processor, robot, and graphical user interface," Schaefer notes. "During inspection, it collects the data from the 2-D image and 3-D data processors, postprocesses the data, and stores the data to hard disk. Graphical and text reports are then generated from these data," he adds.

The high-speed image processor handles the imaging data from the 2-D camera. "It acquires an image from the 2-D camera, processes the image to obtain geometric data on bumps, processes the image to obtain defect data of the die, and transfers the results data to the NT-based host computer," Schaefer explains.

The PowerPC-based multiprocessor acquires data from the 3-D laser sensor, processes the data to obtain geometric data on the bumps, and transfers that data to the NT-based host computer.

FIGURE 4. Bumped wafer is held in place by a vacuum chuck on an x-y-axis air-bearing stage while a TDI linescan camera moves into position above the wafer on a z-axis stage to capture 2-D images. The stage passes the wafer under the camera for image acquisition. Images are processed while the stage moves along the y-axis to the next die. The camera might make 10 to 20 passes across a wafer with the stage indexing up at the end of each row of die to bring the next row under the camera lens.

The software that directs both 2-D and 3-D inspections, image and data processing, and graphical user interfacing was developed by RVSI Electronics. The company also produced the statistical process control and defect-analysis software, which is tailored to a specific type of wafer and loaded into the system as a "recipe" for that wafer before inspection.

The robotic arm incorporates a scanner that detects which cassette slots hold a wafer. Upon activation, the robotic arm grasps the first wafer, places it on the prealigner, rotates the prealigner to the alignment notch or flat to correctly orient the wafer, places it on the chuck, and initiates the vacuum. The 2-D time-delay-and-integration camera performs a quick scan to locate the fiduciary marks on the wafer. Using this information, the angular rotation of the chuck establishes that the wafer's x-axis is parallel to the air-bearing stage's x-axis alignment.

Figure 5. The WS-1000 wafer-inspection system provides a graphical user interface that presents a test operator with such menus as system training, inspection recipe setup, and inspection execution, among others.

Now the vision inspection system is poised for a 2-D scan using either brightfield or darkfield lighting as the stage passes the wafer from right to left under the camera. Images are acquired and processed while the stage moves along the y-axis to the next die. "The system locates the bumps, measures their diameter and roundness, and also looks for nodules (defects) attached to the bumps and bridging between bumps," Schaefer says. Bridging is an undesirable metal deposit that can cause a short circuit.

The 2-D camera might make 10 to 20 passes across a wafer, with the stage indexing in the y-axis at the end of each pass to bring the next row or series of rows under the camera lens.

After completing the 2-D inspection, the system sensors return to the bottom of the wafer, where the brightfield and darkfield LED lighting is turned off. Now, the laser-based sensor is activated to begin a 3-D scan. Accordingly, the stage moves the wafer along the x-axis as the laser scans each row to measure bump height and coplanarity.

"The data are then analyzed for checks against the limits specified by the recipe," Schaefer says, "and each die is logged as passed or failed. The vision inspection system also has an option that inks bad die, but Romega-Thompson says Motorola doesn't use that option now.

Asgari says the biggest challenge faced by RVSI engineers in systems integration was to make sure the process flow moved fast and flawlessly. "This was mainly a software control challenge—to write programs that managed communications among all the elements in the system, making sure they all work together in unison," he says.

Romega-Thompson agrees and notes that Motorola initially encountered a few rough spots, "as with any new tool, requiring software upgrades to correctly classify bumps, missing bumps, and bridging." He adds that after the system is set up for a new job, it is easy to operate.

One criterion for choosing the vision inspection system was its performance in relation to Motorola's gauging repeatability and reproducibility standard. This standard was applied for both 2- and 3-D measurements, "and we were pleased with the numbers we got, which were below 10%. That means the system is able to measure a process variation within a specified limit to within 10% of the total tolerance," says Romega-Thompson.

At the end of a wafer run, the test operator clicks on a "generate report" command, which produces a wafer map showing each good and bad die. After wafer dicing, that map is sent to the pick-and-place station so that only good die are selected.

Romega-Thompson reports that Motorola is considering using the vision inspection system for inspecting "under-bump metallurgy" on stencil-printed bumps. The metal under the bumps is gold on nickel, and the nickel plating is so thin that it has to be inspected optically. "It's not high enough [for another kind of inspection], and that's why this system may come in handy because it provides both 2- and 3-D inspections," he says. "We have to use both inspections to be sure we have good bump classification."

Motorola is also evaluating a WS-2000 system, which incorporates an enhanced laser module that operates at twice the throughput rate for 3-D inspection compared to that of the WS-1000 system. RVSI's Asgari says, "This combination of 2- and 3-D inspections at production rates for complete wafer surface-defect detection, bump-defect inspection, and dimensional metrology is all available in the WS-2000 system."

Motorola Semiconductor Products Sector
Tempe, AZ 85284
Web: www.motorola.com

Robotic Vision Systems Inc.
Hauppauge, NY 11788
Web: www.rvsi.com