MOTION CONTROL and machine vision are used throughout the semiconductor manufacturing process, from monitoring the diameter of ingots as they’re formed from a crystal seed to aligning a die lead frame prior to wire bonding. In nearly every step of the process, motion and vision can be found working together to align, inspect, measure, and identify wafers and die, so the various pieces of production equipment can perform their tasks.
These jobs weren’t always easy or labor-efficient. Our company, Point Technologies, recently applied a much-needed integrated motion and vision solution to a critical area of the semiconductor industry: probe needle inspection. Based in Boulder, Colo., Point supplies precision electrochemical pointing and micro machining services and products for semiconductor, medical and biotech applications.
Eyeing the Needle
Semiconductor manufacturers rely on probe needles for wafer-level testing during initial production phases to ensure that only fully functional product is packaged for final use.
A probe needle is a straight, small-diameter (0.002-0.012 in.) metal wire with one end tapered to a sharp point. Each probe plays the vital role of establishing an electrical connection between the tester and the wafer. Generally hundreds or thousands of probe needles are precisely bent and set into an array on a device known as a probe card. Each probe card is tailored to interface between the specific wafer being tested and the wafer probing machine, or “prober.” During testing, precise needle geometry is essential to ensure test data reliability and consistency.
Most probe card manufacturers use reference diameters to determine where to bend each probe (See Figure 1 below). If the geometry of a group of probe needles is inconsistent, then the reference diameter, as well as the bend angle, tip diameter and tip length, might be inconsistent. A change in any of these variables can cause probe misalignment and result in poor test data consistency.
FIGURE 1: SPECS MUST LOOK SHARP
A probe needle is a straight, small-diameter (0.002-0.012 in.) metal wire with one end tapered to a sharp point used to establish an electrical connection between tester and wafer. Inconsistent probe needle geometry can cause probe misalignment and poor test data consistency.
TO TEST a wafer, both the probe card and wafer are loaded into the prober. Then, the prober drives the probe needles slightly past the point at which the needle tips contact the wafer, causing the needle body to flex and the needle tips to slide across the metal bond pads on the wafer. Because of the nature of this motion, probe tip diameter becomes a critical dimension. The probe tip diameter helps determine the width of the area of each pad that is scrubbed or scratched up on contact with the needles.
The needle body diameter and taper shape help determine how much each needle will flex and the amount of force (known as balance contact force) with which the needles will touch down on the wafer. This is a critical specification set by the wafer manufacturer. It affects probe needle life, and, therefore, probe card life, and determines how well the probe tip will break through the thin layer of aluminum oxide on the metal bond pads.
As a manufacturer of probe needles, Point has quality control inspection standards to ensure proper probe needle performance. The company must measure locations on the needle that define the taper shape, as well as measuring the diameter at specified distances back from the tip, prior to bending. The probe diameter and the taper length (the distance from the tip to the end of the taper) also are measured before bending.
In recent years, demand for wafer production grew, causing an increase in demand for probe needles. Since inspection is a significant proportion of total needle manufacturing time, it was clear to Point’s management that inspection would be a prime candidate for automation.
The Old Days
Before starting its recent improvement project, the company relied on a combination of manual video-inspection systems and several optical comparators to inspect probe needle geometry. “We had two inspection methods,” says Edgar Miller, Point’s quality manager. “An optical comparator enlarged a shadow of a part by 50X, 100X or 250X. An operator would place a fixture on the comparator, and then align and focus the part to be measured. Numerical measurements could be taken using a calibrated ruler, or we could use charts with the required needle shape and tolerance, so the operator could ensure part acceptance.”
Video inspection was an alternate method. “The video inspection system would magnify the part up to 720X,” adds Miller. “The operator would place a fixture on the video inspection system, focus, and align the part. He then would pick points on the taper to compare with the required specifications.”
These inspection processes clearly were slow, labor intensive, required extensive operator training, and were error-prone due to operator bias. They had to go.
Setting Sights on Automation
Point subsequently investigated available options for improving the needle inspection process. “Commercially available equipment that could do the job was cumbersome, slow and too expensive, so our team designed a system to better meet our needs,” says Steve Neely, Point’s semiconductor division manager. “With a clear vision and well-defined goals, our design team won the financial support necessary to succeed and move ahead.”
The team’s primary design goals were to increase measurement throughput, accuracy and repeatability, while reducing inspection time. The new system also would automatically document all measurements onto a shared server, minimize operator training, and maximize operator comfort. Finally, it automatically would plot the measurement data on a graph for comparison to customer specifications, and provide statistical analysis for process improvements.
In addition, the engineering team set out to integrate motion control and machine vision. Up-front work included evaluating the feasibility of using a machine vision system to improve measurement efficiency, and still get repeatable measurements at or under 0.00002 in.
The team initially had little to no machine vision experience, but concluded that it made sense to start with an In-Sight vision sensor from Cognex because it appeared easy to learn and came complete with user software. “The In-Sight system required a minimal investment to allow us to prove that machine vision would work for our application,” adds Neely. “We were able to set up a prototype system to prove the concept in four or five weeks.”
The machine vision sensing system consists of a DSP-based vision-processing unit, high-speed digital camera, onboard light control, and built-in discrete I/O. It also provides a standard VGA output for real-time display (See Figure 2 below), built-in Ethernet communications, and an onboard serial port.
FIGURE 2: OPERATOR ASSISTANCE
The machine vision system includes a standard VGA output for real-time operating data display.
“WITHOUT formal training classes, it took approximately 30 hours to get comfortable enough with In-Sight to set up the inspection routine,” says Neely. “By reviewing the tutorials that came with the system and using the on-phone technical support provided by Cognex, the team became familiar with the vision spreadsheet-development environment, library of vision software tools, and the built-in operator interface.”
|FIGURE 3: ZOOM, ZOOM|
After some lighting and optics experimentation, a high-magnification, telecentric lens that eliminates optical distortions was added to the camera.
Tough this vision sensor proved to be a good validation tool, it wasn’t the right product for the final application. The design team instead chose Cognex’s VisionPro PC-based vision system because it offered higher performance and more seamless integration with the motion control system. “Moving up to the VisionPro platform took a lot more time to learn, and required a couple of weeks of help from a Cognex’s certified vision integrator,” says Neely. “The integrator was instrumental in developing the vision code required, and was well worth the investment, especially since the vision code was used again for subsequent systems.”
Cognex’s PC-based vision system uses a frame grabber in a basic Dell PC and a high-resolution digital camera. “One of the most challenging parts of any machine vision application is selecting the image-formation system to produce a good image,” explains Neely. “After some lighting and optics experimentation, the team implemented a high-magnification, telecentric lens (See Figure 3) that eliminates optical distortions and a 2x2-in. diffuse LED backlight to provide suitable images with the necessary resolution.”
Vision Meets Motion
In fact, the team’s quest to integrate machine vision with motion control led to PointScan, Point’s new automated needle inspection machine (See Figure 4 below). The motion control system for PointScan uses limit switches to define x, y, and z-axis origins, but the majority of the driving input comes from the vision system. Our engineers blended the vision, motion, operator interface, network communications and database using Microsoft Visual Basic 6.0.
“All the logic and control are managed with Visual Basic, which guides the stepper motors to move the camera, and position the needles,” says Neely. “Vision-guided motion applications such as this one require a very fast vision system that will synchronize image capture, analysis and measurement with the motion required to find the needle, focus the needle, and move from needle to needle.”
|FIGURE 4: MAKING THEIR POINT|
The integration of machine vision with motion control led to PointScan, Point Technologies’ new automated needle inspection machine.
PointScan’s motion control system is essentially a tabletop Cartesian robot that uses precision stepper motors to drive three axes of motion. One axis moves the loading table into and out of the inspection area. The other two axes control camera movement. One moves horizontally; the other moves vertically. The motion system combines components that include a three-axis Galil motion controller, precision linear slides, leadscrews and stepper motors from various vendors.
“We chose the controller because the interface with VB was easy, and we have other machines that use the same controller,” says Neely. “It’s better to keep all our production equipment on similar systems, in case we need to switch out parts or replace anything.”
The PointScan system allowed our company to increase inspection requirements to 12 parts per block, compared with the previous limit of three per block, and reduced inspection time from about 4 min. to less than 2 min. It used to take an operator about 20 min. to measure 12 needles and record the data on paper. PointScan measures 12 needles in less than two minutes, records the data, and automatically graphs the data for the operator (See Figure 5 below).
“Product returns based on dimensional errors have been drastically reduced,” beams Miller. “The repeatability is better than 0.00003 in. By comparison, human operators vary as much as 0.0001 in. We gain a huge advantage now because we also record all data and provide inspection reports to our customers.”
In addition, modeling every component using Autodesk Inventor helped to streamline the process, and reduce design time. In all, the team was able to go from concept to working system in about six months. Five systems have been built, and have been used successfully on the production room floor for more than a year.
“The data gathered by the vision system provides great insight into the production process,” adds Neely. “In fact, PointScan provides the feedback needed for operators and the quality team to successfully ‘re-model’ systems and processes with more built-in improvements and efficiencies.”
FIGURE 5: EASY DATA
PointScan’s data collection mode records and automatically graphs the data for the operator.
The design team’s previous experience building multi-axis automation systems made the mechanical design fairly straightforward, but there were challenges. “For example,” says Neely, “due to the very small size of the probe needles and the high magnification of the vision system, we took special care to eliminate both internal and external sources of vibration.”
The camera resolution, optics and working distance needed to achieve the desired field of view (FOV) presented another challenge. “The entire length of part being measured didn’t fit into our FOV at the resolution we needed, so the part must be moved a fixed distance under the camera to complete the taper length measurement,” explains Neely. “This complicated the application because it required combining the information from more than one image.”
Integrating vision and motion also required that engineers carefully calibrate the vision system with the motion system. Optical distortion from the lens, or perspective changes due to camera mounting angle, involves more than a scaling factor that relates pixels to a measured dimension. The vision system tool suite included special algorithms to correct for these errors.
The success of the new machine inspired Point’s management to consider marketing the machine to customers and like-minded companies, but they ultimately decided to keep the technology in-house. “It was designed to interface with our network and job control software,” reminds Neely. “In addition, the market that would require a system built especially for this application is relatively small, and supporting the product would be a new position for us. I believe the machines offer more value to our customers by ensuring a supply of good product than they would if our customers purchased the units themselves.”
Lights, Camera, Inspection: PointScan in Action
PRIOR TO inspection, 200 blank straightened wires are loaded into a transferable fixture or block, and then tapered in a proprietary electro-chemical pointing process. The operator then sets the block of probes on the PointScan loading table, and confirms their name on a touchscreen interface. The specifications for every job are stored in a shared database.
The loading table automatically indexes the probe needles in the inspection area. The camera moves over the top of the needles from the left. When the camera’s field of view intersects the first needle, the table backs up until the point is located. Tightly coupled vision and motion capabilities let the two systems communicate to locate the point, and then center the point in the field of view. To optimize measurement accuracy and reproducibility, PointScan uses an automatic focus routine in which the vision system communicates with the motion system as it moves the camera up or down to focus on the needle. Once focused, the vision system acquires the image, analyzes it, and makes the appropriate measurements. The data automatically is recorded to a network database, and the camera jogs to the next needle to be measured.
As measurements are recorded, PointScan automatically plots the data on the screen, so the operator can see if the results are in spec. At the end of the inspection, or if the operator presses the stop button, PointScan automatically unloads the needles, and the probe needles can be accepted or rejected.
If the parts are rejected, the operator will use the inspection data to adjust the process. If the parts are accepted, a label is printed for future reference to the inspection data. At the end of the job, the system generates an inspection report that can accompany each shipment, and includes all the data for the measured parts, along with the customer’s original specifications for comparison.
|About the Author|