Wire Bond Tester Targets Zero Field Failures
It may soon be possible to non-destructively test wire bond quality in real time during automated wire bonding and “bin” the failures, as is possible in electrical test.
Laura Peters, Editor-in-Chief -- Semiconductor International, 7/15/2008 10:00:00 AM
It may soon be possible to non-destructively test wire bond quality in real time during automated wire bonding and “bin” the failures, as is possible in electrical test. A new product from Hesse & Knipps GmbH (Paderborn, Germany), PiQC, is designed to test the quality of wire bonds and use real-time signals during bonding to selectively rate bond quality according to an index. As explained by Roberto Gilardoni, sales engineer at Hesse & Knipps GmbH, during the IMAPS Advanced Technology Workshop on Wire Bonding held yesterday in conjunction with SEMICON West, even the best bonders cannot account for bad bonds due to substrate contamination, clamping problems or “human factor” issues that occur in a production environment.
The reason why all failures are not caught today is that most quality control measures rely on wire deformation and/or use ultrasonic current as a signal to measure bond quality. These parameters do not catch every failed bond.
Several other metrics currently indicate bond quality, but no one method catches all of the failures. Of course, wire pull tests and wire shear tests are periodically performed for statistical process control. Wire deformation profiles can appear irregular, but a failure, such as the wire not being coupled to the die substrate, might not be found. Inline pull testing, typically used for heavy wire bonding, will find non-sticks and severe heel crack problems, but can overlook poor bond strength. Plus, this test impacts throughput.
The PiQC tester uses a digital ultrasonic generator coupled with a piezoelectric sensor that detects mechanical oscillations at the wedge tip. The sensor is very sensitive to the process acting on the tip of the bonding tool. The system detects resonance frequency, wedge tip mechanical oscillation, friction, transducer impedance, scrub behavior, wire deformation and ultrasonic current. The multitude of signals is statistically analyzed and processed using mathematical algorithms use field-programmable gate arrays (FPGAs) and high-speed control algorithms. The system calculates an individual quality index for each signal, and then combines them to achieve an overall quality index. The system compares the actual signal characteristics to reference characteristics that were learned during a preliminary automated procedure. The quality index varies from 0-100% based on six different input signals and multiple mathematical transformations. The quality index and signal components can be displayed at any time.
After a learning phase, the PiQC system can recognize deviations in real time, which can be classified and interpreted by an operator. Signal deviations can be linked to certain failure modes, enabling process specialists to react more quickly to production issues that affect quality.
Ultimately, Gilardoni said, the program will be an optimized expert system that can detect anomalies in system operation, such as a clamping problem or other need for maintenance. Systems are currently being shipped and evaluated by key customers.
