Search Menu
Photonics Media Photonics Marketplace Photonics Spectra BioPhotonics Vision Spectra Photonics Showcase Photonics ProdSpec Photonics Handbook

Chip Rotation Doesn't Disorient Vision System

Facebook Twitter LinkedIn Email
Daniel C. McCarthy

Imaging integrated circuits is a simple enough task for today's machine vision technology, but problems emerge when a vision system also must identify microscopic details on slightly skewed dies at near-blinding speeds. It is not uncommon for today's wire bond machines to attach dies with pad pitch sizes as small as 50 µm.

appsDoesnt Identifying wire bond leads by geometric pattern matching rather than gray-scale correlation enabled one wire bond machine to locate dies and calculate their rotation using only one alignment point. The increased rate of analysis translated into 12 percent faster throughput for dies with 50-µm pitches. Courtesy of Cognex Corp.
   When Esec designed 50-µm capability into its 3088iP wire bond machine, it required integration of a whole new vision system, starting with a camera that captured a larger field of view at a rate that was 10 times faster. At the same time, the wire bonder increased throughput by 12 percent for devices with high lead counts.
   That increased the pressure on the 3088iP's image processing function, which led Esec to select Cognex's MVS-82400/VME embedded vision system. An onboard MMX processor based on Intel's Pentium III technology enabled Esec's wire bonder to handle visual data 100 times faster than its predecessors, according to Andre Muff, who led development of the machine.
   However, faster imaging is only one side of increased throughput. "We also wanted to increase the robustness of the system," Muff said. "What we were trying to improve was the [mean time between assists]. Our goal was to reduce the stoppages caused by the pattern-recognition system."
   Earlier systems used gray-scale-normalized correlation for pattern recognition, which can be challenged if dies appear rotated in their packs. Gray-scale-based software makes a gridlike comparison of what the camera sees with a reference image of the same pattern. From this, it calculates a point where the X-Y positions of both images best align and determines the pattern's location. If the chip is rotated as much as ±7°, gray-scale-normalized correlation must capture two alignment points -- one to locate the die and another to calculate its rotation.
   Cognex's system instead relies on Patmax geometric pattern matching software, which interprets camera images in terms of geometric functions; for instance, identifying an arrangement of four line segments as a square. Combined with calculations of distance and camera angle, the software can locate a die and calculate its rotation using only one alignment point, making it faster.
   "It's a matter of units per hour," Muff said. "If you only have to align one point, you gain on speed."

Photonics Spectra
Oct 2001
machine vision
Interpretation of an image of an object or scene through the use of optical noncontact sensing mechanisms for the purpose of obtaining information and/or controlling machines or processes.
Accent on ApplicationsApplicationsmachine vision

back to top
Facebook Twitter Instagram LinkedIn YouTube RSS
©2023 Photonics Media, 100 West St., Pittsfield, MA, 01201 USA, [email protected]

Photonics Media, Laurin Publishing
x Subscribe to Photonics Spectra magazine - FREE!
We use cookies to improve user experience and analyze our website traffic as stated in our Privacy Policy. By using this website, you agree to the use of cookies unless you have disabled them.