Pattern recognition for automated wire bonding

The design and development of an integrated pattern-recognition system for transistor/LED wire bonding is presented. The system uses a M6809 microprocessor for computation. Three recognition algorithms are studied extensively, and their relative performances are compared. The first algorithm based on edge extraction and projection is relatively slow but is a candidate for hardware implementation at higher speeds. The second algorithm based on the Freeman tracer is the fastest, but not intelligent enough to identify connected regions. A new algorithm called the `connected-tracer¿ algorithm is proposed, which solves the problem of simple connectiveness with a speed performance of around 0.1 s using a 1 MHz processor. A variety of die samples of different sizes, scribing and attachment methods are tested. The results indicate that the system developed is both reliable and effective for fully automated transistor/LED wire bonding