Neural Network Controlled Gripper for Wire Bonding Machines in Semiconductor Fabrication

Extensive research focus is on the appropriate usage of neural networks in automation industry. This paper presents a neural network based approach for the wire-bonding machine in the semiconductor industry. The lead frame is the skeleton of the IC package, providing mechanical support to the die during its assembly into a finished product. It consists of a die paddle, to which the die is attached, and leads, which serve as the means for external electrical connection to the outside world. The die is connected to the leads by wires through wire bonding or by tape automated bonds. The wire bonding is a thermo-sonic (thermal-ultrasonic) process during which a series of die units in a lead frame are wire bonded with leads. Solenoid operated Index and Eject grippers are used for positioning the lead frame during this wire bonding. Electric current controls the force applied on the lead frame by the gripper. The current required for a force varies with the thickness of the lead frame. Extensive manual work is required in determining the current requirement for each thickness. We propose a neural network algorithm to determine this current for any thickness. The back propagation algorithm is used to train the network with the available data, so that the current requirement for any other thickness can be determined from the trained network.