Design And Implementation Of Semi-auto Probe Bender

Wafer testing is an important precision technology in semiconductor manufacturing. There is no any lead before the chip is assembled. During wafer testing, the electronic circuit of the chip is connected to the Test platform via probe card by the probe tip touching on the chip pad. Cantilever probe card is widely used in semiconductor wafer testing market. And the process of probe bending is the key technology in probe card manufacturing. The precision of the tip bending directly affects the quality of probe card.The traditional manual fixture used for probe bending is introduced in first. In order to overcome the disadvantage of the manual bending fixture, the semi-auto probe bender is designed. The PLC based control system is analyzed in details, for the control of tip length and bend angle of probes. It includes the mechanism principle of probe bending; choose of the mechanical-electrical components; the solution of critical problem for the control system; communication between the controllers; software programming for the motions. Eventually the semi-auto probe bender was implemented on the production, and the tip length and bend angle were precisely controlled.In this paper, we discuss in detail the design of BP neural network including the number of layers, the number of the hidden neurons, selection of the initial values, learning rate and anticipation error. Moreover, aiming at the drawbacks of BP network, we analyze the methods to overcome them. Then, we design two BP network models which are trained and simulated. By contrast, we find that the predictions of two models are both very accurate. However, one model of them is more practicable. With this trained BP network model, the machine parameters matrix is set up for the predetermination in order to improve the yield.Innovation in this project: 1. Automation of probe bending via micro-stepping motor control system overcomes the low precision and weak stability brought by the manual bending fixture. 2. Used the BP neural network to set up the predetermination of machine parameters matrix. Hence the production efficiency and the product yield were improved.