RBFN优化算法在键合机标定中的应用  

Error Compensation Based on RBF in Bonder

作　　者：洪喜[1] 李维[1] 卢佳[1] 白虹[1] 孙海波[1] 

机构地区：[1]中国科学院长春光华微电子设备工程中心有限公司,长春130000

出　　处：《电子与封装》2016年第5期7-9,13,共4页Electronics & Packaging

摘　　要：键合位置精度是衡量键合机性能的关键指标之一。为提高键合工具的键合位置精度,针对键合位置误差的非线性特点,提出一种径向基神经网络误差修正方法。以键合角度与键合点图像坐标为学习样本,以生成最小映射误差为原则调节网络权因子、基函数中心和宽度,建立具有良好泛化能力的误差逼近模型。并根据算法特点提出了一种工程优化方法,在保证算法补偿精度的基础上使得其运算时间也满足工作需要。实际工作表明:采用此种方法可将键合精度提高一个数量级,有效地改善键合位置精度并且很好地解决非线性误差对系统的影响。To improve precision of bonding position which is a key technical index for bonder, a new compensation method based on neural network was proposed. A model based on Radial Basis Function（RBF） was set up, in which the output was bonding coordinate and the input was bonding angle. According to the inhibit condition between the test value and the output of the network, adjusting power factor formula and the center and width of the radial basis function to make the model have a good learning ability and generalization ability. The optimization algorithm reduces the running time to a practical degree but doesn’t decrease the compensation precision obviously. The test results show that the system precision is improved by an order of magnitude by this optimal method and the nonlinear effect on the system is reduced.

关 键 词：径向基函数网络 键合机 误差 修正 精度 

分 类 号：TN762[电子电信—电路与系统] TP183[自动化与计算机技术—控制理论与控制工程]