基于多尺度估计理论的晶圆减薄工艺方差变化检测方法

A Study of the Standard Deviation Change in the Wafer Thinning Process Based on the Multiscale Estimation Theory

晶圆减薄工艺是伴随芯片堆叠技术的发展而出现的新制造过程,其制造质量直接关系最终产品成品率。文章以堆叠芯片晶圆减薄工艺质量参数为研究对象,拟建立监控晶圆减薄工艺质量的完整方法。首先,以该道生产工序质量参数序列建立自回归滑动平均模型,用于表达该道生产工序的质量特征变化。然后,在此模型的基础上,使用多尺度估计理论对该模型进行滤波分解处理,获得质量参数时间序列的高频信号,提取该道质量变异的方差变化。最终,使用统计学上的累积和控制图对质量变异信号进行诊断分析,根据工序方差变化的起始位置,提前发现系统可能存在的质量变坏趋势。经试验数据验证,相比传统的检验方法,该方法有95%的概率可以提前预测产品质量发生变化。

Aiming at the wafer thinning process in memory card products＇ stacked package in its quality control and efficiency improvement, a basic problem in the wafer thinning process is presented by the variation for the measurement of the wafer thinning process. It is critical to monitor the process to detect process changes and further diagnose the process to determine the root causes of the changes. Firstly, a time series ARMA（autoregressive moving average）model has been built on analyzing the equipment productive throughput and operation time between failures data from the factory. The analysis is useful in problem prediction and maintenance. Then, through multiscale estimation theory, the detail coefficients of the data model have been derived. At last, the use of the method is discussed and an example is given. The experimental results reveal that the standard deviation changes of this manufacturing process have been detected in the 95% by CUSUM（cumulative sum）control chart on the detail coefficients of the model, which means the measurement of the wafer thinning process will be worse in the near future.