Improving overlay accuracy through exposure correction using overlay prediction

In recent years, the demand of semiconductor manufacturing productivity improvement is rapidly increasing in advanced packages and stacked sensors. However, these processes require higher precision overlay than ever before, leading to an increased frequency of measurements in the exposure process. Therefore, it becomes more difficult to achieve high productivity and high accuracy simultaneously. To solve this problem, it has been reported about overlay prediction using machine learning, aiming to decrease measurement frequency by Canon. This method uses the correlation between data during exposure and overlay measurement results in the learning phase and predicts the overlay results using exposure data in the inference phase. In this report, we explain the results of improving overlay accuracy by calculating correction values using the previously reported prediction method and applying these corrections to the exposure equipment. In addition to the previously reported overlay prediction, these studies enable productivity improvements in semiconductor mass production processes with stable. These methods are expected to be applied not only to complex advanced packages and stacked sensors composed of different chips but also to a wide range of semiconductor exposure processes. Furthermore, these methods can be implemented on the Lithography Plus solution platform for exposure equipment.