Comparison and Analysis of Different Removal Rate Models in Run-to-Run control of the Oxide CMP process

摘要

The CMP process has been widely applied in the semiconductor industry for both oxide dielectric and metal layerrnplanarization [2]. A typical process goal is to achieve "global" planarization by preferential removal of "high" materialrnon the wafer. The CMP p rocess is subject to different variations, usually caused by tool, product, and incomingrndisturbances [6]. In order to compensate for these disturbances, different run-to-run control methods have beenrninvestigated and reported [10, 5, 9, 1, 7]. In this work, we will focus on the control of the oxide CMP process.rnThe main objective of the CMP run-to-run controller is to reduce the lot-to-lot variation in the post-polish oxide filmrnthickness. Lot-to-lot variations can be caused by both tool-induced and product-induced disturbances. Since the CMPrnprocess includes mechanical abrasion of the surface, the primary tool-induced disturbance is the polish rate drift [4, 5].rnThis drift is a significant source of variation for the CMP process. The product -induced disturbance is due to differentrndevice pattern densities and different levels (e.g. ILD1 vs. ILD2). These differences are also a significant source ofrnvariation to CMP process, but the need to compensate for device pattern dependencies has not been addressed untilrnrecently [9, 3, 6, 7]. The main source of incoming disturbance is the pre-polish film thickness, which can berncompensated by feed forward control, I.e., feed the measured pre-polish film thickness to the controller to calculate thernpolish time. There are other sources of incoming variation, such as film density, but compared to the variation in pre-polishrnfilm thickness, they are relatively small [6]. Therefore, in this work, we only consider the pre-polish filmrnthickness variation for incoming disturbance.rnThe control of the CMP process is poor due to insufficient understanding of the process, degradation of the polishingrnpads, inconsistency of the slurry, variation in pad physical properties, and the lack of in-situ sensors [1]. Most CMPrnprocess control is based on the post-process measurement of film thickness. The state estimator is an important part ofrnthis run-to-run controller. Based on the measurements of pre- and post-polish film thickness, the state estimatorrnestimates the polish rate for the next run and sends it to the controller. The controller then calculates the polish time forrnthe next run, based on the estimated polish rate and the target film thickness. Consequently, the control performance isrndetermined by how well the state estimator can track the polish rate change. In addition to the different estimationrnmethods (e.g. EWMA, double-EWMA), the removal rate model also plays an essential role in the state estimation.