Modeling of Reverse Tone Etchback Shallow Trench Isolation Chemical Mechanical Polishing Terence Gan, a Tamba Tugbawa, a Brian Lee, a, * Duane S. Boning, a, ** ,z and Simon Jang b a Microsystems Technology Laboratories, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA b Taiwan Semiconductor Manufacturing Company, Hsin-Chu, Taiwan In this paper, we introduce a mathematical model for chemical mechanical polishing CMP of reverse tone etchback shallow trench isolation STI structures. We present a detailed formulation of the model and describe CMP experiments using a newly designed STI CMP characterization mask to validate the model. A methodology for extracting the model parameters is also proposed. An improved modeling methodology that incorporates density averaging effects fits the experimental data more accu- rately. Finally, we use the model to predict the effects of pre-CMP step height, pattern density, polish time, pad hardness, and slurry selectivity on dishing and nitride erosion. © 2001 The Electrochemical Society. DOI: 10.1149/1.1348266 All rights reserved. Manuscript submitted June 1, 2000; revised manuscript received November 21, 2000. Shallow trench isolation STI is an enabling process for ad- vanced ultralarge scale integration ULSI technologies. It can be scaled to fine dimensions while maintaining good planarity, good latch-up immunity, and low junction capacitance. 1,2 However, the chemical mechanical polishing CMP process is density dependent, 3,4 and single-step STI CMP can lead to significant oxide dishing and nitride erosion problems. Hence, an additional etchback step is commonly used before CMP to remove the overburden oxide over large active areas. The resulting CMP process is concerned only with removing the excess oxide in the trenches until the oxide reaches the same level as the nitride. This is a complicated process that depends on the relative removal rate of the oxide in the trenches and the nitride barrier in the raised area, which are both density dependent. There are currently no existing models for reverse tone etchback STI CMP, even though models for single-step STI CMP 5 and process control methods exist. 6 In this work, a mathematical model for reverse tone etchback STI CMP based on density and step height dependent oxide and nitride removal rates is presented. The model presented here builds on previous work. Stine et al. demonstrated the strong relationship between CMP removal rate and pattern density. 3 Smith et al. described integrated pattern density and step height models for oxide CMP. 7 Tugbawa et al. developed a removal rate vs. step height model for dual-material damascene CMP. 8 The reverse tone etchback STI CMP model presented here is an adaptation of Tugbawa et al.’s dual-material damascene CMP model, which greatly simplifies the analysis of reverse tone etchback STI CMP. Model Formulation Chemical mechanical polishing achieves global planarity through a combination of chemical action from the slurry and mechanical action from the polishing pad and slurry particles. The slurry chem- istry can be adjusted for varying degrees of selectivity between dif- ferent materials, but the contact between the polishing pad and the wafer is believed to be the main mechanism for removing the local step height to accomplish planarization. In STI CMP, we are inter- ested in changes in step height especially dishing and nitride ero- sion. This makes a step height based model a convenient starting point. When analyzing step height changes, we are not concerned with the details of the physical or chemical aspects of CMP; instead, we seek an effective and efficient model capturing the net effect of all contributing factors. From the measurement point of view, changes in local step height are easily made and used here to de- velop and confirm the STI CMP model. Figure 1 shows the cross section of a reverse tone etchback wafer before and after CMP, and we use it to define step height and nitride erosion as follows. The local step height is the vertical distance between the top of the nitride and the top of the oxide. It is positive when the oxide is above the nitride, and negative when the oxide recesses below the nitride. Negative step height is also known as dishing. Nitride erosion is simply the amount of nitride loss due to CMP. The model is explained graphically in Fig. 2, which shows how the oxide and nitride removal rates on a reverse tone etchback STI wafer change with step height. Polishing progresses from the right to the left of the graph as indicated by the arrows. Following the work of Grillaert et al., 9 if the initial local step height is very large, the pad will not contact the down area nitride initially, and there will be no down area polishing. The up area oxide polish rate is density dependent and is given by the expression RR ox = K ox /1 - , where K ox is the blanket oxide removal rate, and  is the nitride density. As the wafer polishes, only the up areas are polished and this causes the local step height to decrease. At some critical step height denoted by d max , the pad just touches the down area. Addi- tional polishing further reduces the local step height which increases the force of contact between the pad and the down area while it decreases the force between the pad and the up area. This causes the down area removal rate to increase and the up area removal rate to decrease. For simplicity, the model assumes that the down area re- moval rate increases linearly and the up area removal rate decreases linearly with decreasing step height. When the local step height is zero, we postulate that the down area removal rate is equal to K nit , the blanket nitride removal rate. Similarly, the up area polish rate is equal to K ox , the blanket oxide removal rate. Overpolishing causes dishing to occur. As long as the oxide removal rate is higher than the nitride removal rate, the amount of dishing increases with polish time. Eventually, the oxide removal rate and nitride removal rate become equal and steady-state dishing occurs. We denote this point by d ss . From Fig. 2 we can formulate the equations for the oxide and nitride removal rates as follows Oxide removal rate RR ox =  K ox 1 -  H  d max K ox 1 +  1 -  H d max  -d ss  H  d max 1 * Electrochemical Society Student Member. ** Electrochemical Society Active Member. z E-mail: boning@mtl.mit.edu Journal of The Electrochemical Society, 148 3 G159-G165 2001 0013-4651/2001/1483/G159/7/$7.00 © The Electrochemical Society, Inc. G159