Chapter 4 Experiment and Discussion for SiO 2 and Cu Blanket Wafer
4.4 Cu CMP Experiment
The same comparison between dynamic programming and constant removal rate were discussed as we did for the SiO2 CMP experiment. The parameters for the
constant removal rate were also found through the experiment.
4.4.1 Change Down Force as the Admissible Input
The duration of polishing was 118 seconds and the removed thickness was 6000 Å. The required removal rate was 3051 Å/minute. Because of the fixed rotational speed here, the required removal rate was found by changing the down force to be 5.1 psi. From the experimental results which were listed in Table 4.7 and 4.8, the constant removal rate has the better thickness removal but the dynamic programming possesses a little better non-planarization index. However, the difference of non-planarization index between dynamic programming and constant removal rate is very small and it may be considered within statistical error of experimental data. It also shows that the down force is not a major factor to influence the non-planarization index in Cu CMP.
As shown in Fig. 3.9, we see that the non-planarization index data under 3 psi and 5 psi is within statistical error with each other. This explains why there is no degradation in non-planarization index when the down force is decreased from 5.1 psi to 3 psi, i.e. there is no degradation in non-planarization index when Cu CMP process is changed from constant removal rate operation mode to dynamic programming operation mode. The inaccuracy of removed thickness may be caused by the lower predicted value of removal rate on the smaller down force. It means that the model for Cu CMP process needs to be modified.
4.4.2 Change Rotational Speed as the Admissible Input
The duration of polishing was 85 seconds and the removed thickness was 6000 Å.
The required removal rate was 4235 Å/minute. Because of the fixed down force here,
the required removal rate was found by changing the rotational speed to be 50 rpm.
Table 4.9 and table 4.10 show that the dynamic programming operation mode is 24%
less than the constant removal rate operation mode in non-planarization index. The error of the removed thickness may be caused by higher predicted value of removal rate at faster rotational speed and lower predicted value of removal rate at slower speed, there was not enough time to remedy lower removal rate at slower rotational speed by higher removal rate at faster rotational speed. Therefore removal thickness of dynamic programming operation mode is less than that of constant removal rate operation mode. However, it made a significant improvement of non-planarization index through dynamic programming operation of rotational speed. This shows rotational speed is a major parameter affecting the non-planarization index of Cu CMP process.
4.4.3 Change Down Force and Rotational Speed as the Admissible Inputs Simultaneously
The duration of polishing was 60 seconds and the removed thickness was 6000Å.
The required constant removal rate was 6000 Å/minute. We simultaneously increased the down force and rotational speed to be 4.9 psi and 57 rpm, respectively. The experimental results were listed in Table 4.11 and 4.12. The dynamic programming operation still provides 10% improvement of non-planarization index than constant removal rate operation.
4.5 Discussion and Summary
Three cases of CMP process to use down force, rotational speed and both down
force and rotational speed as admissible inputs were examined in this chapter. In the SiO2 CMP experiment, a summary of non-planarization index was listed in Table 4.13.
Non-planarization index of three cases were all improved and the errors of the removed thickness were within 8%. The model could predict the removal rate well. It illustrated that the multi-step SiO2 CMP was feasible to implement. Besides, it could be practiced on IPEC 372M CMP tool.
Slurry chemicals play an important role in the Cu CMP process. The formation of a non-native passivation layer by the passivating chemical (e.g. citric acid) in the slurry, the dissolution of Cu or the abraded materials by abrasives from surface layer are all determined by the chemical environment in the slurry [20]. In the Cu CMP experiment, a summary of non-planarization index was listed in Table 4.14. The result shows that the rotational speed is the main factor to influence non-planarization index.
When we made the rotational speed change, non-planarization index improved 10% at least. It means that the higher the rotational speed, the faster the refresh rate of the slurry underneath the wafer and then increased the removal rate. It may also cause the worse non-uniformity of the slurry to transport on the entire wafer and exercise influence over the non-uniformity of the removal rate. For this reason, the interactions between the mechanical and chemical parameters needed to be considered anew. This also indicated that the current model was not sufficient to describe the Cu CMP. The errors of the removed thickness were above 14% and the maximum error was 25%.
We brought up three ideas to eliminate the error of the removed thickness. The first way is to modify the model and make it more comprehensive and correct. The second way is to get a great number of removal rate data corresponding to every value of the admissible input by experiment and then get a better regression model. The last way is to assemble a sensor to measure the thickness at every stage and we could implement the dynamic programming solution which was illustrated in chapter 3. Mirra, the
CMP tool provided by Applied Materials, had the technology of In Situ Rate Monitor (ISRM). ISRM technology detects film thickness changes during polishing by laser interferometry system that allows the user to precisely defined material removal.
Chapter 5 Pattern Copper Wafer
5.1 Introduction
In pattern Cu CMP, the copper is removed following a three-step procedure as show in Fig. 5.1. In the planarization step, the overburden Cu is removed and the objective is to reduce the step height (difference between high and low features).
This is followed by the planarization/overpolishing step where overburden metal and some barrier are removed. The wafer is further overpolished to remove residual metal.
By intuition, these two defects of dishing and erosion should become more serious as the overpolishing time of the Cu CMP process increased. Therefore, the non-planarization index of Cu CMP must be reduced to minimum to minimize the dishing and erosion of the polished wafer. The pattern dependence of the metal dishing and oxide erosion of Cu damascene structure had been reported by some paper [21][22][23][29]. In this study, we focus on the step height reduction, i.e.
planarization efficiency.
The model proposed by Chen and Lee [24][25] provides an excellent description of step height reduction when the wafer and pad contact each other at any point on the interface. The major assumptions in this model are：1) the pressure difference between the high area and the low area is proportional to the step height, and 2) higher area will experience higher pressure. The model proposed by Fu and Chandra [26]
incorporated the effects of feature scale wafer geometries (e.g., line-width, pitch and pattern density). However, the model was also assumed that the wafer and pad were in contact at any point of the interface. It means that the pad contact with both high and
low areas. It may not be correct when the initial step height is large than pad bending, especially for a hard pad. The Rodel IC1400 pad is composed of the IC1000 stacked on top of the Suba IV and it belongs to a hard pad. Typically, the IC1400 pad is used in pattern wafer experiments. The advantage of the hard pad for the planarization step is that it has small deformations because of its limited compressibility and will ideally touch only the high area to provide good planarity. Generally, if a soft pad, e.g., Rodel Politex Regular E.TM pad, was used to polish the pattern wafer, it will lead to the step topography still exist. Because the deformation and compressibility under the down force of the soft pad is serious, the high and low areas were polished simultaneously.
As a consequence, the step topography still would exist and the goal of planarization was not reached.
5.2 Model Development
Since we analyzed the experimental data in chapter 4, we found that the Luo and Dornfeld equation was unable to fit very well in copper CMP process. Therefore, the power function was used.
b aV KP RR=
The values of the exponents a and b were determined by the method of regression and were 0.38 and 0.85, respectively. As shown in Fig. 5.3, the experimental results can be fitted more accurately as compared with the Luo and Dornfeld equation. The model SSE values for the power function model and the Luo and Dornfeld model were 143270 and 388260, respectively. Then, the power function model is used to predict the Cu removal rate.
To physically explain the development of step height reduction during Cu CMP process, the following assumptions were made：
1. Power function is valid for all polish conditions.
2. Force redistribution due to pad bending is proportional to step height.
3. Critical pressure is determined by the pressure of the low area.
4. The pattern effect is negligible to the rotational speed.
Due to the pad bending, there exists force redistribution. This causes the contact force to drop in the low area. To maintain overall force balance due to the applied down pressure, P0, there is a corresponding rise in the force in the high area. The total area in the high region with unit thickness (length into the page) is (b-a)·1 and the low area is a·1. The modified pressure equations including pad bending effects are
1 and pitch and α is the pad bending stiffness parameter. Fig 5.2 shows the schematic picture. For PL<0, only the high area is polished and we have
b and from Eq. 5.6 and Eq. 5.7 we can formulate the model for the pressure of low area
smaller than zero or equal zero. height. The model switch to the following set from Eq. 5.3 and Eq. 5.4 when the critical pressure is larger than zero
⎭⎬
In the model development, Eq. 5.2 represented the additional force imparted on the high area and corresponding decrease in the force imparted on the low area due to bending of the pad. α is the bending stiffness parameter in the model. It is not known a priori in the model. For the later comparison of the model prediction to the experimental observation, α is used as a fitting parameter.
5.3 Simulation
We used the experimental data from Stavreva etc. [22] and in the experiments, Cu CMP is carried out by the perforated Rodel IC 1000/Suba IV stacked pad. One of the experimental data points was chosen to obtain the α value corresponding to the experimental conditions. Model prediction was compared with experimental data in
Fig. 5.4. The relevant experimental parameters were given below:
Once appropriate models were constructed (e.g. Eq. 5.8-5.13) and the model parameter α were obtained, we were able to analysis the step height for different operation values through the simulation. The polish time and initial step height were set to be 1 minute and 4000 Å, respectively.
The polishing action progresses in both high and low areas. The most efficient method in planarization is to remove the high area only, but this is difficult in a real polishing process. The planarization efficiency which indicates the step height reduction in the CMP process, can be defined as
Area
where △low area is the removed thickness of low area and △high area is the removed thickness of high area.
Fig. 5.5 shows the planarization efficiency can be increased under a lower down force. When the down force is larger, there is the more possibility of polishing the low area and decreases the planarization efficiency. This means that the step height reduction can be improved by polishing at the lower pressure. The method of the dynamic programming could implement this kind of operation. The down force is decreased as polishing continues to the end and planarization efficiency is increased according to the simulation result. More flat surface of wafer is obtained. Moreover,
non-planarization index can be improved by decreasing the rotational speed as discussed in chapter 4. Consequently, the planarity could be improved by decreasing the down force and rotational speed simultaneously.
5.4 Discussion and Summary
In this chapter we constructed the model for describing the step height reduction and obtained the effect of down force on the planarization efficiency. Through the result of simulation we could find that the lower down force can increase the planarization efficiency. The method of dynamic programming could decreases the down force and rotational speed simultaneously and produces the higher planarization efficiency and the less non-planarization index at the same time.
Step height, dishing and erosion are three parameters that are typically used to characterize CMP of pattern wafers. If we could minimize the step height at the end of the planarization step, the high removal selectivity of the barrier layer between Cu and SiO2 could be applied in the planarization/overpolishing step. Dishing and erosion mainly occurs during the overpolished step which is often necessary to assure complete removal of copper and barrier residues across the entire wafer [30][31].
Since the non-planarization index has been improved in the planarization step, the duration of the overpolished step could be reduced and then the goal of planarization is achieved by the three-step procedure.
Chapter 6 Conclusion and Future Work
In this study, the subject was emphasized on the mechanical parameters of the CMP process. The down force and rotational speed were taken as the control parameter. We applied the control method of dynamic programming to the CMP process and experimented with blanket SiO2 and Cu wafers. The impacts of dynamic operation and constant removal operation on the non-planarization index after CMP process were discussed. The non-planarization index could be improved by dynamic programming operation and the rotational speed seems to be a major factor to influence the non-planarization index. The effect of the down force on the pattern wafer was focused on the planarization efficiency and the simulation result was presented. According to the simulation, the lower down force improves the planarization efficiency. Therefore, the dynamic programming operation is a feasible operational strategy for the CMP process to improve non-planarization index and planarization efficiency.
For future works, the following directions can be considered﹕(1) The CMP model proposed by Luo and Dornfeld should be reconstructed or modified in the copper CMP process. (2) The number of experimental wafers should be increased to provide more valid data. (3) In order to apply the dynamic programming operation on the pattern wafer, the performance index has to be reconsidered to improve step height. (4) The experiment can be carried out to varify the simulation result.
Table 2-1 The Parameters of CMP Process
Concentration Chemical Durability/Reactivity Isoelectric point (pH) Wafer Curvature
zeta potential Wafer Mounting Stability of the Suspension Film Stack
Slurry Flow Rate Film Stress Transport Under the Wafer Film Hardness
Temperature Creep
Pressure Work Hardening, Fatigue
Velocity Film Microstructure
Table 3-1 Process Parameters of SiO2 CMP Experiment
Fixed Parameter Value
Down Force 2, 3, 4, 7 psi
Back Pressure 1 psi
Carrier Speed 20, 30, 70 rpm
Platen Speed 20, 30, 70 rpm
Polish Time 1 minute
Slurry Flow Rate 150 ml/minute
Pre-wet Pad Speed 28 rpm
Pre-wet Duration 10 second
Pre-wet Flow Rate 300 ml/minute
Pad Rodel IC1400
Pad Conditioning (after polishing) Value
Condition Pressure 0.3 psi
Platen Speed 30 rpm
Clean Speed 4 rpm
Slurry Flow Rate 150 ml/minute
Duration 30 second
Frequency 1 wafer
Table 3-2 Slurry Formulation of SiO2 CMP
Species Concentration
Commercial Slurry Cabot SS-25 1
Dilution DI water 1
Table 3-3 The Experimental Results of SiO2 CMPto Determine C1 and C2
Set 1 Set 2
Down Force (psi) 3 7
Removal Rate (Å/minute) 618 1440
Table 3-4 The Experiment Data of SiO2 CMP
Down Force (psi) 2 3 7
Rotational Speed
fixed at 30 rpm Removal Rate (Å/minute) 387 618 1440
Rotational Speed (rpm) 20 30 70
Down Force
fixed at 4 psi Removal Rate (Å/minute) 649 883 1431
Table 3-5 The Values of Admissible Inputs by Corresponding Down Force
No. 1 2 3 4 5 6 7
Down Force (psi) 2 2.5 3 3.5 4 7 0
Removal Rate
(Å/minute) 426 521 618 715 814 1440 0
Table 3-6 The Result of Dynamic Programming of SiO2 CMP with Down Force
Phase 1 2 3 4
Duration (second) 178 48 47 90
Down Force (psi) 7 4 2.5 2
Table 3-7 The Values of Admissible Inputs by Corresponding Rotational Speed
No. 1 2 3 4 5 6 7
Rotational Speed
(rpm) 20 25 30 35 40 70 0
Removal Rate
(Å/minute) 639 730 814 893 968 1354 0
Table 3-8 The Result of Dynamic Programming of SiO2 CMP with Rotational Speed
Phase 1 2 3 4 5
Duration (second) 172 19 10 42 91
Rotational Speed (rpm) 70 40 35 30 25
Table 3-9 The Result of Dynamic Programming of SiO2 CMP with Two Inputs
Phase 1 2 3 4 5
Duration (second) 96 49 33 17 42
Down Force (psi) 7 7 4 2.5 2
Rotational Speed (rpm) 70 40 20 20 20
Table 3-10 Process Parameters of Cu CMP Experiment
Fixed Parameter Value
Back Pressure 1 psi
Down Force 3, 3.5, 5, 7 psi
Carrier Speed 30, 50, 70 rpm
Platen Speed 30, 50, 70 rpm
Polish Time 1 minute
Slurry Flow Rate 150 ml/minute
Pre-wet Pad Speed 28 rpm
Pre-wet Duration 10 second
Pre-wet Flow Rate 300 ml/minute
Pad Rodel IC1400
Pad condition No (Manual Brushing)
Table 3-11 Slurry Formulation of Cu CMP
Species Concentration
Abrasive Al2O3 (EXTEC 0.1µm) 2 wt %
Oxidizer HNO3 2 vol %
Complex Agent Citric Acid 0.01 M
Dilution DI water Remaining Balance of Slurry
Table 3-12 The Experimental Results of Cu CMPto Determine C1 and C2
Set 1 Set 2
Down Force (psi) 5 7
Removal Rate (Å/minute) 2942 3508
Table 3-13 The Experiment Data of Cu CMP
Down Force (psi) 3 5 7
Rotational Speed
fixed at 30 rpm Removal Rate (Å/minute) 2522 2942 3508
Rotational Speed (rpm) 30 50 70
Down Force
fixed at 3.5 psi Removal Rate (Å/minute) 2595 4396 5266
Table 3-14 The Values of Admissible Inputs by Corresponding Down Force
No. 1 2 3 4 5 6 7
Down Force (psi) 3 3.5 4 4.5 5 7 0
Removal Rate
(Å/minute) 2257 2445 2620 2786 2942 3508 0
Table 3-15 The Result of Dynamic Programming of Cu CMP with Down Force
Phase 1 2 3 4 5
Duration (second) 68 9 8 3 30
Down Force (psi) 7 5 4 3.5 3
Table 3-16 The Values of Admissible Inputs by Corresponding Rotational Speed
No. 1 2 3 4 5 6 7
Rotational Speed (rpm)
30 35 40 45 50 70 0
Removal Rate (Å/minute)
2445 2852 3259 3667 4074 5704 0
Table 3-17 The Result of Dynamic Programming of Cu CMP with Rotational Speed
Phase 1 2 3 4 5
Duration (second) 41 7 3 7 27
Rotational Speed (rpm) 70 50 45 35 30
Table 3-18 The Result of Dynamic Programming of Cu CMP with Two Inputs
Phase 1 2 3 4 5 6
Duration (second) 31 3 4 4 7 11
Down Force (psi) 7 7 7 7 5 4
Rotational Speed (rpm) 70 50 45 40 30 30
Table 4.1 SiO2 Blanket Wafer Experimental Result by Down Force with Dynamic Programming
Point NO. 1 2 3 4 5 6 7 8 9
Pre-Polish
Thickness (Å) 9009 9004 9016 9003 8997 9031 9059 9009 8998 After-Polish
Thickness (Å) 2999 3918 3361 3841 3805 3979 3204 3724 2807 Removed
Thickness (Å) 6010 5086 5655 5162 5192 5052 5855 5285 6191 Average Removed Thickness (Å) 5499 Non-Planarization Index 435
Table 4.2 SiO2 Blanket Wafer Experimental Result by Down Force with Constant Removal Rate
Point NO. 1 2 3 4 5 6 7 8 9
Pre-Polish
Thickness (Å) 9005 9004 9013 9003 8997 9026 9052 8999 8989 After-Polish
Thickness (Å) 2190 3828 3714 2903 2449 3367 2112 3765 3673 Removed
Thickness (Å) 6815 5176 5299 6100 6548 5659 6940 5234 5316 Average Removed Thickness (Å) 5899 Non-Planarization Index 717
Table 4.3 SiO2 Blanket Wafer Experimental Result by Rotational Speed with Dynamic Programming
Point NO. 1 2 3 4 5 6 7 8 9
Pre-Polish
Thickness (Å) 9008 9005 9017 9007 8999 9032 9058 9006 8996 After-Polish
Thickness (Å) 2122 2820 2584 3230 2185 3347 2916 3553 3351 Removed
Thickness (Å) 6886 6185 6433 5777 6814 5685 6142 5453 5645 Average Removed Thickness (Å) 6113 Non-Planarization Index 518
Table 4.4 SiO2 Blanket Wafer Experimental Result by Rotational Speed with Constant Removal Rate
Point NO. 1 2 3 4 5 6 7 8 9
Pre-Polish
Thickness (Å) 9004 8995 9002 8997 8988 9025 9049 9001 8986 After-Polish
Thickness (Å) 1867 3747 3915 3530 3484 2936 2449 3374 2519 Removed
Thickness (Å) 7137 5248 5087 5467 5504 6089 6600 5627 6467 Average Removed Thickness (Å) 5914 Non-Planarization Index 697
Table 4.5 SiO2 Blanket Wafer Experimental Result by Two Admissible Inputs with Dynamic Programming
Point NO. 1 2 3 4 5 6 7 8 9
Pre-Polish
Thickness (Å) 9013 9012 9023 9013 9003 9033 9057 9009 8997 After-Polish
Thickness (Å) 2400 3673 2569 3525 3231 3789 2961 3315 2820 Removed
Thickness (Å) 6613 5339 6454 5488 5772 5244 6096 5694 6177 Average Removed Thickness (Å) 5875 Non-Planarization Index 487
Table 4.6 SiO2 Blanket Wafer Experimental Result by Two Admissible Inputs with Constant Removal Rate
Point NO. 1 2 3 4 5 6 7 8 9
Pre-Polish
Thickness (Å) 8995 8992 9000 8995 8988 9020 9044 9000 8988 After-Polish
Thickness (Å) 1838 3355 3454 3008 2602 3395 2691 3676 3386 Removed
Thickness (Å) 7157 5637 5546 5987 6386 5625 6353 5324 5602 Average Removed Thickness (Å) 5957 Non-Planarization Index 580
Table 4.7 Cu Blanket Wafer Experimental Result by Down Force with Dynamic Programming
Point NO. 1 2 3 4 5 6 7 8 9
Pre-Polish Thickness (Å)
20000 20000 20000 20000 20000 20000 20000 20000 20000
After-Polish Thickness (Å)
13576 13702 13346 13037 12898 12313 11935 11954 9822
Removed Thickness (Å)
6424 6298 6654 6963 7102 7687 8065 8046 10178
Average Removed Thickness (Å) 7491 Non-Planarization Index 1204
Table 4.8 Cu Blanket Wafer Experimental Result by Down Force with Constant
Table 4.8 Cu Blanket Wafer Experimental Result by Down Force with Constant