Developing a modified ICP process to strip resist without damaging low-k dielectric film

Bob Guerra, Mattson Technology

Inductively coupled plasma ashing experiments demonstrate that lowering the pressure and temperature of the plasma chamber reduces the diffusion of oxygen into the low-k dielectric layer.

To fully take advantage of copper interconnects in semiconductor devices, low-k dielectric materials must be used to reduce interelectrode capacitance. Employing low-k materials with a dielectric constant of 2.5 ± 0.1 at 1 MHz instead of silicon dioxide with a dielectric constant between 3.9 and 4.0 can reduce interelectrode capacitance by a factor of 1.6 if the dielectric layer's integrity can be maintained through subsequent processing steps.

However, standard inductively coupled plasma (ICP) ashing and stripping methods used to remove photoresist can damage the low-k layer through oxidation, which induces water to intrude into the layer. As depicted in Figure 1, this damage mechanism occurs when etching performed to selectively expose the copper interconnect also exposes in the trench sidewall the nitride copper diffusion barrier, the dielectric film, a nitride layer that serves as a stop for the photoresist stripping step, and the photoresist layer itself. As a result of this isotropic reaction, the lateral diffusion of oxygen molecules and radicals often makes the capacitance of the low-k film higher than that of silicon dioxide, essentially defeating the purpose of using such a film.

 

Figure 1: Schematic diagram showing that low-k dielectric films are damaged during standard ICP ashing when oxygen diffuses into the film sidewall, attacking the silicon-to-methyl-group bonds and allowing water to intrude into the film.

Standard photoresist ashing occurs at an elevated temperature of 250°C in a reactive-oxygen atmosphere with a relatively high absolute pressure of 1 Torr. Reactive oxygen (O₂^*) consists of oxygen molecules in an excited state, and the excitation energy promotes the chemical reaction responsible for resist ashing. This oxidation process, expressed in the Arrhenius equation, makes ashing temperature dependent: higher temperatures increase the reaction rate in a nonlinear manner. Hence the need to perform ashing at elevated temperatures. Elevated temperatures, however, increase the rate of oxygen diffusion into the low-k material. Oxygen preferentially attacks the polymer's silicon-to-methyl-group bonds, and the resulting reaction converts SiCH₃ into moisture-absorbing SiO or SiOH. The presence of moisture in a low-k film raises the film's dielectric constant dramatically and causes its destruction.

This article describes experiments that were performed by Mattson Technology (Milpitas, CA) and International Sematech (Austin, TX) on hybrid organic siloxane polymer (HOSP) from Honeywell Electronic Materials (Sunnyvale, CA). The purpose of the experiments was to modify the ashing conditions in ICP chambers in order to reduce the diffusion of oxygen into the low-k dielectric layer while maintaining a commercially viable ashing rate. This reduction was accomplished by lowering the two factors that promote diffusion—temperature and chamber pressure—and by introducing a means other than O₂^* to increase the reaction rate.

Modifying the ICP Chamber

Reactive oxygen is generated in a standard ICP system (shown schematically in Figure 2) by exciting oxygen molecules in a radio-frequency (RF) plasma. Pure oxygen gas is introduced into the top of a cylindrical quartz chamber in which pressure is maintained at approximately 1 Torr by evacuating the chamber using a dry mechanical pump. Inside the chamber, the oxygen is exposed to plasma using a radio frequency of 13.56 MHz and a power level of 700­1000 W. This RF power is coupled into the plasma via an antenna coil wrapped around the outside of the chamber. A Faraday shield between the antenna and the chamber wall ensures that the coupling is inductive.

 

Figure 2: The standard ICP ashing process uses inductively coupled plasma to generate reactive oxygen radicals. A screen neutralizes any ions leaving the plasma region.

The standard ICP process uses excited oxygen molecules rather than oxygen ions (O⁺ or O^­) in the ashing reaction. Ions are removed from the plasma as it flows downward through the chamber and passes through a grounded screen. Thus, reactive oxygen molecules are the predominant species reaching the wafer mounted at the chamber bottom.

The ICP chamber was modified by adding a turbomolecular vacuum pump between the chamber and the dry mechanical pump (which serves as a roughing pump). This modification reduces chamber pressure by two orders of magnitude to 5 mTorr. The turbopump has an inlet diameter of 25 cm and a throughput of 1600 L/sec. It communicates directly with the vacuum chamber through a pendulum valve from VAT (Woburn, MA). High throughput and chamber pressure control are achieved by partially opening the valve. The turbopump and valve can be bypassed by means of a throttle valve, which allows the system to operate in high-pressure mode, evacuated solely by the dry mechanical pump. The turbopump configuration lowers the chamber pressure and therefore reduces the diffusion (concentration) gradient driving oxygen into the low-k material. It also increases the mean-free path in the chamber enough to allow ashing to proceed at high rates.

Low-pressure ashing relies on oxygen ions rather than electrically neutral excited oxygen molecules to provide excitation energy for the ashing reaction. By removing the ion screen at the bottom of the ICP chamber, these ions can leave the plasma generation region near the RF antenna and reach the wafer surface. Providing a separate RF source (again at 13.56 MHz) at a lower power level allows self-biasing across the plasma sheath next to the wafer. The separate RF bias source applies power through its own matching network to the wafer substrate. The RF voltage is applied between the substrate and ground.

During the bias source's positive half-cycle, electrons are attracted across the plasma sheath to the wafer, where they drain off to ground. Because ions are ~10⁴ times heavier than neutral excited molecules, they are also much less mobile, resulting in a much lower ion current during the negative half-cycle. This diode action across the plasma sheath leaves the bulk plasma with a net positive charge. The electric field across the plasma sheath is normal to the wafer surface and charged so that positive oxygen ions from the plasma drive toward the wafer. As these ions encounter the wafer's photoresist coating, they oxidize it and ash it away. The ions' ionization and kinetic energy provide enough activation to allow the ashing reaction to proceed, even at a reduced substrate temperature.

This modified ashing procedure greatly reduces the two process variables that promote low-k damage during ashing: temperature and pressure. The process proceeds anisotropically because, as Figure 3 shows, the dose of reactive ions is proportional to the cosine of the surface normal to the electric-field-driven ion trajectories. Horizontal photoresist surfaces above the low-k layer receive maximum dosage from the ions raining down vertically, while the vertical low-k surfaces adjacent to etched areas receive a much smaller dosage. Combining all these effects makes it possible to ash away photoresist from process wafers without damaging the low-k properties of dielectric films.

Determining the Effects of Different Stripping Conditions on Dielectric Film

Mattson and International Sematech conducted an experiment on several wafers containing copper test features to determine the effects of various photoresist stripping processes on HOSP dielectric film. Ten test wafers containing a conventional silicon dioxide dielectric layer were also processed to serve as control samples.

 

Figure 3: Anisotropic ashing occurs when ions are accelerated by parallel electric-field lines created by the self-biasing of the wafer.

First, Sematech coated low-k dielectric film on the wafers, applied the appropriate nitride capping layers, added patterned photoresist, and etched trenches through the cap and low-k film layers. After the trenches were etched, the photoresist was stripped from the wafers by four photoresist stripping companies using different proprietary techniques and equipment. Sematech then processed the wafers to completion by depositing the diffusion barrier layer, filling the interconnect trenches with copper, and performing chemical-mechanical polishing to remove extraneous copper down to the capping layer.

Two types of test structures—interdigitated comb and serpentine resistor—were inlaid in the low-k film layer to quantify the damage done to the dielectric material during photoresist ashing. Using interdigitated comb structures, as illustrated in Figure 4, the experimenters were able to measure interelectrode capacitances and leakage currents across gaps filled with the dielectric material. Because the damage caused by photoresist stripping can raise the film's dielectric constant, the comb test's low capacitance is a good measure of that damage. Moreover, lower levels of leakage current indicate less damage to the dielectric film.

 

Figure 4: Interdigitated comb structures inlaid in the low-k film layer were used to quantify the damage done to the dielectric material during photoresist ashing.

Using serpentine resistor structures, as illustrated in Figure 5, the experimenters could gauge damage that resulted in low-k film shrinkage or removal. Any reduction in film volume increases the trench width, which increases the cross section to be filled with copper and ultimately reduces the structure's resistance. Accordingly, higher serpentine resistance signals less low-k film damage.

 

Figure 5: Serpentine resistor structures permitted experimenters to quantify the degree to which the photoresist stripping process enlarges the trenches cut into the low-k layer.

Figures 6 through 8 show results of these tests. The abscissa on each of these graphs is a percentage across the data distribution. Thus, the median value for each group of test values is at the 50% level. All of the curves approximate an S shape typical of a Gaussian distribution. Most of the data are clustered near the median, with extraordinary values appearing in the wings. The values taken were the medians (50% probability level) unless otherwise noted.

Figure 6 presents the comb capacitance results for the wafers with dielectric film that had been stripped at the four different companies (identified as processes A, B, C, and D) and for the 10 control wafers. The ordinate is on a linear scale. The comb capacitance for the process A dielectric film was 6.3 pF, which was approximately 15% lower than that of the silicon dioxide film. Processes B and C severely damaged the low-k film, causing its dielectric constant to rise to levels considerably above those of the conventional silicon dioxide dielectric. At 11.2 pF (equivalent to a dielectric constant of approximately 6), the comb capacitance for process B film was the highest of the four test groups, being almost 60% higher than that of the silicon dioxide film. The comb capacitance for the process C film was 9.653 pF (equivalent to a dielectric constant of approximately 5), which was approximately 30% higher than that of the silicon dioxide film. The dielectric film on the process D wafers, which were processed by Mattson using an Aspen III strip system with a modified ICP chamber, had a comb capacitance of 5.68 pF, which was 20% lower than that of the silicon dioxide film.

 

Figure 6: Comb capacitance results for low-k dielectric material that underwent four different photoresist stripping processes (A-D) compared with comb capacitance results for 10 control wafers with conventional silicon dioxide film.

Figure 7 shows the comb leakage current for the four test groups only. The ordinate is on a logarithmic scale. The median values for the process A dielectric film were almost two orders of magnitude lower than those for the process B film. As seen in the comb capacitance data, processes A and D fared considerably better than processes B and C. The process A distribution did not correspond well to a Gaussian distribution; it was skewed to the low-leakage side. While its median was lower than that of process D, its low-side wing was so compressed that the outliers were higher than those of process D. The process A distribution was so extended on the high-leakage side that the whole distribution spread over four orders of magnitude. These results indicate either a problem with the data or a problem with the process's repeatability. While the process D data spread over approximately two orders of magnitude, 80% of the samples (from the 10% level to the 90% level) fell within one order of magnitude. The process D median was more than half an order of magnitude below that of the process C data and a full order of magnitude below that of the process B data.

 

Figure 7: Comb leakage current results for low-k dielectric material that underwent four different photoresist stripping processes.

Figure 8 shows the results of the serpentine resistance test for the four processes. The ordinate is on a linear scale. All of the medians fell within ±13% of the overall median of 21.25 W. Processes B and C again fared the worst, with process C showing the lowest serpentine resistance by far and, therefore, the most dielectric film dimensional change or damage. Process A did considerably better than processes B and C, but also showed a significantly larger distribution range than the other processes. Process D performed better than processes B and C, but worse than process A.

 

Figure 8: Serpentine resistance results for low-k dielectric material that underwent four different photoresist stripping processes.

The experiments revealed a series of problems affecting all four groups of test wafers. First, CMP quality was generally poor, which was attributable to larger-than-normal nonuniformity in the copper plating thickness. While this problem affected the test results, it was not attributable to the operations performed at the four photoresist stripping companies. Second, cap delamination appeared on all wafers, especially around the wafer edges and in the scribe-line structures. The experimenters were unable to explain this problem. Third, because of overpolishing, two wafers from process A had to be scrapped. This fact indicates that the high serpentine resistance of process A wafers may have been a result of overpolishing. Moreover, the wide comb-leakage-current data distribution for process A wafers may have been caused by varying amounts of cap material left on the process A wafers.

Increasing Low-k Material Performance

Mattson conducted two other series of experiments in an effort to increase the performance of low-k dielectric materials: an ICP pressure experiment and a nitride etching experiment.

ICP Pressure Experiment. Tests were performed to determine the effect of pressure on the ICP ashing process. To measure that effect, wafers were prepared in much the same way as the wafers in Sematech's stripping-process experiment and then divided into six groups, each of which was stripped in an Aspen III low-pressure tool at a different oxygen-pressure level between 7 and 150 mTorr. Figure 9 shows the comb capacitance results of this test, in which the 150-mTorr group showed by far the worst results and the 7-mTorr group showed the best.

 

Figure 9: Results of tests showing the effect of reduced chamber pressure on low-k film. Reducing chamber pressure below 150 mTorr reduces damage to low-k dielectric material.

These results confirm that it is advantageous to operate the stripping tool at low absolute gas pressure. While the 7-mTorr group demonstrated the best comb capacitance and serpentine resistance results, it performed marginally better than the next-best group (the 15-mTorr group). The approximately twofold pressure difference between the 7- and the 15-mTorr groups represented a 1.5% difference in comb capacitance (6.6 versus 6.7 pF), while the tenfold pressure difference between the 15- and the 150-mTorr groups represented a 30% difference in comb capacitance (6.7 versus 9.6 pF). The serpentine resistance data showed a similar saturation effect. These results indicate that reducing chamber pressure below 15 mTorr reduces damage to low-k dielectric materials.

Nitride Etching Experiment. All semiconductor fabrication processes involving copper and low-k dielectrics require that a diffusion barrier be placed between the copper and the low-k dielectric. This barrier layer, typically made by vapor depositing nitride material, prevents the diffusion of copper atoms into the low-k dielectric polymer material. But because nitride is an insulator, it must be etched away to allow the copper to make electrical contact with the next interconnect layer. The anisotropic strip system can be used to perform this etch step by adding carbon tetrafluoride (CF₄) to the molecular oxygen entering the plasma generator. Anisotropic plasma etch, combined with fluorine ions from the CF₄, has proven to be quite effective for etching away the nitride diffusion-barrier material without damaging the low-k dielectric material.

Figure 10 shows the results of an experiment that was conducted to verify the effectiveness of anisotropic nitride etch. Figure 10a is a scanning electron microscope image of a wafer cross section before the ashing and etching processes were performed. The micrograph clearly shows the photoresist layer at the top, which is pierced by holes left to permit the etching of the dielectric layer. The dielectric etch has proceeded down to the nitride layer at the bottom but has not etched into it. Figure 10b shows the effects of running the ICP system for 60 seconds with the bias source turned off. Without bias, no nitride etching occurs, but the photoresist material has been removed. Figure 10c shows the effect of a 30-second anisotropic nitride etch with bias turned on. The fluorine-doped anisotropic plasma has effectively removed the nitride at the bottoms of the rather deep wells previously etched through the low-k film layer. A potential problem with etching the nitride layer in this manner is the danger of sputtering copper onto the dielectric walls once the overlying nitride has been removed. Figure 10d shows that by controlling the plasma energy, it is possible to avoid sputtering the copper even after overetching by 100%.

 

Figure 10: SEM images of a wafer cross section showing the effectiveness of using low-pressure anisotropic ICP ashing to etch a nitride barrier layer: (a) before ashing and etching, (b) after 60-second nitride etch with the bias source turned off, (c) after 30-second nitride etch with bias turned on, and (d) after controlling plasma energy to avoid sputtering the copper.

Conclusion

The experiments discussed in this article demonstrate the advantage of using low-pressure anisotropic ICP ashing for stripping photoresist from semiconductor wafers incorporating copper and low-k dielectric materials. The key to success is to modify the ashing conditions to reduce the diffusion of oxygen into the low-k film while maintaining a commercially viable ashing rate. These goals were achieved by altering the ICP chamber to reduce temperature and chamber pressure, which promote oxygen diffusion, and by using an anisotropic method to increase the reaction rate. It has been shown that a strip system with a modified ICP chamber completely removes photoresist while reducing oxygen-induced damage to the low-k polymer material. In addition, the experiments demonstrate that the strip tool used to remove photoresist can also be used to etch away exposed nitride diffusion-barrier material without sputtering the underlying copper onto the dielectric sidewalls.

Acknowledgments

The author would like to thank Thieu Jacobs and Lee Tye, project engineers in the interconnect division of International Sematech, for their help in conducting the experiments and analyzing the data discussed in this article.

Bob Guerra is senior manager of advanced applications in Mattson Technology's strip/etch product group (Milpitas, CA). He holds patents in the fields of semiconductor process technology and polymer materials composition. He received a BS in chemical engineering from the University of California, Berkeley, and has pursued graduate studies in materials science at Stanford University in Palo Alto, CA. (Guerra can be reached at 510/492-6363 or bob.guerra@mattson.com.)

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit

Questions/comments about MICRO Magazine? E-mail us at cheynman@gmail.com.

© 2007 Tom Cheyney
All rights reserved.