Cleaning and Restoring k Value of Porous MSQ Films

Planned increases in IC performance, such as faster clock speeds and lower power consumption, require the use of low-k materials for interconnects in chip manufacturing. Dense low-k materials with k values in the range of 2.8-3.0 are in manufacturing at the 90 nm node. The International Technology Roadmap for Semiconductors (ITRS) indicates that porous low-k materials with k<2.4 will be required for the 65 nm node by 2006. Patterning of porous low-k materials exposes the film to plasma etching, plasma ashing for resist stripping, liquid chemicals for etch residue removal, and water for final rinsing. Typical ash processes result in carbon depletion and modification of the surface properties from hydrophobic to hydrophilic, thereby increasing the moisture sensitivity of the low-k material. The patterning of low-k materials, such as porous methylsilsesquioxane (MSQ), typically results in damage by causing changes in the chemical structure and/or by leaving absorbed chemical residues in the porous matrix. These effects can lead to an increase in k value, poisoning of photoresist in subsequent patterning processes, formation of voids after copper deposition and annealing, as well as reduced adhesion of barrier layers. The uptake of moisture is generally believed to be the cause of the increase in k after etching and ashing due to differences in dielectric constant between the low-k material and water, which are 2.2 and 78, respectively.

Cleaning the surface after patterning presents many challenges. The goal is to develop a chemical process that removes post-etch and post-ash residues that contain carbon, silicon, oxygen and copper without attacking the low-k material.¹ In our work with porous MSQ, we have partnered with five specialty chemical manufacturers to test 19 chemical formulations. Of the 19, five gave acceptable results as indicated by SEM analysis of etched via patterns. These five chemistries, which included both aqueous and solvent formulations, required treatment times ranging from 90 seconds to 15 minutes. Many of these formulations have only recently become commercially available, indicating that there is still a lot of development activity in this area.

Restoring the low-k material after ashing and cleaning may be required to minimize the effective dielectric constant. Therefore, in the work described herein we present a method to restore the k value and surface hydrophobicity of the low-k material using hexamethyldisilazane (HMDS). The proposed treatment is readily integrated into existing wafer cleaning tools.

Residue removal for BEOL applications requires automated tools to be very flexible in terms of the chemical compatibility of the materials of construction, process temperatures and chemical dispense times. Figure 1 shows a schematic diagram of the spray processor used in this study, FSI International's ZETA Surface Conditioning System. It is a batch spray processor that uses centrifugal force for enhanced particle removal and drying. The process chemistry can be dispensed via center and side spray posts from a fresh or recirculated source. The chemicals are stored and dispensed under a nitrogen atmosphere to minimize chemical degradation and maximize bath life. The chemical temperature is monitored at the chemical heater and in the process bowl to accurately control the on-wafer chemical temperature. Temperature and flow can be monitored and compensated using a reaction rate algorithm to accurately control film loss.

1. The ZETA batch processor utilizes centrifugal force, center and side chemistry dispense, fresh and recirculated chemicals, and temperature monitoring at the heater and process bowl.

The integration of low-k interconnect materials offers new challenges for post-etch residue removal chemistries. Typically, the low-k materials are etched in an RIE plasma using a fluorocarbon gas (e.g., CHF₃) followed by ashing in a reducing plasma (e.g., H₂/N₂, H₂/He). The low-k materials are generally integrated with a top-layer hard mask and a bottom copper barrier layer (e.g., SiO_xC_y or SiC_xN_y). In the work discussed herein, the copper barrier layer is opened to copper prior to the residue removal step. Consequently, the residues that remain consist of an organic-species and copper oxide from backsputtered copper as a result of the barrier layer over-etch. To facilitate good adhesion of the copper seed layer, the residue removal chemistry must remove the copper oxide while not attacking the copper, which could lead to undercutting of the barrier layer and the formation of voids. The via stack that is examined here is comprised (from top to bottom) of 1000 Å SiC_xN_y/4000 Å JSR 5109/1000 Å SiC_x N_y/Cu within a 0.25 µm dual-damascene structure provided by International SEMATECH.

SEMs were used to evaluate residue removal and selectivity to copper. Of the formulations studied, the best SEM results were achieved with Ashland EZ Strip 500, ATMI AP-395, EKC 6910, Shipley SAR FO6G5 and General Chemical GenSolv 670.

Figure 2 shows typical SEM images for pre- and post-process via cross sections. The residue is concentrated on the bottom of the via and successful removal was characterized as no observed residue, no bowing of the via, and no barrier layer attack or undercut. Typical chemical dispense times ranged from 90 seconds to 15 minutes at a temperature of ambient to 60°C. Batch spray processing has been successfully demonstrated in dispensing chemistries as low as 60 seconds.² The chemistries were readily rinsed with DI water and did not require any IPA for rinsing or drying. In general, all chemistries had a pH<7 and had a copper and porous MSQ etch rate of <5 and <1 Å/min, respectively.

2. Successful residue removal, characterized by no visual residue, no barrier layer attack or undercut and no bowing of the via, was achieved with the Ashland EZ Strip 500, ATMI AP-395, EKC 6910, Shipley SAR FO6G5 and General Chemical GenSolv 670.

We selected one of the chemistries for electrical parametric testing at International SEMATECH (0.25 µm geometry). A three-way split was examined according to the following: 1) residue removal chemistry only; 2) residue removal chemistry followed by a 120 sec, 400°C bake in a vacuum chamber with 100 Torr H₂/He; and 3) DI water only as a reference. Figure 3 shows the via contact resistance and 360k via chain resistance for the splits. The data show the resistance along the X axis and the cumulative probability distribution along the Y axis. The contact resistance data clearly show a decrease in the contact resistance for the samples processed with the residue removal chemistry compared with the wafers processed only with DI water. Specifically, for a contact resistance of <0.6 Ω, the yields for the chemistry, chemistry + bake and DI water were 98.9, 93.9 and 77.3%, respectively. The data are consistent with the residue removal chemistry removing copper oxide on the via sidewall and via bottom resulting in a lower contact resistance. The probability distribution curves for the wafers processed with the residue removal chemistry appear to have a slope change at 0.35 Ω.

3. The lowest contact resistance is achieved with the chemistry only, whereas the most favorable via chain resistance values are achieved using the chemistry + bake cycle. These results indicate that the MSQ film does absorb the cleaning chemistry and benefits from the 120 sec, 400°C bake.

A wafer map of the data reveals that die positioned close to the center of the wafer dominate the higher resistance values. This is consistent with etch plasma non-uniformities, which tend to result in an "edge-fast" etch. Optimizing the etch chemistry may improve the distribution of the resistance values.

The 360k via chain resistance parametric test can be separated into "pass/fail," such that pass is <1.0 Ω and all others fail. The yields for the chemistry, chemistry + bake and DI water were 33.4, 68.2 and 77.3%, respectively. The large yield increase between the chemistry and chemistry + bake suggests that the residue removal chemistry absorbs into the low-k material.

Many chemical manufacturers show FTIR data to demonstrate that their formulation does not absorb into the low-k material. However, the absorption tests are typically done on as-deposited low-k films, which are hydrophobic and resistant to the absorption of aqueous solutions. Once the materials have been etched and ashed, the surface undergoes carbon depletion and becomes hydrophilic. These materials readily absorb moisture. As a result, a post-residue removal bake may be necessary to remove moisture prior to copper seed deposition.

Blanket low-k films were used to monitor changes in k value and surface hydrophobicity after etch, ash and clean processes. The low-k films were partially etched and then exposed to the full ash process. Although integrated structures typically incorporate a hard mask, blanket films are useful in monitoring what effect these processes have on the exposed sidewall (via, trench), which will ultimately affect copper seed adhesion and device reliability. In addition, there is active development to either eliminate hard mask layers or incorporate OSG materials as hard masks for spin-on materials to lower k_eff. We used FTIR to monitor low-k moisture absorption and compositional changes.

4. Though a broad peak of 3200-3600 cm-1 is generally considered a signature of moisture abosorption, it is important to note that molecular water should also exhibit a spectral peak at 1650 cm-1. Because this peak was not observed, it is believed that the broad spectral peak is caused by surface hydroxyl groups.

Figure 4 shows an FTIR spectrum of an as-deposited porous MSQ film and processed porous MSQ film. The processed sample is characterized by a broad peak centered at ~3400 cm^-1. This region of the spectrum corresponds to hydroxyl (OH) group stretches. Specifically, the broad peak of 3200-3600 cm^-1 is a result of hydrogen bonding between neighboring hydroxyl groups, and the narrow peak centered at 3750 cm^-1 corresponds to an isolated hydroxyl group.³

The broad hydroxyl peak is generally considered to be a signature of moisture absorption. However, it is important to note that molecular water, which is absorbed on the surface, should also exhibit a spectral peak centered at 1650 cm^-1 due to the scissor mode vibration. This peak was not observed in any of our low-k samples, which led us to believe that the broad spectral peak centered at ~3400 cm^-1 is likely caused by surface hydroxyl groups and not "free" molecular water. The processed sample is also characterized by the breaking of siloxane (Si-O-Si) bonds (1055, 1105 cm^-1) and a depletion of surface carbon (1275 cm^-1) yielding a hydrophilic surface.

5. Silyation is commonly used to promote photoresist adhesion. Above ~170°C, hydroxyl groups desorb from silica in a condensation reaction.

The objective of the low-k restoration process is to remove the surface hydroxyl groups and restore the surface hydrophobicity. Previous studies on silica and OSG materials have shown that HMDS reacts with surface hydroxyl groups (Si-OH) to form a hydrophobic trimethyl-siloxy group (Si-O-Si-(CH₃)₃).^3,4 This silyation reaction is commonly used in semiconductor manufacturing to promote photoresist adhesion. In addition, previous work has also shown that hydroxyl groups begin to desorb from silica via a condensation reaction at temperatures above 170°C.³ Figure 5 shows a schematic of the these two types of reactions.

FTIR spectra were examined to determine the mechanism of the surface modification reaction. The region of interest can be separated into functional group vibrations (2800-4000 cm^-1) and silicon bonding vibrations (700-1400 cm^-1). The main functional group peaks of interest included isolated OH (3750 cm^-1), hydrogen-bonded OH (3200-3600 cm^-1) and CH₃ (2975 cm^-1). The main silicon bonding vibrations included Si-CH₃ (840, 1275 cm^-1), Si-OH (960 cm^-1) and Si-O-Si (network: 1055 cm^-1, cage: 1105 cm^-1)

Figure 6 shows the FTIR difference spectra obtained by taking the spectra of the processed sample and subtracting the spectra of the "etch/ash/DI" control sample. The results indicate that HMDS reacts with the isolated hydroxyl group as evidenced by the decrease in the OH peak at 3750 cm^-1 and the concomitant increase in the CH₃ peak at 2975 cm^-1 and Si-CH₃ peaks at 840 and 1275 cm^-1. It is important to note that a sample processed at 400°C without HMDS treatment does not show this phenomena, and therefore, we can conclude that:

• HMDS reacts with isolated hydroxyl groups.
• The isolated groups are not removed via a condensation reaction.

6. The decrease in the OH peak (3750 cm-1) and concomitant increases in the CH3 peak (2975 cm-1) and Si-CH3 peaks (840 and 1275 cm-1) indicate the HMDS reacts with the isolated hydroxyl group. The decreases in the broad OH peak (3200-3600 cm-1) and Si-OH peak (960 cm-1) occurs via a condensation reaction between neighboring hydroxyl groups at 400°C.

Additionally, the "400°C" sample yields a decrease in the hydrogen-bonded hydroxyl region (3200-3600 cm^-1), a decrease in the Si-OH peak at 960 cm^-1 and an increase in the Si-O-Si peak at 1055 cm^-1 (network). This is consistent with a condensation reaction whereby neighboring hydroxyl groups can react with one another to form a siloxane bond. It is important to note that no significant reduction in the broad hydrogen-bonded OH or Si-OH peaks was observed in the HMDS-only sample. Therefore, it is clear that the HMDS and the 400°C bake yield two distinct phenomena. The trimethyl-siloxy groups were stable to 400°C as evidenced by no significant difference in C-H or Si-CH₃ absorption between the "400°C + HMDS" and "HMDS + 400°C" spectra. The HMDS followed by the 400°C bake resulted in the largest hydroxyl group reduction.

We measured the dielectric constant of the blanket films by depositing aluminum dots (1.5 mm diameter, 1.0 µm thick) on the blanket film and measuring the capacitance vs. voltage across the aluminum dot and the backside of the wafer. Assuming an ideal capacitor, the k value is determined by:

k=Ct/AE₀

where C is the capacitance in accumulation, t is the film thickness, A is the capacitor area and E ₀ is the permittivity constant. The C-V measurements were taken at room temperature using an HP4294A precision impedance analyzer at 1 MHz from 40 to -40 V.

The Table summarizes the contact angle and dielectric constant obtained for the different treatments. The as-deposited film has a contact angle of 95-100° and k=2.23. The "etched/ashed/DI" sample clearly shows that the film has been damaged by the etch/ash process, yielding a hydrophilic surface with the largest dielectric constant (3.14). Exposing the surface to HMDS restores the surface hydrophobicity with significant lowering of the dielectric constant (2.57), indicating that k is strongly influenced by surface composition and the fact that the porous MSQ film is permeable. Heating the surface to 400°C without exposing it to HMDS also reduces the dielectric constant (2.54); however, there is no improvement in the surface hydrophobicity relative to the etched/ashed/DI sample. Finally, HMDS followed by heating to 400°C yielded a hydrophobic surface with the lowest k value of 2.43.

We identified residue removal chemistries from five chemical manufacturers that selectively remove copper oxide and polymer residue on porous MSQ. Moisture absorption in damaged low-k materials has been observed via FTIR and electrical parametric analysis, indicating that a post-process bake may be necessary for optimal integration. We evaluated the effect of etching, ashing and cleaning on the k value of porous MSQ films by exposing blanket films to these processes and then depositing aluminum pads on the surface to facilitate electrical measurement of the dielectric constant. As-deposited films had k values of 2.23, whereas the post-etched/ashed samples had k values of 3.14. In addition, the as-deposited surfaces were very hydrophobic and exhibited high contact angles with water, while the etched and ashed surfaces were very hydrophilic with low contact angles with water.

Using HMDS to replace surface OH groups followed by a thermal treatment to remove hydrogen-bonded OH groups, the k value of the etched/ashed film was decreased to 2.43 and the surface hydrophobicity of the film was restored. The HMDS process described herein is readily integrated into existing batch spray processors and could immediately follow the residue removal step.