Rapid and selective post-etch residue removal for Cu and low-k devices
06/01/2005
The adoption of copper and low-k dielectrics has enabled the industry to overcome inherent limitations in aluminum and silicon dioxide materials used in IC interconnects as feature sizes and pitch spacing scale down. However, the use of Cu and new low-k films has also introduced new challenges in cleaning post-etch residues from metal and dielectric layers. This article reviews a series of tests that were conducted to study the inherent cleaning and compatibility characteristics of different chemicals for Cu/low-k interconnects, single-wafer tools, plant safety, and environmental protection.
Integration of Cu wiring and low-permittivity (k) dielectric materials as replacements for aluminum (Al) and silicon dioxide (SiO2) has lowered both the resistivity (R) of the conductor and capacitance (C) of insulator layers in chip interconnects as feature sizes shrink. While Cu and low-k interconnects allow device makers to push performance and meet the goals of Moore’s Law, many additional challenges are being introduced by the changes in materials and processes.
Because Cu will not form volatile metal halides in plasma, it cannot be etched in the conventional manner used with Al technology. As a result, damascene processing has been widely adopted for new interconnect structures, introducing the need for silicon nitride (SiN) or similar barrier layers. This has proven to be problematic for many of the newer low-k materials, which can readily degrade if exposed to excessive plasma processing. Because of this, less aggressive plasmas (e.g., reductive vs. oxidative ash) have been introduced. When coupled with the complexity and diversity of etched materials, this can result in deposition of unwanted post-etch residues (PER). For damascene integration to succeed, a selective method is needed for removal of these residues without corrosion, loss of critical dimension (CD), or increased permittivity.
In an industry increasingly aware of its impact on the environment and the safety of employees, the importance of sustainability in new wafer fab technologies cannot be overlooked. The drive to reduce energy consumption, waste generation, and risks across the semiconductor industry must be reflected in the development of new cleaning compositions. This precludes many organic solvents and any materials that can bio-accumulate or otherwise adversely affect the environment.
Single-wafer revolution
The challenge to clean PER is further complicated by the move toward single-wafer processing for back-end-of-line (BEOL) cleaning. To maintain throughput, the total in-tool dwell time/wafer is often limited to ~120 sec for chemical dispense, rinse, and dry. For current wet cleans with a high solvent content, a rinse of 30-45 sec is required, while more aqueous chemistries are typically ~15-20 sec. Assume an additional 15 sec for drying and the total chemical dispense time is reduced to ~60-90 sec. To increase the speed of cleaning, there generally are two options: Increase process temperature or make the chemical more aggressive. Unfortunately, both of these options can negatively affect the process compatibility with Cu and low-k materials.
One other key question is what are these low-k materials? As yet, no single low-k material has been universally adopted, and many variants from carbon or fluorine-doped SiO(x) (e.g., Black Diamond, Coral, and fluorosilicate glass, or FSG) to organic films (e.g., SiLK and Flare) and “hybrid” materials, like methylsilsesquioxane (MSQ), are available. Porosity is an additional factor that must be considered.
So the challenge is to remove complex PER from a very sensitive substrate using a formulation with a small environmental footprint in approximately a minute; the question is how?
Chemical toolkit
Fortunately, this complex problem can be reduced to a more manageable level by regarding it as a question of selectivity. From a chemical standpoint, the goal is to identify the right functional moieties to target PER while essentially ignoring the rest of the substrate. A series of screening experiments was instigated to study the inherent cleaning and compatibility characteristics of a broad range of different chemicals. Carboxylic, amino, and inorganic acids; amines; and quaternary ammonium salts were tested alongside different classes of organic solvents to provide a benchmark for comparison. From this initial review, the most promising candidate chemicals were advanced to the next stage for further study into the different aspects of activity - cleaning, corrosion, and inhibition. In general, it was evident that cleaning was better at either high (>10) or low (<3) pH while compatibility was typically better at lower pH. The different types of residues targeted are shown in Fig. 1.
 Figure 1. Post-etch residues targeted in tests of removal chemistries. |
To remove the many types of PER that can be present on Cu/low-k devices, a range of different chemical mechanisms needs to be employed, such as halide conversion, chelation, protonation, and dissolution. To maintain a high degree of selectivity, it was also necessary to include an inhibitor that would not retard cleaning. The PER targeted by these various processes is shown in Fig. 2.
 Figure 2. Post-etch residues targeted by different chemical processes. |
The key to this mechanism is to make full use of water’s solvency power. At low pH - in the presence of a suitable halide source - a significant proportion of PER can be removed, as was clear from the use of dilute hydrogen fluoride (dHF) solutions. This simple process alone does not give a broad process window, however, because upstream etch and deposition steps have to be optimized within very narrow tolerances to prevent the formation of harder-to-remove residues. A better solution is to engineer more chemical functionality into the formulation. Based on the initial screening experiments, several interesting components with the desired functionality were selected. Statistical design-of-experiment software was then used to create a second-degree extreme lattice matrix of these components to generate an optimal formulation. This ultimately led to the creation of a new formulation for single-wafer applications, called “S2X3” in the development and testing of its commercialization.
Results and discussion
 |
S2X3 is a highly aqueous formulation designed to selectively remove PER in very short times at low temperatures. A list of basic physical properties of the formulation is given in the table.
Experimental. All etch data was quoted as total loss over a period of 10 min at 25°C and measured on a Nanometrics Nanospec 6100XMP for dielectrics or a Four Dimensions 280SI four-point probe for Cu. Permittivity was measured with a Four Dimensions CVMap 3092A mercury probe system at 25°C over 10 min, followed by a 10 min 200°C bake. SEM inspections were performed using a Hitachi S4500 or S5200.
Cu selectivity. The key to the activity of this chemistry is selectivity - it very rapidly targets and removes residues without damage to either Cu or low-k dielectrics. Obviously, the inhibitor plays a key role here, but in this case the benefit of the novel acid used was twofold because it also increases the removal efficiency of Cu(x)O (Fig. 3). This class of formulation removes Cu(x)O very quickly, on the order of 3-10 sec, yet possesses such a high level of selectivity that total Cu etch over 10 min was constrained to 10-18Å.
 Figure 3. Influence of inhibitor on Cu and Cu(x)O etch. |
For each of these solutions, the ratio and concentration of components - excluding the inhibitor - were adjusted to reduce selectivity, resulting in increased Cu etch and slower Cu(x)O removal times. When the inhibitor was added (in each case at the same concentration present in S2X3), both Cu etch and Cu(x)O removal time were significantly reduced.
Dielectric compatibility. Cu compatibility is only one half of the equation. It is equally important to minimize any adverse impact upon low-k dielectrics. As with Cu, this can be expressed in terms of changes in thickness, but a more useful method is to compare k values pre- and post-clean. As can be seen from Fig. 4, S2X3 exhibits an equally high level of selectivity to several different CVD and spin-on dielectrics as it does toward Cu. A negligible change in permittivity was observed after processing at 25°C for 10 min.
 Figure 4. Change in permittivity post-clean with S2X3 formulation. |
Dielectric compatibility has been confirmed by further tests to quantify any changes in the Fourier Transform infrared spectroscopy (FTIR) spectra of these different materials or changes to thickness and refractive index.
Bathlife. Another critical consideration in developing S2X3 was to ensure that the chemistry was stable. With any halide chemistry, stability is closely tied to pH because the active halide species can change at different pH values. With S2X3 this will not happen. The pH of the formulation is such that any change in composition will not result in a significant change to pH or the active halide species. This has been confirmed in bathlife evaluations for up to 48 hr.
The bathlife has been established under two distinct conditions - open, static at 25°C and 35°C, and closed, dynamic at 30°C - in a single-wafer spin processor with constant recirculation. In both cases, S2X3 retained full functionality up to 48 hr.
Cleaning. The cleaning efficiency of S2X3 has been determined for numerous low-k film substrates, including Coral, Black Diamond, Aurora, FSG, and MSQ. Initially the process window was established in beaker tests; however, the results of testing on an SEZ1300 single-wafer spin processor will be discussed here. The reference SEM images of two substrates are shown in Fig. 5, along with the effects of chemical dispense time vs. temperature and flow vs. chuck spin rate.
 Figure 5. SEM references (top) of process parameters’ impact on cleaning (bottom). |
For the MSQ sample, the heavy PER was readily removed within a 10°C and 90 sec window. Further beaker testing showed no evidence of attack at up to 10 min at 30°C. The importance of flow rate and spin speed was clearly evident when cleaning the FSG wafer. A minimum flow of 1.3L/min-1 and chuck speed of 900rpm was required to insure full cleaning, compared with 1.2L/min-1 and 500rpm for the MSQ wafer. The relationship between flow and chuck speed has not been fully characterized, but it was clear that if either were too low, poor cleaning would result due to lower mass transport efficiency.
Electrical tests. In collaboration with an independent device manufacturer, S2X3 has undergone electrical testing on 200mm FSG and SiOC (k = 2.7) wafers. Four S2X variants were tested at 25°C from 30-300 sec.
In comparison to the process of record (POR) with EKC5930 at 35°C for 10 min, S2X3 gave equivalent yields and via chain resistance at one-tenth the time and fully 10°C lower temperature. At 5 min, all of the S2X variants gave higher yields and comparable or better resistance plots to the POR. S2X3 was able to fully clean these wafers in 30 sec, yet the process time could be extended by a factor of 10 without undue effect, which offers a significant improvement in both throughput and process window.
Conclusion
A new formulation has been developed to address the specific challenges of cleaning PER from Cu/low-k devices on a single-wafer platform at low temperatures without damage to Cu, loss of CD, or increase in permittivity. Extremely high selectivity has been observed for Cu and a range of CVD and silicon-on-glass dielectric materials without compromising cleaning efficiency. Both internal testing on an SEZ1300 and at an independent device manufacturer have shown that S2X3 possesses a very broad process window, allowing for increased throughput while still maintaining high yields. In addition to meeting these critical design criteria, this new aqueous formulation is composed of chemicals that will not bio-accumulate, exhibits a high flash point (>90°C), and is neither toxic nor corrosive.
Acknowledgments
The following are registered trademarks of their respective companies: Black Diamond, Applied Materials Inc.; Coral, Novellus Systems Inc.; Flare, Honeywell International Inc.; Nanospec 61000XMP, Nanometrics Inc.; CVMap 3092A mercury probe, Four Dimensions Inc.; and Aurora, ASM International NV. SiLK is a trademark of Dow Chemical Co. The SEZ1300 spin-processor is a product of SEZ Group.
Chris Reid received his BSc in pure and applied chemistry from the U. of Strathclyde in Glasgow, Scotland. He is an R&D chemist at DuPont EKC Technology Ltd., 19 Law Place, Nerston Industrial Estate, East Kilbride, Glasgow, Scotland, G74 4QL; ph 44/1355-595400, fax 44/1355-595444, e-mail [email protected].
Tomoko Suzuki received his undergraduate degree in zootechnical science and is a research chemist at DuPont EKC Technology KK, Kanagawa, Japan.
Toshitaka Hiraga received his undergraduate degree in control and information systems engineering and is a process engineer at DuPont EKC Technology KK.