Resist Removal Walks a Tightrope
Between limiting damage to low-k materials and silicon removal at the gate, while definitely clearing away all photoresist and its residues, resist removal processes -- wet and dry -- continue to strive to maintain the right balance.
Ruth DeJule, Contributing Editor -- Semiconductor International, 8/1/2008
At the 32 nm technology node and beyond, the most critical front-end-of-line (FEOL) cleans are during ultrashallow junction (USJ) formation, with an International Technology Roadmap for Semiconductors' (ITRS) target of 0.3 Å/clean of silicon loss. Meeting this challenge is most difficult for resist removal following high-dose implants. For stripping in dual damascene interconnect formation using ultralow-k dielectrics, the hurdle comes when k values dip below 2.5, and commonly used solvents or dry strip processes are unable to simultaneously meet cleanliness and k value requirements. Therefore, in both FEOL and back-end-of-line (BEOL) strip/clean processes, the industry is looking at the issues from all angles — from all wet (Fig. 1) to all dry to combining the strengths of each technology.
| 1. New cleaner formulations are under evaluation on single-wafer tools in applications laboratories. (Source: Mallinckrodt Baker) |
Hybrid approach
During USJ source/drain (S/D) formation, doping levels of implanted arsenic (As) and phosphorus (P) species often exceed 1 × 1015 ion/cm2 (Fig. 2), consequently forming a very tough, carbonized, cross-linked polymer crust on the surface of the resist. Because this carbonized layer typically cannot be removed by standard wet chemistries alone, such as sulfuric acid peroxide mixture (SPM) and SC1, plasma is generally used to break through the layer prior to a wet clean (Fig. 3).
| 2. Implant conditions dictate resist removal methods, as indicated by shaded regions typically requiring traditional all-wet resist stripping capabilities (1), some ash-free all-wet stripping (2) and advanced ash-free all-wet resist stripping capabilities (3). (Source: FSI International) |
According to Ivan Berry, director of technology at Axcelis Technologies (Beverly, Mass.), a little known and perhaps more important attribute of plasma is the capability to generate the right chemical bonding at the surface of the junction and, thereby, control the dopant profile, something wet clean approaches cannot do. He deems that "the best results are achieved when you let plasma do what it does best and wet clean do what it does best." Others agree.
| 3. When using a ‘properly’ chosen dry strip process, post-dry strip residues (left) are completely cleaned by a dilute SC1 clean (right). (Source: Mattson Technology) |
For example, FSI International (Chaska, Minn.) extends its wet stripping process to all three layers — "dopant-containing," damaged crust and underlying photoresist — using standard SPM chemistry, an optimized mixing ratio of sulfuric acid to hydrogen peroxide, propriety dispensing technology and a patent-pending method for boosting temperature and reactivity. To maximize the benefits of the peroxide and prevent premature breakdown, room temperature H2O2 is mixed with pre-heated H2SO4 near the point of dispensing onto the wafer. In addition, a catalyst is separately injected into the chamber, which increases the wafer temperature and enhances SPM reactivity.
This process is currently in DRAM manufacturing, but is also amenable to the greater demands of logic, where 90% of implanted photoresist layers can be stripped with this all-wet process; the other 10% requires plasma. "Wet stripping time for medium-level implants [2.5 × 1014 ions/cm2, 40 keV As] is reduced by 90%, from 50 to 5.5 min, in a batch spray system, and previously unstrippable implant levels can now be removed without ashing," said Jeff Butterbaugh, chief technologist at FSI International. However, for beamline-dose levels exceeding 5 × 1015 ions/cm2, plasma is used.
BEOL materials
The introduction of new materials adds another element to the mix. For porous low-k interconnect structures with dielectric constants of ~3, successful integration was achieved by improving plasma ashing and etch chemistries, and making changes to the processes themselves. For example, using a dual damascene partial trench-first patterning scheme, a metal hard mask (MHM) was used to pattern the trench and a resist mask for patterning the via. Because of the MHM, ashing could be performed before the dielectric was actually etched.
Dry strip approaches, however, have not been effective on the more advanced structures containing materials with 25–30% porosity and k values of 2.5–2.3 because of the excessive damage they cause, said Guy Vereecke, team leader of chemical cleans at IMEC (Leuven, Belgium). Therefore, three years ago, work began on wet strip alternatives to remove photoresist on chemical vapor deposition (CVD)-grown silicon oxy carbide (SiOC) and MHM films patterned with 193 nm deep ultraviolet (DUV) photoresist. The scientists studied both wet strip using organic solvents and water.
"The real issue in resist removal is selectivity to the low-k material," Vereecke said. With 193 nm resist, the backbone of the polymer is made out of carbon-carbon bonds — simple bonds that are very tough to break. The goal is to identify the new bonds created after etching the MHM or porous low-k and break them.
The difficulty is illustrated in Figure 4, where a 70-nm-wide patterned titanium nitride (TiN) MHM line on top of a BDII (k=2.5) film is shown following TiN patterning. There was little change in the volume of photoresist on top of the TiN, though solvent dissolution tests were performed under stringent conditions, indicating most of the photoresist was cross-linked (crusted). Nevertheless, there are indications of progress. The team successfully removed photoresist from a metal mask by a wet process in an organic solvent and in the presence of porous low-k materials.
| 4. The challenges of BEOL wet resist strips after MHM patterning are illustrated, as little decrease in the height of the resist on TiN lines is indicated and nearly all the crust remains after a harsh solvent treatment, lending the call for pre-treatments and selective chemistries. (Source: IMEC) |
FEOL materials
New FEOL materials face their own resist removal issues. Hafnium-based high-k gate dielectric films, such as hafnium silicon oxynitride (HfSiON) and hafnium silicon oxide (HfSiO), undergo significant nitrogen loss when exposed to strong oxidizing strip and clean chemistries. Similarly, marked material loss has been observed in some metal gates following O2-based plasma or oxidizing wet cleans, prompting both wet and dry sectors to develop non-oxidizing strip chemistries.
One option is forming gas, a combination of H2 and N2 or H2 with an inert gas, currently being tested at Mattson Technology Inc. (Fremont, Calif.). For some applications, researchers there have looked beyond the standard-forming gas combination of 4% H2 and 96% N2 to formulations containing 10% or more H2 in nitrogen or argon. One study using standard-forming gas chemistries with exposed HfSiON dielectrics resulted in an increase in the density of nitrogen of several percent in the film — unacceptable for production. "We had to go to a new strip chemistry and found that the amount of dielectric constant change in the high-k was minimal," said Stephen Savas, Mattson Fellow and chief surface clean technologist. The absence of oxygen, however, slows the rate of stripping and lengthens process time from ~1 to >2 min.
Minimizing substrate loss
At the 22 nm node, S/D extensions will be shallower than 10 nm, thus severely limiting the tolerance for substrate and dopant loss. In principle, said Rita Vos, senior scientist, FEOL cleaning at IMEC, SPM should be able to remove the photoresist because rather low implant energies and medium-dose levels of ~1 × 1015 at/cm2 are used for extension implantations. The problem lies with the new materials used for 32 nm and beyond. If high-k and metal gate materials are introduced in the gate stack, they will etch very rapidly in the SPM solution. TiN has been shown to etch at >19 nm/min at 90°C, as do others such as AlO or LaO high-k capping layers used in combination with HfO or HfSiO. For these applications, SPM must be replaced, as is the case for high-mobility substrates like germanium, silicon germanium (SiGe) and advanced III-V materials. For instance, in the case of germanium, oxidizing aqueous chemistries can no longer be used because the formed GeO2 is water-soluble. The motivation is clear: Replacements for highly oxidizing aqueous solutions, such as SPM, and for fluorine-based dry plasma that are known to cause unacceptable levels of substrate loss are critical for advanced processes.
On the dry side, advanced ashing based on a non-fluorine-based plasma is under investigation at IMEC in a joint effort with a major dry strip plasma company. And the preliminary results are good: In this all-dry process, the crust and resist were removed with little substrate and dopant loss. The data indicates that the non-fluorine-based plasma can limit loss while maintaining similar electrical performance, when compared with the conditions in which no strip was performed.
Minimal silicon loss was no surprise to Savas. In an internal study comparing 10 cycles of plasma strip/SC1 cleans to 10 cycles of an aggressive standard SPM/SC1, such as those used in memory fabs, there was more than double the silicon loss with the wet chemistries.
The changing face of plasma ashing is caused, in part, by a trend to reduce the plasma component to where it is behaving more like a gas-phase chemical reaction, Berry pointed out. Plasma is traditionally considered an aggressive process, and concerns of plasma-induced damage of the substrate are not uncommon. Plasma penetrates deeper into the substrate than wet chemistries. However, "By tuning the plasma potential, which generates electric fields that enhances diffusion, we can confine the plasma's interaction to the near surface of the silicon," Berry said. The result is a gentler ashing process where so many of the electrons have been taken away that it resembles a hot reactive gas more than a true plasma.
From a wet perspective, wet processing is easier because it provides better selectivity and can be tuned more easily without damaging the substrate. However, the carbonized crust formed by high-dose implants >1 × 1015 at/cm2 is difficult to 'dissolve' in any wet chemistry without oxidizing it. Therefore, all-wet non-aqueous solutions are also being explored because of sensitive high-mobility substrates like germanium, but with physical enhancements to break up the graphite-like surface, Vos noted.
These enhancements include the application of megasonics during the clean or an aerosol spray clean, which bombards the resist in a mist of very small solvent droplets (Fig. 5). Thus far, very encouraging results have been obtained with a cryogenic aerosol clean, which bombarded the implanted resist with CO2 snow particles. When they hit the surface of the wafer, enough damage is created so that the solution can reach beneath to the resist while removing the crust.
| 5. SEM micrograph and high-resolution profilometry (insert) of post-implant resist after high-velocity solid CO2 aerosol treatment shows pronounced curling in the delaminated resist residue layer, indicating the presence of tensile stress. (Source: IMEC) |
Other options
Batch processing has a clear economic advantage, but in resist removal, the benefit is often offset by high defect levels and process control issues. Therefore, particularly for logic applications, single-wafer wet appears universally preferred. Accompanying this is the need to speed the stripping process. One approach is with dry pre-treatments, a short step performed before the wet strip to increase the removal rate. The goal is to perform a wet strip on a single-wafer tool in 1 min.
Plasma and cryogenic aerosol bombardment pre-treatments are also being investigated at IMEC for BEOL resist strip. Both offer advantages and disadvantages. While plasma can be very efficient as a pre-treatment, there are compatibility questions when used on porous low-k stacks. In contrast, no low-k damage is expected with cryogenics, but the benefits need to be investigated.
Green solutions
Environment, safety and health (ESH) have always been a top priority in the semiconductor industry, where cleaning solvents and acids are intrinsic to the fabrication process. The huge number of different chemistries used in the fab are tracked so that chemistries like N-methyl-2-pyrrolidone (NMP), a commonly used wet strip solvent, are flagged and, in this case, on the verge of reclassification to a reproduction toxic band, prompting many IC manufacturers to look for alternatives.
Another popular strip chemical is unique in that a particular blend behaves like a corrosive solvent. One of the environmental headaches is whether to treat it as a solvent or an acid. Certain blends create safety concerns because they can thermally react with other chemicals. And, if it is not properly disposed of and instead is wiped and thrown away, it will smolder and become self-combustible. These and other examples are creating a need for green chemistries and processes.
IMEC's resist removal program targets green solutions, which are based on organics that have a high flash point, so that they are safe. Selected chemistries are classified as neither toxic nor carcinogenic or harmful to the environment. The program includes the development of chemistries in water. Because polymers are insoluble in water, the molecules must be broken down to smaller pieces, making their synthesis more difficult and perhaps a lengthy process. There are many drawbacks to water, while organic solvents are clearly more efficient and intrinsically more compatible with materials like low-k or germanium.
Equipment manufacturers, although traditionally focused on equipment performance, are taking a closer look at environmental issues surrounding their equipment. For plasma, Mattson has been looking at the potential impact of plasma. Using a back-of-the-envelope calculation, dry strip containing oxygen uses a very modest amount of O2 to strip a wafer, Savas said. A forming gas or hydrogen process uses roughly the same amount — ~5 g of O2 produces 0.5 g or less of CO2 release/wafer. He estimates that 30,000 wafers dry stripped in a one month is equivalent to one car driving 1000 miles.
Ironically, alternative green technologies may produce friendlier byproducts, yet prove ineffective at performing the task. For example, ozone, considered as a replacement for sulfuric acid, has the same limitation in removing the high-dose crust. Similarly, super-critical CO2 not only failed to remove the crust, but also posed a considerable safety hazard because of the high-pressure vessel needed to contain the gas.
In addition to an environmental burden, the typical post-high-dose S/D implant ash clean process sequence consumes a large amount of energy, ultimately increasing cost-of-ownership (CoO), said Joel Barnett, senior member of technical staff at Sematech (Austin, Texas) and chair of the ITRS Front-End Processes Surface Preparation sub-Technical Working Group. Eliminating the ash and reducing the temperature of the wet chemical strips, or using just an ash and eliminating the wet chemical clean, could potentially reduce the energy footprint of the strip and simultaneously reduce the number of process steps, substantially cutting processing time.
"The device driver for photoresist removal in the surface preparation section of the ITRS is USJ implants, but we realize that the Roadmap also needs to incorporate the task of removing photoresist after high-dose S/D implants," Barnett said. Perhaps it is also a good time to investigate process-specific energy targets for ESH, as this will give both dry and wet equipment manufacturers reasonable process and ESH targets they can strive to achieve.
