Intel Reveals Patterning, Yield Advantages to SLAM Process
Laura Peters, Senior Editor -- Semiconductor International, 10/1/2002
The via-first approach to dual-damascene patterning requires very high etch selectivity between the interlevel dielectric (ILD) and etch stop layer (ESL), excellent substrate reflectivity control and complete removal of post-etch residues.
In the development of Intel 's 130 nm process, engineers identified several process defects, including a shell-like residue in the vias following trench etch, which caused significant yield loss. Neither a dry overetch nor post-trench-etch cleaning sequences were able to remove the shell defects, which were a mixture of etch residue, polymer and ILD. At the same time, the etching steps required extremely high selectivity (~40:1) between the ILD (FSG) and ESL (Si3N4), a target that is impractical in manufacturing. With insufficient selectivity, yield becomes limited by premature breakthrough of the ESL.
These combined issues drove Intel engineers in Hillsboro, Ore., and material supplier Honeywell Electronic Materials (HEM, Sunnyvale, Calif.) to develop a new, siloxane-based material that could be used to improve patterning performance and yield. Requirements included uniform via filling, improved patterning performance (by easing etch selectivity requirements) and prevention of shell defects.
The result, termed a sacrificial light-absorbing material (SLAM), features excellent gap-fill capability, high absorption of light at the exposure wavelength (248 nm), comparable dry etch rate to the ILD, good etch selectivity to photoresist, and compatibility with standard lithographic processes. Intel presented its findings at IITC 2002 (see "A Novel Approach to Dual Damascene Patterning," IITC Proc., p. 18).
 |
| SLAM-assisted dual-damascene patterning in a six-level copper interconnect (above) allows much better profile control, especially at the trench bottom, relative to the same process using an organic BARC (below). (Source: Intel) |
 |
The material and process is patented by Intel, while HEM is marketing a virtually identical material, DUO 248, to device manufacturers worldwide. The SLAM/DUO approach competes with alternative methods that use organic antireflective coating (ARC) materials, but also bilayer photoresist approaches using, for instance, TIS 2000 resist from Arch Chemicals (Norwalk, Conn.). Toshiba proposed a trilayer approach at IITC (see "A Manufacturable Copper/Low-k SiOC/SiCN Process Technology...," IITC Proc., p. 15).
SLAM was synthesized by adding dye to a member of the siloxane-based family of polymers. The material is deposited after via clean, followed by trench patterning. To keep overall dielectric constant (k) to a minimum, Intel eliminated the buried nitride etch stop layer (k~7) between the via and trench. The use of SLAM allowed relaxed selectivity requirement at the via etch (~15:1) and also less constraint on the trench etch process, giving optimized profiles and better trench depth uniformity (Figure ). "We needed an inorganic film to prevent shell defects so that the plasma would etch SLAM at the same rate that it etched the ILD," explained HEM's Joe Kennedy. With FSG as the ILD, a standard O2 ashing process removes the bulk of the resist and consumes organic content in the SLAM material. A wet remover chemistry (buffered oxide etch, dilute HF or some other fluoride-based wet chemistry) clears the remaining oxidized SLAM.
Feasibility with low-k OSG (organosilicate glass) films has been demonstrated using DUO 248, though the stripping and residue removal steps are different. "SiOC films and DUO are both effectively organosiloxane films, but DUO is less dense and its surface wetting properties are different," Kennedy said. HEM has worked extensively with chemical suppliers to develop cleaning chemistries that would remove DUO selective to OSGs.
Kennedy says there is little difference in resist performance with DUO 248, but more noticeable differences occur with DUO 193 and ArF resists. "Perhaps because these resists are still relatively new and under development, there's more variation from one formulation to another," he said.
Side-by-side performance comparison between SLAM and BARC-assisted dual damascene showed that BARC-coated FSG substrates were as effective at substrate reflectivity control as SLAM (exposure-focus curves showed minimal CD change to focus offset at 20-400 mJ exposure dose). However, only the SLAM process gave defect-free structure following post-trench-etch cleaning. The BARC stack showed premature breakthrough of the nitride etch stop layer. The new process, with improved feature profiles and aspect ratios, provided a high-yielding manufacturable solution.
For additional information on yield management, go to www.semiconductor.net/yield.
