Sub-0.25-micron interconnection scaling: damascene copper versus subtractive aluminum

Historically, the semiconductor industry has made chip speed the focus of its high performance CMOS logic development strategy. For the wires and insulators used in the back-end-of-the-line (BEOL), this has driven the industry to use damascene tungsten chemical-mechanical polish (CMP) local interconnects and vias; SiO₂-based intermetal dielectric CMP planarization; high-aspect ratio aluminum wiring; high density plasma, ozone/TEOS, or advanced spin-on glass SiO₂ intermetal dielectrics; high density plasma reactive ion etching; and excimer-laser DUV lithography. In order to achieve 0.25 μm CMOS performance objectives, the aluminum wire and tungsten via aspect ratios have increased by about a factor of two as compared to 0.50 μm CMOS. This aggressive reverse scaling of BEOL dimensions increases the defect and yield issues associated with the industry standard subtractive-aluminum etch process. We believe that, if subtractive-aluminum wiring is used, the additional scaling required to meet the performance targets of sub-0.25 μm CMOS logic will result in significantly lower yields and increased manufacturing costs. Rather than attempt to drive subtractive-aluminum wiring beyond its reasonable limits, IBM has chosen to employ an additive-copper dual-damascene wiring process for its high performance sub-0.25 μm CMOS logic technologies. In this paper, we discuss defect density, resistance variability, and capacitance variability for 0.25 μm and 0.18 μm CMOS generation subtractive-aluminum and damascene copper wiring