Copper Interconnects: From Aluminum Replacement to Mainstream

In advanced logic chips, interconnect technology is crucial for performance. Transistors handle signal switching, while interconnects transmit power and data. Before copper became mainstream, aluminum was used. Copper replaced it due to lower resistivity (1.68 μΩ·cm vs 2.67 μΩ·cm) and superior electromigration resistance. The Damascene process combined with CMP solved copper etching challenges, enabling its use in sub-90 nm nodes.

Scaling Bottlenecks of Copper Interconnects

As line widths shrink below 10 nm, copper interconnects face rising resistance and reliability issues:

1. Grain boundary scattering: Smaller grains increase electron scattering, raising resistance.
2. Diffusion and corrosion: Copper diffusion requires barrier layers; corrosion requires caps, increasing resistance.
3. Size-dependent resistivity: Nanoscale lines can have resistivity tens of times higher than bulk copper.

Solutions include:

• Grain optimization: Nanosecond laser annealing promotes grain growth, reducing boundary scattering.
• Barrier optimization: Cobalt (Co) liner and cap improve fill ratio and reliability, applied in TSMC and Intel processes.

Next-Generation Interconnect Materials

Copper has intrinsic limits; innovation is key:

1. Pure metals:
2. • Cobalt (Co): Strong EM resistance, weak size effect, superior under 10 nm.
   • Ruthenium (Ru): Low resistivity at nanoscale, high melting point, high reliability, but costly.
   • Molybdenum (Mo): Low resistivity, no barrier layer needed, compatible with ALD, but oxidation needs control.
3. Metallic compounds: NiAl, CuAl, RuAl show low resistivity and high diffusion stability under <8 nm, enabling liner-free interconnects.
4. Topological semimetals & 2D materials:
5. • Topological semimetals (MoP) maintain high mobility even at nanoscale.
   • Graphene + amorphous BN (a-BN) offers efficient 2D interconnects; mass production and process integration remain challenges.

AI-Enabled Material Design

• Accelerated screening: Machine learning simulates experiments, reducing cost.
• Multi-objective optimization: AI predicts conductivity, EM resistance, etc.
• Challenges: Data scarcity, model black-box issues; mitigated via transfer learning, data augmentation.

Industrial Challenges

• New material commercialization must address process compatibility, cost, and reliability.
• Damascene copper process compatibility is crucial.
• Collaboration between material innovation and process optimization will reshape semiconductor manufacturing.