Coupled self-assembled monolayer for enhancement of Cu diffusion barrier and adhesion properties

Background

Cu interconnects need a barrier that stops Cu from diffusing into Si and SiO2 and that keeps Cu adhered to the dielectric. The International Technology Roadmap for Semiconductors cited by the authors (called the 2012 report in the text and listed as the 2013 edition in the references) requires barriers thinner than 2 nm for 22 nm line widths, whereas conventional Ta/TaN barriers suffer from high defect density and become ineffective below 5 nm. Graphene works as a barrier at a thickness of 0.34 nm but is difficult to grow or transfer onto complex structures and adheres poorly to Cu. APTMS SAMs are a candidate barrier and adhesion promoter, but their amine (NH2) end group adheres to Cu more weakly than a thiol (SH) group.

Design and approach

APTMS is first assembled on SiO2 from a 3 mM toluene solution for 30 minutes, and MPA is then attached for 2 hours by carbodiimide-mediated amidation using N-(3-dimethylaminopropyl)-N'-ethylcarbodiimide (EDC) and N-hydroxysuccinimide (NHS), which links the carboxyl group of MPA to the amine group of APTMS. The resulting bilayer binds SiO2 through the silane head of APTMS and presents MPA thiol groups to Cu, with which they form strong covalent bonds. The whole process is carried out in solution at room temperature.

How it was evaluated

Film morphology and thickness were measured by atomic force microscopy (AFM) and ellipsometry, and bonding was examined by X-ray photoelectron spectroscopy (XPS). Barrier performance was tested by time-dependent dielectric breakdown (TDDB) of metal-oxide-semiconductor (MOS) capacitors with Cu electrodes on SAM-coated n++ Si with 100 nm SiO2 at 225 °C and 2 MV cm⁻¹, with more than 15 APTMS and C-SAM samples analyzed on Weibull plots and compared with 25 and 50 nm as-deposited TaN barriers. Capacitance-voltage (C-V) measurements on capacitors with 10 nm SiO2 were made at room temperature under constant voltage stress at 5 MV cm⁻¹. Adhesion between Cu and SiO2 was measured by four-point bending with at least 3 samples, before and after annealing at 400 °C for 30 min.

Key results

The APTMS layer had a root-mean-square (RMS) roughness of 0.40 nm and a thickness of 1.4 nm; after MPA coupling, the C-SAM had an RMS roughness of 0.44 nm and a thickness of 2.0 nm, the 0.6 nm increase matching the thickness of MPA, with a peak-to-valley height below 2 nm. In XPS, the N 1s spectrum of APTMS showed a primary amine fraction of 80%, and coupling decreased the primary amine peak and produced an amide peak at 400 eV and sulfur peaks, the largest pair at 163.5 and 164.7 eV from unbound thiols.

In TDDB tests, devices without a barrier broke down at lower electric field and were excluded. Representative times-to-failure were about 2000 s for APTMS and about 7700 s for C-SAM. The Weibull characteristic time at which 63.2% of devices had failed (t63.2) increased from 0.707 hours for APTMS to 2.931 hours for C-SAM, a four-fold increase that was comparable to the 50 nm TaN barrier in the same experiment. The authors attribute the gain to the chemical interaction between S and Cu and to the thicker bilayer. In C-V measurements, bare SiO2 showed a negative voltage shift indicating Cu ion diffusion, whereas both APTMS and C-SAM layers hindered Cu penetration into the SiO2, and the authors report negligible change in the flatband voltage and C-V curve during the constant voltage stress measurement.

Before annealing, all samples had low adhesion energies of 2.01 ± 0.33 (bare SiO2), 2.82 ± 0.39 (APTMS) and 2.12 ± 0.11 J m⁻² (C-SAM). After annealing at 400 °C, which forms covalent siloxane bonds to SiO2, the adhesion energy remained low for bare SiO2 (2.07 ± 0.30 J m⁻²) and rose to 7.41 ± 0.25 J m⁻² for APTMS and 9.88 ± 0.34 J m⁻² for C-SAM, a 33% increase over APTMS; the corresponding plateau loads were 4.0, 7.5 and 9.1 N. The authors relate this gain to replacing the weak Cu-amine interface, at which annealed APTMS fractures, with Cu-thiol bonding, citing binding energies of 48 kcal mol⁻¹ for thiols on Cu(111) and about 9.4 kcal mol⁻¹ for amine-Cu.

How it connects to other research

The related graphene barrier study by Hong and colleagues, from the group of this paper's corresponding author Taeyoon Lee and cited in its introduction, uses a grown, atomically thin carbon film whose barrier performance depends on grain boundary density; the C-SAM instead builds the barrier from solution-assembled molecules that bond chemically to SiO2 on one side and to Cu on the other, addressing the adhesion problem the authors raise for graphene.