Influence of the plasma on substrate heating during low-frequency reactive sputtering of AIN

Abstract

Reports in the literature over the last several years have pointed out the advantages of using low-frequency power for reactive sputtering. In the 10–100 kHz range the impedance of the poisoned target surface is low enough to prevent the charge accumulation that leads to arcing, while the problems associated with rf power are largely avoided. We have found, however, that for a variety of materials the substrate heating at these frequencies is somewhat higher than that observed during dc reactive sputtering under similar conditions. In order to quantify this, we have calorimetrically compared the incident substrate energies while reactively sputtering AlN with both dc and 35 kHz power. In both cases, the power was delivered simultaneously to two 200 cm2 targets. In one set of experiments, they were operated at a total power of 500 W, a total pressure of 10 mTorr, and a N2/Ar flow ratio of 0.18. These conditions were on the knee of the characteristic flow hysteresis curve for our system, and target voltage feedback to the reactive gas flow was used to maintain stability. The average deposition rate and substrate incident energy were 0.82 nm/s and 20.0 eV/atom, respectively, for the dc process and 0.70 nm/s and 32.8 eV/atom for the ac process. Langmuir probe measurements show that the electron energies in the ac plasmas are somewhat higher than the electron energies in comparable dc reactive AlN plasmas and the differences between the plasma and floating potentials are comparable. The ion densities in the ac plasmas we measured averaged 6.4×1016/m3, which is several times greater than the ion densities in similar dc plasmas. We attribute the increased energies and densities to target voltage spikes as the plasma reignites on each half-cycle, causing rapid electron acceleration in the presheath region. This leads to significantly more plasma bombardment of the substrate in the ac case than in the dc case, which our estimates show can partially explain the energy flux difference.