Electromigration early failure void nucleation and growth phenomena were studied using large-scale, statistical analysis methods. A total of about 470,000 interconnects were tested over a wide current density and temperature range (j = 3.4 to 41.2 mA/μm2, T = 200 to 350°C) to analyze the behavior of the current density exponent n as a function of temperature. The results for the critical V1M1 downstream interface indicate a reduction from n = 1.55±0.10 to n = 1.15±0.15 when lowering the temperature towards 200°C for Cu-based interconnects. This suggests that the electromigration downstream early failure mechanism is shifting from a mix of nucleation-controlled (n = 2) and growth-controlled (n = 1) to a fully growth-controlled mode, assisted by the increased thermal stress at lower temperatures (especially at use conditions). For Cu(Mn)-based interconnects, a drop from n = 2.00±0.07 to n = 1.60±0.17 was found, indicating additional effects of a superimposed incubation time. Implications for extrapolations of accelerated test data to use conditions are discussed. Furthermore, the scaling behavior of the early failure population at the NSD = −3 level (F ∼ 0.1%) was analyzed, spanning 90, 65, 45, 40 and 28 nm technology nodes.Electromigration early failure void nucleation and growth phenomena were studied using large-scale, statistical analysis methods. A total of about 470,000 interconnects were tested over a wide current density and temperature range (j = 3.4 to 41.2 mA/μm2, T = 200 to 350°C) to analyze the behavior of the current density exponent n as a function of temperature. The results for the critical V1M1 downstream interface indicate a reduction from n = 1.55±0.10 to n = 1.15±0.15 when lowering the temperature towards 200°C for Cu-based interconnects. This suggests that the electromigration downstream early failure mechanism is shifting from a mix of nucleation-controlled (n = 2) and growth-controlled (n = 1) to a fully growth-controlled mode, assisted by the increased thermal stress at lower temperatures (especially at use conditions). For Cu(Mn)-based interconnects, a drop from n = 2.00±0.07 to n = 1.60±0.17 was found, indicating additional effects of a superimposed incubation time. Implications for extrapolations ...