Characterization of Electromigration Effects in RDL of Wafer Level Fan-In and Fan-Out Packaging Using a Novel Analysis Approach

Abstract

Electromigration (EM) is an important phenomenon in microelectronics, widely studied and modeled in chip design in the last decades. More recently, the increase of packaging density pushed EM studies outside of the chip. In Wafer Level Fan-In and Wafer Level Fan-Out designs, EM studies have focused on the package solderjoins, due to the poor robustness of solder alloys to EM. Inside the package, the copper traces in the redistribution layer (RDL) have not been considered critical because the minimum cross-sectional area of RDL technology had non-critical current densities. However, the constant demand for packaging miniaturization is requiring even higher RDL densities, with line/space ($\mathrm {L}/\mathrm {S})\lt 5\mu \mathrm {m}$ under development for fan-out and $\mathrm {L}/\mathrm {S}\lt 2 \mu \mathrm {m}$ for fan-in. Due to RDL process limitations, L/S reduction carries a quadratic reduction on the trace's cross-section, which can have a significant impact on EM reliability. Moreover, while IC lines are embedded in a thermally-conductive medium, RDL lines are built on dielectrics with poor heat dissipation and, also critical, fan-out packages have areas/ materials with very different thermal conductivity - silicon (Si), mold compound (MC), which lead to hotter and uneven line temperatures and consequently different EM rates.This paper studies the EM effects in RDL and quantifies its impact on reliability and product life expectancy for Amkor's WLFI/WLFO technologies. The increase of relative resistance, $\Delta \mathrm {R}/\mathrm {R}$, was analyzed on highly stressed RDL Cu traces, built over Si and MC units to test the extreme conditions in fan-out packages. Both continuous and on-off cycled temperature tests were conducted to investigate the thermomechanical stress impact on EM. The results showed very different increase rates of $\Delta \mathrm {R}/\mathrm {R}$ due to the very different thermal dissipation abilities, identifying the need for specific RDL design rules. In the continuous temperature tests, the fairly linear increase of $\Delta \mathrm {R}/\mathrm {R}$ suggested the use of a degradation rate (DR) to characterize the EM effects in a very fast way, instead of determining mean time to failure (MTTF) figures that are time-consuming and depend on arbitrary and heuristic failure criteria (e.g., 20% rise of $\Delta \mathrm {R}/\mathrm {R})$. The linear extrapolation enabled by the DR also allowed the fast build-up of MTTF Weibull plots that would otherwise take several months to complete. A model for the DR, adapted from Black's model, was developed for the statistical estimation of mean DR for a given temperature and current density. In the on-off cycled temperature test, a stepwise behavior of $\Delta \mathrm {R}/\mathrm {R}$ was observed, with general reduction of net MTTF, while DR acceleration was only observed on the MC units, pointing to external thermomechanical-induced effects on the measured $\Delta \mathrm {R}/\mathrm {R}$, which are filtered by the DR analysis.