Last updated 03/05/08 by Daan Hein Alsem (dhalsem@lbl.gov)


FIGURE 1: Scanning electron micrographs of the polysilicon MEMS fatigue life characterization resonator (a) Triangular proof mass (film thickness 2 um) with interdigited comb drive electrostatical actuator and capacitive displacement sensor comb; (b) notched cantilever beam connecting the resonator mass and anchor; (c) device in motion in-situ in an SEM, note the blurring edges on the side and the comb fingers because of the high operating frequency (~40kHz).

Phosphorous-doped (n+-type) polysilicon fatigue resonator devices, electrostatically actuated by an interdigited comb-drive (Figure 1) at ~40 kHz with a load ratio (ratio of minimum to maximum load of R = -1) in ambient air, humid air (>95 %RH) and vacuum (P~2.0 x 10-7 mbar), have been used to measure the micro-scale fatigue behavior of notched cantilever beams (Figure 1B). Figure 2 shows our fatigue data acquired from devices from two different fabrication runs at the PolyMUMPs foundry and the SUMMiT V process. First of all, it shows the applied stress vs. number of cycles to failure (S/N) curve for data from a recent PolyMUMPs fabrication run (MUMPs 50) acquired in ambient air. These are compared with previously obtained ambient air data from the MUMPs run 18 (fitted line). It is apparent that the new results in ambient air closely follow those obtained on the earlier fabrication run. These data prove the repeatability of the fatigue behavior between runs. Furthermore, tests at high relative humidity are performed and these data points are found to lie below the S/N curve for ambient air. Figure 2 most importantly shows that, none of the samples tested in vacuo failed, even after 109-1010 cycles at high stresses (all vacuum data points are plotted at the point where testing was stopped – these points do not indicate fatigue failures, but rather that the device had not failed by this point). Additionally, Figure 2 shows similar behavior for devices from the SUMMiT V process. All these results are consistent with the earlier proposed “reaction-layer” fatigue mechanism, whereby high stresses induce a thickening of the post-release amorphous surface SiO2 oxide layer at stress concentrations such as notches and where stress/moisture-assisted cracking of this oxide layer results in stable crack growth and where, provided the oxide layer is thick enough, the critical crack size for the entire structure is reached inside this layer causing catastrophic failure of the structure.

FIGURE 2: Polysilicon S/N curve.

High-voltage transmission electron microscopy (TEM) images reveal a thickening of the oxide layer at the point of highest stress up to 100 nm for PolyMUMPs specimens fatigued in ambient air. However, this local thickening at the notch root was not observed for fractured (non-fatigued) specimens in ambient air, or for specimens cycled in high vacuum. For the latter specimens, clearly the absence of moisture and oxygen in vacuo prevents subcritical crack growth in the post-release oxide layer. Energy filtered TEM shows that in the SUMMiT V specimens a similar trend can be observed with respect to oxide layer thickness. However, in this case the oxide layer is significantly thinner: ~3.5 nm away from stress concentrations and ~15nm at the notch after fatigue failure. Both type of devices show up to an approximate 4-fold thickening after fatigue, but no thickening after vacuum fatigue attempts.

Both the acquired fatigue data in air, high relative humidity and high vacuum, as well as the TEM micrographs of the oxide layers after failure, provide evidence that “reaction-layer” fatigue is the governing mechanism for fatigue failure in micron-scale polycrystalline silicon thin films.