Time-Lapse Measurements of Stress Relaxation in MEMS Sensor Die Bonds

Abstract

Thermal compression gold bumps have been used to attach high-precision MEMS inertial sensors within hermetic ceramic packages. The bonds can be made at relatively low temperatures, are mechanically robust, and outgas at very low rates in vacuum sealed packages. The thermal expansion coefficients of MEMS die and ceramic packages are not perfectly matched and temperature gradients occur when the assembly is cooled after bonding. As a result, there is considerable residual stress in the bonded assembly, which is accommodated to some extent by distortion of the sensor die. Over time, as these stresses relax, the distortion of the die changes, which causes the spacing between elements of the integral MEMS sensor to change as well. Also, in vibrating instruments, this can change the stress state of the resonant element and cause its operating frequency to shift. An important element of sensor-package design is insuring that stress relaxation effects do not cause the instrument to drift beyond its performance specification limits over a typical lifetime of 20 years. For high precision instruments, this type of performance degradation can be greatly reduced by mounting the MEMS sensor on an interposer structure, which isolates it from package displacements. We have used a silicon interposer to closely match the thermal expansion coefficient of a sensor die and to isolate it from the package by compliant beam elements. The sensor die is brazed or gold bump bonded to the interposer, which is attached to a multilayer ceramic package through bump bonded, beam elements. Even though an interposer greatly reduces package induced strains on the sensor, it does not entirely eliminate them. We have used a phase shifting interferometric system with custom fringe analysis software to measure full-field-of-view with high spatial resolution and nanometer accuracy out of plane, the shape of an interposer-package assembly. The assembly was measured both as built and over the course of several years of aging. After six years, residual stress in the braze material of a chip bonded directly to a ceramic package, relaxes to about half of its initial value. To within the precision that we can measure the residual stress, a similar gold bump bonded assembly fully relaxes within two and a half years. An error analysis of our technique leads us to believe that the measurements are accurate to within a nanometer. In the interposer chip assembly, the observed stress induced deformation of the die is considerably reduced. Over time, as the gold bumps shear, the deflection of the interposer compliant beams diminish, accompanied by some flattening of the die attach area. We have used a combination of analytical and finite element calculations to model these observed stress relaxation behaviors and to derive a stress relaxation curve for a pattern of gold bump bonds.