Component/System Correlation of Alpha Induced Dynamic RAM Soft Failure Rates

Alpha particle induced upset of data stored in dynamic MDS RAMs and CCDs has been well documented. May and Woods (1) of Intel explained how the soft failure mechanism operates in the storage matrix of packaged devices, and presented convincing empirical evidence to substantiate their theoretical model. The semiconductor industry reacted to this phenomenon in several ways: by imposing alpha emission limits on their package vendors; increasing storage capacitance on vulnerable nodes with circuit design and/or processing innovations; initiating research on die coatings to serve as a barrier to incident alphas; informing customers who buy high density dynamic RAM and CCD storage devices in order to build large storage modules, that error correction schemes are mandatory. The bulk of the material so far published pertains to a phenomenological discussion at the device physics level. Adverse effects on system MDBF are only implied. This paper presents soft failure data obtained using low intensity polonium 210 and uranium 238 alpha sources on 4K and 16K dynamic M4S RAM and 64K CCD devices made by various semiconductor vendors. Experiments indicate that certain static NMOS RAM devices may also suffer alpha induced soft failures, particularly when operated in low supply voltage, battery back up data retention modes. Finally, soft failure rates developed by operating a 4M byte dynamic MOS RAM storage module in a read only mode at minimum specified supply voltages are checked for correlation with device level results.