A Statistical Model for Interpreting Hard Disk Drive Stiction Measurements

Abstract

A statistical methodology is presented for predicting drive performance based on fundamental static friction (stiction) measurements. The technique allows the prediction of drive stiction and dynamic friction failures, based on component level spin-stand measurements. We discuss both the fundamental measurement of component stiction and the interpretation of the results as applied to an actual disk drive. The component measurements examine the effects of acceleration, filtering and sampling. It is shown that motor acceleration and the electronic configuration of the test stand affect the stiction measurement, but by proper electronic and mechanical designs this effect can be reduced to an insignificant quantity. The interpretation of component results considers incorporating the effect of multiple heads in a drive, and a statistical model is devised that accounts for both static and dynamic friction variation, along with motor/driver variations. One can predict the probability of a single drive failure or the failure rates of a population of drives, either new or after extensive field exposure. We show that a poorly characterized measurement compared to an arithmetic average of the available motor torque may predict drive failure rates in error by several orders of magnitude.