Designing failure-resistant hard drives requires the use of complicated simulation models, which need extended runs on fast computers. In the worst cases, these can be prohibitively time-consuming. Now, these models can be solved thousands of times more quickly using a technique developed by Wuzhong Lin and co-workers from the Data Storage Institute of A*STAR, Singapore1.
Computational models aid in the design of robust hard drives by allowing the simulation of a shock response without the need for physical experimentation. However, these models can require the representation of hundreds of different parts with tens of thousands of degrees of freedom, or independent variables; the greater the number of variables, the longer it takes to solve the models.
The technique employed by Lin and co-workers, called model order reduction, reduces the complexity of a model before it is solved. They began by constructing a full model of a hard drive consisting of over 3,800 elements and 4,300 nodes, resulting in over 20,000 degrees of freedom.
Rather than simply feeding this model into a computer, the researchers ‘projected’ it onto a much simpler model with ten or fewer degrees of freedom. They achieved this without restricting the kinds of inputs, or shocks, that can be simulated. By comparing numerical solutions, the researchers showed that the simpler model is accurate to better than one part in ten million, while it is faster to solve by a factor of 5,100. And since the publication of their work, they have successfully reduced much larger models, with over 138,000 degrees of freedom, in the same way.
Understanding hard drive failure is becoming more important in view of the increasing use of hard drives in portable electronics, such as laptops and media players, which are more likely to be exposed to shocks. By helping engineers design failure-resistant hard drives more quickly, this research will shorten the lead time required to bring new hard drive designs to market. In addition, hard drives are soon likely to undergo significant evolution, making rapid design tools even more important.
“The research community is developing the underlying science and technology to achieve a ten terabit per square inch density by the end of 2015,” says Lin. “This means many unconventional designs, structures and systems need to be evaluated and optimized.” Lin also points out that the developed methodology can be tailored to evaluate the shock resistance of devices besides hard drives, such as cell phones.
The A*STAR affiliated authors in this highlight are from the Data Storage Institute