Selective MEMS+ model simplifications reduce simulation timeby more than 90% with minimal accuracy trade-off
During a visit to a prospective customer a few months ago, a MEMS design manager told me that her philosophy is that a simulation is not worth doing if it takes more than two hours. I don’t want to focus on whether two minutes, two hours or two days is the right threshold, the point is that all engineers have a time limit on how long they’re willing to wait for simulations to complete. That said, two to four hours sounds about right to me as an upper limit. The question is: what should engineers do when they can’t achieve acceptable accuracy within their self-imposed time limit?
Sometimes the right answer is to buy a faster, bigger computer. Thanks to Moore’s law, computers are continually getting faster and cheaper. It’s amazing how much computing power can be purchased for $5,000 these days. That may be a very smart investment compared to the cost of engineering time, not to mention lost time-to-market opportunity, squandered by using inadequate computers. If only it was this simple. Engineers have this pesky habit of wanting to simulate ever more complex designs with more complex physics, and do it accurately. Thus the expectations for simulation tools continue to outpace the increases in computing power.
Our software uses numerical discretization approaches such as the finite element and boundary elements methods to simulate the behavior of MEMS devices. The essence of these approaches is to divide (or discretize) a complex physical structure into smaller units, or elements. In general, the finer the discretization (the more elements), the more accurate the simulations. This discretization approach is at the heart of most numerical simulation tools. Alas, the more elements, the longer it takes to run a simulation. This implies a trade-off between speed (simulation time) and accuracy. Good engineers understand this trade-off, and spend time optimizing their discretization to achieve an acceptable balance between accuracy and speed. There’s rarely a free lunch when it comes to accuracy.
So, MEMS engineers want to simulate more without compromising accuracy and without spending more time. This engineering fact of life motivates the Coventor development and support teams every day. It motivated us to introduce our revolutionary MEMS+® suite, a different kind of finite element analysis. And it motivates us to continually look for ways to improve our simulation software to achieve more accuracy, speed and capacity. With each succeeding release of our products, we make substantial advances. With MEMS+ 3.0, we introduced new high-order beam, shell and brick elements that provide more accuracy, and we introduced an AutoWizard to automatically connect the elements. With the availability of the new elements and the AutoWizard, we’ve found that users are now creating models that have 10X more degrees of freedom than before, creating a need for more speed and capacity. We will soon announce the 4.0 release of our MEMS+ suite to be followed by the release of the CoventorWare® 2014 suite. Each of these releases will feature significant improvements in speed and capacity which can be traded off for more accuracy or more thorough design explorations and optimizations. We know our customers will appreciate these improvements. But we also know that we can’t stop there. Users of our tools benefit not only from the computing advancements provided by Moore’s law, but also from our continuing advancements in algorithms and methodology.