High performance semiconductor device simulation on shared memory parallel computers

Abstract

The work described in this paper is the evidence that, thanks to parallelization, time can be less of a bottleneck to numerical device simulation. With viable execution times, there is less need for a trade-off between larger structures and finer details to achieve an acceptable accuracy in a reasonable amount. of time. The article shows that Parallel ATLAS runs approximately 3 times faster on a four-processor machine while simulating a wide variety of devices.