1. A robust mesher that intelligently pre-processes layouts imported from other systems to guarantee zero mesh-related simulation inaccuracies.
Layout mesh processing removes artifacts from manufacturing mask generation process such as overlaps, thin polygons, perimeter glitches (see picture above) which can result in inaccuracies, unnecessary memory consumption or non-convergence.
2. A fast and memory-efficient solver based on NlogN matrix compression techniques that tackles problems up to 6x larger than previously possible and solving them 5x faster.
This graph shows the memory consumption vs. the problem size of the new and older Momentum solver. The new solver consumes memory at the rate of O(NlogN) where N is problem size. This is significantly more efficient than the previous O(N2) solver. With 64 GB of available RAM, we can now solve problems 6.6x larger than before.
Not only is the memory consumption decreased, the simulation time is also decreased by up to 10x with the NlogN solver as shown in this graph.
3. Multi-core computer utilization through parallelizing the solve process across all available PC processor cores
In addition to the fast NlogN solver, we can achieve more speedup through solving the problem by breaking it into separate threads for the separate processor cores as shown above. For example, we can routinely get up to 3.5x speedup with a quad-core processor using this multi-threading technique.
4. Distributed simulation over networked computer clusters to perform multi-point EM simulation in parallel with Momentum Turbo.
We can push parallelism beyond multi-core simulation by utilizing multiple computers on the network to perform multi-frequency point simulation in parallel. ADS2008 Update 1 offers economical 8-pack bundle Momentum Turbo licenses specifically for this purpose. This allows idle computers on the network to be utilized for speeding up the EM solution process. In general we have observed a 6x speedup with a 8-node networked computer cluster. The picture below shows a typical setup for distributed computing for high-speed Momentum simulation.
top of page
printer-friendly version
email this page
