PeakView™ electromagnetic solver combines mathematical accuracy, minimal simulation time and enhanced computational capacity in modeling IC devices and interconnect. Our state-of-the-art simulation engine typically runs 10x faster than traditional structure simulators maintaining excellent numeric precision. It is designed to efficiently analyze the complex structures found in modern semiconductor technology.
PeakView™ engine is meticulous in meshing asymmetric and non-rectilinear geometry. Smart meshing strategies are used to optimize computation time and preserve high accuracy taking into account process and design requirements. For example, tall side-walls of advanced process nodes are analyzed with precise sidewall meshing. Skin effects at high frequencies are accounted for with special meshing algorithms that consider current density. Thick metal layers, e.g. AP, are uniquely treated for high quality EM results.
PeakView™’s Method-of-Moments-based field solver is tested for numerical accuracy using rigorous meshing densities at frequencies above millimeter-wavelengths (beyond 80 GHz), and as low as in the MHz range. Our numerical methods are accurate to 4 decimal places and we have been able to achieve 80dB SNR in isolation studies. Physical effects associated with dielectric layers, Epi and SOI substrates are correctly modeled in PeakView™. It allows for easy handling of process corners and temperature coefficients. Our EM solver accuracy has been consistently demonstrated in tape outs, test chips and silicon correlation studies.
PeakView™ is a very fast, multi-threaded EM solver that employs multi-core processors and is able to run simultaneous simulations on different cells. The increase in speed with the number of cores is almost linear with its use of the multi-threading option.
Major Enhancements_4.2.0 and onwards (Speed and Memory)
- Charge-free Speedup
By mathematically reducing the system matrix size for charge panels that are known to have zero charges, we can achieve major speedup. This applies to internal current sheets such as when multi-sheet current with > 2 sheets. This also applies to HFD-L applications where only inductances are extracted to complement existing RC extraction. This is especially important for 5G and other mmwave applications where volume currents simulations are important.
- QuickEM speedup
PeakView introduces a new computational algorithm that can reduce the condition number of problems. This allows for a 50% reduction of system matrix storage and solving time without loss of accuracy. This release ensures that all observed cases achieve the same results as without QuickEM. Now, all major simulation types are with QuickEM turned on by default (Previously only onChipNormal and onchipQuick are with QuickEM turned on).
Following two tables show the improvements on simulation speed and memory with two test cases:
- Hybrid GPU/CPU Solver
To cut the solving time, PeakView has introduced a hybrid GPU/CPU solver which utilizes the advantage of GPU cards and greatly improves the simulation performance. PeakView supports NVIDIA GPU cards with CUDA compute capability greater than 3.0.
With GTX 1080 (version 6.1), an astounding 5x speedup has been observed with enabled quickEM option. Also, please note that GPU versions smaller than 6.x do not support double-precision (DP) arithmetic and hence, cannot be used without quickEM enabled. PeakView supports multiple GPUs, and in theory, a linear increase in speedup is expected with the number of GPUs used. For instance, by using two GPUs similar to GTX 1080, up to 10x speedup is expected.
In order to maximize utilization of computational resources, PeakView™ not only uses multi-core processing capability on standalone servers, but is able to run design jobs on computing farms consisting of multi-core machines as well. In addition to making use of platform load sharing capacity (LSF), our Hierarchical EM (HEM) solution significantly reduces memory requirement and dramatically shortens the CPU time without compromising accuracy.