AFS Multi-Core Parallel for Nanometer Analog/RF Block Characterization

The Problem
As analog and RF circuit designers move to lower process geometries, rigorous block characterization in the context of realistic physical effects becomes increasingly important. At 90nm and below, circuit complexity, circuit nonlinearity, and physical effects—including process variation, device noise, device mismatch, and detailed parasitics—make such characterization a growing computational bottleneck.

Characterization Requirements Increase with Decreasing Process Nodes
    
Traditional Block Characterization
Designers have traditionally run the desired corners, sweeps, and Monte Carlo iterations sequentially on a single machine with a single simulator. Given a practical total runtime limitation of at most a few days, this approach is increasingly inadequate. A distributed approach using scripts, an internal tool, or a third-party tool to distribute the iterations over a number of computers can reduce turnaround time or increase iterations in a given time. However, distributed characterization generally increase use model complexity, introduces turnaround time unpredictability, is subject to network issues, and requires a large number of simulator licenses.

AFS Platform for Traditional Block Characterization
The Analog FastSPICE unified verification platform (AFS) provides nanometer SPICE accuracy 5x-10x faster on a single core than traditional SPICE simulators. AFS also provides the industry’s only silicon-accurate device noise analysis for every type of circuit. With batch mode and seamless integration into the industry’s leading analog design environment, the AFS Platform plugs into current single-simulator or distributed characterization approaches without any changes, providing 5x-10x faster turnaround, 5x-10x more iterations, or any combination of the two.
AFS Multi-Core Parallel (MCP) Block Characterization
The AFS Platform provides a multi-core parallel operating mode that dramatically speeds analog/RF block characterization on shared-memory multi-core systems. AFS MCP automatically runs corner, sweep, or Monte Carlo iterations in parallel on separate cores—each of which runs 5x-10x faster than traditional SPICE. With limited overhead the net speedup versus traditional SPICE on a single-core machine is ~15x-30x on 4 cores and ~30x-60x on 8 cores. This speedup is reduced in cases with significant runtime variance per iteration (e.g., with some RF analysis sweeps). MCP supports all AFS Platform analyses. Running it requires simply specifying the desired number cores—the rest of the inputs and all of the outputs are the same as single-core operation.

AFS MCP Characterization Speedup Examples for Analog and RF Circuits
 
 
AFS MCP Cost and Licensing Advantages
Analog FastSPICE multi-core parallel licensing is very compelling. AFS MCP requires only one additional BDA Token to run on up to 4 cores and only two additional BDA Tokens to run up to 8 cores. Thus, 3 BDA Tokens can perform the same characterization run on a single machine as would require literally 24 to 50 traditional SPICE or “parallel” SPICE licenses on 6 to 50 machines. This advantage applies to any type of iterative run with any type of analysis (e.g., transient Monte

For more information, contact your BDA application engineer or click here for a web request.