May 26 2010
Berkeley Design Automation, Inc., provider of the Analog FastSPICE™ unified circuit verification platform (AFS Platform), today announced the AFS Multi-Core Parallel operating mode (AFS MCP).
AFS MCP automatically runs corners, sweeps, and Monte Carlo iterations in parallel on separate cores on up to 8-core shared-memory systems. The result is 25x to more than 50x higher performance than traditional SPICE on single-core systems with identical accuracy and no use model changes.
Analog and RF circuits in nanometer CMOS processes require rigorous characterization with realistic physical effects such as device mismatch, device noise, and detailed parasitics. With literally hundreds of iterations to run on complex blocks such as PLLs, ADCs, DACs, transmit and receive chains, and high-speed I/Os, design teams can no longer run sufficient characterization using traditional SPICE tools on single-core machines. Distributed approaches increase complexity, require the availability of tens of machines and simulation licenses, compete with other jobs for priority, and are subject to network issues. Single-core AFS Platform speeds up such jobs 5x to 10x, but even that is not enough for this computation-intensive application.
AFS Platform multi-core parallel delivers massive nanometer SPICE accurate characterization throughput on a single machine. It literally replaces a server farm and tens of simulation licenses with a standalone multi-core machine and one to two additional BDA Token licenses. With no change in use model, designers can use AFS MCP in batch or interactive simulation environments. It produces identical results as AFS single-core operation with an additional 3x faster runtime on 4 cores and 6x faster runtime on 8 cores. Available immediately, AFS MCP supports all AFS Platform analyses including DC convergence, transient, transient noise, periodic steady state, periodic noise, and even mixed-signal simulation. It is ideal for circuits sensitive to device noise, device mismatch, and parasitic effects.
"Analog/RF block characterization in nanometer process technologies is a design limiter demanding increased accuracy, performance, and capacity from circuit simulators," said Dr. Xingang Wang, Senior Manager, Skyworks Solutions, Inc. "AFS MCP delivers nanometer SPICE accuracy and allows us to characterize our analog/RF blocks up to 50x faster than single-core traditional SPICE simulators."
"Distributed characterization with traditional circuit simulators is a headache- it increases use model complexity and cost by requiring a large number of simulator licenses," said Mike Kappes, President of IQ-Analog. "The additional performance and licensing model in AFS MCP reduces our cost per run by up to 50% while enabling us to reduce our tapeout risk by running additional characterization samples on our analog/RF IP."
The Analog FastSPICE Platform is the industry’s only unified circuit verification platform for analog, mixed-signal, and RF design. It always delivers nanometer SPICE accurate results, while providing 5x-20x higher performance than traditional SPICE, >10 million-element capacity, and the industry’s only comprehensive device noise analysis. The AFS Platform is a single executable that uses advanced algorithms and numerical analysis to rapidly solve the full-circuit matrix and original device equations without any shortcuts. It includes licenses for AFS Nano SPICE simulation, AFS circuit simulation, AFS Co-Simulation, AFS Transient Noise Analysis, and AFS RF Analysis.
"Competitive analog, RF, and mixed-signal circuits require massive characterization of nanometer physical effects on circuit operation," said Ravi Subramanian, President and CEO of Berkeley Design Automation. “With the introduction of AFS MCP, we are proud to help analog/RF design teams further reduce the risk of non-functional silicon by making it practical to run extensive characterization of their analog/RF blocks with nanometer SPICE accuracy.”
Source: http://www.berkeley-da.com/