Accelerating Accuracy: PrimeSim SPICE Powered by GPU

As advanced nodes and heterogeneous integration drive greater design complexity, circuit simulation teams are under pressure to deliver accurate results faster, often with tighter schedules and limited compute headroom. Meeting these demands requires a SPICE solution that scales efficiently, integrates cleanly into existing flows, and helps engineers converge on optimized designs with confidence. Synopsys PrimeSim addresses this challenge by combining SPICE-accurate analysis with high performance and robust capacity, enabling faster turnaround without compromising fidelity.

Let's consider the following customer case study...

We identified an opportunity to broaden a marquee customer’s use of Synopsys PrimeSim by adding PrimeSim SPICE for circuit analysis and optimization. To highlight the value in a familiar environment, we proposed running one of the customer’s existing test cases using comparable settings and compute hardware similar to those already in place.

As with many simulation-flow migration, the customer naturally wanted clarification on setup effort, migration risk, and how a new simulator would behave in practice. Understandably there was an initial hesitation about introducing a tool that had not previously been part of the existing flow. We welcomed those questions and worked collaboratively to ensure the approach fit within the current workflow and would meet expectations.

Our objective was straightforward: demonstrate that migration and adoption could be completed quickly and with minimal effort, while also delivering faster performance and improved results engineers need to move forward with confidence.

The customer’s design was a post-layout, gigahertz-range phase-locked loop (PLL) containing 200 million devices and a parasitic-heavy netlist (more than 95% RC). In just two days, the setup was ready to run with PrimeSim SPICE — including license installation and management, GPU libraries, tool settings, and expert support to streamline migration. Just as important, the customer did not need to change the input file; the same previous simulator input deck was used with PrimeSim SPICE.

By working with Synopsys, the customer realized two clear benefits:

1. Faster time to results: At PrimeSim’s highest-accuracy setting, runlvl 6, PrimeSim was 2.6x faster than the previous simulator the customer had been using at a setting equivalent to PrimeSim’s runlvl 3.
2. Accuracy of results: PrimeSim delivered improved accuracy in two critical areas.

• Effective number of bits (ENOB): PrimeSim measured 6.6, while the previous tool measured 6.4.
• Signal-to-noise ration (SNR): PrimeSim delivered ~ 41 dB, while the previous tool measured ~ 39 dB.

With the customer’s previous simulator, DC convergence was slow and often appeared to stall. With PrimeSim SPICE, the same job was completed within a few hours delivering a faster turnaround time and strong results on the first run.

The table below summarizes six runs done by the customer on the same design using different compute configurations and PrimeSim accuracy settings.

While results vary by circuit type and design size, this case study illustrates the value Synopsys PrimeSim can bring to large-scale analog simulation. Across CPU-only and CPU+GPU environments, PrimeSim SPICE accelerated circuit analysis, improved accuracy, and reduced time to results. Key takeaways include:

• With 32 CPUs and 4 GPUs at equivalent accuracy settings, PrimeSim SPICE delivered approximately 28x faster overall wall-clock time than the customer’s previous simulator.
• With the same compute configuration, PrimeSim SPICE ran ~3x faster with GPUs than without GPUs on transient analysis, and ~2x faster overall when including parsing, elaboration, DC convergence, and transient analysis.
• At PrimeSim SPICE’s most accurate setting (runlvl 6) on 32 CPUs and 4 GPUs, PrimeSim SPICE delivered 2.62x faster performance than the customers previous simulator running in a mode equivalent to PrimeSim SPICE runlvl 3.
• Along with the speedup, accuracy was also slightly improved .