35 Views

Scaling Physical Verification to Billions of Polygons: Ramtera LVR Demonstrates 96% Cloud Scaling Efficiency

Physical verification is one of the most computationally demanding stages in semiconductor design. As design complexity increases, engineers face growing challenges in verification turnaround time, compute infrastructure requirements and overall EDA costs.

 

Modern physical verification engines must address these challenges while preserving verification accuracy across increasingly complex process technologies.

 

Ramtera Design Automation has reported a new benchmark demonstrating the potential of distributed cloud-based physical verification.

 

Using its LVR physical verification engine, the company checked a 1.35-billion-polygon test layout in 214.8 seconds on 128 AWS cloud servers, comprising 1,024 CPU cores.

 

The experiment used a complete 12nm foundry design-rule deck with more than 15,000 rule checks.

 

Testing Distributed DRC Throughput

The experiment used a post-fill reference layout containing approximately 10.56 million polygons.

 

The block was replicated 128 times to create a 2.42 GB GDS layout containing approximately 1.35 billion instantiated polygons.

 

Each copy was assigned to a separate cloud server, where LVR executed the full applicable DRC rule deck.

 

The infrastructure consisted of 128 AWS r8a.2xlarge instances, each with eight CPU cores and 64 GB of memory.

 

Each server processed its assigned region independently, while external orchestration software launched the verification jobs and consolidated the results.

 

The experiment was designed to measure distributed verification throughput and weak scaling rather than full-chip signoff performance.

 

Near-Linear Weak Scaling

The benchmark evaluated execution across several infrastructure configurations.

 

Configuration

Polygons

Execution time

Eight-core workstation

10.56 million

193 seconds

16 AWS servers

168.99 million

206.5 seconds

128 AWS servers

1.35 billion

214.8 seconds

 

Increasing the number of cloud servers from 16 to 128 increased the verification workload eightfold, while runtime increased by approximately 4%.

 

This corresponds to approximately 96.1% weak-scaling efficiency.

 

The result illustrates how a parallel verification architecture can scale throughput across large compute clusters when the workload can be partitioned into independently verifiable regions.

 

Verification Correctness Across 128 Servers

Scaling physical verification is not solely a performance problem. Verification results must remain consistent as workloads are distributed across many processors and systems.

 

In the LVR experiment, all 128 servers produced results identical to the validated reference output for their assigned block.

 

Each server reported 950 results across 12 rule checks, yielding 121,600 results in the combined database.

 

The reference block had previously been evaluated against Calibre-generated design-kit results, with exact agreement across all 14,731 rule checks included in that comparison.

 

The experiment therefore demonstrates deterministic reproduction of previously validated results across independently executed distributed jobs.

 

The Economics of Cloud-Based Physical Verification

Large-scale physical verification traditionally requires substantial investments in compute infrastructure and commercial EDA licenses.

 

On-demand cloud computing introduces a different operating model, enabling engineering organizations to allocate compute capacity as needed.

 

For the 128-server LVR experiment, Ramtera estimated a total AWS infrastructure cost of approximately $11.57, including server runtime, storage and network address charges.

 

The complete cloud session, including provisioning and shutdown, took approximately nine minutes.

 

This translates to an estimated infrastructure expense of less than one US cent per million polygons processed.

 

The estimate excludes commercial EDA licensing costs and remains subject to confirmation against the AWS bill.

 

While the benchmark does not establish a total-cost comparison with other physical verification tools, it highlights the potential economics of elastic verification infrastructure.

 

Beyond the Benchmark: Physical Verification Across Process Technologies

In addition to its cloud-scaling work, Ramtera reports testing LVR across 30 semiconductor technology nodes associated with GlobalFoundries, IHP and X-FAB.

 

The platform supports DRC and LVS, along with other physical verification capabilities.

 

LVR also incorporates Model Context Protocol (MCP) server support, allowing integration with emerging AI-assisted engineering environments.

 

These developments reflect increasing interest in connecting conventional semiconductor verification workflows with distributed computing and automation.

 

Toward Production-Scale Distributed Verification

The benchmark demonstrates high-throughput processing of independently partitioned geometry.

 

A production SoC presents additional challenges, including cross-partition connectivity, interacting geometries, global density requirements and other context-sensitive verification rules.

 

The test design consisted of identical post-fill blocks, with each block checked independently. Consequently, the result should be interpreted as a distributed throughput demonstration rather than a measurement of complete-chip verification time.

 

Future evaluations on heterogeneous production layouts will help establish how this performance translates to more complex, interconnected designs.

 

Nevertheless, the experiment provides a meaningful example of how cloud infrastructure can be applied to computationally intensive physical verification.

 

Conclusion

Ramtera LVR’s 1.35-billion-polygon benchmark demonstrates strong weak-scaling performance across 128 cloud servers.

 

By increasing verification throughput nearly proportionally with the deployed compute capacity, the experiment highlights an opportunity to reduce verification turnaround time through elastic cloud infrastructure.

 

As semiconductor designs grow in complexity, scalable physical verification engines that combine accuracy, automation and efficient infrastructure utilization may play an increasingly important role in semiconductor design flows.

 

About Ramtera Design Automation

Ramtera develops EDA software for physical verification and physical design. Its products include LVR, a physical verification platform, and PRE, a place-and-route engine.

 

Website: https://ramtera.com

AnySilicon company page: https://anysilicon.com/vendors/ramtera/

Logo Image
Privacy Overview

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.