Arm Validates IP 1,000X+ Faster with Solido on AWS Graviton2

Arm, a global leader in semiconductor intellectual property (IP), develops technology that integrated device manufacturers use to create their own integrated circuit (IC) products, such as system-on-chips (SoC). Best known for designing CPU cores, Arm also offers the critical IP building blocks for SoCs, including GPUs, NPUs, and interconnects. To commercialize its IP, Arm must demonstrate that the functionality of the IP is robust enough to ensure high chip yield and meet the specified goals for power, performance, and area (PPA). In the past, Arm used brute force Monte Carlo simulations to verify PPA and yield; however, the increasing chip design complexities and manufacturing process variation make it too expensive to continue that way.

Arm adopted Solido Variation Designer software from Siemens EDA to verify IP across the full range of process, voltage, and temperature combinations, with variation out to 6-sigma. Now, Arm can verify its IP products to 6-sigma, 1,000 times faster than using brute force Monte Carlo methods, with greater accuracy and coverage. Furthermore, by moving the solution to run on AWS Graviton2 processors, which are based on the Arm Neoverse N1 core, the company reduced costs by 24 percent. 

Arm IP is used across a wide range of diverse applications including IoT sensors and devices, smartphones and PCs, automotive and industrial use cases, as well as cloud datacenters and networking infrastructure. Before licensing and delivering its IP, Arm must simulate the circuit’s design over different manufacturing variations and environment conditions to prove it will function correctly, meet PPA requirements, and yield high-quality products.

Modern process nodes shrink transistor sizes and enable SoCs to deliver higher performance with greater efficiency and power optimizations. As the production environment shrinks, it also reduces the margin of error for ensuring accurate, working products and it requires running an extremely large number of simulations under the brute force Monte Carlo method. This means even more compute resources, more simulation run time, and thus more expense to prove the components operate reliably. Arm knew it needed a way to speed up simulations and IP testing in new production environments without jeopardizing quality or performance.

Solido Variation Designer from Siemens EDA offers a comprehensive suite of tools for variation-aware design and verification. It uses machine learning technologies to deliver orders-of-magnitude fewer simulations with the accuracy of brute force techniques. “We are using Solido Variation Designer on AWS Graviton2 based on Arm Neoverse technology to verify our standard cell IPs to 6-sigma, and we have seen more than 1,000 times faster simulation compared to previously used brute force techniques,” said Andrew Evans, Technical Director at Arm. “Siemens EDA has delivered the perfect combination of advanced technology and great support, enabling us to achieve the accuracy and performance needed across all process technology platforms.”

For over 16 years, the Solido team at Siemens EDA has been a pioneer in bringing artificial intelligence to chip design and developing adaptive machine learning techniques. Solido's breakthrough technology speeds up SPICE-level verification by orders of magnitude, while fully preserving accuracy. This enables chip designers to verify designs more thoroughly and more quickly, and it results in chip designs that have higher performance, lower power, and are cheaper to manufacture. "The Solido technology that Arm is using is the world's most advanced SPICE-level IP verification solution," explained Jeff Dyck, Director of Engineering at Siemens EDA. "Arm has been an amazing technology partner to work with, and we are delighted that Solido technology is helping Arm to produce top-performing and high-yielding IP."

Siemens EDA’s Solido tool helped Arm meet its low-power, high-performance, and area-optimization goals. Arm also needed a cost-effective way to run the simulations it produced. It started off using its own high performance computing (HPC) environment built on an on-premises compute farm, but that proved to be unsustainable from a cost and simulation-time perspective. Instead of just lifting what was used on-premises, Arm adopted services like Amazon Elastic Cloud (Amazon EC2) to optimize for the cloud environment. With Spot Instances, Arm can take advantage of unused Amazon EC2 capacity at up to a 90 percent discount compared to on-demand prices. Arm’s HPC scheduling engine sends jobs to Spot Instances when there is spare capacity and the price charged per hour per instance matches what Arm is willing to pay. Spot Instances work great for customers like Arm who have an unpredictable and variable workload.

Even with server discounts factored in, simulations still make up a big line item in the compute budget for Arm. Cost optimization is always a priority for Arm’s engineering team because more time costs more money, and compute time is determined by the processor running the workload. When AWS offered Amazon EC2 instances based on the Graviton2 processor, Arm saw an opportunity to extend its existing use of Solido Variation Designer by migrating its usage to Graviton2-based instances, which lowered Arm’s costs by 24 percent, reduced total CPU time by 12 percent, and improved turnaround time by six percent.

Using Solido Variation Designer from Siemens EDA on AWS allowed Arm to verify its products more than 1,000 times faster than brute force Monte Carlo, with greater coverage and accuracy. The tool effectively does the math to simulate products before they’re manufactured and ensure the final product will meet PPA and yield goals. The team deployed the tool on Amazon EC2 M6g Instances powered by Graviton2 processors to realize the scale and efficiency of fully managed cloud services with the most efficient compute resources.

But the story gets even better. Curious to know what made Graviton2 so much more efficient? The AWS Graviton2 processor was built using Arm Neoverse N1 cores which enables a highly scalable datacenter infrastructure from cloud to edge. Graviton2 delivers up to 40 percent better price performance over comparable current generation AWS instances for a wide variety of workloads. “When Arm IP is adopted by partners, and we in turn can use it to develop the next generation of products, the entire partner ecosystem benefits from that collaboration,” said Philippe Moyer, VP Design Enablement at Arm. “We call this Arm on Arm.”