Arm and Synopsys Target AI Data Centres With New Platform

Share this article
Share this article
Prioritise Us on Google
Mohamed Awad, Executive Vice President of the Cloud AI Business Unit at Arm
Firms aim to tighten links between CPU architecture, design tools and proven IP to move AI‑ready server chips from concept to production more efficiently

Arm and Synopsys have joined forces on a new Arm‑based server CPU platform designed for the realities of modern AI infrastructure. 

Branded the Arm AGI CPU, the project marries Arm’s server‑class compute subsystems with Synopsys’s end‑to‑end chip design software, pre‑built interface IP and hardware‑assisted verification platforms. 

Arm positions this as the first phase of the next evolution of its compute platform and a step that ties its architecture more directly to production silicon than in prior generations.

Arm is expanding its compute platform to silicon products. Credit: Arm

Why it matters beyond AI teams

While GPUs and dedicated accelerators carry the heaviest numerical loads in AI training, CPUs remain the backbone of the data centre. 

They run operating systems and control planes, schedule and orchestrate distributed jobs, feed data to accelerators and handle networking, storage, security and a substantial share of inference and microservice workloads. 

As AI use expands, the energy footprint of the supporting CPU layer becomes an enterprise concern as much as a technical one. 

Arm aims to improve performance per watt, a metric that can translate into lower power bills, more servers per rack within the same power envelope, and more budget and thermal headroom to allocate to accelerators.

What exactly is being built

The AGI CPU is built on Arm’s Neoverse CSS V3, which provides pre‑engineered building blocks for server‑class system‑on‑chips (SoCs). 

An SoC is essentially a complete computer condensed into a single piece of silicon, combining CPU cores, memory controllers, security engines, I/O and sometimes specialised accelerators. 

Starting from a validated subsystem allows chip makers to skip years of risk‑laden groundwork and focus on the features that differentiate their designs, such as memory bandwidth, I/O choices or accelerator attachment. 

Synopsys supplies the toolchain and intellectual property that turn this blueprint into working chips, along with platforms that let teams validate hardware and software before the first wafers are ever manufactured.

How the design comes together

Designing a data center‑grade SoC begins with logic design and simulation to make sure the chip does what it is supposed to do, long before any physical layout exists. 

That logic is then transformed into an actual floorplan and wiring on silicon, constantly optimised for speed, power consumption and area. 

At each stage, the design is subjected to rigorous checks to confirm that it will meet its target clock speeds, that its power delivery network will remain stable under heavy load and that it obeys all of the manufacturing rules specific to a given semiconductor process. 

Only after passing this, can a design be sent to a foundry for fabrication, a milestone often referred to as tape‑out. 

Youtube Placeholder

By starting from Arm’s pre‑validated Neoverse CSS and running through Synopsys’s mature toolchain, the partners aim to compress schedules, reduce integration surprises and raise the likelihood that first‑silicon works as intended.

The Synopsys toolchain, demystified

Electronic design automation (EDA) is the hardware world’s equivalent of an IDE, compiler and test suite rolled into one. 

Synopsys VCS is used to simulate the chip’s logic, catching functional bugs early when they are cheapest to fix. 

Fusion Compiler takes a validated design and turns it into a manufacturable layout, optimising for performance, power and silicon area. 

IC Validator checks that the physical layout follows the foundry’s intricate manufacturing rules. 

PrimeTime confirms that signals will traverse the chip fast enough to meet timing at the intended clock speed. 

RedHawk‑SC models power integrity and reliability so that the chip remains stable and thermally sound under real workloads. 

Together, these tools give engineering teams the means to move from concept to a production‑ready design with fewer blind spots.

Youtube Placeholder

Speeding up verification before silicon exists

One of the biggest bottlenecks in modern chip programmes is software readiness. 

Data centre buyers need firmware, drivers, operating systems and platform software that are tuned and stable on day one. 

To bring software forward in the schedule, Synopsys provides hardware‑assisted verification platforms that act like high‑speed digital twins of the chip. 

The ZeBu Server 5 emulator lets teams boot operating systems and run substantial workloads against a cycle‑accurate model months before the chip is fabricated, which helps uncover system‑level issues early. 

HAPS prototyping systems use reconfigurable FPGAs to create fast prototypes suitable for extended software development, performance tuning and validation of interactions across the whole system. 

By the time first‑silicon arrives, a significant portion of the software stack can already be shaken down.

What the companies are saying

“Designing data centre silicon for increasingly complex AI workloads requires rigorous validation across the full system,” says Mohamed Awad, Executive Vice President of Arm’s Cloud AI business. 

“The Arm AGI CPU reflects the strength of our SoC design and the effectiveness of our collaboration with Synopsys.”

Ravi Subramanian, Chief Product Management Officer at Synopsys, praises Arm’s ability to deliver “with such ambition and precision”, and describes Synopsys design, IP and advanced verification solutions as “mission‑critical” to the outcome.

Ravi Subramanian, Chief Product Management Officer at Synopsys. Picture: Synopsys

Overall, Arm and Synopsys are tightening the link between CPU architecture, design tools and proven IP to move AI‑ready server chips from concept to production with less risk and better energy efficiency. 

For organisations aiming to scale AI while containing power and cost, this is a development worth tracking as products, benchmarks and ecosystem support come into focus.

Company portals

Executives