The next generation of Arm processors will deliver greater performance, enhanced security, and integrated artificial intelligence (AI) accelerators thanks to an all new microarchitecture, announced earlier this week.

“Armv9 is a rolling program with substantial enhancements to the architecture that we’ll be deploying over the next few years, increasing the computing capability in the widely applicable areas of digital signal processing and machine learning and improving the security and robustness of our systems,” Richard Grisenthwaite, SVP and chief architect at Arm, said during Arm’s Vision Day presentation.

Armv9 is the chip designer's first new microarchitecture in a decade and builds on the capabilities of its first 64-bit architecture Armv8. Arm claims the new architecture will enable the next 300 billion connected devices, which will range from IoT, edge, data center, smartphones, and consumer electronics.

Armv9 also promises to address many of the biggest challenges in computing today, namely confidential computing and AI. And while performance gains remain a key promise of the new architecture, security and AI are for the first time first-class citizens in this generation.

In fact, Arm sees security as computing’s greatest challenge. With Armv9, the chip designer is introducing native support for confidential computing.

“We all remember what happened with Spectre and Meltdown a few years ago and all the havoc that that caused. There’s a lot of things we’re doing with v9 that are going to improve that,” Rene Haas, president of Arm’s IP products group, said.

The Confidential Compute Architecture (CCA) is a hardware-level secure environment introduced with Armv9 that serves to shield portions of code and data from access while in use, even from privileged software. The CCA allows for memory encryption and works by dynamically allocating what Arm calls "realms" to applications. This serves to effectively segregate secure workloads from non-secure workloads.

According to Arm, the CCA could be used in a business application where the realm would protect commercially sensitive data and code from the rest of the system while it is in use, at rest, or in transit.

“I’m confident that every bit of digitally shared data will soon be securely processed on Arm-based technology at some point in its life either at the endpoint, in the digital network, or in the cloud,” CEO Simon Segars said.

Confidential computing has garnered considerable attention over the last few years — especially among public cloud providers — as chipmakers have sought to limit the impact of side-channel attacks like Spectre or Meltdown. Both Intel and AMD have made steady gains in this arena rolling more sophisticated confidential computing capabilities with every release.

“The increased complexity of use cases from edge to cloud cannot be addressed with a one-size-fits-all solution,” said Henry Sanders, VP and CTO of Azure edge and platforms at Microsoft, in a statement. “Arm is in a unique position to accelerate heterogeneous computing at the heart of an ecosystem, fostering open innovation on an architecture powering billions of devices.”

Armv9 also introduces new capabilities designed to accelerate AI and machine learning workloads.

According to Arm, there were more than eight billion AI-enabled voice-assisted devices in use by mid-2020, and estimates that by 2025 more than 90% of all on-device applications will contain some AI elements.

“The ubiquity and range of these machine learning workloads means they are not going to be addressed by a single class of solution, therefore all of our computing devices will benefit from increased hardware support for machine learning in CPUs, GPUs, or in specialized NPU processor elements,” Grisenthwaite said. He added that pending the close of Nvidia's $40 billion bid to acquire Arm, there will be an opportunity to take advantage of their machine learning expertise.

To address the need for onboard AI inference and training capabilities, Arm teamed up with Fujitsu to create the Scalable Vector Extention (SVE), which is at the heart of the Fugaku supercomputer. Armv9 will bring this technology to the mainstream with SVE2, which the company claims will enable a new generation of machine learning and digital signal processing capabilities, such as virtual and augmented reality and image processing.

Arm plans to further extend these capabilities with enhancements to the matrix multiplication instruction set used by the CPU, GPU, and neural processing unit.

Arm expects to realize a more than 30% increase in performance over the next two generations of mobile and infrastructure processors with the move to the Armv9 microarchitecture.

“However, as the industry moves from general-purpose computing toward ubiquitous specialized processing, annual double-digit CPU performance gains are not enough,” the company said.

To address these specialized workloads, Arm has implemented what it calls the Total Compute design methodology, which is designed to accelerate overall compute performance via system-level hardware and software optimizations.

“Increasingly, we’re looking to maximize the performance that the CPU provides through the system design. In fact, we can see more than a generation of CPU that can be gained through pushing the path to the memory from the CPU,” Peter Greenhalgh, VP of technology at Arm, said.

He added that by applying this philosophy across the full swath of Arm’s intellectual property, it will be able to increase clock frequency, bandwidth, and cache size, and reduce memory latency to maximize the performance of Armv9-based CPUs.

The Armv9 architecture is available for license now, with the first round of chips based on the architecture expected to begin shipping by the end of 2021.