Arm’s approach connects efficient AI hardware with optimized software, security, safety-focused automotive designs, partner-built chiplet platforms and company emissions targets. An October 27, 2024 interview with Embedded.com describes these as complementary parts of a strategy to support growing AI workloads while managing power use and environmental impact.
How does Arm aim to make AI computing more efficient?
The strategy spans both the hardware that runs AI and the software that helps developers use it. The interview reports Arm’s claims about its NPU’s performance and efficiency, while describing software work intended to bring Arm optimizations into established AI frameworks.
Ethos-U85: an NPU for edge AI
Arm positions the Ethos-U85 neural processing unit for edge applications such as factory automation and smart-home cameras. Arm reports a fourfold performance increase over its predecessor and 20% greater power efficiency. Configurations range from 128 to 2,048 MAC units, with performance of up to 4 TOPS at 1 GHz. These figures are Arm’s reported specifications in the 2024 interview, not independent test results.
The interview also says the NPU’s standard toolkit is intended to let partners reuse existing assets and maintain a consistent developer experience. That continuity can matter when a product team is adapting an existing design rather than building a new AI software stack from scratch.
#1 Best Overall
KleidiAI: optimization in familiar frameworks
KleidiAI is described as a software layer that integrates Arm optimizations into AI frameworks, including PyTorch and ExecuTorch. Arm’s goal is to help AI workloads run efficiently on Arm CPUs, from cloud data centers to edge systems, without requiring developers to add optimization work for each workload. The interview presents that as an aim, not a guarantee that every model or application will run optimally without developer changes.
What role do Armv9 and security play in AI workloads?
Armv9 combines capabilities for data-parallel and matrix-heavy processing with security features. The interview names Scalable Vector Extension 2 (SVE2) and Scalable Matrix Extension (SME) as extensions intended for those computational patterns. Its listed security features are Confidential Compute Architecture (CCA) Realms, pointer authentication, branch target identification and memory tagging extensions.
Those features are relevant to AI systems because performance is only one design consideration: deployments may also need to protect code and data and reduce exposure to certain classes of memory or control-flow attacks. The interview identifies the capabilities but does not establish that Armv9 is sufficient for every AI workload’s security requirements. Suitability still depends on how a system is designed, configured and deployed.
How does Arm address safety in automotive computing?
Arm’s automotive portfolio is described through three operating modes. The modes offer different ways to separate or coordinate core activity, depending on the safety needs of a workload.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minute| Mode | How it works | Examples or use |
|---|---|---|
| Split | Separates non-safety-critical workloads. | Workloads that do not require the safety behavior associated with lockstep operation. |
| Lock | Runs cores in lockstep for safety-critical functions. | Examples include advanced driver-assistance systems (ADAS). |
| Hybrid | Synchronizes selected logic while allowing cores to operate independently. | Intermediate safety needs, such as lane-departure alerts and electric-vehicle energy management. |
The interview describes the modes and examples, but does not provide certification details or a complete mapping of particular products to automotive safety standards. They should therefore be read as architectural options, not as proof that a given vehicle system meets a specific safety requirement.
What is Arm Total Design?
Arm Total Design is described as an ecosystem for developing chiplet platforms for cloud computing, high-performance computing and AI/machine learning. The interview names Samsung Foundry, ADTechnology, Rebellions, Alcor Micro, Egis, PUFsecurity and SemiFive among its partners.
A partner ecosystem can bring together different capabilities needed to develop a platform, rather than treating the processor IP as the whole product. For data-center and AI/ML designs, that makes chiplet scalability and partner integration relevant alongside throughput and power efficiency. The interview names participants and target areas, but does not give comparable platform benchmarks, schedules or product availability.
What sustainability progress does Arm report?
Arm’s sustainability strategy links partnership with the United Nations Sustainable Development Goals. Kevork Kechichian, Arm executive vice president of solutions engineering, said: “Arm has taken a partnership approach, defined in our current sustainability strategy, aligned to collectively deliver on the United Nations’ Sustainable Development Goals for over a decade.”
The interview reports that Arm reduced its greenhouse-gas emissions by 77% in 2024 compared with a 2020 baseline, used 100% renewable power, and has an absolute net-zero emissions target for 2030. It also cites carbon budgets and hybrid work as measures intended to reduce emissions, including travel-related emissions. These are Arm progress claims as reported by Embedded.com; the interview does not provide a full audited methodology, a scope breakdown or independent assurance for the 77% figure.
Arm also frames the power efficiency of Arm-based devices as a potential contribution to sustainability: devices that require less power can use less operational energy. That is a design-level rationale, not a lifecycle assessment of every device using Arm technology. A product’s overall environmental impact depends on more than its processor, including its manufacture, use and end-of-life treatment.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How do the pieces fit together?
Arm’s approach is not one product or one emissions figure. It combines hardware performance claims, software intended to make Arm CPU optimization easier to access, architectural security features, automotive modes for different safety needs, a chiplet partner ecosystem and company-level emissions actions. For readers comparing these approaches, the relevant measure depends on the application: edge systems need performance per watt and tool continuity; automotive systems need appropriate functional-safety design; and cloud or AI platform builders may also care about chiplet scalability and partner integration.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Recommended Free Tools
