Do Son 
As AI applications at the device level continue to gain traction, with increasing complexity in tasks ranging from photography and voice recognition to video processing and interaction with AI assistants, Arm has announced the enhancement of its computing platform’s AI capabilities in mobile devices through the Scalable Matrix Extension 2 (SME2) instruction set. This advancement empowers developers to significantly boost AI performance without altering existing source code, delivering a high-efficiency, low-power user experience.
SME2, an advanced instruction set within the Armv9 architecture, is purpose-built to accelerate AI workloads such as computer vision, speech recognition, and natural language processing. By executing directly on mobile CPUs, SME2 reduces reliance on cloud computing, alleviating latency and energy consumption bottlenecks in on-device AI deployments.
Notably, with Arm’s KleidiAI software acceleration layer, developers can automatically benefit from SME2’s enhanced performance without the need to modify models or applications. SME2 has already been integrated into numerous mainstream AI frameworks and libraries, including Google’s XNNPACK, LiteRT, and MediaPipe, Alibaba’s MNN, Microsoft’s ONNX Runtime, and even llama.cpp. This widespread integration ensures that developers can immediately leverage SME2 within the existing Android AI software stack.
According to Iliyan Malchev, Senior Engineer at Google Android, SME2 enables advanced language models like Gemma 3 to operate seamlessly on local devices without depending on cloud infrastructure. Benchmarking reveals that Gemma 3, powered by SME2, delivers up to six times faster response times for text summarization tasks—even generating 800-word content using a single CPU core—underscoring SME2’s transformative impact on mobile AI performance.
Devices featuring SME2-enabled Android hardware are slated for release soon, and certain iOS devices have already incorporated support for the SME2 instruction set. With KleidiAI’s automatic matrix computation optimization, developers can effortlessly scale their applications on supported platforms, ushering in the next generation of AI performance with real-time responsiveness, low latency, and enhanced computational efficiency.
Arm reports that more than 9 million applications currently run on its computing platform, with over 22 million developers engaged in building on the Arm ecosystem. As SME2 technology becomes more widespread, an increasing number of applications will be able to execute complex AI tasks directly on mobile devices—fundamentally redefining user expectations for performance and on-the-go intelligence.
Related Posts:
- Google Unleashes Gemma 3n: Breakthrough On-Device Multimodal AI for Smartphones & Laptops
- Google Boosts Real-Time Protection Against Scams and Malware on Android Devices
- Google AI Edge Gallery: Unleash On-Device AI Power on Your Android (and Soon iOS!)
Support Our Threat Intelligence
If you find our CVE report and cybersecurity news helpful, consider supporting our work.
Buy Me a Coffee
PayPal
