Arm Unveils Lumex Architecture for AI-Powered Mobile Chips

Arm Holdings has introduced its latest mobile chip design architecture, named Lumex, which is optimized for artificial intelligence (AI) and designed to support AI functionalities on mobile devices such as smartphones and smartwatches without the need for an internet connection. The Lumex series includes four different types of chip designs, catering to various performance needs, from low-power, high-efficiency versions for wearable devices to high-performance versions for flagship smartphones.

The high-performance version of the Lumex architecture is specifically designed to support large AI models on high-end devices, enabling local deployment and operation of AI functionalities without relying on cloud computing resources. This development underscores the growing expectation among users for AI capabilities in their devices, as noted by Arm's Senior Vice President and General Manager, Chris Bergey.

The Lumex design is part of Arm's Compute Subsystems (CSS) business unit, which aims to provide ready-to-use technology solutions for mobile manufacturers and chip designers. This initiative helps accelerate the integration of new technologies into products, reducing research and development cycles. Arm's long-term strategy involves expanding its revenue streams by offering comprehensive chip design solutions for data centers and mobile devices, and investing in independent chip research and development.

The Lumex architecture is optimized for 3-nanometer process technology, a cutting-edge manufacturing process used by leading suppliers. This optimization ensures that the Lumex chips can deliver high performance and efficiency, aligning with the latest advancements in semiconductor technology. Arm plans to host a dedicated launch event for the Lumex architecture, highlighting its strategic importance in the global mobile market.

In addition to the Lumex architecture, Arm has integrated the SME2 technology into its platform, which has shown significant improvements in AI performance. This technology has reduced the latency of voice workloads by 4.7 times and increased audio generation speed by 2.8 times, enabling real-time AI inference on devices. Practical applications of SME2 technology include enhancing the responsiveness of AI applications, such as smart yoga coaching and large language models, by providing faster and more efficient interactions.

The Lumex platform also features the Mali G1-Ultra GPU, which offers a twofold increase in ray tracing performance and up to a 20% increase in AI workload performance. This GPU is designed to provide desktop-level gaming experiences and faster AI inference, making it ideal for real-time applications. The platform's flexibility allows partners to choose between using it as-is or customizing it to meet their specific needs, balancing peak performance, sustained efficiency, and chip area.

The Lumex platform is supported by industry leaders, including Alibaba, Alipay, Samsung System LSI, Tencent, and Vivo. These partners have validated the performance and efficiency of the SME2 technology, demonstrating its potential to deliver faster and more efficient AI experiences on mobile devices. The platform's developer-friendly design, with the KleidiAI software library integrated into major AI frameworks, ensures seamless cross-platform compatibility and an out-of-the-box AI development experience.

In summary, the Lumex CSS platform represents a significant advancement in mobile chip design, offering enhanced AI performance and real-time inference capabilities. With its flexible architecture and support from industry leaders, the Lumex platform is poised to drive the next generation of mobile AI experiences, providing users with faster, more responsive, and efficient AI functionalities on their devices.