Arm, a U.K. company that develops the core IP (intellectual property) of the application processor (AP), the smartphone "brain," unveiled a new computing subsystem (CSS) platform, "Lumex," on the 10th. Lumex is an on-device (embedded) AI–dedicated platform that seamlessly delivers high-performance AI features such as real-time translation and AI assistants on smartphones and PCs.
The new product bundles Arm's latest central processing unit (CPU), graphics processing unit (GPU), and other key semiconductor designs into a single verified package to help clients such as Samsung Electronics develop AI devices faster and more easily. The new product is part of Arm's strategy to evolve in the AI era from a simple IP supplier into an AI platform company.
Hwang Seon-uk, head of Arm Korea, said at a product launch briefing held at Lotte Hotel in Jung District, Seoul, on the 10th, "We designed a completely new platform that goes beyond existing computing methods to match an era of explosive AI growth," and added, "This is Arm's new platform strategy to help clients bring AI products to market more quickly."
The biggest technical feature of Lumex is that it tailors the general-purpose CPU, which handles a device's overall computation, for AI processing. To do this, Arm fully applied its "SME2" technology, a kind of "dedicated calculator" solely for AI computation. The new "C1 CPU" equipped with the AI-dedicated calculator improved AI performance by up to fivefold over previous versions, and response speed for speech recognition tasks became 4.7 times faster.
Jeong Chun-sang, principal FAE at Arm Korea, said, "In the past, developers faced the burden of shifting work to GPUs or NPUs (neural processing units), which are complex to program, to achieve AI performance," and added, "Now, thanks to SME2, they can implement fast and efficient AI features in a familiar CPU environment without complicated steps."
Graphics processing capability has also improved. The new GPU included in Lumex doubled the performance of "ray tracing," which renders graphics by tracing the path of light, compared with the previous generation. Graphics benchmark performance and AI inference speed each improved by up to 20%, enabling mobile devices to deliver a console-level, highly immersive gaming experience.
Arm's new platform is expanding its ecosystem through collaborations with major technology corporations such as Samsung Electronics and Google. Google adopted SME2 technology in its latest AI model, "Gemma 3," and Alipay used SME2 to improve response times for its LLM service by more than 40%.
Hwang said, "By 2030, SME2 technology will be installed in more than 3 billion devices, adding over 10 billion TOPS (10 quintillion operations per second) of computing performance," and added, "We will actively target the on-device AI market to establish ourselves as an AI platform corporations."