Next-generation design package integrating CPU, GPU, and system IP
ARM Transforms into a Platform Provider Amid the Proliferation of On-Device AI

ARM unveiled its next-generation Lumex CSS (Compute Subsystem) platform on September 10th. This announcement is interpreted as ARM's move to target the era of on-device AI, with an integrated design package strategy that goes beyond simply providing CPU/GPU IP and supports semiconductor companies in quickly designing and releasing SoCs (System-on-Chips).

Lumex CSS is centered around ARM's latest Armv9.3 architecture-based CPU (C1 series), encompassing GPU, system IP, physical implementation, and software stack.
With the Scalable Matrix Extension v2 (SME2) that enhances AI computational efficiency, the Mali G1-Ultra GPU that doubles ray tracing performance, support for 3nm optimized design, and compatibility with major AI frameworks, fabless companies can secure key performance blocks from ARM and attempt to differentiate themselves through their own modules and software.

ARM explained that this can achieve up to five times AI performance improvement with the same power budget.

At the announcement event that day, Hwang Seon-wook, President of ARM Korea, revealed that Lumex CSS will also be applied to Samsung's Galaxy Z Flip series, revealing the first case of it being applied to an actual commercial device.

Jeong Chun-sang, senior FAE at ARM Korea, explained that by adding GPU performance enhancements to the existing NSS function, its usability in on-device AI environments will be greatly enhanced.

ARM Korea CEO Hwang Seon-wook

end="1025" data-start="821">
Industry insiders believe that ARM's launch of this platform demonstrates its strategic intent to strengthen its position as a platform provider rather than a simple IP provider, and to strengthen ecosystem lock-in against alternative ISAs such as RISC-V.

Lumex CSS-based SoCs are expected to be commercialized around 2026, and their potential for expansion beyond mobile devices into various fields such as wearables, automotive electronics, and IoT is noteworthy.