Kioxia has announced its new UFS 5.0 embedded flash memory solutions designed to power next generation AI devices. The storage chips run on the MIPI M PHY v6.0 physical layer to support theoretical interface speeds of up to 46.6 Gb per second per lane. Commercial samples are available now in 1 TB and 512 GB capacities, with mass production scheduled to begin later this year.
Slow storage read performance has historically bottlenecked local AI models on portable hardware. Large language models and real time inference systems require massive datasets to move between memory and compute units almost instantly. According to the official announcement by Kioxia, this hardware update aims to resolve those speed barriers, allowing local AI applications to load faster and run with far less lag. The technology removes the data congestion that previously limited sophisticated on device programs.
The package combines a custom controller with the eighth generation BiCS FLASH 3D memory from Kioxia. Users can expect sequential read speeds peaking at 10 GB per second and sequential write speeds hitting up to 9 GB per second. These transfer rates represent a major jump over previous generation UFS 4.1 hardware. Despite the speed increase, Kioxia managed to reduce power consumption and improve thermal management within the chip.
Kioxia shrunk the physical size down to a 7.5 by 13 millimeter BGA package. This footprint is smaller than older storage models, which helps phone designers save precious space inside tight modern casings. While premium smartphones are the primary target, the technology is also intended for tablets, handheld gaming consoles, virtual reality headsets, and smart security cameras. Live demonstrations of the new storage modules are scheduled to begin next week at the Future of Memory and Storage exhibition.
