The next Universal Flash Storage (UFS) standard has yet to be released, but the standards organization overseeing it has made the uncharacteristic move to disclose its goals for version 5.0, which includes support for AI.

In a briefing with EE Times, representatives from the JEDEC Solid State Technology Association discussed what can be expected from the nearly completed UFS standard, which aims to deliver faster data access and improved performance compared to its predecessor, while maintaining compatibility with UFS 4.x hardware.

Among the features that can be expected are increased sequential performance up to 10.8 GB/s to meet the demands of AI.

To attain those performance improvements, JEDEC is leveraging MIPI Alliance specifications to form its Interconnect Layer. UFS 5.0 incorporates the upcoming MIPI M-PHY version 6.0 specification and the UniPro version 3.0 specification. 

M-PHY 6.0 includes High-Speed Gear 6 (HS-G6), which supports data rates double those of the previous maximum data rate of HS-G5, which will enable a UFS interface bandwidth of 46.6 Gb/s per lane and support up to about 10.8 GB/s effective read/write operation for UFS 5.0 over two M-PHY lanes.

Incorporating M-PHY 6.0 will also enable UFS 5.0 to include integrated link equalization for more reliable signal integrity and a distinct power supply rail to provide noise isolation between PHY and memory subsystem, which makes system integration easier.

Bruno Trematore, co-vice-chair of the JC-64 Committee for Embedded Memory Storage and Removable Memory Cards, said the evolution of UFS has also seen more bandwidth added than necessarily expected or needed by industry, until now. “AI is an insatiable monster.”

Trematore said smartphones are not competing with full-blown servers for performance or capacity, but more AI is being done on handheld devices. “If the memory is faster, you can have a bigger model on it.”

Trematore said mobile devices and automotive has similar requirements from UFS, including the need for signals to travel at high speeds between processors and memory. MIPI is providing the nut and bolts to help JEDEC build the application to optimize this process.

Christian Gyllenskog, also a co-vice-chair of the committee, said the experiences people are expecting from AI-enabled smartphones are also being projected onto the automotive world. “When someone gets in their car and pushes their start button, or turns their key, they expect an immediate response,” he said.

Language translation apps on smartphones are a perfect illustration of the user experience that is expected, Gyllenskog said. “We immediately want to see that the language is translating, that I’m able to communicate with whoever it is who I want to communicate with.”

The gigabytes needed for those models to load are significant, he added. “They’re large compared to the traditional applications. The biggest gating factor is the amount of time it takes for that LLM to load, and that’s where this new UFS 5.0 shines.”

Gyllenskog said the proposed 10.8 GB/s for UFS 5.0 would allow users to load apps with the immediacy users expect.

But smartphone applications are not that big yet. The goal of UFS 5.0 is to have some headroom, he said.

Faster speeds on the link not only address the demands from AI but also help to reduce power consumption, Gyllenskog said. “If you can increase the speed on the link, then that’s less active time for the system. Less active time on the system saves power.”

He said some customers choose UFS because of its power efficiency. It is especially important for small form factor devices relying on batteries—and not just smartphones, but edge devices that can quickly process data when needed and go back into sleep mode. “You want to preserve that power and get all-day functionality,” Gyllenskog said.

Rotem Sela, another vice-chair of JEDEC’s JC-64 Committee, said UFS development has mirrored the development of eMMC when it comes to power consumption. “We always keep power in mind for our ecosystem.”

The standard also continues to provide opportunities for customers to create security strategies, Gyllenskog said, including protecting data at rest.

UFS 5.0 will add inline hashing capabilities, he said. “You’re able to check against a hash to see whether the data that you just read is the data that you thought you had written.”

Sela said inline hashing is more efficient and contributes to overall performance improvements, including acceleration for AI because it is done by hardware, not software.

He said predicting future needs from UFS over the long term is not easy—AI requirements have escalated dramatically and remain unpredictable. “Nobody really knows what the exact requirements are,” Sela said. “We are enabling creativity in the market with these standards.”

JEDEC published UFS 4.0 in 2023 and released a 4.1 update in December 2024.

From EETimes