The Universal Flash Storage (UFS) specification was, from its inception, meant to be road-worthy. Expected to overtake embedded MultiMediaCard (eMMC) in automotive applications, UFS has become the preferred non-volatile storage medium of choice for many smartphones. But as automotive data storage needs become more complex and demanding, it could find itself in a race with Non-Volatile Memory Express (NVMe) SSDs.
The latest and greatest iteration of UFS is version 4.0, published by the JEDEC Solid State Technology Association in August 2022. It has been longer than a decade since it debuted as a faster alternative to eMMC. UFS storage has since taken over the high-end smartphone market, as well as the mid-market, as its cost has come down. Beyond smartphones, it has also been used in some tablets and has become a popular option for automotive-storage applications.
But how far down the road UFS will go in automotive is not entirely clear. While JEDEC remains bullish on the opportunities for the standard, the consolidation of infotainment storage systems coupled with the exponential growth in on-board vehicle data could open the door for NVMe SSDs to displace UFS for some use cases.
eMMC remains a contender
An automotive analogy is apt when comparing UFS with eMMC. The newer interface’s primary advantage is that it is full-duplex, unlike eMMC, which is half-duplex: Essentially, it accommodates one-way traffic only. You can read or write with eMMC—not both—whereas UFS accommodates read and write operations in parallel. This parallelism was forecast to be the reason UFS would eclipse eMMC for smartphone storage, as its performance would be better suited for the expected higher-quality video and faster network speeds on mobile devices.
Aside from dramatically outpacing eMMC when it comes to read and write performance, what is notable about UFS is that the specification is still in development. Although eMMC is still in use, the standard is no longer being updated. But it has not kept vendors from developing devices using that standard where it makes the most sense. Last year, Western Digital debuted its NAND EM141 Embedded Flash Device for applications that require high performance, reliable data storage, a small form factor and efficient power consumption.
While briefing EE Times about the EM141, Eric Spanneut, VP of flash global product management for Western Digital’s flash business unit, said eMMC is still a very popular interface at capacities of 32 and 64 GB and continues to be widely used in automotive applications like telematics, infotainment and advanced driver-assistance systems (ADAS).
The high-end smartphone market, however, is for the most part the domain of UFS, while the PC flash storage market is being served by NVMe SSDs, Spanneut said. The move to a more centralized architecture for automotive systems will serve as a roadmap for where UFS is headed. Like eMMC, UFS has the proven reliability that is highly desired by automakers. With 4.0 just getting going, he said, there is a mix of UFS 2 and 3 in use, which means the landscape is somewhat fragmented. The choice between eMMC or UFS is case-dependent, with eMMC still being good enough as well as cheaper for some applications.
UFS meets many needs
Western Digital has addressed storage needs through its iNAND solutions, while Samsung’s approach has been to provide both the flash storage and LPDDR DRAM memory in a single, UFS-based, multichip package.
Samsung Electronics was one of the first vendors out of the gate with a UFS 4.0 solution, delivering speeds of up to 23.2 Gbps per lane, which is double that of UFS 3.1. That made it ideal for 5G smartphones with capabilities that need large amounts of data processing, including high-resolution images and video, high-capacity mobile game playing, AR/VR applications and automotive use cases like Samsung’s ADAS solutions. Increased multitasking and sensors are key drivers of smartphone memory requirements, resulting in higher performance expectations of both DRAM and flash storage.
Other vendors in the automotive segment, such as Micron Technology, have all the bases covered, with eMMC, UFS and SSD products to address in-vehicle compute, memory, storage and artificial-intelligence requirements. Silicon Motion’s Ferri family also includes all three technologies for a variety of market segments, including mobile, enterprise, industrial and automotive, through its FerriSSD, Ferri-eMMC and Ferri-UFS products.
UFS offers fast boot times for in-vehicle functions that must start up as fast as it takes for the driver to push the start button. However, the computational needs of automotive design have pushed data-storage requirements even further in terms of performance and storage capacity, leading to the adoption of automotive-grade SSDs. Their capacities allow for the consolidation of storage within the automotive system to reduce the overall number of storage devices required.
Like most memory and storage specifications, UFS has sought to double its performance with every iteration, and 4.0 is no different. Among the other improvements are the use of the more efficient VCC = 2.5 V, rather than the 3.3 V supported in the prior version of the standard, and the introduction of multi-circular queue definition for more demanding storage I/O patterns. System latencies are improved by enabling High Speed Link Startup, Out of Order Data Transfer and the BARRIER Command.
One approach to speeding up performance beyond what is outlined in the JEDEC specification is using MIPI’s M-PHY 5.0 physical layer protocol to significantly increase data-transfer rates, which Kioxia implemented early last year with UFS 3.1. Aimed at smartphones, tablets and ultra-portable notebooks, the company touted this approach for UFS devices as being able to deliver performance on par with that of entry-level PCIe 4.0 SSDs.
SSDs could displace UFS
Despite the advances achieved by UFS, Phison Electronics CTO Sebastien Jean told EE Times in an exclusive interview that PCIe/NVMe SSDs are likely to overtake the technology in automotive applications as storage demands grow. PCIe is not limited by only the host being able to initiate a transaction, which makes a huge difference in what you can enable with a UFS bus. “It just doesn’t have the capability of PCIe,” he said.
Jean said that UFS’s lower power consumption isn’t significant enough to make it a better choice than an NVMe SSD. “There’s no real power or performance benefit to UFS. There is a functional deficiency with UFS at a very fundamental level.”
The primary opportunity for UFS today is catering to the smartphone market, Jean added, which is a market comprised of five major manufacturers, with Apple doing its own thing. Smartphones are still dealing with relatively low densities, while the automotive market is becoming more enterprise-like as storage scales up toward more complex topologies. He said smartphones are well-served by UFS, but it’s playing catchup to NVMe when it comes to features required by automotive storage: For example, if there’s a lot of I/O virtualization for an automotive use case, NVMe is the better option.
JEDEC still sees a lot of potential for UFS in automotive and other segments, and the updates in 4.0 reflect the demands in today’s common use cases, including the doubling of the data rate, Hung Vuong, chair of JEDEC’s JC-64.1 subcommittee, said in an exclusive interview with EE Times. “The market demand is still there for bandwidth, so that’s the main change in UFS 4.0.” The addition of multi-circular queue definition is aimed at improved random I/O.
In the case of smartphones, storage performance needs to keep up with networking: A device needs to be able to ingest data as it receives it, whether it’s 5G today or 6G down the road.
“It’s a balancing act, especially for mobile,” Vuong said. Bandwidth and features must keep up with other technology, while being mindful of power consumption in portable, battery-powered devices.
“We can’t just crank up the open valve and let everything out,” he said. “We have to do it judiciously and be battery-efficient.”
While mobile remains a robust market for UFS, he said, JEDEC sees automotive as being a strong segment, a sentiment echoed in the same interview by colleague Bruno Trematore, who helps oversee JEDEC’s JC-64 committee.
Although past iterations of UFS have added automotive-specific features at the behest of OEMs, there is nothing tailored for them in 4.0, Trematore said. Right now, UFS has a strong presence in automotive in-vehicle infotainment systems and driver-assistance systems, such as cruise control, he said, while UFS is overkill for smaller use cases like gateways.
Regardless of how it stacks up against NVMe SSDs, the specification has a solid track record within the automotive industry, an industry that values reliability and longevity.
Vuong said UFS is best suited for applications that have high safety, reliability and data-protection requirements, as well as quick boot-up. Fast booting is an expected characteristic in today’s UFS card applications because consumers have come to expect it from their phones and their cars. UFS’s low-power–consumption profile makes it ideal for IoT edge devices, as its power reliability contributes to overall stability.
Another contender that could supplant UFS and eMMC, especially in automotive, is another JEDEC standard. The Crossover Flash Memory (XFM) Embedded and Removable Memory Device (XFMD) standard was introduced in October 2021 and was designed to serve as a “crossover” between embedded memory and memory cards, such as SD cards or compact flash. Applications include gaming consoles, virtual- and augmented-reality gear and video-recording devices like drones and surveillance systems.
XFMD’s removability is appealing for automotive because it would allow for the flash storage device within an embedded system to be easily swapped in the event of a needed upgrade or replacement due to failure. It could act as a buffer in ADAS storage devices that are continuously being overwritten, which can lead to the flash storage wearing out before a car when more data is written than anticipated.
While UFS’s future for smartphones appears straightforward, the road ahead in automotive is unclear, even if it does indicate new opportunities.
“There is a lot of talk in the automotive industry regarding how the whole car CPU architecture is going to be developed in the future,” Trematore said. “It’s a big question mark there because there are lots of ideas.”
