Samsung has just started sampling its 256 GB SD Express microSD card (aka microSD Express card) with a sequential read speed of up to 800 MB/s through its PCIe/NVMe interface.
The SD Express standard was first introduced in the SD specification ver 7.0 in 2018 adding pads for PCIe/NVMe interfaces in full-sized SD cards before the microSD Express standard was introduced in the SD 7.1 specification the following year. Since then, the SD association further improved the standard promising ever faster SD Express speeds up to 4GB/s and microSD Express speeds up to 2GB/s.
Saying the adoption of the new SD Express and microSD Express standards has been slow would be an understatement, and the Samsung 256GB microSD Express card looks to be the world’s first such storage device. Looking at full-size SD Express cards, I could only find one on Amazon for $59.99 with 256GB capacity, 820 MB/s read speed, and 500 MB/s write speed. It also ships with an SD Express card reader simply because I’m not aware of any devices supporting the new standard… At least those cards are backward compatible so they’ll work on any SD card/microSD card reader but at a slower speed, for example up to 104MB/s with a UHS-I card reader. ADATA also offers two models with 256GB or 512GB capacities at higher prices.
Besides the chicken and egg issue of launching a new storage standard, those microSD Express can get very hot, and Samsung claims its Dynamic Thermal Guard (DTG) technology helps maintain the optimum temperature for the SD Express microSD card even when they are used for a long period of time. Performance-wise, 800MB/s is four times faster than the best UHS-1 memory cards and 1.4 times faster than recent SATA SSDs capable of up to 560 MB/s transfer rates.
Samsung says the 256GB SD Express microSD card will be available for purchase later this year, but did not provide pricing information at this time. The company did say it was developed in collaboration with a customer so we can expect a consumer device such as a smartphone or laptop with an integrated microSD Express reader around the same time.
Via Liliputing
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
One reason why this technology doesn’t evolve fast is very likely because SD cards are among the last rare components to still resist adopting a heat sink, and their connector doesn’t work really well to spread heat. So given that with frequency and/or data rate increases we’re continually increasing the per-chip power draw, it becomes difficult to put a multi-watts controller in a nail-sized chip like this.
It’s kind of amazing how badly it’s gone. I remember reading about SD Express and also SDUC (>2 TiB) back in 2018 and being excited about it.
Not that SD/microSD cards are alone in feeling the heat. We’ve also seen a push for 2580/25110 M.2 SSDs because of connector/form factor heat dissipation issues. Maybe we should be putting storage in PCIe slots instead.
Exactly this. The challenge is twofold: power and density. They’re related in several ways. Density also has the extra challenge of just packing the components into such a small package–the old 8051 type controller that USB and SD have been relying on for many years is going to need a bit of an enhancement to keep up with the PCI-E data rates (both in terms of signaling and in command processing). One can’t just take a normal NVME controller, nerf it a bit and stuff it into this package without grafting on the legacy SD interface–which involves decades of legacy code.
Then there’s the power issue. The limitation for SSDs has always been power dissipation. You see that in every new generation of M.2 device. It takes years for the power draw to come down (if it ever does) as controllers mature in process and in electrical design. Now, scale that down to a uSD card slot from that huge 22x80mm M.2 card (can’t get a proper area of the uSD, but it looks like at least an 11x size difference). And those things overheat without a proper heatsink at just 5W or more.
Making a card like this actually work is quite the engineering challenge and isn’t just a simple enhancement to existing designs. It’s a whole bunch of bespoke engineering. I’m really not surprised it took this long to see one. I stil haven’t seen a device that can utilize such a card. (I’d love to see an SBC that did.)
I’m curious how much work you’ll need on the controller side. Is there some kind of special SD to NVME handoff protocol necessary? Or is it just as ‘easy’ as adding an NVME intercace to the slot itself? It appears they at least use separate electrical contacts for the PCI-E link, but I wonder if any of the existing pins are repurposed when in that mode (requiring SD controller changes as well).
When reading page 6 of this 5 years old whitepaper from the software side of things this will appear just as a regular NVMe device and the HW interface uses both rows of pins but if I understood it correctly w/o the need for special controllers on the host side.
And BTW: an 1.8W limit for these cards is also mentioned as well as optional 1.2V powering with future versions.
> multi-watts controller
At the bottom of https://goughlui.com/2021/02/27/experiment-microsd-card-power-consumption-spi-performance/ there are consumption numbers for some cards while benchmarking them in UHS-I/SDR104 mode.
250mW for a SanDisk Ultra A1 IMO isn’t that bad but with PCIe it’s a different story of course…
Oh nice! Yes, actually what lies in a microSD card is a marvell of technology that end-users completely overlook, and which shows what the IT industry is capable of when facing some constraints. If the rest of the IT industry had to work with similar constraints, laptops would last one month on battery, smartphones would be half as heavy and would be recharged by day light using a solar panel on the back. But drawing watts continues to sell well and fan vendors are here to last.
[ Yes, wondering about all increasing data transfer speeds, but not providing S.M.A.R.T.-like (‘Self-Monitoring, Analysis, and Reporting Technology (S.M.A.R.T., often written as SMART)’) wear control and temperature sensor data to user space? That’s not logical&consistent for the ‘needs of the many’, but maybe ‘fascinating’ (and ‘highly illogical’ with reminisce to Mr. Spock :). ‘dif-tor heh smusma’ ]
If the card is accessed via PCIe then NVMe is being talked and SMART should work as long as the controller manufacturer implements health monitoring. And when accessed as mmc device there’s CMD56 which some manufacturers support (on some cards) but for which no standard format has been defined.
Looks like this with an unsupported card 😉
[ Yes, parsing ‘enduranceRemainLifePercent’ would be simple enough and very useful if file systems do not support read/write summaries, with high bandwidth IO (useful or not towards OS memory cards, but (AV media) sd cards have been announced&provided (1/2TB ~2015) 1TB sizes already since ~5yrs). Only thing missing then would be some (external?) temperature sensor close to sdExpress/PCIe connectors. (Thx) ]
> if file systems do not support read/write summaries
Unfortunately most what happens at the filesystem layer doesn’t mean that much at the flash layer, especially with settings that enforce high Write Amplification. 1 changed byte at the filesystem may result in a whole page being (re)written at the flash layer.
[ Yes, but at least there’s an estimation possible, what’s the amount of R/W io towards a flash storage without each consumer summing up this with guessing from average data bandwidth. Probably a low percentage of consumers know what their file systems block size would be (or if there’s caching partitions from flash or dram involved with SSDs) and if it’s suitable assuming a 512B or 4kB page size (What’s the difference between logical and physical sector size, wrt to write amplification?) for every formatted fs partition? One rough guess for OS (not archived data storage with big files) average r/w data transfer size could be ~2000-2500B each storage read or write command or page access (reasonable?)? ]
> ADATA also offers two models with 256GB or 512GB capacities at higher prices
And there’s also Delock’s 256/512 GB models claiming read/write speeds of 858/804 MB/s.
Im hoping this express sd standard would make pcie more popular around Socs, but seems noone really needs it so far…
Gotta love Samsung’s duality. They make one of the best microSD cards, but you cannot use them in a lot of their phones (and the competition). What I want to see is higher capacity microSD cards, not SSD grade cards. 256 GB might be enough for some, but I think overall lifespan and capacity should matter more than actual speed.
I bet someone at Samsung said “hey, we didn’t use this new standard that came up in 2018”. I highly doubt these can maintain the speed long enough to be worth their price.
Usually speed improves when it’s no longer possible to dump a whole device in one hour due to its capacity. Here it’s probably more or less the same driver for more speed.