Samsung Starts Mass Producing New 512GB NVMe SSD That's Smaller Than a Stamp (pcworld.com)

An anonymous reader writes from a report via PCWorld: Samsung announced late Monday night that it has begun mass producing a new SSD that is tinier than a postage stamp. PCWorld reports: "The PM971-NVMe fits up to 512GB of NAND flash, a controller, and RAM into a single BGA chip measuring 20mm x 16mm x 1.5mm and weighing just one gram, the company said. Samsung says the PM971-NVMe will hit 1.5GBps read speeds and 800MBps write speeds. The PM971-NVMe is built using 20nm NAND chips and includes 4GB of LPDDR4 RAM as a cache. The NAND is triple-level cell but uses a portion as a write butter. The drive will come in 512GB, 256GB and 128GB capacities." While on the topic of hardware, Intel unveiled its Broadwell-E family, which consists of an "Extreme Edition" Core i7 chipset that has 10 cores and 20 threads.

Comments

Stamp-sized SSDs...

[__aaclcg7560]

If the new stamp-sized SSDs are priced like stamps, I'll take a whole book.

Re: Stamp-sized SSDs...

[Anonymous Coward]

They are priced like extremely rare collector stamps.

Re:Stamp-sized SSDs...

[Sir Realist]

Stamp-sized SSDs; that's awesome!

What's a stamp?

Re: Stamp-sized SSDs...

[Applehu Akbar]

They are priced like extremely rare collector stamps.

EXCEPT that the ones with errors do not sell for a premium.

Re:Stamp-sized SSDs...

[jhonkelly]

I agree with you Vinhomes Bc Ninh Rau Dip Cá

Re:Stamp-sized SSDs...

[U2xhc2hkb3QgU3Vja3M]

It's the five Dogecoins you pay when you send an email.

Re:Stamp-sized SSDs...

[CeasedCaring]

It's a small microcontroller

Re:Stamp-sized SSDs...

[Anonymous Coward]

That's a thingy you lick its back side.

Re:Stamp-sized SSDs...

[__aaclcg7560]

What's a stamp?

It's what you put in your grandmother's green stamps book to exchange for a new living room lamp at S&H.

https://en.wikipedia.org/wiki/S%26H_Green_Stamps

stamp

[Anonymous Coward]

but what size stamp? like a postage stamp, or like a FRAGILE red ink stamp?

Re:stamp

[Rockoon]

like a license plate stamp.

The NAND isn't 20nm

[Anonymous Coward]

The LPDDR4 is 20nm. The NAND is their 48 layer vnand per the linked article. Pretty impressive amount of packaging/die stacking going on there.

Re: The NAND isn't 20nm

[Anonymous Coward]

What will this means for prices?

Can one foot into SD card foot print?

Re: The NAND isn't 20nm

[FatdogHaiku]

I just pulled out my calipers and checked, it should fit in an SD card as far as size goes. Will they write a driver for that? One would hope...

Re: The NAND isn't 20nm

[wierd_w]

You can already get 256gb SDcards.

What I am more interested in, is why isnt there a more user friendly SSD socket type out there akin to SDCard.

Your typical SDCard socket has ample interconnect leads and good enough conductivity to be driven similarly to a SATA device, and could even be seen as a SATA device. If this device is that small, the market for feild upgradable SSDs on ultrathin or ultracompact devices at SATA speeds is a deal changer. It makes me wonder why nobody has done it yet.

I could see constraints on SDCard based designs, as the addressing and write optimizing methods arent really tailored to full speed SATA operation, but an actual SSD design in that size range is another creature all together. I brand new SDCard type with no back compatibility driving at full SATA600 speeds would be phenomenal. They would outright dominate the digital camera market, and would enable a whole host of ultra-thin devices to get reasonable upgradable storage.

I say this because at that size, the 6 or so pins on a SATA connector could be upgraded to 8, (for DC power), the whole thing COULD be SDCard sized, AND bitching fast.

Re: The NAND isn't 20nm

[Gr8Apes]

Because NVMe drives are meant to go about 10 times the speed of SATA III. That might be a slight strain on the interconnect.

Re:The NAND isn't 20nm

[U2xhc2hkb3QgU3Vja3M]

Fourty-eight layers? Damn. So the NAND is like.... onions? /Shrek

Re: The NAND isn't 20nm

[ITRambo]

SD cards have processors in them, which affects performance. SSD's can be plugged into a USB 3.0 dock and get about 180-MB/second data transfer, as measured by HD Tune. USB 3.1 docks will be double that speed.

Re: The NAND isn't 20nm

[wierd_w]

A drive going too fast for the interface is a tolerable problem that can be solved later.

A drive that is needlessly slow on the other hand....

Again, the idea here is for a friendly socket, like SDCard, into which such a tiny SSD can be inserted easily. SATA is just a nice industry standard interface with existing drivers that can be leveraged. You see this frequently with M.2 based SSDs. They present themselves as really fast SATA devices on their own dedicated controller. The idea here would be similar.

If you are concerned that the media is too fast for SATA, then give it a whole new interface to use, but be aware that adoption will be tricky.

Regardless, SATAIII speeds are lightyears faster than existing SDXC cards, even in the extended class 3 and 4 varieties. The digital camera market would eat it up like candy.

Offering it as an option on a tablet makes an otherwise unapetizing toy into an interesting prospect for purchase.

etc.

Re: The NAND isn't 20nm

[wierd_w]

The idea here is not to drive an sdcard that fast (that wont work, because that is now how sdcards are designed to work.).

The idea is to create an ssd socket that looks and feels like an sdcard type form factor for practical handling, transport, and installation purposes, but which is actually its own thing inside, capable of SSD type speeds over a very short serial interface. (the interconnect will measure in centimeters! Noise from high speed transmission will be minimal.)

By having a device like that, in such a slim package, many data aggressive devices can be improved-- photographers are the ones driving sdcard speeds. If they could jam an M.2 SSD into a camera, they would do it. M.2 is not meant for hot swapping, but there IS a class of sata III that *IS*. Look it up.

Re: The NAND isn't 20nm

[Trongy]

There are already XQD cards that seem to meet your criteria. Nikon has the interface on their higher end carmeras. The are bigger than SD cards, but more robust looking and still smaller than the old CF cards.

XQD uses PCI express to transfer data so the interface shouln't be a bootleneck at the moment.

My understanding is that SD had a controller on board to be easily imeplemented on portable devices which didn't have a PCIe bus.

Re: The NAND isn't 20nm

[Anonymous Coward]

The main reason is that SD cards is supposed to be used cameras and other portable devices. Updating the SD connector to be yet another SATA variant would mean that a new slower card standard would have to be developed to work with portable devices.
Accessing SD-cards doesn't require external driver circuits and balanced signals. Some ARM controllers have support for connecting the SD directly to the CPU and if it doesn't support that you can still connect it on a simple SPI interface.

There already are memory cards that you can stick directly into the SATA port as you suggest, but since people mainly use memory cards to talk with devices that doesn't support SATA it never took off.
It is not a new idea, it just didn't work that well.

Re: The NAND isn't 20nm

[jon3k]

USB 3.1 will get you over 10Gb/s (>1GB/s) and then there's Thunderbolt 3 as well (40Gb/s, little b) for external enclosures. What is your use case, specifically?

Re: The NAND isn't 20nm

[SuricouRaven]

SD cards actually speak four different interfaces, depending on application. You can run them off an SPI interface if you want to, though it's really very slow. It's still a useful trick because a lot of microcontrollers have SPI hardware support (including atmega/arduino) and others can bit-bang it, so you can interface a microcontroller to an SD card with next to no supporting hardware.

The other three modes are all high-speed modes, but require more elaborate electronics.

Re: The NAND isn't 20nm

[Agripa]

It probably does not matter for the application because people are not aware of the problem and manufacturers do not advertise it but I do not think power loss protection during writes can be implemented in such a small format unless it is added outside of the drive which would require more to the interface specification.

I cant believe its not write butter.

[Anonymous Coward]

mmmmmm, write butter....

Write butter?

[rholtzjr]

Mmmm, might want to combine that with the read toast to compliment the write butter.

Re:Write butter?

[Anonymous Coward]

complement

Re: Write butter?

[Anonymous Coward]

Bitty blitter bot sum butter
"and" she said "this butter's bit-or;
if I put it in my buffer
then my buffer will b@@*&@$_@

Re:Write butter?

[sacrilicious]

Totally cracked me up, thank you

"20mm x 16mm x 1.5mm and weighing just 1 g"

[Nutria]

It would cost a small fortune, but you could easily fit 50TB or more of data in a 1" x 3.5" HDD form factor.

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[NotAPK]

Heat?

They would probably burn pretty hot, but yes, the possibilities are rather startling. It will be an interesting day when SSDs overtake HDDs on practical metrics such as data density, and even more so on price.

Any predictions on when that is going to happen?

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[Nutria]

Heat?

Good point. Could you stack two layers separated by head sinks? You'd obviously need vent holes, though.

It would give 25TB capacity: 512GB/stamp at 5 stamps wide, 5 stamps deep (need some room for control & power circuits) and two layers high.

when SSDs overtake HDDs on practical metrics such as data density

I think that day has arrived.

and even more so on price.

5 years?

Re: "20mm x 16mm x 1.5mm and weighing just 1 g"

[Anonymous Coward]

The density thing happened a long time ago. 3.84TB SFF SSD have been on the market for about a year, with 7.68 due out very soon.

Re: "20mm x 16mm x 1.5mm and weighing just 1 g"

[NotAPK]

Guys, I was asking after "practical density". Of course the chip density is insane, just look at a 256 or 512GB micro SD card!!!!

There must be some practical issues getting in the way, whether it's heat, interconnects, or just plain dumb cost. However, what I was pondering, was when will we be able to buy a 10TB 3.5" form factor SSD? Just as we can for spinning rust.

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[Kjella]

They would probably burn pretty hot, but yes, the possibilities are rather startling. It will be an interesting day when SSDs overtake HDDs on practical metrics such as data density, and even more so on price. Any predictions on when that is going to happen?

There's already 200GB MicroSD cards and 16TB 2.5" SSDs for sale, so I'm pretty sure density is a long won match. Price/GB is another matter, but HDDs don't scale down. I just checked and if you only need 128 GB of space, you pay the same for a HDD as for a SSD. Sure, the HDD will be 500GB but if you don't need it because cloud, streaming etc. you can't save more by buying a smaller HDD. And if you don't want one for a boot drive, you have the budget for a 256GB SSD before SSD+HDD becomes cheaper. I really don't mind HDDs for bulk media, we play movies from discs or stream them over internet connections that are much slower than that.

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[Anonymous Coward]

Heat?

that's what the butter's for. it's a heat sink AND a spread for toast.

Re: "20mm x 16mm x 1.5mm and weighing just 1 g"

[Anonymous Coward]

What about 15TB in 2.5" form factor?
http://arstechnica.com/informa...

Re: "20mm x 16mm x 1.5mm and weighing just 1 g"

[NotAPK]

Holy shit, I stand 100% corrected! Thanks!

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[Anonymous Coward]

Heat?

Wouldn't that be mostly proportional to the data transfer speed? So if you can survive with 1GB/s it'd be around the same as for a single chip. At that pace, filling the 50TB takes almost a full day though.

Of course there is some overhead involved, especially if you just dumbly duplicate all the maintenance logic in every chip, but I'm assuming reasonable design.

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[Anonymous Coward]

Power up time for an SSD is pretty low. You can probably unpower chips not used and still beat the crap out of spinning disk search times.

Re: "20mm x 16mm x 1.5mm and weighing just 1 g"

[Anonymous Coward]

Yes the day has arrived. I recently bought 4 500gb MLC ssd's for $140 each. That's cheaper than 10k rpm media was 3 years ago. That means hdd's are now truely obsolete for fast storage. Though they still can't be touched as far as slow/archival type storage.

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[jon3k]

Samsung already made a 15TB SSD so the demand is presumably there, at least in the enterprise space. Really makes you wonder how much longer spinning disks will be around. 10 years? 20? If Moore's law keeps up, in 20 years we'd have another 13 iterations of Moore's law (lets pretends its not dead now, or by then).

You can get a 1TB SSD for $255 (or less) today from several real name brand manufacturers. That means after 13 iterations of doubling transistor density, we'd have 8,192TB SSD (someone double check my math please). I'm sure there are lots of other issues, like how can we really have that many die shrinks and heat and a million other factors, but I seriously doubt we'll see HDD competing for anything near 20 more years.

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[SuricouRaven]

Moore's law cannot continue for another thirteen iterations for a very simple practical reason: It's difficult to make parts smaller than one atom. Atomic spacing for silicon is 0.5nM - that's the absolute, all-you're-getting limit even if every engineering problem is solved.

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[SuricouRaven]

nm, not nM. Hard to remember all these units!

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[epine]

Am I have a bad morning? This thread is making me want to scream. "Moore's law", "5 years?" Anyone else want to pile in with a tired fatuity? Surf's up, apparently, for a tiny value of "surf".

The following video (from January 2016) contains just about everything worth knowing at this point about Intel's forthcoming phantom memory. Don't even try reading anything else unless you got a bone for chalcogenide chemistry.

Rick Coulson of Intel on 3D XPoint and NVMe

Executive summary: Charge-storage memory is at the end of its rope. There's no longer enough electrons per cell to make the cells reliable. According to Coulson (he passed my bullshit detector with nary a glitch), the typical 3D SSD is correcting about 50 bit-errors per read using advanced error correction, or the devices would not work at all. DRAM scaling is possibly worse.

Resistive memory does not depend upon charge storage, and has completely different physical scaling. Even resistive memories must stack almost out of the starting gate to achieve the scaling velocity that conventional fatuity demands. But the physics and economics of lithography are such that each additional layer suffers from diminishing returns. In fact, the cost of lithography scales as the total amount of surface area patterned, with stacked patterns being ever more expensive the higher you go. There's a sweet spot where the reduction in wafer cost through the use of less wafer pays back the more expensive lithographic process in stacking layers.

It wouldn't surprise me that we're already past that point at 32 layers in SSD, and that what pays off for going even this far are the advantages in density and integration of the surrounding products. (Hence the chips cost more than they could have, but the end product costs less, or justifies a higher per-GB selling price due to some rack-density jizz premium.)

Hard drives have traditionally had the huge advantage of not being a patterned media, hence you get a lot more surface area at a far lower cost. This, too, will change in the transition to bit-patterned media. But this media involves an extremely regular pattern with no fancy layers.

Magnetic Bit Patterned Media Fabrication Using Block Copolymer Directed Assembly

When we finally invent copolymer directed assembly of chalcogenide matrices, then perhaps we can finally dance on the grave of spinning rust.

It pretty much has to be Facebook and Twitter making people so stupid these days. People get so dialed into the present moment where the only acceptable latency is none at all as to completely forget that the past is a large object. Just look at the archival data generated by gene sequencing, EOS, deep sky surveys, CERN, and gravity-wave telescopes and then ask yourself whether an 8 ms cold latency over raw data is too big for you.

With ZFS, 1500 TB is around the present practical limit for one pool (200 drives at 8 TB each, per the Oracle Cloud guy, early 2016) and it's still just 8 ms sector latency (depending on what else you're standing behind). How about you sit there and watch while some lithographic robot patterns out 1500 TB of charge storage cells? Ring me when it's done. I've got more pressing things to do.

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[slashrio]

...head sinks...

You mean your head stinks?

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[slashrio]

Don't worry, 'they' will find something.
Like multi-bit storage in programmable quantum dynamic wave function overlap on atom scale or so...

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[Nutria]

Yes. Sorry.

Re:"20mm x 16mm x 1.5mm and weighing just 1 g"

[Sir Holo]

Heat?

They would probably burn pretty hot, but yes, the possibilities are rather startling. It will be an interesting day when SSDs overtake HDDs on practical metrics such as data density, and even more so on price.

Any predictions on when that is going to happen?

It has. DARPA. At all length scales, save for the smallest. I apologize for spawning the "thermal via" frenzy of the past few years. Sorry.

Chipset?

[johnsmithperson123]

No new Intel chipset today. A ten-core CPU, though. Actually we've been moving towards smaller, single "chipsets".

Re:Chipset?

[JackAxe]

I'm personally happy Intel didn't go with a new chipset, since my board is a X99. This makes upgrading in the future more affordable.

Intel?

[Anonymous Coward]

What does this have to do with intel? Is it because of the pointless comment at the end of the summary, which has nothing to do with the article, and has already been posted separately just a few articles down? Is someone just a big Intel fanboy here?

Re:Intel?

[Anonymous Coward]

The editor is an idiot. In the front page alone he's done this to two other articles.

Oh Great

[Anonymous Coward]

It'll only take 15 years to finally put one of these in a smartphone.

But will it support ads?

[manu144x]

Will the firmware randomly create ads?

Re:But will it support ads?

[Anonymous Coward]

It will do that only after the device has been obsoleted by next model. With Samsung the firmware updates may also brick the SSD and the company denies of any responsibility if that happens, so it is unlikely the device will survive into the ad creation era.

Comment removed

[account_deleted]

Comment removed based on user account deletion

Mobile?

[Anonymous Coward]

How long before I can get this in my phone? 128gb is starting to feel cramped!

-1 Offtopic for BeauHD.

[dinfinity]

While on the topic of hardware, Intel unveiled its Broadwell-E family, which consists of an "Extreme Edition" Core i7 chipset that has 10 cores and 20 threads.

Stop. This. Stupid. Shit. Now.

Re:-1 Offtopic for BeauHD.

[Qzukk]

In sort-of related news, the new Windows 10 update will allow you to have big, long filenames on your itty-bitty stamp-sized SSD.

Re:-1 Offtopic for BeauHD.

[U2xhc2hkb3QgU3Vja3M]

In not-quite-but-still-sort-of related news, "the iPod has no wireless and has less space than a Nomad. Lame."

 

Re:-1 Offtopic for BeauHD.

[SuricouRaven]

Presumably so that anyone who hasn't upgraded to Windows 10 will have a good reason to do so if they don't want their software crashing upon encountering long filenames.

Re:What's a stamp?

[Anonymous Coward]

The thing you do with your foot when you wanna make a point?

A kick in the face?

bad product.

[Anonymous Coward]

20 threads is strange limitation. Do not buy this!

Re:bad product.

[hcs_$reboot]

20 threads is the number of threads handled by the CPU itself, quite a lot! The OS handles its own threads and processes (which are in the hundreds/thousands usually).

Re:bad product.

[Blaskowicz]

It actually runs a special version of MS-DOS which can run up to 20 programs concurrently (up to 20 users).
In that mode though, each program is limited to 64K memory. One trick is to reserve a 64K ramdisk for several programs to communicate with although you can run 19 programs then. Failing that, just plug null-modem cables between some of the COM ports on the back of the PC and call it done..

Re:bad product.

[Anonymous Coward]

You forgot the "FILES=" setting you need to add to CONFIG.SYS. I'd suggest FILES=200.

Re:bad product.

[U2xhc2hkb3QgU3Vja3M]

Why the fuck do I still know that shit in 2027?

Re:bad product.

[U2xhc2hkb3QgU3Vja3M]

Oups, I meant 2016!

2016!!

Re:bad product.

[hcs_$reboot]

Oh that's why the CPU has an integrated floppy, that makes sense!

4 GigaBits of RAM

[Anonymous Coward]

Capitalization is important. Standard architecture for a SSD. I suspect it needs a heatsink.

Younger readers may not understand this

[InterGuru]

Zits' Jeremy "What's a stamp?"

http://zitscomics.com/comics/s...

Smaller? Not quite

[Anonymous Coward]

1.5mm sounds as high as at least 10 stamps glued on top of each other.

512GB=4096Gb?

[Anonymous Coward]

Is this a real, true-to-God 4 Tera-bit SSD? Or is it just a 64 Giga-BYTE SSD?