Cerabyte posts video of ceramic storage system prototype: 10,000TB cartridges bombarded with laser rays,may become mainstream by 2030

[mariushm]

From : https://www.techradar.com/pro/video-of-ceramic-storage-system-prototype-surfaces-online-10000tb-cartridges-bombarded-with-laser-rays-could-become-mainstream-by-2030-making-slow-hard-drives-and-tapes-obsolete

 

 

Summary

 

 

Cerabyte has released a video showcasing the potential of its long-anticipated ceramics-based data storage system that promises to revolutionize how organizations store data on data centers in the future.

In contrast to data usually stored on the best hard drives and the best SSDs of today, Cerabyte wants to use ceramic material, combined with glass, to hold mountains of data. For instance, it wants to build palm-sized cartridges that can store 10,000TB of data. 

[...]

This demo system comprises a single read-write rack for storage accessibility as well as several library racks. The firm used only commercial off-the-shelf equipment to build it.

 

Each cartridge comprises a data carrier that uses a layer of glass, similar to Gorilla Glass by Corning, with a thin, dark ceramic layer as the data storage medium. These cartridges are kept in a robotic library. When data needs to be written, the cartridge is moved from the library rack to the read-write rack. The cartridge is opened and out comes the data carrier, which is positioned on a stage.

Data is written by two million laser beamlets that punch QR code-like nano-scale patterns into the surface of the media. The laser pulse is sharpened by a digital micromirror device, and shaped by microscope optics onto the surface of the data carrier. This process imprints holes – or no holes – onto the surface layer, which represents binary information. 

[...]

 

More in the article and video

 

My thoughts

 

Doesn't look like it's discs, so less of a risc of exploding ceramic discs... hopefully it becomes reality, unlike the Hyper CDROM one of my fellow countrymen invented but never made reality - https://en.wikipedia.org/wiki/Hyper_CD-ROM - or those quartz "5d" many layers optical discs that never became reality -  https://en.wikipedia.org/wiki/5D_optical_data_storage - 

 

[porina]

An interesting technology for sure, but "mainstream" sounds a long stretch for what looks like an enterprise level bulk storage system. I don't see anywhere saying it is RW so it is more for WORM archival uses. Use in consumer tier would be challenging due to the sizes involved. Dust would be a major problem and may demand a lot of ECC to get around.

 

Costs are questionable too. Following the links there is a claim by 2030 they can get it below $1/TB but capacities are 10x lower at claimed 2025 launch. For comparison, I think we're currently looking at ball park <$20/TB for HDs and <$50/TB for SSDs going by best price/capacity. And that is at consumer one-off pricing. I'm sure large corporates could get better rates.

[leadeater]

I'll take that bet, won't be generally available by 2030. Stuff like this always takes way longer.

[leadeater]

28 minutes ago, porina said:

I think we're currently looking at ball park <$20/TB for HDs and <$50/TB for SSDs going by best price/capacity. And that is at consumer one-off pricing. I'm sure large corporates could get better rates.

Our price for 7.2k RPM NL-SAS 12TB is $48/TB (includes 12 bay SAS enclosure) with HPE onsite support for 5 years. Larger capacities should be a little cheaper and SATA interface variant of same disk also a little lower. Enterprise HDDs aren't actually the cheapest, although a bulk order of desktop HDDs would if one were to do that but not sure why. Only Backblaze has been silly enough to use WD Green HDDs in the datacenter (they don't anymore).

 

Amazon Seagate EXOS 16TB SATA - $15.86/TB

[porina]

1 hour ago, leadeater said:

Enterprise HDDs aren't actually the cheapest

It depends on the criteria but I have a pair of "Enterprise" Toshiba 8TB drives in my NAS, along side another pair of their actual NAS offerings, as they were what's cheapest at the time I got them. So they're not necessarily that far off unless you have more demanding needs than just bulk capacity. Personally I aim for 7200rpm CMR as baseline.

 

1 hour ago, leadeater said:

Amazon Seagate EXOS 16TB SATA - $15.86/TB

Which is <$20/TB as I wrote. I got lazy. I looked at one UK seller and converted in my head. So not far off. SAS versions don't cost that much more (<+10% in UK).

[soccermomisqueen]

This kind of reminds be of Project Silica that some people from the Microsoft Research team have been working on for a few years. Here is an overview page from Microsoft's research site: https://www.microsoft.com/en-us/research/project/project-silica/overview/

The other part that excites me about these glass technologies is that if the hardware gets out and is in use, and somehow Microsoft/Cerebyte go out of business, the media for these data storage methods are essentially extremely refined silica glass (I realize that there is more to it than that, but for arguments sake). This could allow a new company, or even third parties to get involved in the making of this glass in a (somewhat) trivial level (cause there will definitely be money for it).

[Doobeedoo]

Impressive for new archival storage. Nah this ain't mainstream centered. Requires lasers and all and can't see it replacimg say USB drives like that. We'll see how it all goes. 

Speed is looking very good. Though for immense capacity it's not as fast growing for full drive pass vs something of today. But yes this is archival focused tape replacement. 

[05032-Mendicant-Bias]

This looks interesting, it's the spiritual successor of CD Roms and DVDs, with optical equipment to read write from cheap (?) mediums.

 

10PB on a palm sized cartridge sounds like it can have lots of use cases.

 

I'm not sold that this solution will become competitive with Flash storage for consumers, ever. All that optical equipment sounds expensive, requiring lots of precision machined parts and high quality lenses and lasers to work. I don't see that equipment being cheaper than a flash chip. Ever.

 

I can see this solution becoming cheaper per TB at the PB and EB scale on datacenter.

[leadeater]

2 hours ago, porina said:

It depends on the criteria but I have a pair of "Enterprise" Toshiba 8TB drives in my NAS, along side another pair of their actual NAS offerings, as they were what's cheapest at the time I got them. So they're not necessarily that far off unless you have more demanding needs than just bulk capacity. Personally I aim for 7200rpm CMR as baseline.

 

Which is <$20/TB as I wrote. I got lazy. I looked at one UK seller and converted in my head. So not far off. SAS versions don't cost that much more (<+10% in UK).

Yea I just mean when purchased for their wider actual usage they come with extra support which basically doubles the cost. You can buy them like you did but unless you are self sparing etc then you'll have support. HPE/Dell etc won't even sell you them without some level of support included pushing the price up and Dell EMC/NetApp etc won't sell to you at all period without support.

 

Support, software licensing etc etc kind of kill off price comparisons in enterprise storage.

 

Even then there is a wide range between "NAS rated" and "Enterprise" i.e. Seagate Ironwolf vs EXOS.

 

Any technology that will effect profits etc will just get  massaged away in other fees, customer never wins haha

[Needfuldoer]

And then we never heard about it again, just like all the other experimental "breakthrough" technologies in power efficiency and storage density.  

[bcredeur97]

This is great for archival! And it should be resilient to Solar Radiation (since it's physically etching into ceramic). 

But yeah being WORM kinda stinks. But I'm ok with that for archival.

[wanderingfool2]

4 hours ago, leadeater said:

I'll take that bet, won't be generally available by 2030. Stuff like this always takes way longer.

Surely not, I mean look we have Holographic Versatile discs becoming a thing...oh wait....how about InPhase Technology that surely...oh wait.

 

I agree with what you said, but I think I would personally go further; I'd bet this won't be available or maybe in like 20 years.

 

 

 

It always bothers me when companies like this come out with fuzzy information "GB/s" and things like that without any context...to me it seems like they don't actually know, because maybe they haven't done the testing yet (but is theoretical)...or does it have the general issue like tapes where data stored on tape is generally slow for initial access.

 

With that said, I see this technology as being problematic in terms of where it fits in (or rather how many clients would want it).  While I do think the purpose could change based on what it's actually capable of; I'd be worried at the moment that the only "benefit" over something like tape storage is the overall cost.

 

The way I look at it, there are many cases where you don't want your data stored forever.  Financials, you would probably only want to keep things until it's statute barred otherwise if something like the IRS does an investigation and then requests the older data you would be capable of complying.  Long-term storage of client data is a problem in regards to the right to be forgotten in the EU.  Some of them require a specification on the length the backups will be kept to the user.  I'd imagine that it might run afoul if you said the backups remain forever.  Then if you do need to destroy the data it's pretty easy with a tape...you simply put the thing in a degauser or the many ways of destroying the data on there.

 

 

If they are able to somehow access data with minimal latency penalty though, then I could see it being used as some types of long-term storage (like lets say Netflix and video content).  Your primary servers have lets say the first 10-20 seconds of each video already buffered then when the user clicks on it it quickly reads the rest.

 

[Kisai]

Optical storage has always been "the cheapest" next magnetic storage. Optical has only ever worked as a WORM (Write Once, Read Many) solution. CD RW and DVD RW discs exist, but almost nobody uses them that way because it's cheaper to just just burn a new disc. Why pay $20 for a RW disc when you could pay 0.20 for write once one.

 

That's probably what's going to happen here, if it isn't vaporware.

You'll be able to get a 10PB ceramic storage for probably about $100, at some point, but magnetic storage will still be used in servers, and SSD's will be used in portables/laptops for power usage reasons and thin-ness.

 

I'm also a bit worried about the fact it's optical at all, because CD/DVD/Blueray has always had a moving parts problem and this sounds like a moving parts problem at a precision level that possibly stomping your foot might kill the cartridge. We will have to see. 

[porina]

3 hours ago, leadeater said:

Yea I just mean when purchased for their wider actual usage they come with extra support which basically doubles the cost.

I think you're over thinking it. Company made claims on future cost/TB. My brain says, what is current cost/TB? I look at a list of big-ish HD and estimate the value point. Repeat for SSDs. That's it.

 

Support contracts I had experience with in the past on specialised test equipment. Initial hardware purchase price was nothing compared to the ongoing software support costs. Maybe if/when this new storage goes into volume production, they'll also sell the system with a fat support contact. But there'll still be a headline media cost.

[leadeater]

2 hours ago, porina said:

I think you're over thinking it. Company made claims on future cost/TB. My brain says, what is current cost/TB? I look at a list of big-ish HD and estimate the value point. Repeat for SSDs. That's it.

 

Support contracts I had experience with in the past on specialised test equipment. Initial hardware purchase price was nothing compared to the ongoing software support costs. Maybe if/when this new storage goes into volume production, they'll also sell the system with a fat support contact. But there'll still be a headline media cost.

Yep but the issue is that these costs saving on the media side doesn't mean it'll get passed on to the customer. This definitely isn't consumer technology so it's a relevant aspect. It's the same reason why almost everything has moved to per core licensing and per TB. As we made hardware technology cheaper and started purchasing less all the vendors changed how they charge for their products and that's made many things unaffordable which is a huge issue. Products companies could afford before have actually become unaffordable after licensing changes and the move to consumption/capacity licensing.

 

It wasn't that long ago you didn't have to pay per TB to use the physical storage in your fancy storage appliances, now that's the exception to the normal.

 

That's why for something like this the $/TB raw figure isn't that useful, nobody is going to be paying that to actually use it.

[Forbidden Wafer]

This vs microfilm storage

[StDragon]

12 hours ago, leadeater said:

I'll take that bet, won't be generally available by 2030. Stuff like this always takes way longer.

Probably way sooner. There's deep pockets for this level of 24/7 video surveillance archival...

https://www.nature.com/articles/d41586-023-03745-5

AI changes everything, and will quickly accelerate R&D to market.

[williamcll]

What's the write speed for these?

[soccermomisqueen]

38 minutes ago, williamcll said:

What's the write speed for these?

There is a good chance that it is atrociously slow (much like a DVD or CD) since you are physically changing the structure of silica glass. That is not to say that it will be as slow as a DVD or CD (if it was, it would probably never take off in any meaningful way because a petabyte could take months), but for long-term WORM archival of data, speed does not necessarily matter, but rather reliability matters more.

[wanderingfool2]

6 hours ago, StDragon said:

Probably way sooner. There's deep pockets for this level of 24/7 video surveillance archival...

https://www.nature.com/articles/d41586-023-03745-5

AI changes everything, and will quickly accelerate R&D to market.

AI really wouldn't be advancing this type of technology really...perhaps in the future miniaturizing it with new discoveries but that kind of thing is still years out.

 

The biggest hurdle from getting from R&D to market in cases where it's coming in to supplant is manufacturing; and specifically either cost or making things in volume (although those often go hand in hand).

 

They claim though that they are essentially using already existing products essentially to do this though...at which point it really is, have they fine tuned it enough to show that it does what they claim it does and does it cost less than the competitors (or offer a benefit that additional costs can be ignored).

 

Their claim will be that the benefit of it never degrading and not costing in terms of power after it's written will reduce overall costs and be a stable medium.

 

13 hours ago, Kisai said:

You'll be able to get a 10PB ceramic storage for probably about $100, at some point, but magnetic storage will still be used in servers, and SSD's will be used in portables/laptops for power usage reasons and thin-ness.

Honestly though, if it ever hit that kind of pricing it would probably create disruptions as at that stage it might be worth it for backups for smaller businesses.  Especially if you consider it would be resistant to ransomware types of attacks, since it can't really write overtop of/delete the old data.

 

9 hours ago, leadeater said:

Yep but the issue is that these costs saving on the media side doesn't mean it'll get passed on to the customer. This definitely isn't consumer technology so it's a relevant aspect. It's the same reason why almost everything has moved to per core licensing and per TB. As we made hardware technology cheaper and started purchasing less all the vendors changed how they charge for their products and that's made many things unaffordable which is a huge issue. Products companies could afford before have actually become unaffordable after licensing changes and the move to consumption/capacity licensing.

Although, thinking about this and reading more into it; I think the target is more of the massive datacenter ones...so it might be down to $/TB and energy costs; as places like Google has the pull to be able to just buy the drives without requiring necessarily a vendor support contract. 

 

The thing is this realistically is going against either LTO or if they are able to access the data in a reasonable time it's competing against cold storage.  In the case of cold storage, I'd argue that $/TB even if you could buy drives for sub $20/TB; one has to factor in needing parity, backups and replacement costs.  You could probably speak to this better @leadeater, but I know in a 5 year period I had to replace 2 of 14 SAS drives in my SAN (and like 3 of them were showing signs of issues).  Not sure really about drives in cold storage though, as I haven't done research into that kind of thing before.  So the way I would look at it, a $/TB/year metric would be more appropriate anyways.  Either way, the cost I don't think would be below the $20/TB @porina not when you factor in replacement costs (even things like the added heat so you need to run more HVAC units)

 

Actually side note, how many rewrites can LTO's do before getting degradation

[Mark Kaine]

19 hours ago, Doobeedoo said:

Impressive for new archival storage. Nah this ain't mainstream centered. Requires lasers

you know that "lazers" are ancient tech, right? 

i dunno, sure it won't replace usb sticks anytime soon ,but that's not what its designed for either,  this seems just an evolution of traditional storage devices (aka hard-drives) 

im getting MOD flashbacks here, curious about data retention... MOD disks were nigh indestructible ...

[porina]

13 hours ago, leadeater said:

That's why for something like this the $/TB raw figure isn't that useful, nobody is going to be paying that to actually use it.

3 hours ago, wanderingfool2 said:

Either way, the cost I don't think would be below the $20/TB @porina not when you factor in replacement costs (even things like the added heat so you need to run more HVAC units)

While I agree the effective cost will be something else, I don't think this changes my look at media cost. I get it, there will be costs beyond media, but that applies now too. The scaling between media cost and shall we call it full operating cost may not be known for a while, but media cost claim is all we have for now.

 

Given the context this new tech might go more into bulk archival storage, maybe looking at HDs and SSDs was off on my part in the first place. What media is used for that? Tape? What's the $/TB of that?

[Doobeedoo]

2 hours ago, Mark Kaine said:

you know that "lazers" are ancient tech, right? 

i dunno, sure it won't replace usb sticks anytime soon ,but that's not what its designed for either,  this seems just an evolution of traditional storage devices (aka hard-drives) 

im getting MOD flashbacks here, curious about data retention... MOD disks were nigh indestructible ...

I know, but don't see this tech in consumer space, maybe similar type in future when computing or just memory is more optical. This is not an SSD replacement what I wanted to say. But at

current long term archival like tape. For HDDs they're kinda almost dead for consumers really, aside for some needing dozens or hundreds of TBs for personal use and don't need speed. But general buyers just use SSD though.

[tikker]

Yes please. If it is somewhat affordable this could solve a lot of data storage issues in my field.

[wanderingfool2]

9 hours ago, Mark Kaine said:

you know that "lazers" are ancient tech, right? 

i dunno, sure it won't replace usb sticks anytime soon ,but that's not what its designed for either,  this seems just an evolution of traditional storage devices (aka hard-drives) 

im getting MOD flashbacks here, curious about data retention... MOD disks were nigh indestructible ...

Higher powered lasers have regulations surrounding them; and I'd guess this probably has to have more power given that instead of doing a single point they are writing data in a 2d space all at once by the sounds of it.

 

 

 

My guess on how this is working (or rather the parts already on the market).

 

Femtosecond laser (lots of places manufacture this)

DLP projector mirror for laser control (Technology that's essentially from projectors)

And high resolution cameras.

 

The biggest thing is how they move around the ceramic disc to write more data on it...after all, the way I mentioned above if they used a commercially available mirror would have 4k resolution (there might be 8k ones out there, but lets just say 4k).  At a 4k kind of resolution that means they could blast ~7.9 megabits of information at once.  Although they are using QR so there is actually error correction built in so there is actually less than 1MiB of actual usable data.

 

At this stage then it's just about scaling.  If they are able to get lets say an 8k chip, that would be ~4MiB per pulse...but if there is a 16k mirror that exists on the market though that would make it ~16 MiB per pulse.  My guess on what they would do is move the ceramic linearly along at a steady speed; so that they can pulse the data onto it (where you get many multiple pulses per second)  Now a DLP is capable of moving at 1440hz at least apparently.  So if they could maximize that; it puts it at ~5GB/s (including all the overhead stuff though)