Toshiba Unveils First FC-MAMR HDD: 18 TB, Helium Filled | Hacker News Reader

Toshiba Unveils First FC-MAMR HDD: 18 TB, Helium Filled

anandtech.com

112 pointsrbanffy83 comments

For those who haven't been following HDD closely.

Energy Assisted Magnetic Recording, EAMR is a term that sums up both MAMR, Microwave Assisted Magnetic Recording and HAMR, Heat Assisted Magnetic Recording.

And this isn't the first 18TB HDD, there are 20TB HDD already. Nor the first MAMR drive. Western Digital Ultrastar DC HC550 [1] claims that title.

MAMR will bring us some incremental gains, both WD and Seagate believes HAMR is the right path in the long term. Seagate shipped [2] their first HAMR HDD only last month and my bet is that it will take another year before it trickles down to consumer market. As of the last investor conference both Seagate and WD are still aiming at 50TB HAMR HDD in 2026.

Seagate is also working on Dual-actuator Mach.2 technology, which will finally put the HDD transfer speed close to the maximum SATA 3 6Gbps speed. Two of these in RAID on a NAS we can finally saturate the 10Gbps Ethernet.

Seagate also expect to use HAMR across most range of its product to amortised the enormous R&D and component expense with HAMR. After all we have been talking about HAMR for nearly a decade. Which suggest while we might be getting bigger drive in the near future. Price / GB will likely stay flat for another few years. ( Of course that is also dependent on how fast NAND could drive its cost reduction. )

[1] https://www.westerndigital.com/products/data-center-platform...

[2] https://www.tomshardware.com/news/seagate-ships-hamr-hdds-in...

Is transistor density somehow holding back magnetic storage density too or something else?

Magnetic storage density is held back by how small you can make a bit (magnetic domain) and still be able to write it AND not have it succumb to random flips due to thermal energy. EAMR works by temporarily making tiny magnetic domains of a hard to flip material easier to flip while under the influence of additional energy.

No, AFAIK the limit is caused by various tradeoffs between the media material and read/write head design (the Magnetic Recording Trilemma).

http://www.fgarciasanchez.es/thesisfelipe/node5.html

This seems to have come out of no where.

Helium is an old trick, but MAMR / HAMR (energy assist technology, either microwave or heat) was a big announcement from Seagate and WD.

I didn't know Toshiba was also working on the tech.

HAMR/MAMR have been well understood and just on the road to commercial viability since at least 2014 (probably long before, but that is when I came across them). Toshiba being such a small percentage of the disk market makes their later deployment of this stuff easy to understand simply from the standpoint of having a smaller research budget.

A subsequent technology to expect is the separation of magnetic domains into small islands on a non-magnetic substrate, rather than the current approach of applying the magnetic field to regions of a continuous all-magnetic medium. This would be paired with heating, in what I only remember being described as "heated dot magnetic recording".

That was where the technical roadmap for HDDs ended last time I checked. After that I guess platter additions will carry us to the end of the decade.

I remember hearing somewhere that helium will slowly leak through whatever vessel you try to contain it with. I wonder if that (is true, and if so whether it) imposes some kind of maximum life on these drives. Or maybe this is only hydrogen that is this problematic.

> I wonder if that (is true, and if so whether it) imposes some kind of maximum life on these drives.

It does. Most helium-filled drives expose a SMART attribute (ID 22) for how much helium is remaining.

No expert, but it seems to make sense that helium is particularly hard to contain based on molecular diameter:

  Hydrogen 2.75 Angstroms 
  Helium 2.18
  Argon 3.67
  Oxygen 3.64
  Nitrogen 3.64
  “Air” 3.74

Yet it takes it weeks or months to leak out (I suppose it'd be more accurate to say 'slip between' the other molecules) of an extremely thin mylar balloon. I am sure it's an engineering challenge to keep it inside a small metal box, but it seems possible, at least. After all, the reserves we have have just been stuck in the ground, without managing to leak into the atmosphere.

Fascinating - I had expected helium to be a larger molecule than hydrogen.

The given value for hydrogen is for hydrogen gas (i.e. H₂), while helium is just a single atom.

That's what you would expect, though. Helium gas is "He1", hydrogen gas is "H2".

I think that was their point: hydrogen gas is diatomic, so it makes intuitive sense for it to be larger than helium

There's a detailed explanation here: https://chemistry.stackexchange.com/questions/115363/is-a-hy...

Edit: Oops. Still relevant since it's H2 in play here, but yes, not hyrdrogen vs helium.

I think that explanation is only about Hydrogen molecules and hydrogen atoms, not Hydrogen vs. Helium.

There's a free textbook available here [1] that explains that atomic sizes decrease across the periodic table and increase down. In short, it decreases across a horizontal period because more electrons are in the same orbital shell (really a cloud), increasing the negative nuclear charge of the cloud, which is more attractive to the positive atomic nucleus, so the whole cloud is squeezed together, making the whole atom smaller.

[1]: https://chem.libretexts.org/Bookshelves/General_Chemistry/Bo...

This seems unintuitive as I would expect one electron by itself to be more closely attracted to a positive nucleus instead of when there are two or more fighting to remain in "orbit".

Your definition would make sense to me if you can think of electrons adding to an over all single electrical charge instead of individual electrons, but in that case, I have to ask, what is the original reason a single electron is repulsed by the atom. I.E. why does it stay so far away until more electrons are added. What is the force keeping electrons from falling into the nucleus in the first place and have an "orbital" or "cloud" at its set distance (like one Angstrom or Bohr)?

My instinctive guess is that it has to do with the spins of two electrons "balancing" each other keeping them closer to the nucleus.

My understanding is that spin leads to a slightly different electrical force between up and down spinning electrons due to weak force symmetry breaking or singing to that effect. It's more like they have tiny electrical differences rather than canceling out any charge unless I'm mistaken.

I'm getting this from memory of my high-school chemistry classes, so be careful.

Back then the explanation was that since the He nucleus has twice the positive charge as the Hydrogen one, it'd bring the electrons a bit closer. Both atoms have a single S orbital, He with two electrons.

Also, atomic Hydrogen won't remain atomic for long - it'll quickly connect to another Hydrogen so it's S orbitals share electrons.

Again, from memory of my high-school chemistry classes (30+ years ago).

Wouldn't the real problem be other things leaking back in? If the helium just leaves, that's a vacuum. If only helium can pass, then it's going to stay helium in there.

Helium can leak into and between crystals in the metal inside the container and container itself. That process changes dimensions and mechanical properties of construction.

In this particular case I see it as a planned deprecation, sort of.

drives like this will typically be rated for a 5yr operational life.

If there is leakage, I wonder how long it would be before the data could not be recoverable? Could one transplant the platters and 're-charge' the Helium? Or just somehow recharge the old disk? Because unless / until we get some cheap lots-of-bit storage, all the 'stuff' we're generating will not be, in effect, archivable. Is this right?

Yay back to syringes and weird jigs and refilling our computing accessories with weird chemicals.

Dunno, but disks are consumable items - don't kid yourself otherwise.

I expect some drives to fail within 5 years, but I have many that still have data from much longer ago.

It would be a disconcertingly new phenomenon to find every drive in some old archival NAS failed simultaneously, or every one In a collection of external drives, because all drives last exactly 5 years with little variation, even when switched off, due to helium loss.

I believe helium is used due to it being more effective than air at removing heat. Vacuum is significantly less effective (as in, no convection in a vacuum)

I heard that it was because helium filled drives have less drag on platters (thus, are more efficient) than air filled drives.

At data copy rate of 100 MB/s taking a backup from one of these drives takes 50 hours.

Besides the fact that your BW is off a factor of 2.81, which would bring it to ~18.65h, note that:

* That's why they are often rather on the receiving end of backups ;-)

* Are often in RAID10 systems, for redundancy/uptime + performance, roughly doubling their bandwidth

* Delta syncs/backups exists and thus normally only something like tens to hundreds of GiB need to be actually read, that can be done i way less than an hour

Depends naturally on the whole setup and actual workload, but if you have a high data rate change between backup times you may want to go for a SSD only system using ZFS or Ceph or the like (again depending on actual use case) which could do both cheap differential syncs.

Where did you get that? The specs table says:

Sequential Data Transfer Rate (host to/from drive) 281 MB/s

For comparisons sake, my 14TB Toshiba MG07ACA14TE drives get effectively ~260MB/s outer rim to ~180 MB/s inner rim.

In a usual medium-to-large cloud configuration where 3-5 redundant copies of each chunk/file/object are randomly and thinly spread across hundreds or thousands of drives, a loss of a single 18TB drive would result in a few minutes of «redundancy repair». No RAIDs would be harmed in the process.

Is MAMR energy assist similar to SMR - requiring nearby information to be rewritten?

It shouldn't require re-writing adjacent tracks or sectors. Heating up a portion of the platter makes it easier to change the bits, but the magnetic field doing the actual writing is still pretty narrow. If MAMR/HAMR has any significant effect on nearby data, I'd guess it would be similar to DRAM rowhammer: repeated writes (and thermal cycles) to one location may cause nearby locations to have increased error rates, but that's not the same as SMR directly clobbering adjacent tracks.

Can someone explain why this FC-MAMR tech is significant?

MAMR isn't necessarily significant per se. It's the energy assist in general (be it heat or microwaves).

By heating up a small dot on the disk, that small section will change it's bits easier. As such, we can shove more bits into the same area.

Energy-assist hard drives is the next step, leading to more and more bits in a smaller area.

Energy-assisted magnetic recording is the next step in the hard disk industry roadmap and it's believed to be necessary to increase capacity beyond 20 TB.

What happened to memristor based storage revolution ?

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4479989/

Speaking of storage revolutions, there also was the promised MRAM revolution. That never materialized, but at least they're shipping useful niche products.

Blame HP for repeatedly over-promising on that one.

This seems like a terrible use of helium from a sustainability perspective. There are some things that helium is truly essential for, like low temperature cryogenics and neutron detection. Our supply is finite and largely tied to natural gas reserves. When helium leaks out it floats into space and is gone forever. Why does the atmosphere in the drive absolutely have to be helium? Why not any other mix of inert gases at optimal pressure for a thin air cushion?

Helium is also literally the leakiest thing you can try to contain. It will even diffuse out through the metal itself given enough time. Not trying to go full tinfoil hat here, but filling equipment with helium in this way seems like an effective way to cap drive life by time elapsed rather than amount of actual use.

The amount used in these drives is minuscule, akin to a single party balloon. Given these drives are to be used for many years at a time, I'll go after them after I manage to stop all the birthday balloons.

Does anyone have a number on how much Helium is produced in the Earth's crust by radioactive decay?

Not enough. It is generally known that we are using helium faster than the planet can naturally produce it. We currently get helium from natural gas deposits. In the future we might end up puling it from the air, but that would be very expensive. For many use cases it might be easier to switch to hydrogen.

For those interested in SETI, excessive atmospheric helium might be a detectable techno-signature, a sign that someone is digging into deep crust presumably for hydrocarbons.

When we use the reserves completely and prices sky rocket, probably unnecessary entertainment use of it would stop.

I am sure we are using it faster than we are extracting it, and it'll be interesting to see what happens when selling hydrocarbons no longer offsets the cost of extracting it, but I was thinking about how much is actually produced.

Just use hydrogen for the party balloons

That'd be an exciting party!

A party balloon full of helium is a minuscule amount?

Note this is liquid helium in one operational MRI machine:

"Let’s say you have an MRI scanner with the OR76 magnet with the full capacity of 1800 litres of helium. Your system consumes approximately 4% of its helium capacity per month, that sums up to 48% per year (4%12 = 48%). That means that 864 litres are consumed per year (1800 litres 48% = 864 litres)."

https://lbnmedical.com/liquid-helium-in-mri-machine/

That is liquid, so for the gas we can multiply 864 by about 1000. So this MRI machine is using 86,000 liters of helium gas per year. That is what, 40,000 party balloons? Say each machine does 1000 scans per year, that is a good-sized cluster of party balloons for every scan. So yes, a single party balloon pales in comparison to the amount of helium being used for MRI scans.

(There are near-zero helium MRI machines, but there are also plenty of older machines in use out there.)

Many modern liquid rocket engine powered space launchers also use quite a lot of helium for tank pressurisation in flight. Helium being inert with very low melting point is ideal for this use case, especially with cryogenic propelants.

With the most modern generation of reusable space launchers such as SpaceX starship an alternative solution is being worked on, possible thanks to both fuel (methane) and oxidizer (oxygen) they use being cryogenic.

Instead of pressurizing with helium they heat up a bit of the cryogenic liquids using the rocket engine heat, let it expand into gas and use it to pressurize the tanks, no helium required!

One of the reasons is that for fully reusable space launchers the main cost is the propelant and helium in the amounts needed would be a big part of that!

In some cases the helium cost is comparable to that of the propellants!

Both are negligible compared to the total cost of space launch currently, but once reusability and operations become more streamlined, they start mattering more.

> (There are near-zero helium MRI machines, but there are also plenty of older machines in use out there.)

For the purpose of worrying about future supply, we probably want to compare to those. A balloon is a tiny waste compared to an MRI that vents helium willy-nilly, but it's a much larger waste compared to an MRI that gives half a care.

Thanks, this thread just made “number of party balloons” my new favorite unit of measure

balloon gas is typically about 90% helium and is generally a recycling of helium expended in industrial applications.

> "Let’s say you have an MRI scanner with the OR76 magnet with the full capacity of 1800 litres of helium. Your system consumes approximately 4% of its helium capacity per month, that sums up to 48% per year (4%12 = 48%). That means that 864 litres are consumed per year (1800 litres 48% = 864 litres)."

I know this is napkin math so it doesn't really matter, but a 4% loss times 12 months would be 1-(.96^12) = 39%. Unless of course they're topping up the helium every month.

I'm not sure that the percentage matters in cooling. This stuff is cooling working parts (magnets). Boiling is therefore related to the amount of heat needing to be absorbed rather than the amount of helium in the tank. A half-empty machine would probably need to boil off just as much helium per time/use/work cycle as a full one.

The other error I now realize is that these HDDs probably have much less helium than any party balloon. They appear to be below atmospheric pressure and are so small that they are probably less than a tenth of a liter each. So one MRI scan is probably the equivalent of hundreds of these HDDs.

Superconducting magnets don't produce heat. (They are superconducting) All the heat comes from the resistive feeder cables and the walls of the cryostat.

Loses a percentage of its capacity, not a percentage of the amount of helium it has.

Backblaze did a blog post about helium drives in 2018:

https://www.backblaze.com/blog/helium-filled-hard-drive-fail...

Some interesting stuff in there, like the SMART data element for helium status. And this quote, after they had 3 years in with helium drives:

"To date only one HGST drive has reported a value of less than 100, with multiple readings between 94 and 99."

But, then, later, this quote:

"My prediction is that the helium drives will eventually prove to have a lower AFR. Why? Drive Days."

Would be cool if a Backblaze employee could chime in with what they've seen since.

>Would be cool if a Backblaze employee could chime in with what they've seen since.

can't you check their follow-up reports they've posted? I believe that they break out AFR numbers by drive model so it shouldn't be too hard to separate out the helium drives from the normal drives.

> Not trying to go full tinfoil hat here, but filling equipment with helium in this way seems like an effective way to cap drive life by time elapsed rather than amount of actual use.

Recognizing planned obsolescence is not tinfoil hat territory. It's a rational and predictable commercial practice equivalent to rent-seeking, you're not crazy.

Hard disk drives are a commodity component and the main buyers of He nearline drives are the most sophisticated companies in the world. It’s not a trick dude.

Thought i would expand on this with some order of magnitude leak rate guesses. For a standard small lab vacuum system with viton o-rings something like 10-7 mbarliters/sec (same as atmcc/sec) would be considered acceptable. Assuming the drive is pure helium at half pressure with this leak, the loss would be on the scale of 3 cc/year. The rate will decrease as the partial pressure decreases, but once this becomes significant im guessing the drive is already toast.

I would also note that the helium will leak out before atmosphere leaks in. If the drives are well sealed they should fail due to lack of pressure causing head crash.

I think helium is used for its low kinematic viscosity. That lets the R/W head fly closer than, for example, air. The other inert gases (Ar, Ne, etc.) have KV values higher than air and would give no advantage. He is better for thermal reasons, too.

Counterpoint: new large markets & high tech applications for helium will drive up the price. At this point consumers might scale back the balloons.

Wow you should definitely reach out to the thousands of career hard disk engineers to inform them of your idea using different inert gases. They probably haven’t considered that one.

Tbh I would like to see some calculations of how much helium there is in the drive, a quick bit of napkin math makes it seem like you can make around 5*10^10 drives with the gas from a single field in the US on a single day, assuming that it is at 1 atm pressure, though I don't think that would move much more than one order of magnitude at most. Though your other point still stands.

Note that right now we're burning the natural gas either way. How much of the helium we get out depends upon market use of natural gas, and market prices about whether we even bother to capture the helium from natural gas supplies that are rich in helium.

What gas would you recommend and how does it compare? This would be easy to settle using data.

my first thought was what happens when it leaks out?

If the internal pressure in the device is lower than one atm, the leakage is easier to contain because it's not helium wanting to go out, but air pushing to come in.

Eventually it will happen and the internal drive atmosphere will be compromised to a point it doesn't work anymore. By then, it should be safe enough to open the drive in a clean Helium-filled room and reseal it.

Or just read the data, store it someplace more durable, and put the drive in a museum.

I'm no physicist, but I don't think diffusion works that way.

Even if the internal pressure is 1% of a standard atmosphere, if the density of helium inside is higher than the density of helium outside, the helium will still "want" to leave the box.

So it becomes read only when helium leaks out?

I though it would just lead to more heat, but reading this[1] from Seagate it seems indeed possible that the drive will not function if too much air gets in:

Furthermore, the use of a higher number of thinner disks is not possible because the windage induces turbulence and makes 7200-RPM tracking impossible at the desired TPIs. The only way to reduce windage in air-filled HDDs is to significantly lower the spindle speed (RPM). However, the largest part of the business-critical market is not willing to accept lower RPMs due to the performance and throughput loss. This means that helium HDD technology is the only viable path forward for delivering higher capacities because it will allow for an increased number of thinner disks.

That said, they also go into the efforts they do to seal the helium, so hopefully the helium stays put...

[1]: https://www.seagate.com/files/www-content/product-content/en...

I'll interpret that as: The drive could be designed with a failsafe lower-RPM mode but it's not worth the engineering budget to allow more than one speed.

Especially looking at WD's "5400 RPM Class" thing as evidence that it takes effort to change the speed of a drive.

I have no idea. May become unreadable if the head can't read because the disk is too far (or if it's too close and it crashes on the disk)

Sounds like planned obsolescence.

It’s not planned obsolescence to include Helium when the medium requires it. If you want 18 TB, you’re going to need Helium (from what I know), but if you want less, you might not get it. Planned obsolescence is about planning (hence the name) to make a device worse to encourage upgrading. The device going bad after a while simply by its nature just makes the obsolescence aspect an added benefit.

Is it planned obsolescence that alkaline batteries go bad if left unused for a while? That moving parts in a car can wear out? I wouldn’t say it is.

This class of drive will typically have a design life of 5yrs at 100% duty.

100% duty cycle on the spindle motors, perhaps. But they're definitely not rated for actively performing IO 100% of the time.

Reply on news.ycombinator.com