Hard disk internals

[ewx]

As well as the spare disks mentioned in the previous postings I have a number of broken disks. I’ve been meaning to disassemble one for a while and today I found the time to do so.

This is a 4.5GB Seagate SCSI disk. It’s probably been sitting in a junk box for more than a decade and I can’t remember in what way it failed. I don’t even have a SCSI adapter any more so it’s not practical to find out.

The controller board came off easily; it could easily be replaced with one that wasn’t broken, if there was still important data on the disk.

There’s actually no need to remove the controller board to get at the platters, though. Instead the tape seal around the edge must be removed and the lid then taken off. The seal didn’t really want to peel off and in the end I sliced it down the middle with a knife.

The screws holding the disk closed were hidden under the label. They also require a hex bit, fortunately I got one a while ago, I think to get into one of Apple’s computers. Obscure bits seems more justifiable here.

The disk platter is a pretty good mirror. It also rapidly collects dust, so don’t open a disk with data you care about on it unless you can keep the local environment very clean.

Can anyone explain the characters on the metal component at the top left? They could be “&OLC” but I’m wondering if they’re not actually Latin script at all.

Detail of the read/write head assembly. There are two arms visible to the left; an upper one with nothing attached and a lower one with a read/write head attached. Below the top platter is a second platter and three more arms. There are four read/write heads in total, one for each side of each platter. The assembly could support two more platters, for a maximum capacity of 9GB.

Detail of the read/write head itself. You are looking at a device sat on top of a mirror, so you will need to ‘divide by two’ to make sense of it.

Damage to disk surface incurred while I was figuring out how to get the disk open.

I’ve been attempting to extract the platters this evening, with no luck. This youtube video shows someone else successfully disassembling a similar design but I can’t get past the first step.

I also had an SSD lying around in the broken disks box.

SSDs are, frankly, a lot less exciting inside. The chips are Micron MT29F32G08CBABA MLC NAND flash chips of 32 Gbits each, 16 on each side giving 128 Gbytes. The device advertizes 120Gbytes which leaves 8Gbytes headroom for the controller’s wear levelling and garbage collection algorithms.

The controller. The disk went from working fine to refusing to show up on the SATA bus overnight (actually, worse: it hung the BIOS on one machine), so I suspect its this component that failed rather than the flash (although I suppose it’s possible that the controller loads its firmware from flash).

More pictures.

Comments

[tigerfort.livejournal.com]

My brain initially suggested that the writing on the magnet-casing was &02C, but I suspect that's a delusion because I associate &[hex] with storage of various kinds. (I can't imagine why you'd want to write an address on there.) It seems more likely that it's 802C, which could easily be either a part number or the ID of the production machine that magnet assembly came out of, both of which are more plausibly useful during manufacturing.

[sweh.livejournal.com]

Have fun with the magnets; they're pretty strong!

[crazyscot.livejournal.com]

What are you having trouble with - getting the platter assemblies away from the drive housing, or getting the spindles off the platters? I've done all this (in order to put the platters beyond use by mangling them, which was great fun :-) ) though it was a few years ago; I'm having trouble remembering whether I needed to do anything special...

[ewx.livejournal.com]

Separating the platters from the housing looks conceptually simple; there's a nut that needs to be undone on the back. There's not much clearance around it though and I've not yet found anything that can get in to grip it.

What I was mostly attempting was despindling (per the video link) and that has so far gone nowhere. I suspect the real answer is a specialized tool.

[crazyscot.livejournal.com]

Memory says that the armature assembly effectively holds the platters in place vertically, and once that was off (which was a bit of a pain on some of the drives) the platter stack just slides off its axle. I don't remember despindling at all, so I suspect it was easy once I had access to both sides of each platter.

[crazyscot.livejournal.com]

for "a bit of a pain" read "partially destructive", at least to the heads and arms...

[ewx.livejournal.com]

Evidently your drive had a different design l-)

Success!

[ewx.livejournal.com]

The answer turns out to be: ignore the middle and instead apply leverage to the outer rings until the top one goes ping and flies across the table.

Re: Success!

[armb.livejournal.com]

[Like]

[baljemmett.livejournal.com]

Yep, SSDs are definitely not as funky internally - but I believe you're right about the controller firmware living on the bulk flash. There should be several copies strewn about the place, though, so a terminal failure would probably be upstream of the storage medium...

In fact, my understanding is that mechanical HDDs do the same thing - the ROM in the controller holds a basic bootstrap firmware that has enough smarts to spin the disk up and load the real driveware from the platters (again, it has several copies to choose from). A chap on the EEVblog forums designs the electronics for these things and has posted a few times about some of the fascinating hairiness that goes on behind the scenes - alas I can't find the post I'm thinking of, but this one (http://www.eevblog.com/forum/reviews/quantum-xp32150-scsi-hard-drive-(1995)-teardown/msg366313/#msg366313) has an overview of the factory formatting process and mentions the driveware living on the disk.