This post is a bit of a follow-on from last salvage posting. The undisclosed location provided many bits and pieces, including a large number of hard drives from (mostly) HP branded computers. Well, I suppose provided might not be the right word … after all, I did work hard to remove them by hand from the multitude of chassis.
Luckily, I was the one who had ended up with the drives in my hand as I’m not particularly evil. I had decided that it would be great to use these drives to try and understand their reliability through reading SMART data as a “statistical” exercise, and good to do a full random fill and verify to check the condition of the drives for potential re-use as a backup medium.
The list of drives salvaged and the results of testing the drives were as follows:
All drives were SATA except one. Out of the haul of drives, the majority were 250Gb units. The Seagate units were 7200.10 and 7200.12 series drives with varying firmwares, some of which were HP-specific. This suggests that the 7200.11 drives may have failed and been returned due to brick bug. Two WD RE-series drives were found, and were likely used in the same system as the SCSI drives discovered earlier as they had the same caddy design.
The drives have gone through varying numbers of hours, most spending at least 20,000 hours in service. Of the drives, all of the Samsung units survived and were recertified successfully. The Seagate units were a troublesome bunch by comparison, with one drive completely unusable, and most of the drives with some level of reallocated sectors, and others degrading and growing bad sectors through read-back tests and reporting read errors.
One of the WD units were also having issues with read errors after a full write and throughput issues, the graph below obtained over USB 2.0.
The interesting thing is that the Seagate drives seem to have a different reallocation strategy than prior drives. Instead of making reallocations and pending sectors when data is very likely to be lost, these drives seem to do it almost pre-emptively as sectors get weak and no data loss usually occurs. I’ve experienced this with two Seagate (HP OEM) 250Gb 7200.12 drives myself.
I decided the worst WD and Seagate (unusable) drives would be good candidates for a teardown, just because they were not otherwise useful.
Seagate Barracuda 7200.10 250Gb ST3250310AS
We start the teardown with the Seagate which originally was thought to be undetectable, but stayed online just long enough to get its SMART data retrieved. It’s a hopeless drive, so it’s a great candidate for taking apart.
The drive is a 7200.10, which is before the brick bug 7200.11, and is Made in China. The PCB on the rear is relatively small, with a similar dimension to the PCBs on Seagates to this day.
It shows a little resemblance to the Maxtor 6E040L0, which was another “thinner” OEM optimized drive, featuring one platter and one head. The smaller dimensions might have a benefit in terms of storage space occupied in warehouses and in shipping.
The ends had the interface block and the serial number label as usual. Turning the drive around and removing the PCB shows that there is only a small intervening cellophane sheet and a thermal pad for the main SoC, but it’s conducting heat to the cellophane sheet which would have a decent thermal resistance. Hardly optimal.
One thing to notice is that there are no cut-outs on the side, and there doesn’t seem to be any taped cut-outs on the top. There are no cut-outs on the bottom either, so it seems that this drive is a self-writing servo design where some seed tracks are laid on the media before or during assembly and the drive “fills in” the rest of the tracks on its own without the help of a servowriter. There is one hole to the chamber, but this might just be for evacuation of the inside air so that the atmosphere is all filtered through the breather hole.
The connections to the motor and to the head stack are made by springy contacts.
The PCB has a very low number of components on it, thanks to an SoC-based design. The ST branded SoC handles almost everything, with another ST branded motor controller and Hynix cache RAM being the main chips on the board. There is a crystal package, a bit of serial flash memory, some power conditioning components and the regular passives. There seems to be a mass of 1 ohm resistors which might be to limit the inrush current to motor windings. Near the top right screw, there seems to be a mounting point for a piezo vibration transducer that was not fitted.
The top lid cover is very basic, and only supports the head stack from the top. A clear “silicone rubber” style gasket is used all around, but also above the depression used to “wedge” the lid on the case with the magnet assembly. From the looks of this gasket, it might have just been a bead of liquid silicone that has set.
Behold the glory of the drive. It’s a simple one-platter affair with two heads. The spindle is secured with three screws. There is an airflow-based head unlocking mechanism, so the head is retained securely in the landing zone in the center. There is a recirculation air filter visible in the corner.
From the side, we can see the fact that it is one platter, but also, we can see the connectivity to the head being done by a patterned “rigid” transmission line.
The actuator arm uses a copper-coloured coil, as expected, and a “bumper” silicone rubber can be seen to prevent head smashes into the ends of travel. The small magnets were positioned on both sides.
The head slider has a very angular design, which is quite sophisticated in its design compared to the rectangular simplicity of early hard drives.
Six connections are used to the head, all of which are actively connected, thus implying the use of active fly height control. Two connections for read, two for write, and two for the head heater, which controls the air temperature around the head to control the fly height.
The simplified three-screw mount can be seen disassembled, where the underlying platter’s ID mark and laser textured anti-stiction landing zone can be seen.
The breather air filter is below the platter itself, deep in the tub.
Western Digital Caviar Blue 160Gb WD1600AAJS
The second drive in the teardown sequence is the Western Digital Caviar Blue 160Gb unit. This one exhibited continual pending sector growth and extremely slow throughput. The Caviar Blue series is seldom used in my builds, as it is normally reserved for mainstream performance applications where storage capacity is not a primary concern.
From the outside, it very much resembles a modern WD drive, along with the wedge shaped PCB in the rear. The unit already has a servo trackwriter cut-out in the bottom, sealed over with thick aluminium tape.
Despite the OEM nature and low cost, this drive was a full sized unit, unlike the Seagate above. It’s probably because WD only make full sized units, just for standardization reasons.
As usual, the interface is visible on the end, along with the serial number label at the other. The notch in the front of WD drives is seen as well, which might be there for production handling reasons.
The PCB itself is separated from the rear with a cream coloured piece of foam, with some evidence of accumulated heat.
Again, the resulting PCB is quite similar to Seagate’s effort – an ST Smooth motor controller, a Marvell SoC with external crystal, and a Winbond cache buffer RAM. Also, near the screws is a mounting for a piezo vibration sensor that was not fitted. The use of springy contacts for the head stack is similar, although the gender of the motor connections is reversed.
Aside from that, the rear of the tub is pretty much sealed, with only one piece of tape closing up the servo track-writer access cut-out. The motor is connected by a stuck-down piece of printed flexible cable. As with other drives from WD, the motor is held to the tub by “ribs”, which are thicker on this drive implying a smaller platter chamber.
Removing the label from the top lid shows the use of the label to “seal” off the breather air supply “tunnel” which lengthens the path and reduces the chance of particulate matter going to the filter.
The filter itself is mounted on the reverse side of the lid, held by adhesives. Only the head gimbal assembly is secured to the top lid, the spindle is left unsupported from the top.
The drive utilizes a one platter design with just one head. It also utilizes head ramp loading technology, which keeps the heads off the surface of the disk during stop and start, and transport, thus reducing the risk of media damage and head damage. The top surface of the media, seen here, is actually a bad surface, as the outer edge is lighter in colour and seems to be unevenly coated. A traditional spindle clamp using six screws is used.
Due to the ramp load and unload, the head itself is exposed and visible in the parked position. A different, optimized angular design can be seen, along with six connections implying active fly height control.
With the ramps removed, you can see the “tang” at the end of the arm which is engaged by the ramp to lift the heads off the drive. As the tang keeps rubbing on the ramp on every load and unload, this can cause contamination of the platter chamber with the particles liberated by the friction, which is why there is a ~300,000 load-unload rating for most drives.
A silver coloured coil can be seen, with the rubber bumper sleeves and just one magnet was used on the lower side.
There’s the whole head assembly, which could have accommodated another head in its present incarnation. The head pre-amp can be seen directly mounted onto the flex cable, as is usual for drives after the late 90’s.
The tub itself featured an abridged spindle motor and raised tub floor, and was designed only to accommodate one platter.
It was interesting to see how the Seagates don’t seem to have held up as well as the other brands in the batch of salvaged drives. Reallocated sectors, and read errors were experienced on a few drives. Considering their age and the environment in which they were used, I suppose it is not unexpected.
On the whole, the two vendors seem to have their own design choices in the drives they supplied, and various cost-optimization measures are taken. As they are relatively modern drives, both drives seem to employ active fly height control, but only the WD employs ramp load-unload technology whereas the Seagate still utilized contact start stop. The Seagate seems to have some self-writing servo system, whereas the WD still seems to utilize a more traditional servo-writer system based on the shell design. Both designs were made to handle one platter and up to two heads.
Next time, maybe I won’t haul as many drives home. It was unbelievably heavy!