Question Legacy Marvell HDD controller specs

[SIC66]

Hi guys,

I have been making some die shots of old Marvell HDD controllers, like this 88i6632 :

https://imgur.com/a/ZCkZOv7

View: https://imgur.com/a/ZCkZOv7


I tried to find even basic specifications of this chip to compare it to other chips, but I am not able to find anything.

Do any of you know a way to dig up some specs of these legacy chips?? It would be greatly appreciated!

 

[fzabkar]

That's interesting, thanks.

AIUI, these chips are usually custom designed for each HDD manufacturer and each family of HDD. I believe Marvell has a selection of building blocks which are selected and configured by the designer.

For example, there should be an ARM core, DSP, static RAM, embedded flash memory (optional), read and write channel, DRAM controller, etc. IIRC some Samsung product manuals had block diagrams of these MCUs, but I've never seen a datasheet.

See ...

Spinpoint M9T Product Manual REV 1.0:
https://www.seagate.com/files/www-c...ducts/spinpoint-m-series/en-us/100736114a.pdf

 

[SIC66]

That's interesting, thanks.

AIUI, these chips are usually custom designed for each HDD manufacturer and each family of HDD. I believe Marvell has a selection of building blocks which are selected and configured by the designer.

For example, there should be an ARM core, DSP, static RAM, embedded flash memory (optional), read and write channel, DRAM controller, etc. IIRC some Samsung product manuals had block diagrams of these MCUs, but I've never seen a datasheet.

See ...

Spinpoint M9T Product Manual REV 1.0:
https://www.seagate.com/files/www-c...ducts/spinpoint-m-series/en-us/100736114a.pdf

Thank you for your reply.

I suspected they were custom made, so datasheets would be hard to find. I have a lot of HDD PCB's and I can indeed not find the same chip on disks from different manufacturers.

The ARM core is easy to see, but hard to identify which core exactly.
Same goes for the memory, the yellow and blue memory banks looks different, but I can not identify which is RAM and which flash. This HDD had no EPROM on the PCB.

This chip (88i6540) uses a slightly modified core and has no EPROM on the PCB, hence one type of on-die memory (RAM)

https://imgur.com/a/VS73yvc

View: https://imgur.com/a/VS73yvc

I expect the yellow banks on the first image to be the embedded flash judging by surface are on the die compared by the other die.

My other die shots can be found here, if anyone is interested: https://www.flickr.com/photos/sic66/albums

 

[fzabkar]

Would you consider adding your collection of HDD PCB photos to this database?

http://www.hddoracle.com/viewforum.php?f=112

My "photos" are actually CCD scans.

How do you produce the die shots?

BTW, newer HDDs have dual cores, AIUI.

Sometimes the core version is printed on the chip. Otherwise you might be able to identify it by examining ROM dumps from firmware databases.

http://files.hddguru.com/

I have written tools that parse and decompress these ROMs. You can find my tools at hddoracle.com.

 

[SIC66]

Sure, I can upload photos of my PCBs, but I have not made most of these photos yet. How do I check they are not already uploaded there? Some models have many different PCB versions too..

HDDs are not a special interest for me. I have boxes full of all kind of electronics and chips, it just seemed like a fun project to compare 15 Marvell HDD controllers.

I think it would be best to find some chips where I know which ARM core they use and use that to compare to chips like these HDD controllers.

The die shots are made by first removing the silicon from the plastic package with hot acid. Then etching away the top layers of the silicon with some other chemicals.

The images are made with a microscope, stitching together many separate photos. These HDD controller chips are just 4 by 5mm, so it takes only 6-20 photos or so to make them.

 

[SIC66]

I tried using a scanner on a Brother printer to image some PCBs, but the depth of field is way too low. Parts of the PCB that are lifted slightly from the glass as very blurry. That would mean using my DSLR camera to make images of 100+ PCBs. I'm afraid that would just take up too much time.

 

[fzabkar]

It looks like you have a CIS scanner. Mine is CCD. If you are located in Australia, I could do the scanning for you, or I could buy the PCBs if you would be willing to sell them.

 

[SIC66]

I'm from the Netherlands, pretty much the other side of the world
I am not willing to sell them at the moment, sorry. Half of them also have the HDD still attached.
Maybe I will find some time to use the camera to make the images anyway. You need both sides of the PCB?

I changed job a year ago and I am no longer allowed to take the HDD's and PCBs home, they have to be destroyed.

 

[fzabkar]

Both sides would be better for circuit tracing. Thanks.

Writing zeros to the drive is all that it takes to render the data unrecoverable. I wish employers could be made to understand that. :-(

 

[SIC66]

You can still recovery most of the data if you do just 1 pass of zeroing, it takes a few passes to fully destroy the data.

Our customers are very big corporations and they simply want drives to be shredded when their lease is terminated. We sometimes have to destroy very expensive hardware that has no ability to even retain data, like RAM or entire controllers for enterprise level storage systems. Customer is king..

 

[fzabkar]

Nobody can recover your data after a single pass of zeros. You have been caught out by an Internet myth based on Gutman's paper many years ago. Recording densities are now a gazillion times greater than they were then.