2 chips replaced - ESMT 5ns to Hyundai 5.5ns.
Location is front, U5 and U7.

Heh. It's worse

Given the 0 errors on bank1 63:32 and almost the same errors on 31:0

I think it's safe to say you broke 64:95, 96:127
So at least you've identified those channels.
Most likely the 2 chips you removed were fine.
bank 0 is most likely the front - which had no errors.
Bank 1 is most likely the rear.

I'd guess the soldering didn't go right.
Try putting back those 2 chips, check the mats result is similar to the first, and then remove all the rear chips.
You should get 0 errors , 16mb.

If yes, add 1 chip at a time to the back and test between.

myne wrote on 2025-07-22, 13:02:

Given the 0 errors on bank1 63:32 and almost the same errors on 31:0 […]
Show full quote

Given the 0 errors on bank1 63:32 and almost the same errors on 31:0

I think it's safe to say you broke 64:95, 96:127
So at least you've identified those channels.
Most likely the 2 chips you removed were fine.
bank 0 is most likely the front - which had no errors.
Bank 1 is most likely the rear.

I'd guess the soldering didn't go right.
Try putting back those 2 chips, check the mats result is similar to the first, and then remove all the rear chips.
You should get 0 errors , 16mb.

If yes, add 1 chip at a time to the back and test between.

Is it possible that the BGA balls may be cracked for those channels ?

Yeah, but it's far more likely the ram, transistor, cap, resistor or other is bad.

There were 9 working chips.
Now there are 3.

btw, here's the datasheet for a similar SD chip.
I counted 25legs per side, making it a TSOP-50.
Assuming I can count, pin 18 is the chip select which as noted can be used as a bank select.

I don't know HOW this is done on these cards, but it's possible that that is the main issue.

https://docs.rs-online.com/bd22/0900766b8129ef4a.pdf

I think the process going forward should be fairly straightforward.
* Remove the 2 replacement chips.
* MATS. Confirm that bank0 has 2 chips with total failures and the others have returned to their previous level of disfunction. If yes, bank0 = front, bank1 = rear.

ASSUMING bank0=front
* return the 2 original chips
* MATS. to confirm it's back to the original level of broken
* remove all rear chips
* MATS to confirm 0 errors as a 16mb card
* Trace pin 18 on rear for each chip, to attempt finding something broken.

Wouldnt be unusual for it to be a combination of both issues, some dead ICs and a cracked memory channel, see it on modern cards quite a lot, you fix the dead ICs and boom you now have to reball the core to fix the damaged memory channels.

Final update, at least on the Leadtek: pads ripped upon chip restoration.

I'm waiting on another 256SDR with good RAM (Guillemot branded, which is a match to my Area 51 project) but a different issue (broken cap causing restarts, probably due to dirty power)

Eh?
What happened?
Overcooked the solder?

myne wrote on Yesterday, 12:26:

Eh?
What happened?
Overcooked the solder?

I assume the pads couldn't resist all the solder heating cycles - they already went through three heat cycles before I started the RAM chip replacement process to rule out any connecton issue. The heatgun must've been the final straw, I guess.

I'll check the 2nd card once it arrives. That one keeps rebooting past the XP logo due to broken C71 (which also took the pads with it...). If the RAM checks out fine on that one, I'll hopefully attempt a pad transplant from the Leadtek, since I did find where the pads go through on the other side using DMM on continuity mode.

Also the chips for the Leadtek were quite rare to source - 5ns. The one I'm waiting on (Guillemot) uses 5.5ns, which I recall having a few more of.