[Q] eMMC crash - possible reasons and solutions

Hello everyone.

I've been looking around here for some time, reading all that suff about eMMC chips burning on Desire S. That fact dissapoints me as I was aiming to buy the gadget myself. However, I didn't find any general solution or even investigation of the case, so I'm trying to develop some kinda stuff. Let me summarize main points that we have so far.

1) The faulty guy is usually Samsung eMMC-type BGA chip KLM4G2DE (2 Gb NAND flash), however Sundisk chips were also found to burn.
2) The problem is rather hardware than software dependent as it is observed without any corellation to hboot/flash installed.
3) It was noticed that in many cases eMMC fault followed extraction-insertion of battery after phone freeze.
4) HTC doesn't recognize this as defective case and no improvements to hardware are made in new revisions of motherboard (MB) as there have been cases (at least one) when the same phone after warranty repair crashed again after some time.with the same eMMc chip installed
5) Other phones with the same eMMC installed (e.g. Sensation) doesn't experience same problems.

What can I deduce out of all this stuff and my own experience?
As soons as the case seems to be non-software dependent it should be the chip or some other hardware that drives it wrong. As soons as the chip itself seems to be OK (see 5) I beleive that it is poor motherboard design that burns the chip down. eMMC is rather bomb-proof architecture combining the memory itself and the memory controller on the same crystal. Two major ways to drive it wrong are:

1) Supply incorrect clock pulses to clock bus
2) Supply incorrect power (current/voltage/voltage slope) to memory and/or controller

The first assumption seems not very possible as clock usually comes from centralized source controlled by oscillator. If the clock is wrong, the emmc fault wouldn't be the only problem
The second point seems rather reasonable as Samsung eMMC power-up guide (see file attached) directly points out the importance of accurate power supply (especially power-on slope!), otherwise memory faults are inevitable.

That's all I can deduce so far, unfortunately there's no photos/schematics of desire s on the web to analyze the connection of emcc chip to MB. What can I suggest to prove/disprove all the stuff I wrote:

1) Can someone brave disassemble his Desire S and make high resolution photos of both sides of motherboard? This may help in further analysis.

2) Can someone even more brave and being on close terms with oscilloscope try to measure power-up voltage slope on Vсс and VccQ inputs of eMMC chip(see document attached) ? May be we are just having one of the issues described in the document.

UPDATE
I found the datasheet four our chip, find it attached to this post

From my experience the eMMC in the "fried" cases is not actually faulty but simply does not allowing write access. Usually it is accompanied by /cache or /cache + /data corruption. When only /cache is the problem it is fixable, but when /data is affected there is no way to write a bit on the internal memory.

Unfortunately I have no confirmed explanation to this...

Just in theory when updating several apps from the Market (or other activities requiring use of /cache partition) it is possible that the /cache is filled with data and the device stucks at a point when it has no more space available to write. Rebooting to recovery and wiping /cache solves the problem. But if in that moment, when the app is downloading to /cache and another app is written from /cache to the /data partition at the same time, disconnecting the power source (battery pull) can interrupt the process making this partition unavailable (example: if you take out your USB flash drive from the PC while writing data on it there is a great chance to destroy it - tested myself ) The ext4 file system provides a protection for such cases by the way it is managing the writing process - reference here.

In my opinion all of the bricked devices (famous "fried eMMC") reported in this forum are easily repairable with JTAG and a skilled technician, but unfortunately there are no such cases reported here. Personally I do not have the knowledge, equipment and intention to do such experiments myself.

This is my logic based on my observations while trying to assist people in this forum to solve this issue. For some of them it was successful, for others - not.

I hope that my post will make a contribution to the general picture.

Regards,
Stefan

Also, strangely, a common way to trigger the 'dead emmc' is when you hit 'update all' apps in the android market, regardless of what ROM/market version your on..

Surely it must be a combination of a software fault/problem too?

amidabuddha said:

Just in theory when updating several apps from the Market (or other activities requiring use of /cache partition) it is possible that the /cache is filled with data and the device stucks at a point when it has no more space available to write. Rebooting to recovery and wiping /cache solves the problem. But if in that moment, when the app is downloading to /cache and another app is written from /cache to the /data partition at the same time, disconnecting the power source (battery pull) can interrupt the process making this partition unavailable (example: if you take out your USB flash drive from the PC while writing data on it there is a great chance to destroy it - tested myself ) The ext4 file system provides a protection for such cases by the way it is managing the writing process - reference here.

Click to expand...

Click to collapse

I agree with that. Most plausible theory i read about eMMC fries/corruptions. You've been helping people get out of this crap (i call it crap because it's HTC's fault, cheaping out on us) for a long time now.

What i don't understand is this - when downloading multiple apps from market (i had 2 - Angry birds, ES file explorer), the phone goes to sleep mode, and NEVER wakes up. No ADB, no 3 key combo, NOTHING. This leads to an unavoidable battery pull, which results in corruption like you said above. Why does the phone enter this "Sleep of Death" if you may call it? What the hell is the problem? Also, why not other HTC devices? (for that, i guess the answer is the unique slide out battery, only in 2 other devices - Bliss and Radar, whose battery can't be removed). If we can solve the "Sleep of Death" mystery, we'll get this issue out.
(More info - http://bit.ly/rXhDRR and http://bit.ly/v3lsS6 )

Also, DS (and all new HTC devices) are EXT4 by default. Flashing (not s-off'ing, but flashing a ROM after that) changes back to EXT3. That is probably why my phone survived the battery pull i did as said above. (It was freshly s-offed by XTC Clip, stock ROM).

Finally, i think this only happens to devices with already screwed eMMCs. Mine survived many battery pulls after the first one. The screwy ones are region specific. I never read anyone from India had the issue.

levantine said:

Hello everyone.
1) The faulty guy is usually Samsung eMMC-type BGA chip KLM4G2DE (2 Gb NAND flash), however Sundisk chips were also found to burn.
2) The problem is rather hardware than software dependent as it is observed without any corellation to hboot/flash installed.
3) It was noticed that in many cases eMMC fault followed extraction-insertion of battery after phone freeze.

Click to expand...

Click to collapse

Just leaving this here for Google'ability. I've experienced a similar case on a Samsung device, equipped with a MAG8DD moviNAND (manuf. date ~10/2009). After pulling the battery the moviNAND died.