Hacking a BFL Jalapeno to 32GH and beyond....(???)

[lightfoot]

I have thought about the BE's, but they seem like odd chips, less documented than the BFL stuff and to be honest it might not be worth the cost of repair for 30-40gh. However I have been very busy this week with a separate consulting project, so it's not dull.

Also got in a broken jally that is weird. I think the FETs are shorted to hot, as the power on the 1 volt rail is all over the place. Will be working on it more this weekend, it's weird.

C

[1714968040]

1714968040

[ManeBjorn]

Dude congratulations on the consulting job. 
I have been recommending you all over the place.

FET's shorted that does sound odd.  I wonder what failed to make that happen.
Keep us posted.

I have thought about the BE's, but they seem like odd chips, less documented than the BFL stuff and to be honest it might not be worth the cost of repair for 30-40gh. However I have been very busy this week with a separate consulting project, so it's not dull.

Also got in a broken jally that is weird. I think the FETs are shorted to hot, as the power on the 1 volt rail is all over the place. Will be working on it more this weekend, it's weird.

C

[lightfoot]

So what's been going on...

Got three jalapenos in as donations, all of which have interesting little problems.

The first is pretty unusual: It looks like the FETs have shorted out. Symptom is 2-3 volt pulses on the 1 volt line that heat up (and I believe have destroyed) the hashing chips. I think the culprit is a short of the high fets *or* a failure of the lower voltage FETs.

The way the power supply works is that the 1 volt line is attached to the 12v line for a short time, then to the 0v line for a longer time. The capacitors and the the chips smooth out the resulting voltage, and the monitoring circuit picks this up as a stable 1 volt. What this also means is that the bottom chips (the 1v) are on longer, but the top chips (12v) have a higher inrush current. One set of FETs is going to take a lot more current than the other, this might be the reason the "hot FET" issue popped up in the beginning, and why BFL put bigger FETs on one side (and beefed up all FETs on the singles). Mystery makes sense.

However if the FETs fail to gate interesting stuff will happen. If the high FETs short, then you're going to have 12 volts on the 1v rail but the bottom fets should short as well. Might result in blown FETs. If the bottom FETs open then nothing is there to prevent the high rail from blowing up the chips.

This would explain why trying to remove chips from this board and put them on others does not work. Nasty failure mode.

The second one had one chip hashing at about 4gh. The second chip looked good, but was not working. Tried a reflow, chip would warm up but still nothing. Removed it from the board, held it in finger and thumb...

And it cracked apart. Not the normal shatter into little bits, a clean break in half. I think this one was a victim of a heat sink screwed down too hard; what happens is that the pressure torques the case around the BGA balls, and if one of them is a tiny bit higher you get a fracture inside the chip and failure. Only way to see it is to pull the chip and try bending it.

Third unit is another one that is shorting out, will take it apart and see if I can replace the FETs to get it going. But I'm guessing it has the same failure mode as the first. Drat.

Anyway, things keep rolling along, we keep hashing, and I dream about 300gh jallies in my sleep. What a neat thought...

C

[lightfoot]

Update: I just had that shorted jalapeno erupt in a mini fireball on the bench. Cool!

Apparently removing the incinerated FETs allowed 12 volts to go through the little caps. Each cap lit up like a little sun on 2 amps of power, then the next one went off. Counted three before I could crowbar the supply.

Wow. Like little arc lamps on your jalapeno!

Note: This one was a June 2013 model with little FETs. I think the fan threw a blade, and the unit overheated the FETs and started a small fire on the board before the power supply blew. Very interesting.

C

[ManeBjorn]

Sounds like you might need a fire suppression system in your work room.

Update: I just had that shorted jalapeno erupt in a mini fireball on the bench. Cool!

Apparently removing the incinerated FETs allowed 12 volts to go through the little caps. Each cap lit up like a little sun on 2 amps of power, then the next one went off. Counted three before I could crowbar the supply.

Wow. Like little arc lamps on your jalapeno!

Note: This one was a June 2013 model with little FETs. I think the fan threw a blade, and the unit overheated the FETs and started a small fire on the board before the power supply blew. Very interesting.

C

[tiozes]

Hi, I have 2 butterflylabs single sc 60 gh / s, I think this post is aimed for the jalapeno, but it could make the single 60gh Oc / s?

Forgive my English I'm Spanish.

regards

[lightfoot]

Update: The fireball is now hashing away at 5gh. Bit slow, but I have seen worse.

What did I do? Why dig out the fire, like a coal mine fire. Literally, I dug through the board substrate, removing layers till I got rid of the ground plane for the right side FETs. Then removed all the caps in the region, removed the right side choke, and fired up the unit.

Came online with the left side FETs. Put heat sinks on the FETs to control temps, put a big fan on pointing *DOWN*, and it's hashing away at 5gh.

Yes, anything can be fixed if you have enough insanity. But still the little fires were pretty.

And no, I am not turning my back on this one. It will go into the museum of weird stuff.

C

[lightfoot]

Sounds like you might need a fire suppression system in your work room.

I do have a fire extinguisher, and yes when I saw the first fireball I thought "grab the extinguisher". But this was merely a 12 volt, 2 amp mini fire, not a 300v 10,000 amp one. With a 300 volt fire (or a 36v fire if you have several thousand amps of power) you literally throw the fire extinguisher *AT* the IGBTs so that it's mass will blow the parts far way enough that the plasma arc will go out.

Been there, done that.

So the unit stayed on the heat plate (which is all about heat) till I was able to dig the fire out.

C

[ManeBjorn]

That is cool.
You should have taken before and after pictures.

Update: The fireball is now hashing away at 5gh. Bit slow, but I have seen worse.

What did I do? Why dig out the fire, like a coal mine fire. Literally, I dug through the board substrate, removing layers till I got rid of the ground plane for the right side FETs. Then removed all the caps in the region, removed the right side choke, and fired up the unit.

Came online with the left side FETs. Put heat sinks on the FETs to control temps, put a big fan on pointing *DOWN*, and it's hashing away at 5gh.

Yes, anything can be fixed if you have enough insanity. But still the little fires were pretty.

And no, I am not turning my back on this one. It will go into the museum of weird stuff.

C

[ManeBjorn]

I was part of the Fire Protection unit during my summer off from college at a PaperMill.  We had an electrical fire it bizarre to say the least.  Hair standing on end in my turnout gear trying to get around it to cut the power while it melted the steel frame to a multi-ton dryer unit.  Not fun.
I can imagine the plasma bolts you saw on the other one were spectacular.  Run, Run!!

Sounds like you might need a fire suppression system in your work room.

I do have a fire extinguisher, and yes when I saw the first fireball I thought "grab the extinguisher". But this was merely a 12 volt, 2 amp mini fire, not a 300v 10,000 amp one. With a 300 volt fire (or a 36v fire if you have several thousand amps of power) you literally throw the fire extinguisher *AT* the IGBTs so that it's mass will blow the parts far way enough that the plasma arc will go out.

Been there, done that.

So the unit stayed on the heat plate (which is all about heat) till I was able to dig the fire out.

C

[lightfoot]

That is cool.
You should have taken before and after pictures

Got em. And movies. Pretty.

C

[ManeBjorn]

Now that rocks.
You should post that. 

That is cool.
You should have taken before and after pictures

Got em. And movies. Pretty.

C

[lightfoot]

Well, I thought about it, but the problem is these situations are so far beyond normal it's not funny. Posting videos would give the impression that this kind of thing can happen normally, when it really is a test bench sort of problem.

The most a normal jally would do is smoke the FETs then fail the power supply. And since it's in a case it wouldn't damage anything else (which is probably what happened). So I'm posting as a warning to other people doing work on stuff; be careful and always take proper precautions.

It's never dull.

C

[DanielVG]

Is there anybody here who can flash these devices?

I have a faulty one, would like to have it checked/repaired.

send me a PM if interested, European if possible.

[bcp19]

Hi, I have 2 butterflylabs single sc 60 gh / s, I think this post is aimed for the jalapeno, but it could make the single 60gh Oc / s?

Forgive my English I'm Spanish.

regards

This isn't really an overclocking thread, the OP started adding chips to his Jalapeno to increase speed.  The Single is already at the maximum number of chips, so there is no way to increase from that.

[MrTeal]

The way the power supply works is that the 1 volt line is attached to the 12v line for a short time, then to the 0v line for a longer time. The capacitors and the the chips smooth out the resulting voltage, and the monitoring circuit picks this up as a stable 1 volt. What this also means is that the bottom chips (the 1v) are on longer, but the top chips (12v) have a higher inrush current. One set of FETs is going to take a lot more current than the other, this might be the reason the "hot FET" issue popped up in the beginning, and why BFL put bigger FETs on one side (and beefed up all FETs on the singles). Mystery makes sense.

Interesting note for those that are interested to follow up on what lightfoot said. The highside (top) FET doesn't actually conduct any higher peak current than the lowside (synchronous or bottom) FETs, and the average current is actually quite a bit lower than the lowside FETs because the duty cycle is quite low. That is why you often see designs with the lowside FETs have a higher current rating and lower resistance.

The power dissipated in each of the lowside fets is pretty simple to calculate, it's just (1-Vout/Vin) * Rds(on) * [(Io/n)^2+(Ir*nphase/n)^2/12], so the percentage of time the lowside is on, times the resistance when it's on, times the current. n there is just the number of lowside FETs in total (6 for the Jalapeno and LS), nphase is the number of phases (2 for the Jalapeno and LS), and Ir is the ripple current. That's usually designed to be about 20%-50% of the output current, and depends only on switching frequency, output voltage and inductance. For the BFL design Ir is ~6.8A. For the lowside FETs the power dissipated is directly proportional to the resistance (Rds(on)) of the mosfets, so you often see a lot of big, beefy FETs used since they have lower Rds(on).

The topside mosfets are a little trickier, since there's two things to calculate. The first is the conduction losses, which is the same formula as for the lowside fets except instead of using (1-Vout/Vin) (0.917 for the 1V from 12V that BFL uses) you use Vout/Vin; in this case 0.083. Obviously if you used the same FETs for the top as you do for the bottom the dissipation from conduction would be a lot lower, 11x lower in this case. However, you never use the same FETs in the top because there's another kind of losses that the highside FETs have to deal with, conduction losses.

The conduction losses aren't from the steady state P=I^2R losses you get when the FET is fully turned on and there's very little voltage across it. Real MOSFETs aren't perfect switches though. When you switch on and off the highside FET, the change doesn't happen instantly, so there is a time when you have a lot of voltage across the FET and have current flowing at the same time, so you burn up a bunch of power there.
The formula is P = 2 * f * (Vcc*Io/n) * Rg * (n/nphase) * Ciss. f is the frequency (300kHz for BFL), Vcc is 12V, Rg is the total gate resistance of the driver and FET, and Ciss is the input capacitance of the FET.
What's interesting about this is that the n term cancels out; adding more highside FETs per phase doesn't actually decrease the switching losses per FET. This is because even though if you run 2 or 3 FETs in parallel each of them sees less current, the time it takes the driver to fully turn on the FETs is 2 or 3 times longer due to the input capacitance being higher. That also means that if you run 3 of the same FETs for the highside, your total switching losses actually go up instead of down like they do with the conduction losses. That's why you generally always see designers use lower current rated FETs with higher Rds(on); those ones have a lot less input capacitance so your switching losses are lower. It's also why you see arrangements like this one, with one highside and two lowside FETs.


Pulling up some calculations I did on the BFL PSU at 1V and 100A output using the new "cool" BSC014/BSC0902 pair (which is what's in my Single), the conduction losses per lowside FET were 0.52W, or 3.12W for all 6. The conduction losses in the highside FETs is 84mW each and the switching losses are a relatively massive 2.2W per device or 13.5W in total. At 100W output, the BFL PSU (at least on paper) would dissipate 16.6W in the FETs and 3.25W in the inductors, so about 20W in total, or 83% efficient.

As an interesting note, if they would have used a transistor better suited to the highside with a lower Ciss like the BSC052N03LS, those hot highside FETs might run a lot cooler. Those would have 0.93W of switching losses each, and with 1 the conduction losses would be 1.64W each, with 2 the conduction losses would be 0.41W each, and with 3 the conduction losses would be 0.18W each. If you ran with two highside BSC052's, not only would your per device dissipation go from 2.28W to 1.34W, but since you only have two of them the total highside dissipation would drop to 5.35W from 13.5W. You should give that a shot next time you're replacing the FETs on a Jalapeno.

[lightfoot]

This is why Mr. Teal builds boards and I fix them. Excellent explanation, and thank you.

C

[freddyfarnsworth]

complicated What is so cool tho, is the end result is even more info for a learner like me.

Thank You Mr Teal...

In a year or two I will understand every word.

I gotta get one of these machines..

[lightfoot]

However I will say this: The BFL units are putting a fair bit more than a few watts of heat from the FETs. There's also the issue of imbalance due to thermal conductivity issues at work here.

Personally I prefer IGBTs, or single power FETs.

C

[freddyfarnsworth]

However I will say this: The BFL units are putting a fair bit more than a few watts of heat from the FETs. There's also the issue of imbalance due to thermal conductivity issues at work here.

Personally I prefer IGBTs, or single power FETs.

C

Can you show the difference in pix ?

IGBTs, or single power FETs

What they look like.

All I know is the old NPN sets from old motherboards.