Hacking a BFL Jalapeno to 32GH and beyond....(???) - page 5.

MrTeal

legendary

Activity: 1274

Merit: 1004

Quote from: lightfoot on January 21, 2014, 06:11:59 PM

Sort of. Here is one of the chips with the top... Removed.

Note that the curled up thing on the top is the chip itself, with little tiny solder dots on it. Also note the mask that connects it to the BGA and the smaller than hair *wires* that go from the chip to the BGA board.

What's happening here is that heat from the top does not flow through the chip carrier; it gets stuck in the top chip. And heats the solder balls under it to the point where they expand and explode out the sides of the chip (seriously I had a little ball of solder on top of the chip). I can't believe that most BGA chips have lots of little wires in there; maybe they do.

The new model seems to be the chip right on top of the BGA bottom. Heat flows through, there doesn't seem to be solder inside, and they never give me a problem.

I think I need to do this with a home-made reflow oven and not air heat. I can only get the bottom up to 375 or so; one option would be to direct the 400c air heat to the bottom of the board under the chip and try to bring that region to 400+. Or something. But man is it a problem to work with these chips.

C

I'm reasonably certain those wires are the top traces of the substrate, and they just lifted off when you overheated it.

Edit:

lightfoot

legendary

Activity: 3094

Merit: 2239

I fix broken miners. And make holes in teeth :-)

Sort of. Here is one of the chips with the top... Removed.

And here is a closeup with a 6x loupe in front of the camera:

Note that the curled up thing on the top is the chip itself, with little tiny solder dots on it. Also note the mask that connects it to the BGA and the smaller than hair *wires* that go from the chip to the BGA board.

What's happening here is that heat from the top does not flow through the chip carrier; it gets stuck in the top chip. And heats the solder balls under it to the point where they expand and explode out the sides of the chip (seriously I had a little ball of solder on top of the chip). I can't believe that most BGA chips have lots of little wires in there; maybe they do.

The new model seems to be the chip right on top of the BGA bottom. Heat flows through, there doesn't seem to be solder inside, and they never give me a problem.

I think I need to do this with a home-made reflow oven and not air heat. I can only get the bottom up to 375 or so; one option would be to direct the 400c air heat to the bottom of the board under the chip and try to bring that region to 400+. Or something. But man is it a problem to work with these chips.

C

MrTeal

legendary

Activity: 1274

Merit: 1004

Quote from: lightfoot on January 21, 2014, 04:34:39 PM

Wrecked another old-style chip in testing; even with a completely perfect reballing and placement, the board was shorted out, and the problem was the chip. I could see a tiny amount of solder around the top of the chip. So I broke the chip open.

These chips are way, way different from what I thought. I was expecting a simple interface, but there are tiny little wires in there going to a chip assembly that is even *smaller* than the BGA. It's like a pico-BGA or something.

I have no idea anymore what BFL was thinking. But a warning: It is not possible to drop the chips into place by using heat from the top; I will have to explore using heat from the bottom if even possible.

Very interesting.

C

I'm not sure what you mean there. The shiny part is the actual silicon die, and it has little solder bumps that attach it to the substrate which is just a tiny PCB. That's the bottom of the BGA. On another note, the new packages and substrates that BFL uses are amazing. The old style (below) is a pretty typical BGA substrate

The new package is altogether different, and looks incredible. I'd really love to see actual test results, showing the performance differences between them.

lightfoot

legendary

Activity: 3094

Merit: 2239

I fix broken miners. And make holes in teeth :-)

Wrecked another old-style chip in testing; even with a completely perfect reballing and placement, the board was shorted out, and the problem was the chip. I could see a tiny amount of solder around the top of the chip. So I broke the chip open.

These chips are way, way different from what I thought. I was expecting a simple interface, but there are tiny little wires in there going to a chip assembly that is even *smaller* than the BGA. It's like a pico-BGA or something.

I have no idea anymore what BFL was thinking. But a warning: It is not possible to drop the chips into place by using heat from the top; I will have to explore using heat from the bottom if even possible.

Very interesting.

C

helipotte

hero member

Activity: 650

Merit: 500

Pick and place? I need more coffee.

Quote from: MrTeal on January 21, 2014, 12:08:45 AM

Quote from: lightfoot on January 20, 2014, 11:43:50 PM

Quote from: helipotte on January 05, 2014, 02:11:27 PM

Lightfoot, I just got 4 jalapenos and I have gotten the hackit-to-go-faster bug. Already changed out the 7k resistor with a larger one to get more voltage to the

ASICs. This reduced the errors by 1% and increased the speed by 500Mhs. Thinking about putting maybe two more chips in each unit. Do you know where the early type

Speaking of which, what resistor are you swapping out to boost the power, and do you have a digikey/mouser part number? My 8 chip jally is actually humming at 32gh/45c due to the water block and now I'm thinking..... Just a bit.

C

I don't know if there's been a new update that's changed the designators, but on the old boards R11 and R12 controlled to voltage. The formula is Rtop(R11, 7k) = Rbottom (R12, 10k) * [(Vout - 0.6V) / 0.6V], or Vout = (Rtop/Rbottom+1)*0.6V
Currently it's set to 1.02V, so if if you want to change it to 1.1V change R11 to 10k*[(1.1-0.6)/0.6] = 8.33k. Not a standard value, but 8.2k would work well enough.

That's it. 8.2K is what I have been using. I get at least another 1Ghs out of a jalapeno.

Unrelated, but I have a Single with a chip that shows all zeros in bfgminer. Chip 0d. Been that way since I got it. is there a way to turn it on? Is it just bad? Can the chips be disabled in hardware? This unit already hashes at 69Ghs but having that chip online would be nice. Any ideas? Huh

I have actually "looked" inside and counted 16 chips so I'm sure one is just turned off/dead. Thanks for any insight.

MrTeal

legendary

Activity: 1274

Merit: 1004

Quote from: lightfoot on January 21, 2014, 12:50:55 AM

Quote from: MrTeal on January 21, 2014, 12:08:45 AM

Quote from: lightfoot on January 20, 2014, 11:43:50 PM

Quote from: helipotte on January 05, 2014, 02:11:27 PM

Lightfoot, I just got 4 jalapenos and I have gotten the hackit-to-go-faster bug. Already changed out the 7k resistor with a larger one to get more voltage to the

ASICs. This reduced the errors by 1% and increased the speed by 500Mhs. Thinking about putting maybe two more chips in each unit. Do you know where the early type

Speaking of which, what resistor are you swapping out to boost the power, and do you have a digikey/mouser part number? My 8 chip jally is actually humming at 32gh/45c due to the water block and now I'm thinking..... Just a bit.

C

I don't know if there's been a new update that's changed the designators, but on the old boards R11 and R12 controlled to voltage. The formula is Rtop(R11, 7k) = Rbottom (R12, 10k) * [(Vout - 0.6V) / 0.6V], or Vout = (Rtop/Rbottom+1)*0.6V
Currently it's set to 1.02V, so if if you want to change it to 1.1V change R11 to 10k*[(1.1-0.6)/0.6] = 8.33k. Not a standard value, but 8.2k would work well enough.

Hm. I may try this on my 5 chip jally (the one I just can't seem to add chips to). What types of resistors are those little things (soic or something); I feel weird wiring a 1/4 watt axial resistor in there. :-)

C

They're just 0603 resistors according to the Altium files.
http://www.digikey.com/product-detail/en/RC0603JR-078K2L/311-8.2KGRCT-ND/729777
http://www.digikey.com/product-detail/en/RC0603FR-078K45L/311-8.45KHRCT-ND/730339

or just get a kit

http://www.ebay.com/itm/New-SMD-SMT-0603-1-144-values-resistor-kit-144-X-100pc-14400-filled-BOX-ALL-/170869001211?pt=LH_DefaultDomain_0&hash=item27c8960bfb

lightfoot

legendary

Activity: 3094

Merit: 2239

I fix broken miners. And make holes in teeth :-)

Quote from: MrTeal on January 21, 2014, 12:08:45 AM

Quote from: lightfoot on January 20, 2014, 11:43:50 PM

Quote from: helipotte on January 05, 2014, 02:11:27 PM

Lightfoot, I just got 4 jalapenos and I have gotten the hackit-to-go-faster bug. Already changed out the 7k resistor with a larger one to get more voltage to the

ASICs. This reduced the errors by 1% and increased the speed by 500Mhs. Thinking about putting maybe two more chips in each unit. Do you know where the early type

Speaking of which, what resistor are you swapping out to boost the power, and do you have a digikey/mouser part number? My 8 chip jally is actually humming at 32gh/45c due to the water block and now I'm thinking..... Just a bit.

C

I don't know if there's been a new update that's changed the designators, but on the old boards R11 and R12 controlled to voltage. The formula is Rtop(R11, 7k) = Rbottom (R12, 10k) * [(Vout - 0.6V) / 0.6V], or Vout = (Rtop/Rbottom+1)*0.6V
Currently it's set to 1.02V, so if if you want to change it to 1.1V change R11 to 10k*[(1.1-0.6)/0.6] = 8.33k. Not a standard value, but 8.2k would work well enough.

Hm. I may try this on my 5 chip jally (the one I just can't seem to add chips to). What types of resistors are those little things (soic or something); I feel weird wiring a 1/4 watt axial resistor in there. :-)

C

MrTeal

legendary

Activity: 1274

Merit: 1004

Quote from: lightfoot on January 20, 2014, 11:43:50 PM

Quote from: helipotte on January 05, 2014, 02:11:27 PM

Lightfoot, I just got 4 jalapenos and I have gotten the hackit-to-go-faster bug. Already changed out the 7k resistor with a larger one to get more voltage to the

ASICs. This reduced the errors by 1% and increased the speed by 500Mhs. Thinking about putting maybe two more chips in each unit. Do you know where the early type

Speaking of which, what resistor are you swapping out to boost the power, and do you have a digikey/mouser part number? My 8 chip jally is actually humming at 32gh/45c due to the water block and now I'm thinking..... Just a bit.

C

I don't know if there's been a new update that's changed the designators, but on the old boards R11 and R12 controlled to voltage. The formula is Rtop(R11, 7k) = Rbottom (R12, 10k) * [(Vout - 0.6V) / 0.6V], or Vout = (Rtop/Rbottom+1)*0.6V
Currently it's set to 1.02V, so if if you want to change it to 1.1V change R11 to 10k*[(1.1-0.6)/0.6] = 8.33k. Not a standard value, but 8.2k would work well enough.

lightfoot

legendary

Activity: 3094

Merit: 2239

I fix broken miners. And make holes in teeth :-)

Quote from: helipotte on January 05, 2014, 02:11:27 PM

Lightfoot, I just got 4 jalapenos and I have gotten the hackit-to-go-faster bug. Already changed out the 7k resistor with a larger one to get more voltage to the

ASICs. This reduced the errors by 1% and increased the speed by 500Mhs. Thinking about putting maybe two more chips in each unit. Do you know where the early type

Speaking of which, what resistor are you swapping out to boost the power, and do you have a digikey/mouser part number? My 8 chip jally is actually humming at 32gh/45c due to the water block and now I'm thinking..... Just a bit.

C

lightfoot

legendary

Activity: 3094

Merit: 2239

I fix broken miners. And make holes in teeth :-)

Sounds good. I just posted the pics on the other thread. I think this might be possible to do, at least to 4 more chips. This is going to be fun; I'm going to screw up a perfectly working single.

Anyone from BFL still read this; if so are there any schematics for the bigger single boards around?

C

Drug5bitz

full member

Activity: 123

Merit: 100

Love imgur.com

It is my love for photos and memes. Great community or just great photo hosting. Share your stuff to the world or keep it relatively close to yourself.

ManeBjorn

legendary

Activity: 1288

Merit: 1004

I use photobucket.
I don't now though if it will do what you need or not though but it may be worth a look.

Quote from: lightfoot on January 20, 2014, 03:28:28 PM

Ok, I'm building a thread for the single. With a speed of 8 I have it running at a solid 28gh, and with a resistor in line with the big fan I have the noise level down to quiet with a temp of 65 on the board. Not too bad. The trick is to keep the fan over the chips running at full speed; the end fans just need to get the air out of the box and around the heat sinks.

https://bitcointalksearch.org/topic/m.4624265

On a side note it looks like Flikr is being an ass and doesn't want to give me the raw embed URLs for my photos. Anyone got a better site to share photos?

C

lightfoot

legendary

Activity: 3094

Merit: 2239

I fix broken miners. And make holes in teeth :-)

Ok, I'm building a thread for the single. With a speed of 8 I have it running at a solid 28gh, and with a resistor in line with the big fan I have the noise level down to quiet with a temp of 65 on the board. Not too bad. The trick is to keep the fan over the chips running at full speed; the end fans just need to get the air out of the box and around the heat sinks.

https://bitcointalksearch.org/topic/m.4624265

On a side note it looks like Flikr is being an ass and doesn't want to give me the raw embed URLs for my photos. Anyone got a better site to share photos?

C

vapourminer

legendary

Activity: 4326

Merit: 3519

what is this "brake pedal" you speak of?

Quote from: lightfoot on January 18, 2014, 12:52:51 PM

Right. Also remember that with FETs the loads are not spread symmetrically. The ones I have sinks on tend to be hotter on 4 chip diamond designs, the other FETs really don't get very warm because they are not carrying too much of the load.
...
That little heat sink on the center chip is there for the FET gate driver control; it's a high frequency chip that is driving the gates directly for all FETs.

ah shiny. I was worried they changed it; apparently not.

with the hot mosfet units Josh recommended heatsinking the 1850 anyway as its runs hotter in that design. when I flashed it to 8.2 I just put a ubar fan on it and called it a day. been 6 months so its worked heh.

Ill order a hot air rework station, some heatsinks, and some chips and start practicing then.

godsdog

newbie

Activity: 17

Merit: 0

Quote from: vapourminer on January 18, 2014, 10:19:02 AM

Quote from: godsdog on January 17, 2014, 10:02:02 PM

exhibit B: 4 chip jalapeno 2 installed by lightfoot with heatsinks on 2 power FETs,1 gate driver(copper)

http://i.imgur.com/fYexzqJ.jpg

there are 6 MOSFETS under the 2 bigger copper heatsinks shown on the above image (not two as quoted above)

fixed it

godsdog

newbie

Activity: 17

Merit: 0

bfgminer stats

On a side note ♪ I changed power supply from bfl replacement 13v 6amp to 12v 10amp overnite from my last post as the heat on the psu was uncomfortable to me probably 130º f+
with the 12v it, a little more than piss warm.
also running it sideway's with cheap usb laptop cooler blowing the bottom plate it get's 5 º cooler it varies from 55 c to 64 c but will stabilize to 55 c if irun it without interuption....
http://http://i.imgur.com/00plNq1.png

lightfoot

legendary

Activity: 3094

Merit: 2239

I fix broken miners. And make holes in teeth :-)

Right. Also remember that with FETs the loads are not spread symmetrically. The ones I have sinks on tend to be hotter on 4 chip diamond designs, the other FETs really don't get very warm because they are not carrying too much of the load.

Once you get to 6 chips you will start to load up the other set of FETs, at which point you put sinks on everything. Get to 7-8 and you had better have sinks on both sides of the board; they do wick a lot of their heat into the board itself.

That little heat sink on the center chip is there for the FET gate driver control; it's a high frequency chip that is driving the gates directly for all FETs. As you put more and more power on the board, the driver on that ship has to work harder to switch the FETs. So putting some cooling on it with anything over 4 chips is probably a good idea, very important with 6-8 chips.

The right way to design the board would have been for that to drive a 2708 FET driver/op amp, that is what we use for IGBT circuits. But that's ok, we do with what we have.

C

Drug5bitz

full member

Activity: 123

Merit: 100

Quote from: vapourminer on January 18, 2014, 10:19:02 AM

Quote from: godsdog on January 17, 2014, 10:02:02 PM

exhibit B: 4 chip jalapeno 2 installed by lightfoot with heatsinks on 2 power FETs,1 gate driver(copper)

http://i.imgur.com/fYexzqJ.jpg

hmm. this image does not bode well for upgrading my jally. my board is different(?). there are 6 MOSFETS under the 2 bigger copper heatsinks shown on the above image (not two as quoted above) on mine. and 6 more along another side.

mine is a very early jally, board says BitFORCE SC rev C.

I think you might be overreacting as even though you can't see it there are 6 fets under those two sinks. Every board i have seen or heard of have 6 in on corner and another 6 in the other corner. The difference is the "Hot mosfet" design where they're all the same size. I'm assuming he's summing up the sink count. Also all of my boards including the blackfriday special are rev. c just my observations.

vapourminer

legendary

Activity: 4326

Merit: 3519

what is this "brake pedal" you speak of?

Quote from: godsdog on January 17, 2014, 10:02:02 PM

exhibit B: 4 chip jalapeno 2 installed by lightfoot with heatsinks on 2 power FETs,1 gate driver(copper)

http://i.imgur.com/fYexzqJ.jpg

hmm. this image does not bode well for upgrading my jally. my board is different(?). there are 6 MOSFETS under the 2 bigger copper heatsinks shown on the above image (not two as quoted above) on mine. and 6 more along another side.

mine is a very early jally, board says BitFORCE SC rev C.

ManeBjorn

legendary

Activity: 1288

Merit: 1004

Awesome!!!

Quote from: godsdog on January 17, 2014, 10:02:02 PM

A satisfied lightfoot client here! submitted for your approval :
exhibit A:Ezminer stats
exhibit B: 4 chip jalapeno 2 installed by lightfoot with heatsinks on 2 power FETs,1 gate driver(copper)

2 morrow will post bfgminer stats with rma power supply 13v 6a ,going to run my other 4.2gh jalapeno with this !....2nite

thank's lightfoot u da man Cool

Topic: Hacking a BFL Jalapeno to 32GH and beyond....(???) - page 5. (Read 54272 times)