Dabs
Legendary
Offline
Activity: 3416
Merit: 1912
The Concierge of Crypto
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 06:13:18 PM |
|
The part about defective chips, or replacing individual chips intrigues me. If Klondike is a 16 chip board, and 1 or 2 fail or die, there should be a way to identify which one, and there should be a way to replace only those. Sockets are an option. Desoldering and soldering on a replacement chip can also work (all these cell phone electrician repair guys can do it.)
If you can identify or single out a chip, that would be great. If it's physical damage, that should be easy enough. If not, you're going to play mastermind or jigsaw on your board trying to figure out which ones are good and which ones are dead.
|
|
|
|
jorgebuck
Newbie
Offline
Activity: 15
Merit: 0
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 06:18:19 PM |
|
Wouldn't it just be the one not giving off the same amount of heat?
|
|
|
|
loshia
Legendary
Offline
Activity: 1610
Merit: 1000
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 06:21:36 PM |
|
The part about defective chips, or replacing individual chips intrigues me. If Klondike is a 16 chip board, and 1 or 2 fail or die, there should be a way to identify which one, and there should be a way to replace only those. Sockets are an option. Desoldering and soldering on a replacement chip can also work (all these cell phone electrician repair guys can do it.)
If you can identify or single out a chip, that would be great. If it's physical damage, that should be easy enough. If not, you're going to play mastermind or jigsaw on your board trying to figure out which ones are good and which ones are dead.
+1 I do think that 2 dbg pins are for that but we have to wait for docs. It will be great if we can do it. I do not know how chips are tested in FAB but it can happen that some small amount are "dead born" or damaged during the assembly. We have to have a way to diagnose this
|
|
|
|
volosator
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 06:37:57 PM |
|
Wouldn't it just be the one not giving off the same amount of heat?
Chips can fail in a number of different way. There is simply no binary state for this (1 for working, 0 for not working). For instance, chip can give out same amount of hashes therefore showing similar heat signature, but producing only invalid shares? I don't have access nor funds to get high-res thermal camera! I'd like to see small LED right next to the chip that shows it's logical state, LED should be controlled by software and used as ID when you looking for it. A couple of control points, like ones on Raspberry Pi right next to each chip is another good idea. Socket mount approach also sounds very good. That would save a lot of time looking for qualified tech in the area (I recently had control board repaired on my HVAC unit, a tech claimed he had 30 years of experience of repairing PCBs; He could f%ck up a cup of coffee - how skilled he was). Yes, it will add cost in the beginning, but it has a potential of home replacements in the future. Holes for heatsinks mount, 2 per each group of 4 chips would be another thing - what if you want to overclock the chips assuming you can deal with heat? Talking about overclocking, there is got to be proper power reserve on board for this purpose.
|
|
|
|
loshia
Legendary
Offline
Activity: 1610
Merit: 1000
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 06:45:00 PM |
|
Socket mount approach also sounds very good. .....
+1 But such pcb socket could cost 100+USd x 16 chips...we can only dream. And we will say goodbuy to cooling as long chip is contacting socket only not pcb. pcb socket is good solution for a test single chip board or something like this. I like the led idea in general for production PCB for sure. We have to be able to identify broken/defective or malfunction chips soldered on production PCB. Bkkcoins can you put a comment here. 10X
|
|
|
|
|
loshia
Legendary
Offline
Activity: 1610
Merit: 1000
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 08:48:18 PM |
|
You made my day ! Maybe i can get 6 meters raw profile, that would bring the price down to 3-5 EUR per 100x100 heatsink. You are talking about TO3 package which ha better thermal resistance right. Only holes are needed correct 10X
|
|
|
|
BkkCoins (OP)
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 09:02:39 PM |
|
What if a chip burns out? How do I replace a chip without damaging the rest of the board? Will board support overclocking the chips?
Your local appliance repair electrician will be able to do it or put you in touch with a local SMT guy who can. I'm interested in the overclocking aspect too, I would imagine a socketed clock chip will be the way to experiment with this. I dare say there will be threads popping up on here when we get to that point... There is enough extra power on board that it should handle 300 MH/s, beyond that I don't know. No need for changing crystals. Well, there isn't a crystal but an oscillator package. But besides that there is a PLL that multiplies the 32MHz to 256 - 300 MHz on chip (or at this point I believe there is since we don't have docs). This is a good thing as running a 300MHz clock all over the board needs more care than running a 32MHz clock. I've completed the conversion to Kicad now. It looks ok and I have 4 layers so I'm starting to work with that. The conversion was a major pain because the program I used buggered up a lot of stuff that I spent hours fixing and re-trying. It worked fine initially but when you need annotation to work so that changes can be brought from schematic to board it was such a fail. I manually fixed this all up and have tested annotations. Fingers crossed.
|
|
|
|
Knecke
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 09:11:02 PM |
|
You made my day !
Maybe i can get 6 meters raw profile, that would bring the price down to 3-5 EUR per 100x100 heatsink. You are talking about TO3 package which ha better thermal resistance right. Only holes are needed correct 10X For both profiles, yes hole patterns are needed. the profiles have a thermal resistance of ~ 1K/W, and with a fan you can lower it to 0.5K/W. I think i could produce 5,000 - 10,000 heatsinks, in this case the hole patterns, a week if there is demand.
|
|
|
|
loshia
Legendary
Offline
Activity: 1610
Merit: 1000
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 09:17:03 PM |
|
You made my day !
Maybe i can get 6 meters raw profile, that would bring the price down to 3-5 EUR per 100x100 heatsink. You are talking about TO3 package which ha better thermal resistance right. Only holes are needed correct 10X For both profiles, yes hole patterns are needed. the profiles have a thermal resistance of ~ 1K/W, and with a fan you can lower it to 0.5K/W. I think i could produce 5,000 - 10,000 heatsinks, in this case the hole patterns, a week if there is demand. Count me in i will need about 130 as a start i hope ![Wink](https://bitcointalk.org/Smileys/default/wink.gif) and i am in EU so shipping should be no issue Thank you very much! PS: Bkkcoins do you think it is a better idea to mount two additional heatsinks on top of dc/dc converters also?
|
|
|
|
ewibit
Legendary
Offline
Activity: 2955
Merit: 1050
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
April 30, 2013, 09:19:04 PM |
|
I think i could produce 5,000 - 10,000 heatsinks, in this case the hole patterns, a week if there is demand.
+1 ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif)
|
|
|
|
scotjam
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
May 01, 2013, 01:10:40 AM |
|
I'm very interested in this - please put me on the list
|
|
|
|
BkkCoins (OP)
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
May 01, 2013, 01:22:47 AM |
|
PS: Bkkcoins do you think it is a better idea to mount two additional heatsinks on top of dc/dc converters also?
I don't think that will be needed or of much benefit. Like the ASICs the buck regs all have large thermal GND pads under them to make contact with the PCB. I'll add thermal vias under to help heat get to the back side. One thing I'm thinking is that it may be necessary to put a thin copper shim under each ASIC / PWR area to mate the exposed copper better with the heat sink. Or maybe better, thermal compound. The reason is that I won't be able to use blind vias for the other layers without greatly increasing the board cost. So the bottom will have solder mask in non-contact areas, which may cause non-optimal mating of the heat sink, to ensure the exposed vias don't contact the heat sink. This is similar to the thermal pad you see on CPU heat sinks (though in that case there are no vias to worry about).
|
|
|
|
sveetsnelda
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
May 01, 2013, 02:13:28 AM |
|
Also very interested. I'd need 40+ boards.
|
14u2rp4AqFtN5jkwK944nn741FnfF714m7
|
|
|
samurai1200
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
May 01, 2013, 04:18:20 AM |
|
PS: Bkkcoins do you think it is a better idea to mount two additional heatsinks on top of dc/dc converters also?
I don't think that will be needed or of much benefit. Like the ASICs the buck regs all have large thermal GND pads under them to make contact with the PCB. I'll add thermal vias under to help heat get to the back side. One thing I'm thinking is that it may be necessary to put a thin copper shim under each ASIC / PWR area to mate the exposed copper better with the heat sink. Or maybe better, thermal compound. The reason is that I won't be able to use blind vias for the other layers without greatly increasing the board cost. So the bottom will have solder mask in non-contact areas, which may cause non-optimal mating of the heat sink, to ensure the exposed vias don't contact the heat sink. This is similar to the thermal pad you see on CPU heat sinks (though in that case there are no vias to worry about). Couple other ways to go about this that I can think of right away: 1) Raised mating surface on the heatsink, such that the rest of the heatsink sits above the pcb. HASL finish and put thermal compound, that will gain you a few extra mils of clearance. 2) HASL finish + thermal pad, like http://www.digikey.com/product-detail/en/T68-2A-120-90-0.08/1168-1758-ND/3460915
|
|
|
|
salty
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
May 01, 2013, 08:36:30 AM |
|
Socket mount approach also sounds very good. .....
+1 But such pcb socket could cost 100+USd x 16 chips...we can only dream. And we will say goodbuy to cooling as long chip is contacting socket only not pcb. pcb socket is good solution for a test single chip board or something like this. I like the led idea in general for production PCB for sure. We have to be able to identify broken/defective or malfunction chips soldered on production PCB. Bkkcoins can you put a comment here. 10X You'd use an oscilloscope for finding the fault. Really, if you can't do this stuff properly, find a *good* tech locally who doesn't charge the earth, it's going to be better than putting a partially populated board in a skillet and cooking all those valuable asics, it's common sense. I'm really not trying to talk down to anyone! f.e. I build my own audio gear and service a nearly priceless vintage sound desk, and I know my limits, I'm taking this project to the local SMT guy.
|
|
|
|
Bicknellski
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
May 01, 2013, 09:38:40 AM |
|
You made my day !
Maybe i can get 6 meters raw profile, that would bring the price down to 3-5 EUR per 100x100 heatsink. You are talking about TO3 package which ha better thermal resistance right. Only holes are needed correct 10X For both profiles, yes hole patterns are needed. the profiles have a thermal resistance of ~ 1K/W, and with a fan you can lower it to 0.5K/W. I think i could produce 5,000 - 10,000 heatsinks, in this case the hole patterns, a week if there is demand. Nice.
|
|
|
|
sensei
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
May 01, 2013, 12:21:49 PM |
|
There, I'm finally on. This board is exactly along the lines of what I had imagined. I have 500 chips on order and plan on more. Regarding the use of a socket for the chip is a non starter. The chip has a thermal ground pad underneath that must be soldered to the board. This area should be filled with thermal vias to aid in heat transfer to the opposite side of the board to the heat sink. https://en.bitcoin.it/wiki/File:Avalon-A3256-Q48-foot.pngI would use thermal tape along each heat rail (row of chips). The heatsink I think should be milled so that it is off the surface of the board between the heat rails. There is no need to transfer heat to the remaining board area.
|
|
|
|
BkkCoins (OP)
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
May 01, 2013, 12:52:01 PM |
|
There, I'm finally on. This board is exactly along the lines of what I had imagined. I have 500 chips on order and plan on more. Regarding the use of a socket for the chip is a non starter. The chip has a thermal ground pad underneath that must be soldered to the board. This area should be filled with thermal vias to aid in heat transfer to the opposite side of the board to the heat sink. https://en.bitcoin.it/wiki/File:Avalon-A3256-Q48-foot.pngI would use thermal tape along each heat rail (row of chips). The heatsink I think should be milled so that it is off the surface of the board between the heat rails. There is no need to transfer heat to the remaining board area. Agree with all that. I was planning to put an exposed pad under each Quadrant of ASICs approx. 30mm square. So there would be 4 per board and then 2 for the buck regs which would be smaller. This method keeps the heat sink just slightly clear of the board using either tape or compound. I think we can avoid milling if the tape pad is thick or if a thin copper shim is adhered to either surface. The problem with using the entire surface, other than heat transferring from ASICs to other components, is that it would demand blind vias for the internal layers which is costly and best avoided IMO.
|
|
|
|
sensei
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
May 01, 2013, 01:37:31 PM |
|
I'd like a controller with a CAN bus. We can put a 3 pin header on board that will allow us a communications channel for debugging.
|
|
|
|
|