Topic: DIY PCB with AVALON: "The Quarter Stick" - Needs Help! - page 8. (Read 89497 times)

I'll join others here in saying that it's going to be much more cost effective to use a switching regulator with integrated MOSFETs. The NCP3170 looks appropriate.

As for C6, C7 and L2, these appear to form a low-pass LC filter. The corner frequency of such a filter is 1 / (2 * pi * sqrt(L * C)), and the filter is second-order. So for example, if L = 100 nH and C = 44 uF, the corner frequency is 75 kHz. This means the 642 kHz input ripple current is reduced by a factor of (642 / 75)^2 which is about 70. How low do you want the input ripple current to be?

We could just keep it them in a vat of liquid nitrogen, no?

Of course its possible.... but how big do you want your PCB to be? The problem here is that each avalon chip essentially generates 2 watts of heat... that heat has to go somewhere. The current design has chips mounted opposite of heatsinks, which lots of copper & vias thermally connecting the two.

I imagine produce development will follow that of GPUs... as we (the third party manufacturers) are designing upgrades to the reference design (cooling, power management, etc), Avalon/others hopefully will be designing their next gen chip that runs faster and cooler.

And so it will be until the end of time.

Out of interest... it looks like the Asics and hashrate in network will grow rapidly in the near future. So the miner will become bigger over time. So i wonder is it possible to create a pcb with maybe 100 avalons on it? Maybe 200, on each side of the pcb? I only mean that the future probably will move into that direction. Simply because 2,5GH will mean nearly nothing at some point.

So will this be possible? I believe the demand will be high from the start. And production most probably will be way more cost efficient.

Seeking PCB designer TODAY:
https://bitcointalksearch.org/topic/wtb-want-to-buy-2-custom-pcb-designs-187408

Announcing a new thread to discuss my project below:

Codename: Klondike. https://bitcointalksearch.org/topic/klondike-16-chip-asic-open-source-board-preliminary-190731

I've just found this Eagle to Kicad converter.

It worked ok on my board (without having thoroughly checked everything yet). It converted the footprints and board together.

I also have a schematic converter I used that I had to update a bit. That's worked too.

So I'm thinking I'll go forward on Kicad with 4 layers and see how it works.

For anyone who might be interested in Kicad make sure you haven't missed the libraries at http://smisioto.no-ip.org/elettronica/kicad/kicad-en.htm. I haven't used them yet but there are decent 3D models there for many parts that could be used to render a 3D preview.

As of Eagle 5 (have not tried with 6), you could edit and save (but not run any CAM jobs) on a larger-than-free-version board if it was created by a licensed version of Eagle.

My main career has been systems software dev. I'm retired now living in Thailand. I've only ever done electronics for personal interest. I have a PIC programmer and tools for that so will do the software.

If you wanted to use the full Eagle to do a 4 layer design based off what I've already started here I could easily put my work on my github and you could fork it and work on it. I'd only ask that you keep it open and contribute back. Given what I've done, unless there's glaring faults, I think the 4 layer work won't be too difficult. It'll be a lot easier with layers each for 1.2V, data, 3.3V, GND.

I may still do a Kicad 4 layer version but mostly because I like to do this stuff, and be able to edit, which I would lose with not having the full Eagle.

PM me to discuss if you like.

Ya, saw that. I was going to move the caps and maybe push the ASICs in towards the center a bit. But looking at what I'll need for getting data connected I'm not sure I'll proceed. Another idea I had was removing 2 chips per quadrant, leaving a 4 chip board. While this would not be a final board it could allow me to get a board working @$25 useful for software debugging and then finish with a 4 layer final design.

My approach was to get a board done and order 10 ($25). Even if it turns out to be no good for final mining due to 2 layer power problems I could use it to develop and test the PIC code.

I've also thought about what I could stick into the ASIC spots that could act as a simulator. Maybe a small 8 pin PIC hand wired to the pads. It could rx/tx dummy data. The key is to debug as much as possible before real ASICs arrive, even if it involves funky workarounds. Since we have to wait so so long I want to be confident that when they get here I can drop them in and be running.

If youre trying to learn KiCAD, I would just re-route the board to learn the tools. Anyway, I see a problem with your 1.2V rail. You have a choke point for core current going to the upper left quadrant chip.



Couple Q's:
Who will be doing the software for your project?
Do you need help with the PCB layout? I'll be doing my own designs and I have a full copy of Eagle, as well as a slew of my own test and measurements tools. If you need I can help you out, could be mutually beneficial (information for services).

I'm skeptical too. But I'm exploring it fully before giving up. I've actually done one quadrant of 4 chips so far, and it was difficult. The next step of running data signals is pretty dubious and I'm feeling 90% likely I'll end up with 4 layers.

My two motivators were cost and not wanting to learn Kicad. I have to stick with no-cost software and Eagle doesn't allow 4 layers. I've converted my schematic to Kicad but in order to convert the board it looks like I need to adapt the ULP myself. For me boards would cost $2.50 each 2 layer vs. $9.00 each 4 layer (qty 10). I want to spend the minimum up front on prototype boards.

Here's screen grab of the one quadrant I've done, 1.2V in bright red.
(I know there are some fixes to do).

Limited power from a USB port. 2.5W tends to be the most (5V @ 500mA) you can pull from a single USB port. There isn't really a hard upper limit, but I believe 500mA is the lower limit.

Very good aproach for core power; I will take a look. I´m thinking about the Murata DC-DC converters because I don´t wan to lost time in PSU issues. I prefer to work in the protocol and mining software when the Avalon specs arrives. I´ll use the same PIC as you in order to work with the same software. Thank you for share your thoughts

This is a long thread, not sure if anyone mentioned this but:

with the single chip honey bee USB's being made, any thought on maybe sticking 4 Avalon chips on a USB PCB stick?

a lot of us will be getting chips so why not make some 1Gh USB sticks?




 

I'm skeptical that you're going to be able to put 16 chips on a 100x80mm board @ 2-layers. It's going to be very difficult to supply/return 24A around the board properly with signaling and +3.3V chopping up your copper pours. I'm thinking 4-layer will be necessary for such a design.

My 16 chip board will be using 2 x IR3895 16A buck regulator ($2.87) for 1.2 supply. And an MCP16321 3.3V 1A buck regulator ($1.55). Both available from Mouser.

There is a MCP16322 2A buck regulator that would be ideal for a small 1 chip design along with an LDO linear 3.3V reg. I wouldn't go with a 32 bit PIC for the 1 chip design. It's overkill in what it needs to accomplish. I'd use the small/cheap PIC16F1454 USB 8-bit, 14-pin ($1.15) for this.

I'm using it's big brother, the PIC16F1459 20 pin ($1.62) for the 16 chip board but only because it has enough pins that I figure whatever the Avalon docs tell us it will likely have enough to work. I also use 2 pins for fan controller, though that may go unpopulated. The PIC has a temperature sensor on board.

I only pull 12A from each IR3895 keeping it in the more efficient operating area and not pushing it's heat dissipation. It's possible to push the board to 20 chips but it pushes the reg limit and I'd have to move to a board size beyond what the free Eagle allows. I'm also looking at Kicad for a 4 layer design if my 2 layer board proves fruitless. Right now I have the 2 layer board figured out but I consider it a bit non-ideal. The main problem has been getting 3.3V in to the chips without cutting off the main 1.2V supply. I've managed this but some pins don't have as large supply traces as I'd prefer.

I wouldn't limit yourself to hand soldered parts for this. Just read up and learn the home reflow techniques. You're going to need that for the ASIC anyway and most buck regulators have ground pads for heat dissipation. It's also feasible with solder paste and SMD rework station (hot air) but much easier and more reliable with an oven/hot plate.

You are right, 1.5A is very tight. I have looked some other converters but they are more expensives. for example:

http://www.ti.com/product/lm20343

The app notes have the PCB layout too

I have used monolithic linear power ICs (www.linear.com) in some of my proyects but they are a more expensives too

I like the idea of a 10 chip design and the power supply module.
Still need 3.3 volt supply of some sort.
The 16 amp 1.2 volt module might be a little marginal if we try to push the operating frequency.
Is a PIC chip necessary ? Might it be possible to just connect the 10 chip miner with a level translator to a standard RS232 COM port ?
Keep the good ideas coming!