Quote from: sidehack on May 15, 2015, 12:14:17 AM

So, general testing update.

I was worried about comms - I've never been fully satisfied with the UART level shifter I had on there. And based on I think on vs3's comment that USB lines could be finicky, I was a bit skeptical about comms. So I peeled off the CP2102 on a stick I had working moderately, and ran it off the Novak USB adapter because that had proven itself on the breakout boards. Same performance. Rigging up a diode-drop level shifter for the RX line (to bypass the actual level shifter chip) gave me a bit better logic level hysteresis (and I might end up using it on the final since it's simpler, cheaper and more resilient than the level shifter chip) but the same overall performance. So then I says, well what if it's regulator noise hosing up the chip? I was seeing about 200mA draw being pulled by the whole hub and occasionally around 400Mh reported speed with sporadic shares reported, so obviously something wasn't right - I should have seen 600mA draw at least at the numbers I was running. But the breakout board, with identical circuitry, worked just fine - except its regulator was more distant than and also orthogonal to the ASIC, which was itself under a large aluminum heatsink.

So I yanked an inductor on a moderately-functional board, wired up some pins to drive it off my test schfifty-three board (the one rigged up for the dual-chip testing with the original breakout boards) an inch or two away (on the other side of a ground plane from the ASIC) and BOOM within a second the chip was too hot to touch, current meter was pegged and I saw about 8GH reported on cgminer.

So it looks like I need to do some shifting around with the regulator circuit and see what might be done about isolating RFI from the ASIC. An initial test will probably be to stick a grounded metal shield between the inductor and the ASIC. Past that, maybe drill out the inductor pads and mount it on the back of the board, to see if putting it on the other side of a ground plane from the ASIC will do the trick. If necessary, we can make a double-sided board with the ASIC and its heatsink on the back. This might help make it a bit more compact, and though double-siding sucks we need to get good at it anyway for the TypeZero boards. Hopefully sandwiching the ASIC between ground planes and a heatsink will be enough to RF-isolate it from neighboring sticks' inductors in a fairly dense hub setting.

It's getting on quitting time now, but I might have more news tomorrow.

Edit - so shielding added a bit more stability, not a lot. I went ahead and stuck on a heatsink all proper, and it basically did things just as well as the shield. We noticed a lot of ground noise on the board that wasn't present with the external regulator attached. Tomorrow I'll probably drop on a higher inductor and see if the reduced ripple current cuts down on interference. If something like that won't solve the problem, we gotta go for a board redesign with better ground plane islanding - probably in conjunction with a larger inductor. The U2 inductor is 4.7uH to our 0.47uH and it's pretty much right up against the ASIC with no performance problems. We couldn't do with a 4.7uH inductor though, since we need to be able to handle probably 10A there aren't gonna be many options which are power-efficient and also not huge or expensive. But I'll test with one anyway, stolen off a roasted Block Erupter since we have a bunch of those. Should be good for 3A, which will do 600mV 125MHz testing.
The thing is running off the schfifty-three power board right now, and its own stock heatsink, on Eligius at the 1BURGER address. Should be at 125MHz churning out around 7GH. I'm really hoping I don't have to do another board design, because that'll shift things back another two weeks or so.

WOW!!! Things are heating up Grin

I personally don't mind if it lasts a month longer or not and i believe most contributors in here share my voice Cool

Keep it on sidehack yeeeaaahhhh!!!