Bitcoin Forum

So, there's sensible use of time, then there's this. Test socket for bm1385 chips. Not quite sure if anyone else has done this yet, they're only 0.4mm pin pitch... still not sure if it will work. The 'connector' is most of an hdmi mini plug. The tricky bits are going to be local decoupling, xtal and and the power connection, I think. That and the 1.8v IO voltage, but I think I have a trivial solution to that (at least for testing).

https://i.imgur.com/yrrWbzO.jpg

Anyone else been down this rather painful route before?

Cheers,

Allan.

They do make actual QFN test sockets.  Kinda pricey, especially for 0.4mm, but if you lurk on eba+aliexpress long enough one should come along for under $50.  Or else make your own using ENIGold PCBs from eg OSHPark (https://oshpark.com/).

But... why?

I mean are you going to desolder all the chips from your PCB, test them individually, then solder them back on?  The SMT pads are incredibly fragile, even a very skilled operator with the right equipment will damage the board ~5% of the time in a multi-cycle rework operation like this.  Multiply that by 45 chips/board and you're not likely to end up with any more working boards than you had to start with.

All? No, of course not, that would be silly :)

However, the nice people at Bitmain put nice, high-current connectors on the back of each chip in the chain. So, power up 1 set of 3 (the set at ground - .666v). If it starts hashing at the right rate (you can also test clocks, which appear to come up with the 3.3v supply) those are not likely to be faulty. Power up the next set, and so on. Move on down the board until you get a failure. At that point, you can guess that one of those 3 chips is faulty.

When you know the faulty set, you can follow the clock through and isolate the individual faulty ASIC or board section.

Desoldering them on the target board isn't too bad - just don't try and remove until the solder is fully molten (you don't care about the chips you are removing here). You do need to care about the decoupling caps, but you can test those after you have the chip back on and the heatsinks still off.

The Bitmain boards are one of the best I have worked with for a long time - the board its-self is quite high-temperature capable. This would all be easy if it wasn't for RoHS :)

The source board is a problem, since you DO care about the chips - so you need to heat up using a quite controlled profile, but even so, as you said, these chips are fragile. Hence the need for a test socket...

This is assuming that the level shifters never fail, but I've not seen that yet.

I'm intending to write all this up and probably Youtube it, not that it helps many people as you've said - since it requires quite an insane level of luck, skill and equipment to do. That is, if I get time to do it, which is in rather short supply.

I certainly see why Sidehack doesn't want to touch these chips. You wouldn't think that .1mm is all that different, but... Ugh. And that damn black goop!

Cheers,

Allan.

Of course, there is always the option of https://shop.bitmain.com/productDetail.htm?pid=00020160416053348855GEgP3cx0067E (https://shop.bitmain.com/productDetail.htm?pid=00020160416053348855GEgP3cx0067E)

Which I didn't realise at the time.

Well heck, that almost makes it too easy.

$1.8 for a single S5 chip? Wow.

S7 chip, but yeah. That is an impressive price, given they wanted $2.50 for BM1384 at quantity 100k. I guess they must have a reel of old stock laying around.

Certainly is. Also didn't realise that the BM1385E+ was a thing...

Currently going through the nightmare that is salvaging chips and managing to knock off some of these 0402 caps. The feeling when you correct a tombstoning one with just the right application of a pair of tweezers is amazing :)

On another question, if you're willing to answer, with the compac / 2pac - I was wondering if the driver you've been using happens to support the BM1385... ;)

Cheers,

Allan.

A lot of the register and protocol info was gleaned from BM1385 datasheet, so a lot of it should work right off. If I remember right from kernel logs there's some differences in chip detection polling. Why, you thinking of rigging up a USB adapter to drive board sections for faster troubleshooting?

Don't talk to me about tombstone. About a third of the lost time I've had with 2Pac manufacture was fixing tombstoned or dry-pad 0603 parts because the minion took an unreasonably long time to figure out how to paste properly or keep the pick-and-place calibrated. And then sometimes forgot how to do things right. Yesterday's average across six panels was about 8% of the total small parts that needed manual fixing once it came out of the oven. Pretty sorry for a production line. I'll be happy if I never see another tombstone, but hey at least they're not 0402. I refuse to work with parts that small.

Exactly. I have a couple of old slightly modified (okay, hacked apart and overclocked - kis3r33 modules were a godsend) Eruptors which I'm now willing to salvage - with the current difficulty, they can't seem to even reliably report enough diff 8 shares to register a chunk of their hashrate. Not that it's really a big proportion of my mining these days ;)


Once you get the process down, it's not too bad. But seeing a whole batch needing rework must be slightly disheartening. Hopefully said minion volunteered?

I sent out a call for help and he's the only one that responded so I didn't have much choice. Yesterday's problem was he forgot to clean solderballs (coming through vias from the ASICs on the backside) off the PCBs before stenciling, so the stencil was raised off the board and paste came through incredibly gloopy. Lucky he didn't tear up the stencil any more than he already has. Unlucky that he did six of them and noticed the crap job but didn't try to figure out what was causing it. His inability or unwillingness to think through things is the biggest disappointment, especially for someone two months from graduating with an Engineering degree.

I prototyped a BM1385 Compac last year but between not knowing if the chip was any good, not having a decent driver and not really wanting to work with epoxied-on 0.4mm pitch parts it was readily abandoned.

Ugh. Laser cut Kapton stencils work quite well and are cheap enough to be mostly disposable, don't know if that's an option for you. Well, at least he won't make that mistake again.

And on the fresh chips plan, it would seem to be a failure.

"Can not buy internal product ×"

Well, thanks, Bitmain for allowing Google to index that part of your site, then. I was all excited for a minute :(

Kinda figured that would be the way of it. If you're in need, I have a board here that was working up until it fell off a shelf and busted some chips off, so the remaining 40 or so are probably good. I hot-air'd the chipside heatsinks off but there's still that bastard black epoxy to deal with.

...
Tell me about it...
After decades of me being the um, only technical brains, of our company we finally hired another EE last year. I'll give him an 'A" for enthusiasm but as far as any understanding how to actually apply whatever he was taught in college...  ::) Then there is the issue of hopefully passing on to him at least *some* what I've learned over the past 1/2 century +...

First thing I did was give him my ancient dog-eared copy of The Radio Amateur's Handbook I got in 1965 when I was 9 to fill in much of what should be Electronics 101 but was never covered in college.

I should know better than to expect much from a guy in his twelfth semester for one degree. In the same time I got two engineering, one science and almost a couple nontechnical minors. Half my senior design class had never even used a soldering iron. After two months this guy is just almost worth the $8 an hour I'm paying him. Almost.

NotFuzzyWarm, I would love to hang out in your shop for a while and just absorb. Wish I could afford to employ an old salt who knew his business but y'all always want stuff like pensions and insurance and right now I'm just trying to keep the lights on.

Speaking of keeping the lights on.... sidehack have you got a paypal donations address? I sold my non working green compac to Dibblah to try and fix/tinker with and im happy to donate the amount to whatever fund is in most need. Its not much but its something. Please PM me if you have a paypal address for donations.

Sidehack appears to have done essentially all of the work which enabled me to be able to even think about repairing these boards.

It's just a hobby for me, but as gt_addict says, if there's a paypal address, I will send a little bit your way.

Also, I should take up hobbies which take less effort and bending over a damn paint stripper because my hot-air station broke. ;)

In other news, today's reflow didn't work. Definitely the second chip dead (good clock in, nothing out). Still not sure if it's an "I broke the chip" or not getting a good suck-down onto the board. Those thermal vias make the ASIC behave unpredictably when soldering - they're even worse than power QFNs because the pad is so damn large.

Also, WHY is their placement outline off by one row of pins? That's just... Argh.

Cheers,

Allan.

Check the 2Pac sales thread for address anything.

The bitshopper sticks I'm working on now, I went with a pretty cheap board house because it's such a small batch. The silkscreen registration on them is terrible, pushing a millimeter off on both axes. Bitmain is known for maximizing profits, they probably opted for the lowest bidder who may not care so much.

Wait, what work did I do that did anything for an S7 board? Only S7 work I did was PIC firmware for volt setting.

Trawling through a large number of threads gives me the power layout, the VCCIO layout and the general 'chain' strategy. Most of the interesting ones I remember were your posts :)

Even though it was bm1384 stuff, it still seems to directly apply (apart from the integrated LDO)

Just to confirm, I thought I'd ask -

Yes, i agree with mjosephs,  I'd rather make pcb from a pcb manufacturer (https://www.seeedstudio.com/fusion_pcb.html) than buy from ebay or aliexpress.

The real problem is re-placing QFN chips just sucks. With BGA you have nice little balls that you can place on the pads, preheat, heat, and they drop with a nice little thud, but QFN is a crapshoot. Either there is too much solder on the pads, not enough, almost enough and you have to drag solder around the edge with an iron, and then you pull up a pad and life just loses meaning.

Bigger question: Is it worth it? Replacing the QFN48 power chips on a KNC controller board is worth it (although I prefer swapping the BGA FPGA by a long shot and have never screwed one up) but is it worth it for a S7?

Yep. And the leadframe on this damn chip is quite thin, so there's essentially no wicking or heating from the top. Hey, everyone needs a hobby ;)

Biggest problem I have at the moment is I have no way to prove that the chip I'm putting in worked _before_ I touched it.

Again, these PCBs are quite good - I haven't seen a delamination without extreme heat being applied. And the "interesting" pads have external access - technically you can rework, but it's a real pain at .4mm.


I agree completely. It's not a business, it's a hobby. The only bit that's easy and worthwhile is Frankensteining the 14v supply (boost modules on ebay are really cheap and it doesn't appear to be a high current or particularly sensitive rail). Also, why the heck does that area of the board always corrode?

The 14V only sources power to local LDOs on the top couple nodes; the rest are fed by Vcore from nodes higher up the chain. Those LDOs provide 1.8V for control and IO and 0.9V for the PLL. Total current requirement per chip is almost nothing, on the order of 10mA. On the 2Pac I use a resistor divider for the 0.9V and it works fine. As long as your boost rail is at least about 1.5V higher than the Vcore of the top node (since ground reference there is about 0.7V lower and a good 1.8V LDO will work at 2.1-2.2V) and it can source ~100mA it should work. You're right, that is easy.

So...

Because everyone deserves to laugh at someone else occasionally... I present to you, the worlds least efficient stickminer.

33w at the wall, one linear supply powering a SMPSU powering a single bm1384. Hash rate of 11Gh/s or so at an approximate efficiency of...

0.03GH/J

Err... There is a point to this, trust me.

https://i.imgur.com/saIXivj.jpg

Right. Finally got another few minutes to look at this.

https://lh3.googleusercontent.com/LwFGnYJ_ZAJJGS9bmSvpmBLZ9qW1FbnyvMxRueCCxtyzT4lIeKMiOsSUi-KrQEfeQyqSgGxXJapAdLthqfcl8HTkMmYQrDynrQZmvvfHn0SM_6d2Vi7nS3Ec9KIAL7S2BdQoqltYEOnSexAFkJG_NPmxSqfAMUUWCId4rfFhbNduOsLUrAVYe5b8m8Favk8VyJ1uiuR4sA_D7K66_uHVl5jxdlPXsLGEttCKUlRitZF95hRnxnGi0-_NEVXEf_XjKLdgzoNMg-H5mHSBGgFnmDxsnG6i-Ma39NasX8AoBUjJRbV9_F2_LPYcAiE_9rXF2kP3dSnU4wPF4yG8KMKUEIPHpeCM4ua9T7oRqvS8szWguzA2fQvNdhhaWPto_F8TIy2iz6NGM2XU03i24TKGJIWWSEBlMEzv8pT7G7Xg-1PFCOnynjR-xaVNVZbQqqMZ-pNgKmwknlYCJHS9pbjzxwXxpRQn_KQeLOLS6H1-5eNdEfM933-MT0ylnDMe_vwloIBjI9A0zI-1UotaFc6xKQ2w_FIBp99qYqeM0r39CPLBmJ6GbOSXeJKSrjAD81NGf3413SnEeDzFdKdYzqtRwygBcrtZblTYN5-flav-v3wL9OZhDqk7H6CsA1zDzr3KzflXHcYtRDEJtXXGY5juZcSPuROw_1GtbutSu-4_HQ=w361-h641-no

CP2104 does 1.8v IO voltage natively. Of course, it's never that simple - the dongle I bought bridges 3.3v and the IO voltage under the chip. Aaaargh.

Reflow, slice, reflow, kynar, variable IO voltage yay.