Small-scale BGA soldering

Do you try to do it on your own? How do you know all the balls are soldered correctly? X-ray checking is not a problem if the production scale is high, but what do you do in the case of prototypes or batches of

Reply to
Piotr Wyderski
Loading thread data ...

Hardly anybody x-rays in production. We have an optical inspection gadget, but that's more for process monitoring than for inspecting each unit.

formatting link

BGAs turn out to be far more reliable than leaded parts. So we mostly assemble and test.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

This sounds like a "young man's" job. Somebody with steady hands and perfect eyesight.

Or... and older engineer with a ready bottle of aspirin. :)

Sounds like a headache in the making to me. Whenever possible, I prefer to prototype in through-hole. (Ack - I'm a dinosaur!) :)

Reply to
mpm

The parts are getting so small that everyone needs good optical aids.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

I do too, depending of course on why I'm doing the proto. Sometimes you have to do real SMD prototyping, like that 120-ps single-diode TDR that ChesterW and I did early this year--0402s and 0603s in mid air. Good for the patience, that. ;)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

Eh, green with envy...

Far mor reliable even though there is no thermal stress relieve mechanism? Electrically far superior than anything else, but mechanically it is a disaster waiting to happen. IMVHO.

Best regards, Piotr

Reply to
Piotr Wyderski

But in real life, BGAs are amazingly reliable, far better than leaded parts, both at initial production and in the field. Our BGA per-pin initial failure rate is ballpark 1 PPM, and I don't think we've had any field failures. I didn't anticipate that myself.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

Good to know, maybe I should give them a chance. So now a practical question: IIRC you are using ZynQ in your devices. Do you design your own PCBs for them or use the ready-made modules like the Zeds? If the former, then what number of layers do you consider to be the bare minimum to route all that insane CL* package signals reliably? Can you do it on 4 layers?

Best regards, Piotr

Reply to
Piotr Wyderski

On Thursday, October 12, 2017 at 12:36:28 PM UTC-7, Piotr Wyderski wrote: ...

ts.

One of our designs did have trouble after several hundred cycles that requi red a package modification.

kevin

Reply to
kevin93

we make our own and use 10 layers, I doubt 4 layers are realistic, before you done anything you need 3+ different supplies, and a bunch length matched wires to the DDR RAM

for low volume I think it is had to beat a module

formatting link
a module with 1GB RAM, flash, ethernet, etc. for $99, the Zynq alone at digikey cost +$60

Reply to
Lasse Langwadt Christensen

Holy Cow, assume there was no question... Thanks, all!

The problem with modules is that they contain a lot of things you don't need, but do not contain the ones you really do need, so the resulting system is a Frankenstein wired of random boards. But if you say 10 layers, then well, what can I do..?

Those are nicely stuffed, but like the Zeds, they are more dev boards than embeddable modules. Trenz also has a lot of interesting modules, but for some reason he doggedly pursues only the 4x5 form factor and I simply have no room for a horizontally-mounted module. So for me this seems to be the only game in town:

formatting link

but I am concerned about the long-term SO-DIMM contacts reliability.

Best regards, Piotr

Reply to
Piotr Wyderski

Depends very much on the size, and whether you use underfill or not. With good thermal management and good underfill, you can make flip-chips of 20-22 mm square without ripping the corner leads off the base level metal (BLM). Organic mezzanine packaging makes this easier.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

We've done both, a couple of uZed boxes, and ZYNQ on our own board.

formatting link

The board I'm working on now is 10 layers. We were hoping for 8, but the data lines out to a *lot* of fast parallel DACs couldn't be routed on 8.

formatting link

That's a ZYNQ 7020, 484 BGA, with a 500 MHz DDR DRAM. It boots Linux off the SD card in about 1 second. It worked first board spin... I was shocked.

It might have worked on 8 layers, but might have taken another month of layout.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

We did a single DDR ram on our ZYNQ board. One DRAM doesn't need the awful address and data line terminators and associated term supply that two or more chips would need; I think we only terminated the single clock pair.

I could share those schematic and layout snippets if anyone is interested.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

Underfilling and similar thermal stress management techniques are acceptable if your name is Samsung or another IBM. Otherwise how do you know you're doing it properly? I'm afraid that a similar phenomenon to the fabless semiconductor companies is about to happen: smaller shops will soon be unable to manufacture their own designs and centralized assemblers will emerge, similar to what TSMC is in the silicon world.

Best regards, Piotr

Reply to
Piotr Wyderski

Smaller shops probably don't need BGAs with thousands of balls.

Normal BGAs with a few hundred balls don't need anything so elaborate.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

Fun one.

formatting link

--
Best Regards, 

ChesterW
 Click to see the full signature
Reply to
ChesterW

At 120 ps, 0805s or 1206s would be fine.

I bet the schematic is equally ugly.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

es you

ood

Absolutely, except in a 10-pF single-diode sampler with an 0402 Schottky di ode.

You'd be wrong about that. The sampler is the little bit centering on the s ilver semi-rigid coax. All the schmutz around the outside is support stuff-

-the Tx and sampling pulse generators, delay scanner, and so on.

The actual sampler is nice and clean, has an RMS jitter less than 10 ps end

-to-end (0.1 inch gauge length on an air line), and costs $2.

Plus you're just crabby about not getting to Truckee for the weekend. ;)

Cheers

Phil

Reply to
pcdhobbs

That is no justification for such a hideous breadboard. If I was that ugly, I wouldn't work at all.

Here's my old 2-diode sampler. 70 ps, 5 GHz, hand etched.

formatting link

I recently did a PCB of a TDR based on a d-flop 1-bit sampler [1], but I haven't had time to fire it up. I'd expect maybe 50 ps, after a bit of digital processing.

I think Joerg did a single-diode TDR recently, too.

With a bit of deconvolution math, all you need to do is make a really ugly sampler and, as long as part of the step response is fast, you can beautify it in the PC.

Oh no, I much prefer staying at home in the smoke, repairing the rot on the deck, standing on a ladder ratcheting lag screws over my head.

[1] sort of like the dflop phase detector.
--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.