8
\$\begingroup\$

see this question.

OP wants to breadboard a PC motherboard design, which is, er, funny, but raises the question: how are motherboards designed? Breadboarding, with incremental improvements, is out, you have to build a PCB from the start. How are these evaluated and debugged? I guess you can't simply tamper with it, nothing would work anymore. Is this really first-time-right?

\$\endgroup\$
4
  • \$\begingroup\$ I don't know, but I think it heavily relies on analog and RF circuit emulations. \$\endgroup\$
    – jippie
    Apr 15, 2012 at 16:49
  • \$\begingroup\$ @jippie: without a doubt. But do I have to see it as: simulate, design, manufacture? Most commercial designs I'm aware of go through a few design iterations before they're ready for manufacturing. \$\endgroup\$ Apr 15, 2012 at 16:59
  • \$\begingroup\$ Adherence of copper to the circuit supplied is of course covered by design rule checks. "Correctness" largely relates to timing issues. Modern PCB suites have extensive emulation of physical parameters and the integral ability to route to balance or optimise eg differential timings or circuit impedance included so the testing and layout aspects are integrated. I know somebody who does motherboard development Depending on the answers received I could pass the question on. \$\endgroup\$
    – Russell McMahon
    Apr 15, 2012 at 17:00
  • 1
    \$\begingroup\$ PC motherboards are somewhat of a special case. Because they are produced in volume and in rather fixed configurations there is a large effort by CPU and chipset manufacturers to ensure that the routing is as straightforward as possible (to require as few PCB layers as possible). While a PC is very complex, the end goal is known to everyone at every step along the way. PCBs which serve some novel purpose are much more complex to design and verify. \$\endgroup\$ Apr 15, 2012 at 21:58

1 Answer 1

15
\$\begingroup\$

A while back I hired an EE straight out of college. He asked me something like, "how do you prove out your designs? Do you breadboard them? Wire-wrap? How?"

My answer was, "Well, we build a PCB and if it works we go into production and ship it!"

Many circuits simple cannot be prototyped without making a custom PCB. Anything high-speed, low-noise, high switching currents, etc. are difficult or impossible to make any other way. Sometimes the logistics alone get in the way. You cannot, without a custom PCB or some crazy adapter/socket, prototype with a 1000+ ball BGA. Even if you could, you wouldn't want to. The odds of making 1000+ connections without error is very low. Doing it twice is almost impossible.

Making custom PCB's for prototypes seems expensive, but it isn't. Not compared with paying someone to make and debug several prototypes using breadboards or wire-wrap and even then ending up with something that is of questionable reliability.

So professional EE's usually go straight to designing a PCB and then doing everything they can to make sure that PCB works the first time. It almost never works the first time, but the closer it is to working the better it is. Then the board is modified (a.k.a. re-spun) and built again. With any luck a PCB goes through 2 or 3 respins before going into commercial production.

I recently did a custom PCB that used an Intel Atom PCB and all the usual PC type things. My approach was no different with this PCB. I designed it, tried very hard to get it to work the first time, and built it. The Rev 1 PCB's mostly worked, but has some minor issues. The Rev 2 board worked perfectly.

\$\endgroup\$
2
  • 5
    \$\begingroup\$ How do you find the cause of a not-working board? On high density designs probably half of the signals are not reachable (BGAs, inner layers...). How do you measure things like SSN? \$\endgroup\$ Apr 16, 2012 at 6:52
  • 4
    \$\begingroup\$ @FedericoRusso Debugging a board like this is very difficult, which is all the more reason to try very hard to get it right the first time. As long as you are not using blind/buried vias then all signals are reachable. Even so, you have to plan ahead with debugging in mind and possibly change the circuit accordingly. For SSN, pay very close attention to decoupling caps and power/gnd routing-- then hope you don't have a problem. SSN is normally only an issue on wide addr/data busses, so you output a worst case test pattern and look at the signal with a fast scope. \$\endgroup\$
    – user3624
    Apr 16, 2012 at 15:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.