Slashdot Mirror
Motherboard Design Process
MBRFan writes "MBReview has posted the latest revision of their 'Motherboards - The Designing Process' article. This article covers the design process for modern motherboards, and also goes over some of the most common components that can be found. Definite read for information lovers, though beware, it'll take a while to read!"
Seems to be something which I dream of.
Wrong placement of capacitors and stupid placement of ports.
Before I threw out my P2B-DS motherboard when it stopped working. I might have been able to do something about it. :\
if they also gave a little insight into what goes into the manufacturing process as well. After all, you can design the best board there is, but if you have trouble making it in large quantities, it's not going to be successful. Cost and ease of manufacture should be factors in any engineering decision.
Monstar L
I used to design motherboard power system components, and the author spends a good bit of time talking about that. That is actually the most complicated part of the board design, as it is not at all automated. Most component vendors try to sell a complete solution to the motherboard vendor, easing their job somewhat while helping the sales of the component vendor.
One particularly interesting item of note: all those capacitors the author describes are absolutely crucial, and together form one of the largest cost items on the board. The system is tested using a processor-vendor-supplied "load tool" which simulates the worst case load transients one can ever expect to see. Most of that testing is done by the power system component vendor and then provided as a block to the motherboard vendor. Most motherboard vendors have no idea what they are doing there.
In fact, a lot of the foreign manufacturers (no names) cost-reduce their designs by simply pulling out caps until the system blue screens. Then they put the last one back in and sell it. Intel is the only manufacturer I am aware of that actually sells the worst-case performing design.
Note that I am only aware of products related to Intel-type motherboards. I never worked on the othe stuff.
Just my $0.55 (US inflation, 1774-2008, for $0.02)
I've heard stories of crappy motherboards blowing capacitors. Which is why its always better to get the 150 dollar one than the 30 dollar one. Because the 30 dollar one is effectively made out of scrap metal...
All in all a motherboard is a complicated piece of electronics so it isnt surprising that bugs sometimes creep in. As with software I expect it is the quality of the engineers working on it that is the ultimate deciding factor in the quality of the final product.
gmail invite
Every single component on a motherboard can be a magnificient piece of work, but it deals only with a limited number of variables. A CPU or chipset component ultimately only deals with bus protocols and internal design. The same goes for other components like memory or harddrives.
But everything comes together through the motherboard, and that (in my eyes) makes it the hardest part of a system to design well, considering the number of variables. A truly well done MB design is really a piece of art.
Yes... but does your motherboard manufacturer follow best practices like Feng Shui?
Take-off every
LOL, I'd like to see them try and make a single-layer motherboard!
...comes with the wisdom of the Far-East, yessir!
What, you haven't heard of the Feng Shui Motherboard?
I only post comments when someone on the internet is wrong.
That article kicks ass. I feel smarter already. I never knew EMI and line noise was such a problem for MoBo designers.. What a bitch that must be
IANAMBD, but...
... I.e the thickness of the "Prepeg"
"trace" == Metal tracks carrying the current
"Prepreg" == insulating "silk" layer separating the metal tracks running above and below.
"static impedance" == "guess" they mean parasitic capacitances formed between the tracks crossing over each other, separated by the "Prepeg".
"Er, or dielectric constant" == See physics book for parallel plate capacitor, the 'E' is the "greek epsilon"
"height of the trace" == the plate separation
.
The guy who wrote this article has a glittering career in the Patent business.
Graphic cards hitting capacitors.
IDE and Floppy conectors on top of each other.
Fan connectors way at the bottom and not enough of them.
I have to be a Chinese contortionist to pulg things up once the motherboard is installed in a case.
Because stage one of the design process is always going to be "what form factor we gonna go for?"
if it's ATX then 99% of the components can only go where they go, PCI slots have to go where they do, CPU has to go at the top cos pci slots extend to the bottom on cases, ide connects towards the "front" of the pc, psu cables near the top cos psu makers save money by shaving cable lengths, ram is an awkward shape so it can only go where it does, and all the i/o can also only go where it goes.
ok that leaves "some" slack for placing shit like fan headers, but bugger all for power devices like capacitors etc.
it's not like they can rats nest all the components and do auto routing AND auto placement to minimise pcb size or track length.
as another poster says, multi-layering as also inevitable unless you are going to use pc104 style stackable cards to get your extra routes on.
no, sadly too much of the design process is fixed by form factor, and too many mobo manufacturers are thus left with no real design decisions except..
"can we use six 20 cent capacitors instead of twelve 90 cent ones?"
and
"should we dye this one red? or black? or blue?"
end
http://slashdot.org/~GuyFawkes/journal
£12.95 will get you a (Pcchips KT266A SKT A DDR266 ATA133 AGP Sound LAN USB 2.0 ATX Retail Box) ebuyer
This is not a recomendation of this particular motherboard, I simply took the item that was the top of the list. When you see something like this for this price it makes you wonder how they make their money. I know the quality is not great, it will probably die before too long, but thats not the point.
For that price how can someone make a motherboard that works even once?
PC motherboards typically have 4 - 6 layers. That's pretty complicated. Think about routing all of the signal lines of the PCI (64-bit) bus and memory bus. If you have AGP nad PCI-X, add those in ther too. Think of all the signals going from your IDE ports and SCSI ports (if you have them) to your chipset. If you have an opteron, you have in the neighborhood of 940 pins. They all get connected. That's a lot of nets and a lot of routing!
And, it's a good thing to reduce board layers to a certain point because adding layers gets expensive! While it does complicate the process, the process of routing the board is a one time process and thus a one time cost. Once the board is routed and tested, it's off to manufacturing where the cost of extra layers is recurring. It does complicate the manufacturing process in that now you have to glue together more layers after routing them. This is why more layers get so expensive. Those layers have to line up. With nets getting as small as they do, there is almost no room for error in lining these layers up. Very difficult. The more layers you add, the more likely you are to have layers aligned improperly. Someone has to eat the cost of those boards that cannot be sold and it wont be the manufacturer.
Roughly speaking, more PCB layers makes it easier to do a complex design, but harder (read: more costly) to do the manufacturing.
The mobo market is very competitive, so shaving a single $ of the price of manufacturing a board, is profit/market advantage for a mobo maker. So the mobo maker will try to keep the number of PCB layers to the absolute minimum needed for the design. AFAIK, many mobo's use 4 layers.
To save the other metric readers the trouble googling, 1 mils = 1/1000 inch = 2.54e-5 metre. conversion
was that Intel would sell any component vendor down the river AND shoot their grandma to save a nickle per board....
When did this occur? A couple of years ago, there were some bad caps on the market after an employee stole *part* of the dielectric formula when he left one cap company for another. The new company used his process/formula without knowing it was incomplete, an dthe result was bad caps on boards from quite a few vendors.
I don't recall whether the vendor was found culpable, or just the employee.
I know it was covered on slashdot...
And if the author uses the word "basically" one more time, I'll throttle him.
Those who fail to understand communication protocols, are doomed to repeat them over port 80.
one of my friends graduated MIT when i was still a sophmore in HS twiddling with microcontrollers. i asked him what he could do with his CompE degree; he told me could build a motherboard. i was amazed, but then again i already knew i would be a CompE anyways. anecdote over.
with more and more cores becoming more and more embedded, sometimes i think there's a small possibility users might be able to again develop their own systems. alright, so i'll probably never build a motherboard for an intel desktop processor, but perhaps some hyper-evolved ARM or powerpc-derivative isnt entirely out of the question.
one of the classic challenges of motherboard design has always been where do you put all these signal paths. we have two solutions in the pipe aimed squarely at this: PCIe and FB-DIMM, both of which drop channel width dramatically.
i'm just a computer engineer (may '05 baby), they dont teach us much about the black art of high speed signalling, but somehow BGA gives me faith. it may mean you have to have your 6 layer board shipped to you, but that seems like an OK tradeoff for being able to design some amazingly high speed hardware without being a signal-foo master.
of course & as usual, the reality is probably highly departed from my little dream world. but its a nice thought, and i still think it might be possible. at least if i can afford some good PCB design software. that, i believe, will be the crux of the issue.
myren
"PC motherboards typically have 4 - 6 layers. That's pretty complicated. Think about routing all of the signal lines of the PCI (64-bit) bus and memory bus. If you have AGP nad PCI-X, add those in ther too. Think of all the signals going from your IDE ports and SCSI ports (if you have them) to your chipset. If you have an opteron, you have in the neighborhood of 940 pins. They all get connected. That's a lot of nets and a lot of routing!"
That's why you're going to see more and more of a serial design. PCI-X and SATA are more serial than parallel. With moores law and improved manufacturing technology. Having the signals come out in a serial fashion is easier. And it makes board manufacturing, and design easier.
The article starts off interestingly, discussing
the nutty gritty details of designing the PCB.
But as it reaches the area of components, it
becomes a second grade electronics text book.
The author does not give any hint on why you need
so many different capacitors(shapes/sizes), and
why they are needed even if they are simply
paralleled. There is also no context of
noise due to simultaneous switching(except for
the cross talk etc.).Except for the first couple
of pages, the article becomes a definition for
electronic components, instead of how and why
they are used for motherboard design.
For those who does not know a cap/inductor, it
could be a good read, but not for me(being in
the field of VLSI design for the last 14+ years).
The article could be named as:
Basics of electronic components used in a mother
board design and PCB design.
...beware, it'll take a while to read!
Doesn't this kinda rule out the article as a Slashdot topic? I mean, what self respecting Slashdotter does a RTFA anyway?