About motherboard building

[ktalap]

Hi, y'all. That's a question to people who have built a motherboard for ARM, RISC5 or x86 architectures before. Where do I start, besides me being an ECE major, and what should I focus on? The computer architecture books, while relevant don't seem to hit in the bull's eye.

[tggzzz]

Make sure you understand "signal integrity" and "power distribution network" theory and practice.

Those topics are too large, complex, and subtle to explain in a forum thread. There are many books and articles on the subject. Consider starting with Bogotin's Rules of Thumb on EDN.

[ktalap]

THANK YOU!

[Andree Henkel]

I´d advice to select a "daughterboard" which come in several formfactors, and contain the CPU, RAM, Flash
this components are today mostly BGA, very fine pitch, Multi layer boards >=6 Layers, a lot of nets that need length matching, diff pairs and maybe controlled impedance

and design not a "Mother"-board, but a "Carrier"-Board containing Power and Connectors / Peripherie to outside which can be 4..6 Layers and Compnents with pitch that can be still handled maanually if you got steady hands.

[Doctorandus_P]

Start by figuring out what you want to build, and whether you can also buy the actual parts you need. Not all parts are available in small quantities and for "reasonable" prices.

I'd say have a look at Olimex. They both sell ready made ARM / Linux SBC's, have the complete design open sourced as KiCad projects, and they also sell bare Allwinner chips if you want to make your own. At least start by downloading the PCB project from gitlab and studying it in KiCad. It gives you a good Idea of the complexity involved, and whether it's a task you want to do yourself.

[mctaylor]

Where do I start[?]

I would start with building a simple embedded system with a low-speed 8-bit microcontroller, such as one from the Microchip (formerly Atmel) AVR family. This is on the order of complexity of an Arduino Uno, but with you designing its support circuitry and interfaces to the real-world.  This task can be as simple or complex as you want. You could make a simple clock for $10-20 USD of parts. But it should be a complete thing. A clock, a kitchen timer, bat detector, daylight recorder. It needs to have a purpose, and to be able to fulfill that purpose.

From there you could move onto a single board computer (SBC) using a low-speed 8 or 8/16-bit microprocessor, such as the Western Design Center's (originally MOS Technology) 6502, or a Zilog Z80 family or derivative (Z8, Z180). As many 8-bit microprocessors are available as a through hole part, if you wanted to, you could build it on solderless breadboards. It is also simple enough to be debuggable without a large array of expensive high speed test equipment.

Then I'd ensure you are familiar with high speed design, power distribution, and EMC (electromagnetic compatibility).

Pick a processor family, then a particular device, read its datasheet, sketch out the block diagram of the support circuitry, buses, memory map. Then design, simulate, build, and debug those circuits. Write some firmware.

[ktalap]

So it's better to design a carrier board which should not be that large (smaller than 10cm^2) and will be made on 7-10 layer PCB. Then the rest of the functionality can be implemented on a 2-6 layer PCB, to which we mount the carrier board. That's a much better approach than having a whole motherboard have 8 layers. I did some research designing motherboards is not that hard I think, it's like lego.

[ftg]

Robert Ferenec has a nice 1hour and 52 minute interview video with Istvan Nagy about designing motherboards:

[tggzzz]

I did some research designing motherboards is not that hard I think, it's like lego.

It very much depends on the processor, the number of pins, the PSU requirements, and particularly the edge speeds. Arduino != FPGA/x86

[Andree Henkel]

ktalap, you clearly underestimate the complexity.

My advice was: you buy a daughterboard and try to design a carrierboard.

But ofcorse you can go route as suggested by others: use a µController-Board or designing a µController onto a PCB as well

[ktalap]

Yeah, I did some research again. Making an actual motherboard is like 200k or more USD. Even if you bring it down to 100k, you can still buy like 4 really powerful double CPU servers (256 core, 512 threads) with that. It doesn't make any sense to create your own one, unless you are actually in the business of motherboards. I mean, yeah, you will have constraints, but it's actually not even a problem, unless you are trying to build a neutrino-atomic computer that is a size of flash memory stick and have the power of RTX 4090, did I say that it would be able to fly on its own too?

Now it's another question if you can bring the cost down to 25-50k somehow, that would actually be viable. But bringing the cost down 8-fold, is a lot to ask. Manufacturing these, is not a problem, it's not that expensive either, carrier-board + the rest of the functionality will cost you ~1000-1500$ total a piece, even in small volumes. But the cost of the design, that's where the problem lies. While it's non-repeating cost, you would want to design all kinds of motherboards and improve them iteratively, which is why it has to be low.

So yeah, I will just buy a carrier board from lattepanda or something, when I need custom work. Otherwise, it's even more fun, just buy a server mother board with many PCIe ports and you can do A LOT with them. Additional NVMes or something else, you can design it in such a way that it will have flexible busses going towards places you need etcetera.

Thank you for your advice.