[Letux-kernel] my own JZ4780 board

[H. Nikolaus Schaller]

Hi Ivan,

> Am 04.04.2021 um 20:37 schrieb Ivan Gagis <igagis at gmail.com <mailto:igagis at gmail.com>>:
> 
> Hi Nikolaus,
> 
> cool, so you have done the real working devices which run on > 200MHz frequencies, so you have probably done all that impedance matching stuff etc. You are tough, as you are doing the stuff I gave up on doing :).

Well, it was quite some work. Imepdance matching is a smaller problem the smaller the device is. If component distance is less than 10% of wavelength it becomes more and more invisible.

> 
> You replied to some middle message from the email thread, probably you have seen the whole thread, but just in case, I have opened the sources I had managed to create, those are available here:https://gitlab.com/igagis/cranberrypi 
>  <https://gitlab.com/igagis/cranberrypi>
> 
> As far as I know, EAGLE is not a free software, right?

No it isn't. But the data format is. It is XML. So even in 50 years someone can write an import tool to whatever is needed.

And, when we started (ca. 10 years ago) there was no KiCAD (well I think it did exist but had no functionality). So with KiCAD we would have needed to wait. Usually you need working tools first :)

For the GTA04 and the Pyra projects we did even develop our own CAD-support tool which can read in EAGLE files and present them for much easier on-screen editing as EAGLE does. Finally it is reexported and one can send an EAGLE board file to manufacturing.

Basically this CAD-support tool is another choice besides KiCAD and EAGLE. But it is not a complete solution with schematics entry, placement, component library, gerber export etc. Just some important parts to get an 8 or 10 layer layout done.

> My idea was to work with free software only, so I did not even think of using something like that.

Well, that makes life unnecessary difficult :) IMHO, use a proprietary tool with open data format is equally good. Then, if support is stopped, you can still rescue your data.

> Perhaps, it is possible to export schematics from KICAD to EAGLE and do the PCB routing in EAGLE. But I'm not sure how much sense there is in that :). And I'm not sure that the schematics I developed are good, it was my first take on this kind of systems.

Usually such projects run in iterations.You develop initial schematics. While doing the layout you regognise that some chips should better be replaced. Or pins should be connected differently. Or you find that there is no space for all the blocking capacitors you have planned for. Then you update schematics.

At some stage, you have a first layout which passes all design rule checks. Then you build a prototype. Just to recognise that two pins are swapped in the footprint definition. Or that an input is an output. Or you did not recognise the importance  footnote on page 3241 of the technical reference manual.

Then, you update schematics and layout again for a next version...

Therfore it is not a smooth process doing schematics and layout in different tools.

> Though, I had designed much simpler devices based on ATmega8 and ESP8266, and those projects succeded.
> 
> As for 4 layers vs 10 layers, I started doing the MPU->DDR routing and doing it with just 4 layers didn't look impossible to me, though far not easy, of course.

Yes, the pure DDR routing is 4 layers. But to get them connected to the omap it did need laser drilled vias. This can not be done on 2 layers on one side and 2 on the other. They must all have 100µm distance.
For symmetry reasons you have 4 + 4 layers. Which is good because you usually want to connect all the gpios etc. which cross the DDR chips. So we ended up in 5+5 layers for the Pyra processor board.

> Also, the JZ4780 design guide from Ingenic mentioned that minimum number of PCB layers is 4, so it should be doable with 4 layers.
> Also, my understanding is that their reference jz4780 board uses 4 layers, but I'm not sure about that.

It is difficult to find out without damaging the board. X-Ray might be an option...

> Why I wanted to use jz4780 ? Well, I must admit, it was only because it has MIPS architecture,  I thought that there is a lack of MIPS devices in single board computers market sector right now, so the idea was to fill that gap, possibly. Haha, big plans :P.

:)

> Of course it was just a hobby, I did not plan to earn anything on that, nor that I planned to achieve some practical tasks, so buying CI20 is not an option. The thing was in doing own single board computer.
> 
> It does not look like I will get back to that project. I still have 6 pieces of jz5780 chips lying at my electronics parts storage, so in case someone plans to do something with jz4780, I could possible share those 6 pcs. :D I also have a bunch of DDR3 chips and some other stuff I was buying for the project which I won't most likely ever need now.
> 
> Thanks for mailing to me, I'm still subscribed to the CI20 mailing list and it is pleasant to see some email activity from time to time, that there are some people still playing with these, already dead technology, boards :).

Yes, for me it is fun developing and fixing Linux for these and keep it running for other projects.
And, it is a test bed to keep compilers and other knowledge up-to-date.

Also just fun without specific goals for these boards.

What I also had tried is to get PVR/SGX for the jz4780 back and working. And get some parts upstream. But that is an equally big project and somehow spare time is getting scarce more and more.

So I can fully understand that you have abandoned your project. I might have done the same.

BR,
Nikolaus


> 
> Br,
> Ivan
> 
> воскресенье, 4 апреля 2021 г. в 19:19:31 UTC+3, Nikolaus Schaller: 
> 
> Hi Ivan,
> unfortunately I did miss your mail 12 months (minus 2 days) ago...
> You may know that I am the designer of the Openmoko GTA04 board (OMAP3) and the Pyra-Handheld (OMAP5).
> I have done both in EAGLE years ago but I have no experience (and no plans to switch to and invest time) with KiCAD.
> By the way: the jz4780 and the omap5432 (dual core Cortex A15) have the same package size.
> So they are not exactly interchangeable (partially they have different interfaces) but it is not impossible to design a jz4780 based processor board for the Pyra-Handheld...
> Beware: the omap5432 board had needed 10 layers and 4 on the top side had been needed to connect the DDR to the processor. Some others had to be used for wide trace power distribution.
> And the PCBs need laser drilled micro via technology (which is something you can't buy around the corner and has a very high project setup cost - several 1000$ for 1 unit. Or 10. Or 10000).
> Yes, it is challenging but can be done. If money is available.
> Production is also challenging.
> But there I now have equipment for SMD. A small Pick&Place machine and a vapour phase soldering machine.
> I use them for prototyping fine pitch BGA devices. So it is basically possible with this equipment to build some prototype boards.
> The question of availability of chips has already been addressed and it may be possible to get a handful. But AFAIK it is not possible to source them from the manufacturer. For omap5 we can still buy chips from TI (although they are not on public offer).
> A bigger question is how much time (and money) to invest. And there - even where I really like the jz4780 more than the OMAP5 - it is questionable if it is a good idea.
> The jz4780 compared to omap5 is slower (roughly factor 4) - but stays much cooler.
> Both use the PVR-SGX with very problematic setup of kernel drivers and user-space.
> The omap5 has some more modern interfaces (e.g. one USB3 OTG port).
> Nevertheless, both are old designs if we compare to ARM Cortex A73 to A78.
> So for your media center idea, I'd suggest to just buy a CI20 board since it fulfills almost all your requirements instead of designing a new one (even if it is a nice fun and learning project).
> Best regards,
> Nikolaus

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.goldelico.com/pipermail/letux-kernel/attachments/20210405/695ac10f/attachment-0001.htm>