exxos's DFB1 trials

[exxos]

A bit more progress. Each power pin now has its own capacitor and there is now a Bulk capacitance in the centre of the FPU.

The someewhat bizarre crazy routing is the GND traces I slapped in. This will be replaced with a poly fill, if I ever figure out how to actually do that in KidCad

Capture.PNG (127.12 KiB) Viewed 1426 times

Also with all the oddities with the clock driving and oscillators am thinking of just driving the FPU with a "Tinylogic" buffer. Which will be driven from 5V.

It is one possible reason why the FPU may work better with a external oscillator because the PLD is only driving the FPU from the 3.3V supply. When running at 50MHz speeds that voltage is going to be likely less as well. While theoretically there should be no problem in doing that, I don't want to just leave it to chance that it does not have any sort of effect in not having a proper 5V driven clock.

I am not totally sure what I'm going to do with the master clock which drives the PLD. I think currently it is driven from a 5V oscillators, so moving it to a 3.3V one would match what the TF536 does.

However, the TF536 actually uses a 100MHz OSC. In terms of driving the CPU clock, it goes through a internal FF which would actually buffer the CPU clock. I'm not actually sure if the 50MHz ends up being basically a "pass through" the PLD, like a straight wire in and out ?

I could just add a buffer on the 50MHz OSC anyway and have done with it. Though with timings been so borderline I don't really want to do too much in case it has a detrimental effect. Though I guess I could put jumpers on it I guess. Choices choices..

[Badwolf]

@Badwolf (Or anyone for that matter who may know)..

On the FPU A0 and SIZE are both tied to VCC but via a solder jumper pad. Is there any specific reason for keeping those pads because I don't think anyone ever got the trace ordered any modifications to it ? Am thinking of removing the jumper pads...

The jumpers are there in case the intrepid modder wishes to use a 16 bit bus on the FPU (I wanted to keep my options open) or somebody (ie. me) might want to use the FPU socket for an expansion and wouldn't appreciate a tied-high A0...

For use as designed only, they're not needed.

BW

[Badwolf]

Got somewhat distracted in tidying up some of the other traces because there is a lot of "V" shaped bends which I am also trying to iron out as much as possible. A lot of traces have had angles removed and lengths of the traces shortened. Also there have been a few roots which could easily be optimised. It kind of turned into a bit bigger job than I was planning to do

Hmm, I shouldn't have many v-shapes in signal traces. Unless they were going into pads or vias, possibly. I'm more laissez-faire on the power traces, though.

It looks like the board was originally autorouted or something but I thought Kicad did not have a proper autorouter ?

*super-offended*

I don't think there's any auto-routeing on this board. It's probably just the sum total of my PCB routeing training was a few minutes in OrCAD in 1995 before making a single-sided blinky whilst on work experience at UKC.

BW

[Badwolf]

However, the TF536 actually uses a 100MHz OSC. In terms of driving the CPU clock, it goes through a internal FF which would actually buffer the CPU clock. I'm not actually sure if the 50MHz ends up being basically a "pass through" the PLD, like a straight wire in and out ?

I could just add a buffer on the 50MHz OSC anyway and have done with it. Though with timings been so borderline I don't really want to do too much in case it has a detrimental effect. Though I guess I could put jumpers on it I guess. Choices choices..

The PLD swiches the clock up and down so it's not a 'straight through'.

If you're thinking about duty cycle from the fact it's 3V3 against 5V, then I'd suggest use a 3V3 main oscillator into the CPLD but I don't think there's much you can do on the way out. Whatever you choose there will have a 5V versus 3V3 level sensitivity. TF536 drives alright direct from the CPLD (but, of course, doesn't have the FPU which is perhaps more sensitive).

BW

[stephen_usher]

KiCAD doesn't have autorouting itself but it can utilise other auto-routing programs by exporting and then importing the Spectra Session.

I use the Java based Freerouting which does a good job if you leave it to run for hours to finish, though it can get stuck now and again.

All my Freerouting boards have worked first time (as long as I got the schematic correct!).

[exxos]

If you're thinking about duty cycle from the fact it's 3V3 against 5V, then I'd suggest use a 3V3 main oscillator into the CPLD but I don't think there's much you can do on the way out. Whatever you choose there will have a 5V versus 3V3 level sensitivity. TF536 drives alright direct from the CPLD (but, of course, doesn't have the FPU which is perhaps more sensitive).

Yeah I'm going to swap to a 3.3v OSC. Not sure if it's worth switching to 100mhz ?

Yeah the TF536 does direct drive from the PLD. But would more problematic cpus work with a 5v drive.. we just don't know, same with the FPU. Though I'm leaving no stone unturned with everything FPU related. I want my 50MHz .. minimum

[TotO]

KiCAD doesn't have autorouting itself but it can utilise other auto-routing programs by exporting and then importing the Spectra Session. I use the Java based Freerouting which does a good job if you leave it to run for hours to finish, though it can get stuck now and again. All my Freerouting boards have worked first time (as long as I got the schematic correct!).

The autorooting is good enough for through hole parts. The problem is always the same, they not try to route full buses first and easily fail when using SMD circuits with a lot of pins. Last but not least, the final result is far better when done by hand, the pride too.

[stephen_usher]

Yeah, through hole is easier as you have loads of pre-made vias it can use.

Still, I don't care about pride, I just want it to work and not have to click endlessly making lines where I may miss one.

[exxos]

What's even better than a ground plane ? TWO ground planes!

Capture.PNG (136.47 KiB) Viewed 1332 times

Seems I now know how to do polyfills!

[Rustynutt]

I certainly value bus acceleration over CPU acceleration. If you made me choose between my ct60e and my Phantom, I'd still choose Phantom. There are just too many advantages. Higher video resolutions, higher frequency audio playback in ace tracker, overall higher system performance.

@alexh I don't think there has been any documentation released for how to populate & use the booster section of the ct60e. If there was I wouldn't need to use a separate phantom board.

Noticed looking at my populated CT63, it uses a 40MHz oscillator. I've not used the feature.

I've tested my CT60e and CT63 with a modded bus, the upper limit seems to be 46/23MHz, same as I'm getting with the Afterburner. But that's not changing clocks after the units are up and running, it's from a cold boot.

You are right about the accelerated bus though. A stock Falcon using a 60MHz COMBEL clock is insanely fast, TC mode is as fast as 256. It's only when you really need horsepower the CT is missed.