TF536 + 68000 relocator and ROM board project.

[stephen_usher]

I wonder what would happen if you switch the CPU power rather than control lines.

[derkom]

I wonder what would happen if you switch the CPU power rather than control lines.

This is exactly what I tried when first playing with this idea in the Stacy, and it will not work. You have to switch the bus control stuff.

[derkom]

I was worried about that. I'm guessing the long parallel data lines may be generating cross talk. Also I couldn't work out how to tell Kicad to make the power and ground rails wider.

Someone who actually knows what he's talking about can probably give a better response here, but my understanding is that at least with something as slow as 8 MHz, crosstalk isn't really an issue, because magnetic fields that generate crosstalk are produced only when the voltage is changing, and since all the address and data lines share the same clock, they're all changing at the same time, so any crosstalk is generated only during this transition, but since 8 MHz is fairly slow, the time every line will stay high/low is very much longer than the transition time, and therefore crosstalk is irrelevant.

However, since the 8 MHz clock is a sine, it's in transition all the time, and therefore will generate magnetic fields and potential crosstalk at all times. For this reason, in my switcher design, I've got CLK some distance away from everything else, and when it has to cross the bus, it does so only at 90° angles, where crosstalk cannot be induced.

I have no idea if this design actually accomplishes anything, but my switcher is definitely rock solid. (I have not tried running Elite however.)

If you feel like cutting up one of your boards, you could try rerouting the clock on a bodge wire to see if it makes any difference.

It is probably also a good idea to look at your power situation. The TF536 already has an auxiliary power connector, so it's easy to feed some power in there, but since that doesn't fully fix things for @PhilC, then there's likely something else going on. Interesting that the Frontier crash seems to be repeatable. I've got the Stacy in too many pieces right now to give Frontier a test on mine.

[stephen_usher]

Yeah, your design is far more compact so far less likely to have those sorts of issues.

Once I have the header pins tomorrow I can look at the data/address and other lines on the oscilloscope to see how noisy things are. I'm guessing blind at the moment.

I don't s'pose you know if just turning off the power to the not in use CPU would be acceptable. It would definitely simplify everything and lower the power draw too.

[derkom]

I don't s'pose you know if just turning off the power to the not in use CPU would be acceptable. It would definitely simplify everything and lower the power draw too.

Will not work. See above.

[stephen_usher]

OK, fair enough. I wonder what would happen if you did both. Hmm....

[derkom]

OK, fair enough. I wonder what would happen if you did both. Hmm....

That's not something I have tried, but when I did bring up the power idea in the first place, @terriblefire pretty strongly recommended against it. So I suspect that doing both isn't wise/won't work, but I haven't actually tested it. Should be pretty easy to test if you want, just by bending out the VCC pins on the 68000.

[stephen_usher]

OK, I may change things so that /HALT is asserted on the out of service CPU so it's not trying to run at all (I don't think that you're asserting /HALT are you?)

Anyway, we actually don't know what the problem is currently, so better gather some evidence first.

[derkom]

(I don't think that you're asserting /HALT are you?)

Nope.

[stephen_usher]

OK, if needs be I can rig up a /HALT line switcher using a GAL, as it needs to be bidirectional.

I can use the switch line used on the 157s, have /HALT, /HALT1 and /HALT2 as inputs and generate /HALT1 and /HALT2 from the input /HALT and /HALT output from the /HALT1 and /HALT2 inputs. I shouln't get any feedback issues.

It would be messy on the current board but would only need two cut traces.