Could you mean J5?
There is a "bypass" line to the LM380s (pin 1) that can turn off sound and might be what you're experiencing.
Can you measure the voltage on pin 1 of U20 & U21 when it's working/not working and see if there's a correlation?

Sorry - I should have been more clear. I know you said that you've narrowed it to the MPU, but I was talking about U20 & U21 on the SB-100. When I've had SB-100s that were intermittent with sound/no sound, it usually came down to either connectors (which you said you've reflowed) or the bypass on the LM380s on the SB-100. When the machine boots up, the SB-100 sound is suppressed until it sees a valid "write" to address 0x00A0. That triggers transistor Q6 on the SB-100 to un-bypass the amplifiers.

So, my first step in diagnosing would be to see if I have about 8V going between Q6 and U20 & U21 on the SB-100. If that's happening, then I know that the SB-100 is getting all the address, R/W, and VMA information that it needs to properly work.

This is a good place to measure the output of Q6
IMG_3143 (resized).jpeg

This is what the voltage of Q6 should look like when the machine resets and then plays its starting tune
IMG_3145 (resized).jpeg

Yes, it sounds like the "bypass" line of the LM380 is never lifting, so the amps aren't putting out sound.
The bypass signal is generated by Q6 on the SB-100 (which is a 2N3904), and that's controlled by one of the JK flip flops on U6.
The next part is difficult to pin down without a logic probe.
Depending on your board revision, your board might be clearing the flip flop (bypassing the amp) with a low signal on RESET (A4J5:pin25).
If you have a logic probe, I would look for these signals on the SB-100s U6:
1) U6:Pin 10 should be high almost immediately after power goes on
2) U6:Pin 6 should be low at power-on and then go high after the 7th flash of the MPU
3) U6:Pin 9 should be low, but pulse whenever there's a sound call (these pulses are in the microsecond range, so you need a logic probe or scope to see them)

If you see all those signals, the issue could be Q6 on the SB-100. If you don't see the pulse on U6:Pin 9, then the issue could be U1 or U2.

All of this assumes that you have good connection between the SB-100 J1 and A4J5 pins: 10, 13, 15, 17, 23, 26, 25, and 27 (all of these are involved with the bypass signal to the amps)

Anyway, my prime suspect at this point is U6. I've seen those go bad before.
If you want a cheap and dirty solution, you can clip the end of R68 that's closest to U6 and tie it to 5V. The machine will make noise at startup until the 7th flash, but the sound will be okay after that. Of course if that doesn't work then you definitely have a bad Q6.

No worries - let me know how it turns out.
If you're ever interested in modding, I did a re-code of Trident last year that I think turned out well:

Cool. Two questions:
1) What Rev of the board do you have? There are some pictures here: https://www.techdose.com/projects/Stern-Pinball-SB100-Sound-Board/346/page2.html
2) Are those readings the same if you have sound / no sound?

Weird. I would probably bench test U1 and see if that's the issue or if perhaps a trace got opened/shorted.

The last thing that made sense to me in this thread was:
> there is 0 volts on the U20 and U21 pin 1, but when the sound is working there is 8 volts

That really suggests that there was a problem with either
1) the cable (or connector pins)
2) Q6
3) U6
4) U2
or
5) U1

But for any of those things, I would have expected the signals on U6 to be different with sound / no sound. The signals you have on U6 (or had before the U1 issue), don't make any sense to me with a board that sometimes works...

I'm still not understanding why you're not getting a pulse on pin 9 -- seems like you've eliminated all the possible causes.
Regardless, it really looks like your U6 is doing the right thing now.
When U6:Pin 6 is high, it basically turns Q6 into a short, which mutes the LM380s.
Then, after the 7th flash, when U6:Pin 6 goes low, Q6 doesn't short the bypass of the 380s, so you should get sound.
Like you said earlier, when you have 0 voltes to pin 1 of U20 and U21, you have no sound. 8 volts and it works.

I know you've replaced Q6 already, but I'm wondering if you would consider running an experiment. The top leg of Q6 (blue arrow in this picture)
IMG_3143 (resized).jpeg

Is the final stop in the chain of components that bypasses the LM380s. If you could pull just that one leg (the top one), the machine should start making noise as soon as you turn on power. Basically, there would be nothing telling the amps to stay quiet until everything is booted.

If pulling that leg reliably gets you sound, and U6:pin 6 goes from high to low after the 7th flash, then the only problem could be R67, R68, or Q6.

Normally, these cards are not nearly this difficult to diagnose!
I had one catch on fire and it was easier to fix than yours.

Screen Shot 2021-07-29 at 9.21.06 PM (resized).png

Actually U6 pin 6 will go low almost immediately after power on. It's done in the first handful of instructions executed on boot-up.

Hi Quench - here's a video of what happens on my Trident's REV C-1 SB-100 sound card (pin 6 goes low only after the 7th flash):

I think I understand why you think that it would go low immediately. There's a line very early in the code that sends a 00 to M00A0
CLRA *15E4: 4F 'O'
STAA M00A0 *15E5: 97 A0 '..'
(the machine boots to 15E0)
But, as you can see in the video, that write instruction doesn't trigger U6.
Rev C-1 of the sound card doesn't use the /RESET line, even though the schematics show it. I think there's an RC circuit on there building charge for a moment that delays the rise of pin 10. It's just a guess, but that's my only explanation as to why the card ignores the first write command.

Ah, thanks for that. Good call. Yeah the longer delay of the RC circuit clearing U6 on power-up probably makes that very early write to $00A0 ineffective.

I'm pretty sure I have a rev C SB-100 schematic somewhere. I'll try and dig it up tomorrow to scan.

That's great news! Congrats!

Congrats again, then!