0
\$\begingroup\$

I have an old Macintosh IIfx and it originally runs with 80 MHz DIP-4 crystal oscillator (CPU speed is half of this, so 40MHz). So far I have used 100 MHz crystal oscillator to overclock without troubles but I would like to push it a little further.

It seems crystal oscillators over 100MHz (like 105, 110, 120) are not cheaply available so I was wondering using a Si5351 or similar to have a programmable clock. Right away there are couple issues: IIfx uses 5V for the crystal oscillator and Si5351 has 3V output so some kind of buffer is required. Also, Si5351 seems to produce quite many harmonics on the signal.

Is Si5351 anyhow feasible choice to replace a crystal oscillator on a computer? Or what would be the preferred way to create variable oscillator for frequencies from 80 MHz to 120 MHz?

Update: Managed to get Si5351 and SN74AHCT125 combo to work at 110MHz. Soldered on a protoboard with short connections. Most likely this would be able to produce higher frequency clock signal but the target machine was not stable anymore due to other limitations.

\$\endgroup\$
7
  • \$\begingroup\$ Terminology note that a crystal is not an oscillator. A crystal is a crystal and won't oscillate on its own. An oscillator is an entire circuit that may or may not contain a crystal that actually does oscillate. You need to know which it is your require. Neither a MEMs oscillator nor a crystal oscillator (both of which are entire circuits) can replace replace a crystal which is not an oscillator. \$\endgroup\$
    – DKNguyen
    Commented Feb 12, 2023 at 19:53
  • \$\begingroup\$ @DKNguyen They say it's a DIP-4 crystal oscillator, which sounds like a complete oscillator to me. I've never seen a crystal packaged with more than two pins, at least not through-hole. \$\endgroup\$
    – Hearth
    Commented Feb 12, 2023 at 19:55
  • \$\begingroup\$ @Hearth If so, then they may be able to replace it if other specs match. I was mainly double checking the OP meant what their words actually meant. I guess another thing to consider is whether the oscillator output required is a square wave or sinusoidal. \$\endgroup\$
    – DKNguyen
    Commented Feb 12, 2023 at 19:56
  • \$\begingroup\$ The original type is what you get with image search "dip 4 crystal oscillator". Pins are GND, 5V, OUT and Disable. Most likely the same type as in here: righto.com/2021/02/teardown-of-quartz-crystal-oscillator.html \$\endgroup\$
    – Petteri H
    Commented Feb 12, 2023 at 20:22
  • \$\begingroup\$ I've used it to clock ASICs and FPGAs, so no reason it would not work for an old fashioned CPU. But overkill but it would work. \$\endgroup\$
    – user1850479
    Commented Feb 12, 2023 at 21:16

2 Answers 2

Reset to default
2
\$\begingroup\$

You're getting into some serious hacking here, but why not?

Probably the best approach if you can find a home for it is to build a little mezzanine board. Steal 5V from someplace, regulate it down to 3.3V or 2.5V to feed the Si5351, pick off the output that you want, amplify it back up to 5V, and feed it to the appropriate spot on the board.

Note that you're going to have to be very careful of signal integrity -- you'll want to make sure to use at least some semblance of a transmission line from your board to the main board. I'd look at the board and make a guess at the clock line's impedance, or just assume \$100\Omega\$. Then I'd try to match that with the transmission line from the daughter board. If it's just an inch or less you could probably get away with a twisted pair.

Note, too, that you may wish to approach this project expecting to fail -- this would be a hard enough (and questionable enough) modification for the board's design team to pull off that no sensible manager would punish them for failure. You should plan on not being downcast if you don't even succeed at making it work at the system's design frequency on the first try.

\$\endgroup\$
3
\$\begingroup\$

The preferred way is a PLL design to digitally select dividers to multiply a stable Xtal. Realize that 5V logic is pushing it above 100 MHz and increases temp rise which raises logic impedance and thus slower risetime.

This is possible with the Diodes Inc. PT7C4511 PLL Clock Multiplier using a linear but more noisy 10 to 50MHz VCO to drive it and choose 16x using a selected multiplier.

Integration will be require caps, clock multiplier, VCO and short paired connections with gnd.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.