CDC421xxx Fixed-Frequency Clock Generator outputs wrong frequency if input clock isn't stable when the chip powers on

I recently encountered a case where the TI CDC421xxx clock generator outputs the wrong frequency.  I am designing an application that requires a high-frequency LVPECL reference clock input that must operate in an environment pressurized to several thousand PSI.  That rules out the use of any standard hermetically-sealed oscillators, which will implode under such extreme pressure.  To solve this problem, I turned to a combination of the TI CDC421xxx with the XIN 1 pin driven by a 3.3V LVCMOS clock oscillator IC and the XIN 2 pin grounded.  Immediately we hit a problem where my design would only operate properly if powered by a supply with a very slow voltage ramp-up.  We spent a week chasing our tails trying to make all the varous power supplies on the board (7 in all) power up in different sequences and at different rates, but made no progress.  I finally broke down and did what I wish I'd done first: verified that the CDC421xxx was outputting the correct frequency.  It was, but only if my design was powered on with the supply that had a very long voltage ramp-up.  Unfortunately, the real supply used in the final system powered on 10x faster, resulting in a broken design.  As I thought about what might be causing this odd behavior, it occurred to me that perhaps the CDC421xxx locks to its reference clock before that clock is actually stable.  I emailed TI tech support for confirmation, but after a couple of days of no responses from them, they dumped my support request onto this Wiki.  Such is life when you aren't buying 1 million units of a particular component.  So here I am, documenting this issue for any engineer that comes along after me and encounters the same problem.  That'll be $500, TI. I accept cash or beer.

After some testing, I was able to confirm my suspicion that the CDC421xxx prematurely locks to its reference clock.  Since both the CDC421xxx and the LVCMOS crystal oscillator reference clock are both powered from the 3.3V rail, and since the CE pin isn't really a chip enable but a buffer enable (I know the datasheet says it's a chip enable, but I tried it), the only solution to this problem is to delay power-on of the CDC421xxx until the LVCOMS clock being fed to the XIN 1 pin is stable.  I quickly settled on powering the CDC421xxx through the Micrel MIC2014-0.5YM5 fixed-current power switch due to it's tiny size, minimal external component count, and the presence of an enable pin which I could hold off for 15 ms with a simple RC circuit while the LVCMOS oscillator powered on and stabilized.  This power hack worked exactly as I'd hoped, and now my design powers on and operates perfectly, regardless of which power supply is feeding power to my board. 

Note to TI: please add a note in the CDC421xxx datasheet informing customers that if they drive the part with a CMOS crystal oscillator, that oscillator needs to be powered on and stable BEFORE the CDC421xxx is powered on.