In thinking about DFM (design for manufacture), you always, of course, think about shaving pennies here and there.

The prototypes use an LTC2050 chopper amp as the buffer between the DAC and the oscillator. That chip is really expensive. It's $2.90 for one, $2.33 each for 100 and then it jumps down to $1.30 each for Q:500.

I spent a little time looking around DK this evening and came across the AD8538. It's not quite as zero-drifty, but for this application, I'm not entirely sure how important that really is. What this design requires is excellent short-term stability of all of the components. The long term stability will be taken care of by the feedback loop with GPS.

The big spec differences: The AD8538 slew rate is faster and the gain bandwidth is lower. Neither of those matter, as the DAC is only every going to change once every ~50 seconds or so, and the amp is being used at unity gain. The input offset voltage is 10 times worse - 5 µV instead of 0.5 µV. But our DAC step voltage is ~50 µV (3.3 / 65536), so I'm not sure that matters so much.

But where the rubber hits the road is that the AD8538 is $1.52 for one, and has a much nicer volume discount structure, with the price dipping under $1 at Q:250 and a full reel of 3000 at 64¢ each.

I think it's worth building another prototype with one to see if it's any less stable.