Tony, that is certainly compact but I have about 1.5 sq in of space available (as per the DIP TL494) plus I might be able to do away with the LM317 altogether now.
Anywhere from 11.5 to 13.5V would be ok for the application. (12V relays involved).
At the same time I'll bump the inductor to 1mH to handle ripple from low loads (150mA) as it's size and cost are similar to the 680uH unit.
While the newer SMPS IC's really minimise external component sizes for portability and are battery optimized, the 7X to 10X cost factor over the NCP3063 is hard to justify once you have the PCB space available (for inductors & caps) and mains power to use.
Just for completion. I got the paralleled mmbt3906s to work fine with only a 30°C heating by implementing an NPN to discharge the PNP bases quickly and prevent active mode heating. So the circuit is technically a success and can be implemented. However, I'll probably stick with the NCP3063 solution as it's simpler and certainly less cumbersome albeit it adds to the BOM.
Here's the simulation schematic and graph showing only a fraction of a uS delay in switching. The actual circuit also behaved the same.
It's useful but I didn't attempt it as 2n2222 is not used in the design only 2n3904s so the BOM issue only justifies the 2n3906.
EDIT: The concept is valid and I could replace all of the 2n3904 with the 2n2222 as the costs are very similar. So your suggestion is valid and I'll keep it in mind for future needs.
One area I was looking at is a capacitive base drive rather than ganged resistors to conserve on heat generation in the base resistors. While certain cap values in the 1 nF region for the 170Khz Freq worked well. A single cap. value couldn't cover the entire range of loads without either causing under saturation or over saturation and missed pulses.
Well I have a hybrid design using both thru hole and smd parts totaling around 200+. Perhaps 25 unique parts so far. I am concerned about manufacturing and limiting the number of reels needed for automated SMT assy. It's not very high volume so having to buy 4000 count reels for 1 or 2 instances per device can add up and make up front assy. costs significant that take a quite a while to be amortized. So I guess the penalty can run into the multi hundreds of $$ of start up overheads.