Here's another cap charger circuit. After blowing up a number of LTC chips, burning fingers, and sampling about a dozen tiny transformers for flyback converters, we revisited the simple home-made peak-current-controlled buck converter. It's a boost-doubler, a Bubbler Converter.
The main problem with a simple boost converter is the huge turn-on inrush into Cload, which will be 4000 uF in real life. This circuit mitigates the inrush by precharging the cap through R4 and limiting surge current with C1.
Another nice feature is the variable peak switch current achieved by the TH computing network. That lets us reduce the initial-boost 9-volt supply current when VP is low, and ramp it up later. This circuit is simple and dirt cheap but has lots of knobs to turn. Efficiency looks good, too, even with the snubber.
Note the second-order filtered current shunts, for Spice snooping. And the much-criticized efficiency calculator.
A lot of switcher chips are so smart that you can't actually understand what they do. And the behavioral models lie, which leads to painful burned fingertips. We can understand this one. The LTC models should include a red LED on the schematic symbol that lights up when some subtle spec is violated.
That IRLML part is interesting. IR claims 100 K/W and 1.3 watts dissipation in a SOT23, which I hope is real. IR is notorious for wild thermal claims. Their data sheet footnotes AN-994, which claims 169 K/W, already optimistic. I'll breadboard this for sure.
Coilcraft is great.