In trying to get to my goal of 2A output power, I've run into my old nemesis again: heat.  After getting stable operation at 1.8A continuous output current, I thought I could get to 2A with minimal effort just by changing some current sense resistors.  But this doesn't seem to be the case.

The main culprit is the asynchronous charger's Schottky diode.  For all other power components, it's pretty much up to me how much heat they generate.  If it gets too hot, just get a MOSFET with lower RdsON, or an inductor with lower resistance etc.  But for the Schottky diode, it's another story.  Sure, you can get a diode in a larger package, allowing it to dissipate more power safely, but it's still going to generate the same amount of heat, because due to physics the diode will always drop voltage.  So it will always dissipate power (P=V*I).  Worse, the amount of voltage it drops goes up with higher current as well.

Now 2A doesn't seem to be that much current, but keep in mind that I'm aiming for 2A output current at 5V, which means that with a battery voltage of minimum 3V, and counting on converter efficiency around 85%, the diode sees average currents of (2A*5V)/(3V*0.85)=3.9A.  Now I understand why, even though the power components are external, the CN3801 datasheet indicates 4A maximum output current.  How much power is dissipated?  At 4A, the SS56 Schottky diode I use drops around 0.6V, burning off 2.4W.  As it gets hot, it starts to exhibit an appreciable reverse current as well, leading to even more losses.

The mechanism preventing me from getting to 2A due to this heat seems to work like this: as the circuit (mostly the diode) heats up, some of this heat warms up the CN3801, eventually at high temperatures making the current sense voltage reference drop.  Nominally 120mV, I've seen it go as low as 95mV when drawing high currents.  This will in turn limit the amount of current drawn (for the same sense resistor, lower current will generate this lower sense voltage).  So the whole thing is self-limiting: as it gets hot, it will limit the current.  This could be considered a feature I guess, but the bottom line is that even when you lower the sense resistor, as things get hot the sense voltage drops enough to negate the lower sense resistor.

I've been considering possible solutions and what this means for the project:

1. Is there a way to turn this into a synchronous converter (replace Schottky diode by MOSFET) with some clever circuitry and if so, how much circuitry and will it fit?  The chip has no complementary output for this so it will have to be a hack.
2. My simple topology of running all current through the charger is showing its limits.  Another topology would be to only run the charge current through the charger and provide a parallel path for the load current directly from the power input.  Downside is that now the 5V converter needs to become step-up/step-down, since the input voltage can be as high as 24V and the battery voltage as low as 3V and both need to convert to 5V.  A step-up/step-down immediately pushes you into chips that are in the several-dollar range.
3. A spec fix: instead of 2A max, advertise the system as 1.8A max.  The heat still bugs me, and will limit the current even lower at higher temperature.  2A is also a nicer number, and what the Raspberry Pi foundation recommends for a Pi3 power supply.  On the other hand, 1.8A is double what the current #LiFePO4wered/Pi3 can do without active cooling, so it's still a pretty nice upgrade.

I would very much like user feedback on this.  How much current do you need?  The current design works well at 1.8A, and can probably go into production this fall.  Completely changing the topology will get me past 2A, but is a big redesign that will likely not reach production this year.  Would it be worth it to delay and get the spec I wanted?  It would likely also conflict with other goals such as making it cheaper...