Embedded infrastructure, part 2: Power
The problem
I started out building tiny little USB 2.0 powered FPGA/MCU dev boards that didn’t even get warm to the touch. But the power budget of something like a 48-port Ethernet switch, no matter how efficient you make it, is going to vastly exceed the 2.5W I can pull from a passive USB wallwart. The trigger crossbar prototype on my bench now is pulling just shy of 10W and the switch will will need ~30W for the 4x 12-port baseT line cards plus whatever the FPGA based switch engine and management system uses.
I’ve used 5/12V barrel jacks for some stuff but they have an annoying tendency to come loose and aren’t the best for higher power levels anyway.
A direct mains inlet is an option, but comes with more annoying safety concerns (and regulations if I want to make these for anyone but me to use). So that’s not my first choice.
USB PD is the trendy option these days, but requires a fairly smart power brick for each device. I am planning to put a bunch of these gizmos around my lab and if I could run them off a common power supply that would be nice given my limited mains power capacity.
The plan
There’s another option popular in the datacenter/telecom world: 48V DC. I decided to go with this route - specifically +48V DC on a six pin Molex Mini-Fit Jr connector with the negative side at earth ground potential, since there’s commercially available power bricks that supply this but it’s also straightforward to adapt to e.g. a rack wide DC bus system in the future if I wanted to go that route.
48V is a bit high to feed directly to most of the DC-DC modules I use for my projects, so I designed a reusable intermediate bus converter to step the 48V input down to 12V on an 8-pin Mini-Fit Jr. The IBC also provides an always-on 3.3V standby rail to power supervisor/management logic and a bunch of I2C sensors.
Current state
The first generation IBC, after a board spin to fix a stupid layout error that slipped through design review, is functional but has annoyingly high standby power consumption (around 3W) causing awful efficiency at light loads.
I’ve got a second generation of the IBC at fab now that’s about 40% smaller PCB wise and should have idle power closer to 0.5W, greatly improving low-load efficiency.