| Electronics > Projects, Designs, and Technical Stuff |
| PMIC choices. |
| (1/2) > >> |
| dmills:
So I have a rather butch fpga, with the associated many, many power rails, because that's how those things roll. My problem is that all the PMIC sand I can find that looks suitable needs configuration over I2C before it will actually work, which means another little processor to set the power subsystem up before I can even boot the main system, I would really like to avoid this (Even the ones with built in eeprom need that configured, so another manufacturing step). What would be ideal is a PMIC that is resistor programmable, instead of relying on the I2C bus for basic setup, I2C is fine for monitoring, but I need to get the power on, at defined voltages without needing to program the damn chip. Something current, available and widely used would be ideal. Has anyone got any good suggestions? Infineon tell me that they can do us a custom pre programmed part, but we don't really do the volume to make that work commercially. I have a vague recollection of a TI part, but can I find the thing now, can I hell. Anyone? Regards, Dan. |
| exe:
I'm no pmic experts, I'm just curious how many pmics you need, what's the input voltage, output voltage, and what's the current requirement? |
| T3sl4co1l:
Well yeah, if you're shopping for all-in-one digital controlled regulators like what's used on laptops and cellphones, you'll get exactly that. And these usually do have an auxiliary processor, hypervisor, or some sort of low-level startup to do basic control functions like that. If you're using a blind and dumb FPGA alone, yeah, it's not for you. :) Why not a dumber multichannel (or several duals, or many singles) regulator that's just set with resistors? You'll find these under the regular category, of course. If you aren't optimizing every last watt, you'll likely find it's worthwhile to use a switching reg for the biggest rails (e.g., 5V, 3.3V, 1.0V), and LDO the rest (e.g., 2.5V, 1.8V). Or if one is nearby and doesn't need fantastic regulation, even just a series diode to drop the difference. There's also chicanery that can be done with the regulator topology itself, for example tapped or multi-winding inductors to deliver supplies related by simple ratios. 5 and 2.5V can be reasonable this way; obviously, the diode drops and stray inductance matter: cross-regulation tends to be poor. If you end up with a somewhat too high voltage and LDO it down anyway, well heck, that's just fine. Tim |
| dmills:
Raw input is 24V, which I figure to buck down to 5V as a first step before getting into the annoying shit. 1V @ 30A (PL, PS Core, BRAM). 1.8V @ 10A. 1.5V (DDR3), probably 2A or so. 0.75V (buffered divider for DDR Vtt). Then a mess of really weird ones for MGT transceivers, must be really low noise, so probably linear regs, but current requirements for these are proving difficult to track down... 1.05V (Yes, really). 1.2V 1.8V 1.8V (Separate reg or noise filter for clock generator). Then the non FPGA stuff, 2.7V (To be LDOd to 2.5V for some 12Gb/s reclockers). 3.3V for misc shit. 15V for a really annoying series led backlight on a front panel. I guess 1.5, 1.3, 2, 2.7, 3.4 would be reasonable (all these are semi low current rails), getting the butch 1V and 1.8V as well would be a bonus, but I think those are going to wind up being some sort of polyphase thing probably from TI, Empirion, or AD (Who have managed to fuck up Linears web site in remarkably short order, grumble). I know roughly what I need, it is filtering down the ways to get there that is causing me a headache. |
| T3sl4co1l:
Ah, and what tolerances? How many of those can overlap? Like the filtered 1.05 might possibly be Vcore with some LCs. Tim |
| Navigation |
| Message Index |
| Next page |