MIC5270 negative LDO misbehaving

Must be too much ESR: "High-ESR capacitors may cause instability. Capacitors with an ESR of 3? or greater at 100kHz may cause a high frequency oscillation." Low ESR out to 100KHz means a high frequency low loss type.

If the sawtooth is stable it's unlikely to be a coupling issue.

We've seen that, a linear reg making sawtooths. I think we used some other Micrel reg that was bad news.

Maybe the current limiter circuit is less well-designed than the main loop?

Did you try the Williams stabilization technique, namely using a huge aluminum lytic?

Here's my latest idea about supplies for my DDG rev C.

Designing this is like solving a 7-sided Rubics cube in the dark. I'll have 17 regulator chips, four of them duals, and a couple of references.

The big Zynq is rude and crude and draws nasty current bursts as code executes. The +1 volt switcher was oscillating at 80 KHz and made the Arms intermittent. The trigger FPGA is super sensitive to supply noise. Both have sequencing rules. There are thermal issues. The pick-and-place machine has run out of feeder slots, so I can't just add a lot of parts.

A couple of places, we are using a 56uF polymer cap in series with a 1 ohm resistor, as a damper. That's crazy but works.

Switchers driving switchers have fun negative-resistance issues.

My LTM8078 positive to dual negative switcher idea has occasional hangup issues too. Not in Spice of course. Hence the 79M05 and the question mark.

Not complaining, of course.

Try a cap across "R2" (in the datasheet circuit)? Or RC, or...

Tim

-- Seven Transistor Labs, LLC Electrical Engineering Consultation and Design Website:

That chip has a real high theta-JA. Could be an oscillatory partial thermal shutdown effect. Datasheet is mostly static so no telling.

Yeah, right, the unlimited capacitive load capability. I have a follower based on the LM8262 supposed to be a rail splitter. 0.1% thermally coupled resistor array at the input with 0.05% matching, 100uF ceramic at the output and the resulting voltage is off by 30mV, 4x the max from the datasheet. Consistently wrong on two prototype units.

Will need to disassemble it part by part to find the core issue, but I don't see any oscillations.

Best regards, Piotr

Thanks.

Loop instability usually leads to oscillation at some frequency near the unity gain cross, with a fairly well-defined period and a distorted sinusoidal waveform.

This one is a sawtooth whose frequency depends very strongly on the load, and whose waveform is very far from periodic. That's why I was looking for some interaction with the (much faster) switching junk on the LDO's input.

The maximum frequency sawtooth I've seen is around 20 +- 20 kHz. (It really is very irregular.)

Tomorrow I'll disconnect it from the switchers and run it off a lab supply.

Cheers

Phil Hobbs

(Who would like to think that the other technical things he has to worry about are more important, except that they aren't.)

I did all the usual loop compensation things--no joy.

Thanks

Phil Hobbs

-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510

My sympathies. BTW amps used with giant output caps can still be oscillating internally--try putting sense resistors in the supply leads, and see.

Re: capacitive loads in general.

There are op amps, e.g. my long-lost fave LM6361, which have their compensation cap close to the output, so that a capacitive load basically just overcompensates the amp rather than adding an additional feedback pole.

Cheers

Phil Hobbs

Thanks. The sawtooth happens even at very light loads.

Cheers

Phil Hobbs

Perhaps their claim that "The output does not require ESR to maintain stability" is overly optimistic, and they do require some.

Sylvia.

Many RRO amps have one end of the comp cap on the output pin. Or two caps, c-b on PNP and NPN common-emitter output transistors. So adding an external cap to ground just makes the single pole slower. Nice for making precision power or reference rails.

LM8261 is specifically c-loadable. OPA197 is happy with, say, 5 uF or more. AD8565 likes big caps.

I wonder if one edge is an internal RF oscillation, and the other edge is recovery.

Squegging?

Science teaches us to trust

-SW

And hence I am puzzled with what I see. I admit I didn't have time to debug this hopefully minor issue, but this little subcircuit should have worked correctly from the very beginning. The part is genuine, and the meter readings exceed the datasheet worst cases many times. Weird.

I will follow Phil's piece of advice.

Best regards, Piotr

I am about to use that part so thanks for the alert. Have you eliminated input bias current, datasheet worst case is 2700nA?

piglet

No, my dear piglet, not at all. Moreover, I have just identified this as the root cause. Holy cow, didn't know they are still making parts like this.

The follower is otherwise very stable even with a 100uF capacitor, so I dismiss the alert. The "high impedance" input was biasing this precious precision divider.

Best regards, Piotr

A quick and dirty fix comprised of two 0603 1.2k/0.1% resistors put in place of that thermally coupled 0605 array has resulted in a spot-on VCC/2. Luckily, there are arrays in this range, so no new board spin is needed. The opamp offset is negligible.

For the last few years I have been using mostly CMOS opamps and lost my vigilance. I have just received a very valuable lesson and appreciate you being a part of it. :-)

Best regards, Piotr

Often when this happens, it's because the crossover is much higher (low MHz?), and the poles are so far right that it overshoots deep into cutoff. So you end up with a relaxation oscillator. You have to load it hard enough (whether by brute-force load current to get it back to threshold sooner, or with a big enough cap (give or take ESR?) to absorb the huge output current peak) that that nonlinearity goes away and the actual loop response becomes apparent.

That's the usual thing with LDOs (sometimes it's even in the datasheet, you often see a sawtooth bounce on the light-load transient plot), and a lot of other regulators (I've seen it plenty of times with TL431 regulated flybacks). But if you can't load it (or unload it, for that matter) enough to get into the near-linear range, that's a bit of a problem...

Also, it seems to be bipolar, so RF rectification isn't a bad guess.

On an only somewhat related note, I've see integrated switching regs that went cuckoo under induced noise. Specifically, a few kV of 1.5/50us surge, in a wiring harness an inch above the board. Those were some oddball regulator from Intel (via acquisition from a brand I don't remember), high frequency, integrated inductor, tiny. Best I can tell, the ADJ pin (which was hardly exposed on the board, maybe 0.01pF to the offending wire?) was sensitive to that tiniest bit of overvoltage and shut itself down. Symptom was, CPU resetting erratically. Great, huh?...

Tim

Is there anything special about the load that the device is facing - extremely fast switching times for instance (even at low current).

The other thought is if it is load current related does the behaviour become any more regular as the DC load current is steadily increased.