Jon Dutra on ensuring your power supply design is stable

My buddy Jon Dutra is a field applications engineer over at National Semiconductor. Previous to that he worked at Linear Technology. He knows a lot about power supply design. He sent me an email last year about board layout guidelines and you all seemed to like it— it sure got a ton of hits. Here is a write-up we did about his methods to achieve stability in your power supply design

Is your power really stable over line, load, temperature, and time? There is the expensive, time-consuming way, by looking at the loop gain frequency and phase response with a network analyzer, and there is the "quick check" discussed here.

This is a question anyone shipping a product in volume should be concerned with. The passive components around a regulator IC can have a profound effect on stability. This includes both big components like the inductor and output caps and the smaller components such as feedback networks and compensation components. Changing compensation always requires additional testing.

Compensation is one determinate of the loop gain. Inductor value and ESR (equivalent series resistance), capacitor values and ESR, modulator gain, even the resistance of MOSFETs also affects loop gain. That buzzing noise you might hear is a warning of a serious problem, indicating oscillation or loop instability. You can often see this oscillation by looking at the switching waveform. Is it jittering?

To find out, solder a 200 to 1K-ohm resistor to the switch node and probe the free end. The resistor keeps the switching currents from going into the scope input capacitance, so the other scope channels are not corrupted as much. Now look at Vout on the second channel. Set that channel to ac coupling, then gain it up. Do you see any oscillations there too? Even quicker, make an inductive probe with 6 turns of wire around the end of your little finger, solder that coil to a piece of coax, then feed that to your scope preferably with 50 ohm termination on the channel. Put the coil on top of the main inductor for a quick look at the switching waveforms. The other channels will not be corrupted with switching noise with this method. It is a great way to trigger the scope off the channel.

Changing a comp cap to 0.1 uF may be OK on this unit ( For example) , but may be too close to the edge when building many units.

A quick test to determine power-supply transient response.

First make a "power resistor oscillator" ( PRO) with a 555 timer driving a MOSFET, at about 100 Hz frequency. Then put a non-inductive load resistor in series with the drain of the MOSFET, so that the load increases 25% or so when the MOSFET is on.

Channel one of the scope goes to the power supply output, Vout, and is ac coupled at a 0.5 V / division amplitude. Channel 2 of the oscilloscope goes to the MOSFET gate, and you set at 5V / division. Use this for the scope trigger. Set the horizontal sweep to 5 mS / division as a good starting point.

Look at the shape of the transients, both when the load increases and decreases. Does it settle quickly? Great, but if it rings 3 times, that is not so good. OK, now carefully use a hair dryer or heat gun to heat up the board. Do not get things too hot; you can damage the circuit. Does the transient waveform still settle quickly? Awesome. Now cool down the board with Freeze Spray. Does the transient output waveform still settle quickly?

OK, now go to the other extreme of input voltage. Does the waveform look OK under the same procedure? Then do the test under the extremes of high and low load current. If the transient waveform looks OK in all the tests, then you have a product that will not come back for loop stability issues.

If the waveform rings on either edge or exhibits overshoot or if it oscillates or if it sags excessively then you should look at the circuit with a critical eye. The problem could be loop compensation. It might also be caused by current limiting, either in your circuit, or in the input supply to your circuit that is going into current limit.

These tests are not exhaustive, but they are a quick reality check as to whether you have a power supply design which is on the edge, or has plenty of gain and phase margin.

Further tests to determine power supply stability.

Four-corner testing is more exhaustive, but you can ferret out hidden problems. For instance, today I had a 3.3 V output supply, with just a 20-ohm transient load, and it had more than 500 mV of deviation at the output. This is unacceptable. I fixed the board by changing compensation values and gave it back to the customer.

Now how do you fix it? Well, the "quick fix" is easy. First solder a capacitance substitution box (or an RC box) in parallel with the suspect capacitor. Does changing the value change the transient response? If no, then look somewhere else. If the transient response does change with the capacitance value, then you have found a component that directly affects loop gain and phase. Note that it is very likely not the only component that does so.

Now you can change the compensation capacitor value to tune for a transient response that is much better. This may not be an optimum fix. If you are going into production, as a bare minimum, you should test it in the 4 corners, and over multiple units.

The four corners are

·        Minimum Vin, minimum load with the transient load 0% to 25% load, or 10% load to 35% of full load.

·        Minimum Vin, maximum load with the transient load 75% to 100% of full load.

·        Maximum Vin, minimum load with the transient load 0% to 25% of full load, or 10% load to 35% of full load.

·        Maximum Vin, maximum load, with the transient load 75% to 100% of full load.

Ideally you should do these four corners at different temperature extremes as well.

In summary, a transient load can be made with a LMC555 timer, a MOSFET and a power resistor. Every bench should have one. It will show marginal stability by ringing. Just trigger on the MOSFET gate, then ac couple to the output to look at the transient response.

If anyone can give you good advice on making your power supply work it is Jon. Remember, a switching voltage regulator is a servo system. You really should have an idea of the stability of the system in order to feel confident about the design. Now that we use big ceramic capacitors, many power supply circuits can get really twitchy and unstable. And as your electrolytic capacitors reduce in capacity over time or temperate, well that causes anther set of stability concerns. With Jon’s tips you can at least have some assurance that your design will work OK in production and in the field.