Power Supply Design

Question: Is a minimum load necessary on a switching-mode power supply?

Original Question: Recently I have been trying to use an SMPS (a widely available PC power supply) for my hobby work. I find that the SMPS starts up, only if I apply a minimum load of about 0.5 ampere on the 5 V output. Is this the normal operation of an SMPS? Is it necessary to ensure a minimum load, just to ensure that the supply kick starts?

Answer: When a switching-mode power supply with an output LC filter operates with sufficient load to keep the instantaneous inductor current above zero, the filter averages the switching voltage waveform on the filter input. When it operates at lighter load, the voltage tends to rise to the peak value of the switching waveform. (See the waveform in the tutorial on a heater, where the average value is Vin*D and the peak value is Vin.) The feedback loop may be able to compensate for some of this but usually not all of it, especially if multiple levels are controlled by one feedback loop. It is customary to include overvoltage sensing or protection on all outputs and if the voltage rises on any output, these may be triggered. Also, it may be more difficult to keep the power supply stable at light loads.

For these reasons, an internal load is often added to the power supply to prevent these problems. When an external load can be depended on to provide a minimum load, this internal load can be removed resulting in higher efficiency and more load capability. Hence, some power supplies do depend on an external load being present for proper operation.

This is apparently what is happening in your case.

Since my overarching concern is field reliability, I usually try to design so no external load is required for the supply to operate and regulate -- using housekeeping or other circuits to provide the necessary minimum load.

Others may have a different perspective on this topic. Comments are always welcome.

Comment by Ray Ridley, 09-07-2000:

There are many other implications behind what you are asking here, beyond the normal "does it work" question, and the regulation issues you raise.

Many times I have seen (and built!) converters which have chaotic operation
at very light load due to noise tripping of the current-mode controller. The output is tightly regulated, and the current is bounded, but the duty cycle is chaotic. [More on chaos - JF]

There are also some topologies - such as the phase-shifted full bridge that can become unreliable at light load, and failure rates increase if they are operated there.

And then the self-powered controller which cannot generate enough bias from an auxiliary winding to keep the bootstrap off. It either burns up the bootstrap device, or cycles on and off.

Ray Ridley, Ridley Engineering, has been a past sponsor of this website.

Comment by James, 9/11/00:

New perspective: ...In a power supply with multiple outputs, all the outputs must have the same charge and discharge constant of RC to keep the proportions of voltage. The main output is sampled, if the load on the main output is too light, the other output voltages would be too low. Especially, the value of supply for regulator IC, if lower than Vstop of the IC, the power supply would not work. So, we say that a minimum load is necessary to waken the power supply.

This thread is expanding into the area of cross-regulation, a complex subject that is mostly beyond the scope of this question. However, it might be useful to look at critical inductance and its impact on cross-regulation as related to the above comment.

Consider a buck converter with an LC output filter. At full load, the filter averages the switched waveform. As the load is reduced, averaging continues until the load is reduced to the critical inductance value, Lc.

Buck Converter


Switch Waveform


Voltage Vs Current Showing Inflection Point at Critical Inductance Lc

In the region beyond critical inductance, Lc, the duty cycle of the switch is a weak function of the load and the power supply tends to self regulate for loads down to the Lc point. Lc is the point where the current in the inductor drops to zero each cycle.

Two things happen at this point.

First, the system is reduced from a second order LC system to a first order RC system. C is the output capacitor and R is the load. This is the RC time constant referred to in the comment above.

Second, the output voltage begins to increase from the average value to the peak value, Vin, of the switching waveform as discussed in the original answer above. This voltage does not rise in the closed-loop system because the feedback loop reduces the duty-cycle to maintain regulation (dotted line). However, if any other outputs are cross-regulated to this output through the turns ratio of the power transformer, the shorter duty cycle would drop the voltage in all these outputs. The bias to the regulator IC is often obtained from a cross-regulated winding and at light loads this might drop too low for operation as stated in the comment.

A suggestion is made in the comment that the RC time constant of all the outputs should be the same. I'm not sure if this helps or is even practical.

The reason I'm not sure it works is that the shape of voltage versus current for each output is shaped by the switching waveform and the critical inductance, independent of RC (assuming an LC filter on each output). It is prudent to keep these shapes the same on all outputs. This is often done by winding all the inductors on one core with the same turns ratio as the power transformer. Since all currents now contribute to lowering the critical inductance current, the power supply can operate in the continuous current mode down to a lower current. The turns-ratios have to be exactly matched or some strange things can happen in the control loop (added right-half-plane zeroes).

If C instead of LC filters are used, then there is probably merit in trying to match the RC's, but, since the R is controlled by the user, not the designer, this may not be practical.

In any event, the discussion has now entered the realm of cross-regulation, a complex subject that will eventually be discussed as a problem in the problem/relevance/solvability/solution section.

I like the concluding sentence in the comment, "So, we say that a minimum load is necessary to waken the power supply." I've never heard it put this way, but it is often true.