There are a variety of reasons why a system designer may want to consider paralleling of DC power supplies. Some of these are related to the bill of materials and logistics issues, others are focused on satisfying system current, performance or reliability objectives.
On the non-design side, the ability to parallel supplies may allow a single supply model to be used singly or in combinations across a broad product line. This can simplify sourcing, increase per-unit volume, and streamline inventory management.
The technical reasons to consider paralleling supplies are more complex, of course. First, using parallel supplies can be a form of 'insurance' in case the product actually needs more current than budgeted, perhaps due to unavailability of lower-power components or new features and capabilities added by marketing. Second, parallel supplies may support N+1 and even N+2 redundancy to safeguard against single-point failures, or to enable hot-swapping of a failed supply without system impact. Third, it permits the use of a known, proven supply with well-understood features, characteristics and form factor, thus reducing design-in risk and uncertainty. Finally, it allows for 'heat spreading' by adding flexibility in physical placement of the power converters, if a single higher-capacity unit would dissipate too much heat in a highly localized area.
The flexibility and potential benefits offered by the paralleling of supplies brings an obvious question: can any supply be used, as-is, in parallel configuration? The answer is 'no'. It depends on the design of the supply, the technique used to connect the supplies, and the reason the supplies are being used in parallel.
The most obvious and simplest way to hope to put supplies in parallel is to simply tie their outputs together. In general, this won't work, as each supply has its own output voltage regulation, and so would be trying not only to maintain this regulation versus changes in load, but also attempting to regulate against the closed-loops of the other supplies.
For supplies which include their own traditional internal error amplifier and reference, just placing multiple supplies in parallel is not an effective way to make a high power array. Parametric differences from supply to supply will always cause one supply – the one with the highest output-referred reference voltage – to carry all of the load current, while all of the remaining supplies will carry no load.
In this case, as the load exceeds the capability of this 'lead' supply, it may enter a constant-current limit mode (which may or may not be a rated mode of operation), or it may interpret the overload as a fault and shut down. Depending on the supply in question, these responses could lead to overstress, especially if they occur as part of regular operation in the application. Further, for cases where the supply shuts down due to an overload, the supply in the array with the next-highest reference voltage will be forced to carry the entire load, and will similarly shut down. This will quickly lead to collapse of the entire supply rail.
One way this direct-connect topology can work well is if one supply is set to constant-voltage (CV) mode and the others are set to constant-current (CC) mode, but at slightly higher output voltage; note that not all supplies allow choice of output mode. The supplies which are set to the higher output voltage will provide constant-current output, and each of their output voltages will drop until it equals the output of the CV supply. The load must draw enough current to ensure that the supplies which are in CC mode must stay in that mode. Note that use of the two modes does mean that the multiple supplies are no longer strictly identical, thereby negating some of the advantages of the parallel configuration.
The direct-connect approach is viable if the supply is specifically designed to support that topology, or if there is a single control-loop error amplifier which feeds the error signal back to all of the other supplies, so they share the load. However, the latter method also requires a "share bus" for the control signals from the master to the slaves.
Another approach adds small ballast resistors in series with each supply’s output, to equalize the distribution of the load current among the supplies in the array even when their control loops are seeking dissimilar output voltages, as shown in Figure 1. The ballast resistors create some loss of load regulation, depending on the spread of setpoint errors that the ballasting intends to overcome. However, these ballast resistors also dissipate heat, which degrades system efficiency.
Figure 1: One sharing approach is to use relatively low-value ballast resistors on each supply’s output, but this has issues due to resistor-related dissipation and overall efficiency.