How should engineers choose a power supply?

With the popularity of electronic products, it is not uncommon to see devices that can be powered by multiple power sources. For example, industrial handheld instruments or portable medical diagnostic equipment are powered by batteries most of the time, but once plugged into an AC adapter or USB port, they draw power from the AC adapter or USB port, which not only charges the battery but also powers the system. At the other end of the mobile system, large high-availability server racks have at least two power supplies to keep the server running when any one power supply fails. Storage servers use supercapacitors as backup power to achieve clean and error-free shutdown when the main power supply is disconnected. Of course, some servers use high-current main power supplies and low-current auxiliary power supplies. All of these systems face an important task, which is to choose one of the various available power supplies to power the system load.

Hidden Problems in Power Multiplexing

The task of selecting the right power supply for a given environment may sound simple and easy, but the consequences of an incorrect selection are serious and can cause system failure and damage to the power supply. If the voltage applied to the power supply output is high, then switching between power supplies operating in parallel can cause current to flow back into the power supply. Some power supplies will fail if they encounter energy return, interrupting the control loop and causing overvoltage at the power supply input terminals, which may cause capacitors and other components to burn out. There is also a risk when switching between parallel power supplies that all supplies may be disconnected from the output for too long, causing the output voltage to drop, the system to reset, or the system to operate improperly. A third problem occurs when the voltages between the power supplies are close. Some comparator-based control methods introduce an oscillating mode that continuously switches between the power supplies, so that the switching between the power supplies requires careful design.

Same power supply

Let's start with the simplest case - a system powered by two identical power supplies. Identical here means the same nominal voltage, with variations of typically a few percent within the power supply tolerance. This situation occurs in high-availability servers that have two or more redundant power supplies to provide uninterrupted operation in the event of a failure of any power supply. In such systems, a simple approach is to select the power supply with the highest voltage to power the system. Two diodes are connected with the anodes of each power supply connected to each power supply and the cathodes connected together to form a so-called diode "OR" circuit, which implements the function of powering from the higher voltage power supply (see Figure 1). This circuit also works properly when only one power supply is connected. When there are two supplies, the diode of the higher voltage power supply is forward biased and the other diode is reverse biased.

Figure 1: A diode-OR circuit of two power supplies delivering power to a load.

With multiple cards in modern servers, power can easily exceed kilowatts, so the 12V DC power supplies must provide 50A to 100A of current. Using plain old diodes, even Schottky diodes with low voltage drop, diode-ORing two 12V supplies like this would present a daunting thermal management task if not impossible, because a 1V drop across the two diodes at such high currents dissipates a lot of power, e.g., 50W at 50A. So ideal diodes with much lower voltage drop are needed. Once again, as with many other circuit problems, MOSFETs come to the rescue. A MOSFET, with a sense circuit, can act like an ideal diode, turning on a very low voltage drop when forward biased (input higher than output) and turning off when reverse biased (input lower than output). The ideal diode voltage drop can be reduced to 1/10 that of a normal diode, so power dissipation is reduced to a manageable 5W. Such an ideal diode-OR circuit is easily implemented with single or parallel N-channel MOSFETs with an RDS(ON) of 2mΩ. Figure 2 shows one such circuit and its IV curve. Linear Technology's LTC4352 controls an N-channel MOSFET to achieve an ideal diode function. When two such circuits are connected in parallel, an ideal diode "OR" circuit is formed, which can be used in redundant power supply systems. Linearly following the voltage drop of the MOSFET in a certain proportion ensures that the power supply does not oscillate and switches smoothly, while the fast turn-on and turn-off time of 0.5μs minimizes the output voltage drop and reverse current.

Figure 2: LTC4352 ideal diode with UV/OV and its IV curve.

The functionality of an ideal diode is beyond the reach of passive diodes. The LTC4352 operates as an ideal diode only when the input is within the valid range set by the undervoltage (UV) and overvoltage (OV) thresholds. The STATUS# pin provides a status signal to the downstream circuitry that the MOSFET is on or off, and the FAULT# pin indicates whether the MOSFET is off due to a UV/OV condition or because the MOSFET is resistive or open, resulting in an excessive voltage drop, the latter of which alerts you to an impending fault before it occurs.

Let's share the load

Diode-OR is a “winner-takes-all” system where the highest voltage supply provides all the load current. If both supplies deliver power equally to the load, the thermal stress is split equally between the two supplies, and the reliability of the power system is greatly improved, and the life of the supplies can be extended. However, many load-sharing circuits that regulate the supplies are plagued by loop oscillations. The load-sharing control loop interacting with supply variations complicates the problem. This is where the ideal diode concept can help. By adjusting the ideal diode voltage drop to compensate for the difference in supply voltage, the output voltages of the two ideal diodes can be made equal. Adding sense resistors between these two equal points and the shared load ensures that the currents flowing from the two supplies are equal or proportional. The LTC4370 diode-OR current-sharing controller uses this method of current-sharing for two supplies (see Figure 3). This method can compensate for supply voltage differences of up to 600mV, which means a ±2.5% tolerance for two 12V supplies or a ±6% tolerance for two 5V supplies.

Figure 3: The LTC4370 balances a 10A load current between two diode-ORed 12V supplies. Current sharing is achieved by adjusting the MOSFET voltage drops to compensate for supply voltage mismatches.

Different power supplies

In the server example above, the diode-OR and load-sharing methods work well when both power supplies are the same. However, these methods are not suitable for battery-powered systems, where the input comes from a battery, an AC adapter, or a 5V USB power supply, that is, the nominal voltages of these power supplies vary greatly. In some cases, supercapacitor backup power is also involved. Therefore, a more general solution is needed that does not simply work by weighing the power supply voltage. This solution is called a priority power processor. The basis of this solution is that the power supplies in a battery-powered system have a priority order. Usually, the AC adapter is ranked first, and the system draws power from the AC adapter as long as the AC adapter is present. Each power supply must have a certain valid voltage range (to detect the presence of the power supply) and a priority. If a power supply is present, it will be considered for powering the system according to its priority. The LTC4417 priority selector selects only one of the three power supplies based on the valid voltage window and priority, connecting it to the output (see Figure 4). Careful switching is performed to avoid connecting the two supplies together, and the power supply is only connected to the output when the output voltage is lower than the input voltage. This minimizes or eliminates the reverse current flowing back to the power supply. This also enables controlled fast switching to limit output voltage drop and inrush current.

Figure 4: The LTC4417 3-supply priority processor.

in conclusion

Depending on the type of power supplies used in the system, the first thing to do is to choose the right solution for power multiplexing. The options are diode "OR" (with or without load sharing) and priority supply processors. Regardless of which method is chosen, selecting the correct power supply to power the load requires careful design to avoid destroying the entire system. Reverse current flowing back to the power supply and output voltage drop must be minimized to avoid causing oscillatory switching back and forth between the power supplies. The solutions presented in this article solve these problems in an elegant way.