Power Supply: Better controlled switch concept for desktop PC power supplies

The power supply of a desktop PC, sometimes also referred to as a Silverbox (Figure 1), supplies all the power needed in a desktop PC. During normal operation a number of DC power supply voltages have to be provided: a 12, 5 and 3.3V supply voltage, a 5V standby supply, and a low current, less accurate –12V supply. The 12, 5 and 3.3V supplies must each be capable of supplying 20A or more with a voltage accuracy of ±5%. The average efficiency per today at maximum load is about 70 percent, so when 300W is delivered to the load about 100W of power is wasted and converted into heat, which is subsequently removed using heat sinks and fans.



The 5V standby supply must be active in both normal and standby modes. In both modes it must be ±5% accurate, and able to supply a current of 4A or more.



During standby mode, all supplies must be switched off except for the 5V standby supply. New energy requirements such as energy star (EPA), and 80PLUS (Ecos consulting) demand a standby power of less then 2W for a desktop PC power supply. When the standby supply has to comply with the ‘Blue Angel’ specification, it must even have an efficiency of at least 50 percent at a load of 0.5W.



Recently there has been a trend towards higher power densities, and more silent power supplies. These require efficiency improvements because there is a limit in the number of fans that can be applied.STANDARD TOPOLOGY

The topology that is most widely used for a desktop PC power supply is illustrated in Figure 2. This is a single transistor forward converter with multiple output windings and coupled coils. It also has an extra reset winding to reset the transformer, which has the same number of turns as the primary winding. The reset voltage is equal to the input voltage, so the maximum drain voltage of the primary switch can be more than twice the maximum input voltage. This results in a breakdown voltage of the primary switch of about 900V. Only one of the outputs can be regulated by the duty cycle of the primary switch. The other voltages are then determined by the turns ratio of the transformer windings and coupled coils.



Coupling of the coils at the secondary side helps in reducing cross regulation, but this can still be insufficient to achieve the required accuracy.



An additional problem is that often the ratio of the number of turns does not always fit. This necessitates the use of an extra post regulator for the 3.3V supply to obtain the required accuracy. Presently, this is a linear regulator, which leads to significant additional losses. Schottky diodes are used at the secondary side to reduce the forward voltage drop across the diodes. This could be further reduced by straightforward synchronous rectification. But when post regulation remains necessary to achieve the required accuracy, it is questionable if this can efficiently be applied in this topology.



The only way to disable the main power supplies (±12, 5 and 3.3V) during standby mode is to switch off the primary switch. As a result, for the standby supply a separate converter is necessary. For this, a flyback converter is usually used.



BETTER CONTROLLED SWITCH

The main arguments for introducing a new concept for a desktop PC power supply were to improve design, and allow higher power density and efficiency for future requirements with minimum cost increase. These can be achieved by the introduction of extra switches as illustrated in Figure 3. These switches can block the voltage in both directions, but are four times as large as a single MOS transistor for the same on resistance. The other disadvantage is that the switches need to be controlled and are more expensive than Schottky diodes.



However, by implementing these switches a number of cost reductions and advantages can be realized, leading to overall cost optimization.



When all the outputs run in continuous conduction mode (CCM), a simple expression for the duty cycle holds:







N is the winding ratio of the transformer; a fixed value. So to make the output voltage constant, the duty cycle of the primary switch should be inversely proportional to the input voltage. This is called feed forward regulation. The rectifying diodes at the secondary side are replaced with bi-directional switches. These provide extra fine regulation and allow operation in discontinuous mode, where the duty cycle is smaller than in CCM, and is dependent on the load current. All outputs are now regulated individually at the secondary side. At the primary side resetting the transformer is done with an active reset clamp.



Together this creates a number of advantages over the standard configuration:



• The breakdown voltage of the primary switch can be reduced to about 650V.



• The post regulator and the separate standby converter can be completely omitted. The standby output has been integrated in the main converter.



• The outputs can now be individually controlled to almost any value equal to or lower than the original value without the presence of the bidirectional switch, providing more accuracy and flexibility.



• There is no cross regulation problem anymore, so one output can run at no load while another one can run at maximum power without sacrificing accuracy.



• By using the active reset clamp, the maximum duty cycle can be increased to about 70 percent, reducing the secondary voltages.



• With feed forward regulation heavy load transients cannot cause overshoot of the drain voltage of the primary switch, which does occur in a conventional regulation when the duty cycle changes faster than the reset voltage can be adapted.



• In combination with replacement of the secondary rectifying diodes by bidirectional switches, the active reset clamp mechanism enables zero voltage switching (ZVS) at the primary side, reducing EMI and eliminating switch on losses of the primary switch.



• The reset winding can be left out, simplifying the design of the transformer, and removing the unwanted parasitic capacitance between the primary and reset winding.



• The number of secondary windings can be reduced because multiple outputs can be taken from the same secondary winding, simplifying the transformer design.



• Replacing the coupled coil with individual uncoupled coils also improves flexibility and ease of design.



• Efficiency can easily be improved by using MOSFET with lower RDSon and by replacing the free wheel diodes by MOS switches as well.



• Higher switching frequencies are easier to apply because of ZVS at the primary side and because the individual regulation loops at the secondary side can be much faster than the conventional regulation loop, which has to cross the isolation via a relatively slow optocoupler.



STANDBY OPERATION

To comply with the latest environmental requirements, like the earlier mentioned Energy Star and 80PLUS requirements, the standby supply must have a standby power less then 2W. This can be obtained by minimizing the magnetic field of the transformer. This is because the core losses increase more than quadratic with magnetic field. So the magnetic field must be reduced at low loads, which is done by reducing the on time of the primary switch.



One way to reduce the on time is to take the 5V standby output from the secondary winding of the 12V output. The voltage at this winding is higher, so the duty cycle can be smaller. Measurements showed, however, that this was still not enough to reach the target specification. So the on time had to be reduced even further at low load.



At a certain load the standby converter starts to run in discontinuous mode where the duty cycle depends on the load current. The duty cycle will then reduce when the load current decreases. The solution is to reduce the on time of the primary switch even further in standby mode by slowly reducing the on time when it is longer than the on time at the secondary side. When the load increases and the on time at the secondary side must be longer than the actual on time of the primary switch, the on time of the primary switch is quickly increased again. In this way the on time of the primary switch is always adapted to the minimum value that is needed.



It is also necessary that ZVS or near ZVS is guaranteed all the time, so a minimum on time is used in combination with valley detection of the drain voltage of the primary switch. When the minimum on time is reached, the losses are reduced even further by means of lowering the switching frequency.



CONCLUSION

With the newly introduced controlled switch concept, it is possible to make a desktop PC power supply that is easier to design, is more flexible and more accurate. Due to the minimum number of external components, the integrated standby functionality and the zero voltage switching, a design can be made with minimal increase in cost.