The voltage regulator circuit gets the +12 V voltage present on the ATX12V or EPS12V connector found on the motherboard and converts it to the voltage required by the component that the voltage regulator is connected to (CPU, memory, chipset, etc). This conversion is done using a DC-DC converter, also known as switching-mode power supply (SMPS), the same system used inside the PC main power supply.

The heart of this converter is the PWM (Pulse Width Modulation) controller. This circuit generates a square-wave signal that will drive each phase, with the duty cycle from this signal varying depending on the voltage that the circuit wants to produce (duty cycle is the amount of time the signal stays on its higher value; for example, a signal with 50% duty cycle will spend half the time on its lower value - usually zero volt - and the other 50% of the time on its higher value -- which means +12 V on the case of the voltage regulator circuit. 

The value of the output voltage the voltage regulator circuit must produce is read from the CPU “voltage ID” (VID) pins, which provide a binary code with the exact voltage that must be supplied. Some motherboards allow you to manually change the CPU voltage inside the motherboard setup program. What the setup does is to change the code that is read by the PWM controller, so the controller will change the CPU voltage according to what you’ve configured. Even though we are talking about the CPU, the same idea applies for the memory and the chipset.

The DC-DC converter is a closed loop system. This means that the PWM controller is constantly monitoring the outputs of the voltage regulator. If the voltage on the output increases or decreases the circuit will readjust itself (changing the frequency of the PWM signal) in order to correct the voltage. This is done through a current sensor, since when current consumption increases the output voltage tends to decrease and vice-versa.

In Figure 17 we have the block diagram of a PWM controller usually found on the CPU voltage regulator circuit (NCP5392 from On Semiconductor). On this block diagram you can easily identify the voltage ID pins (VID0 through VID7), the loopback pins (CS, Current Sensor pins, located on the left side) and the outputs to drive each phase (G pins, located on the right side). As you can see, this integrated circuit can control up to four phases.

click to enlarge
Figure 17: PWM controller.

Each phase uses two transistors and one choke. The PWM controller does not provide enough current to switch these transistors, so a MOSFET driver is required for each phase. Usually this driver is made with a small integrated circuit. In other to cut costs some manufacturers use a discrete driver using an additional transistor on very low-end motherboards.

In Figure 18, you can see the basic schematics of one phase from a motherboard (the loopback connection is missing on this diagram) driven by an NCP5359 MOSFET driver. The driver and the MOSFET transistors will be fed by the +12 V voltage provided on the ATX12V or EPS12V connector (where it is written “10 V to 13.2 V” and “4 V to 15 V”). You can see on this diagram the two MOSFETs (the top one is the "high side" and the bottom one is the "low side"), the choke and the capacitors. The loopback signal will be provided by linking two wires connected in parallel to the choke to the PWM controller CS+ (CSP) and CS- (CSN) pins. The PWM pin is connected to the PWM output provided by the PWM controller and the EN pin is the “enable” pin, which activated the circuit.

click to enlarge
Figure 18: Phase simplified schematics.

As you can see in Figure 17, there is one PWM output for each phase. As explained, the PWM signal is a square waveform where its width (duty cycle) changes depending on the voltage you want (that is why this technique is called Pulse Width Modulation). Assuming that the output voltage is stable, all PWM signals will have the same duty cycle, i.e., the size of each “square” on the signal will be the same. These signals will, however, have a delay between them. This delay is also known as phase-shift.

For example, on a circuit with just two phases, the two PWM signals will be mirrored. So while phase 1 is turned on, phase 2 will be turned off and vice-versa. This will ensure that each phase will work 50% of the time. On a circuit with four phases, the PWM signals will be delayed in such way that phases will be activated in sequence: first phase 1 is activated, then phase 2, then phase 3 and then phase 4. While one phase is turned on all others are turned off. In this case, each phase will be working 25% of the time.

The more phases you have, less time each phase will be turned on. As explained earlier, this makes each phase to dissipate less heat and each transistor to work less, which provides a higher life-span to this component.