Driving High-Performance ASICs and Microprocessors with Multiphase DC-DC Converters

Multi-phase DC-DC converter leads

Today's high-performance ASICs and microprocessors have been widely used in various fields such as industrial control, communications, and aerospace. However, due to their high power consumption, sometimes up to 150W or even more, for a power supply voltage of 1V to 1.5V, the current required by these devices can easily exceed 100A. This causes large size and weight of the equipment and a series of problems in application. Therefore, how to solve the power supply solution for these devices is a new problem faced by designers.

The use of multi-phase DC-DC converters for power supply is a new type of high-efficiency power supply technology. Why? This is because a multi-phase DC-DC converter can be designed using a scalable power controller chip. The clock generator based on the PLL (phase-locked loop circuit) on the controller chip enables multiple devices (high-performance ASICs and microprocessors) to work synchronously, and its scalable architecture allows several controllers to be connected in parallel and work synchronously. Based on this, the design scheme of multi-phase DC-DC converters (topology, input ripple current, output ripple voltage, MOSFET, inductor selection, heat dissipation, etc.) and design examples are introduced.

Multiphase Topology Advantages

The power of the commonly familiar single-phase step-down regulator (converter) is not strictly limited, but when the load current rises to 20A to 30A or more, the single-phase buck regulator becomes incapable, and the multi-phase converter will have obvious advantages. These advantages include: low input ripple current and less input capacitors; the output ripple voltage is reduced due to the equivalent doubling of the output ripple frequency; the temperature of the components is also reduced because the losses are distributed among more components; and the height of the external components is also reduced.

The multiphase converter is essentially a multi-channel buck regulator working in parallel, that is, their switching actions are synchronized, and the phase difference between them is 360/n degrees, where n is equal to the number of phases. Although the parallel connection of buck regulators makes output regulation a little more complicated, this problem is easily solved using a current mode controller because this controller can regulate the current in each inductor in addition to regulating the output voltage.

Input ripple current

When selecting input capacitors, the key issue is the handling of input ripple current. The use of multi-phase topology has greatly reduced the input ripple current, so that the input capacitor of each phase only needs to handle a lower amplitude input current pulse. In addition, the phase deviation also increases the equivalent duty cycle of the current waveform, thus producing a lower RMS (root mean square) ripple current. The ripple current values ​​listed in Table 1 illustrate the reduction in ripple current (from 31.6A for single-phase to 11.2A for 8-phase) and the saving of input capacitors (from 11 for single-phase to 4 for 8-phase).

Ceramic capacitors with high-K dielectrics not only have high ripple current handling capabilities but also allow for a very small PCB (printed circuit board) area. For example, the rated ripple current of each 1812-shaped ceramic capacitor is as high as 2A to 3A. For designs that require lower costs, electrolytic capacitors are a good choice.

Reduce output ripple voltage

For ASIC and microprocessor core power supply, voltage accuracy is usually required. Large load current steps require the output capacitor to have extremely low ESR (equivalent series resistance) to reduce transient voltage, and the output capacitor is also required to have a large enough capacity to absorb the energy stored in the main inductor when the load jumps downward. Organic polymer capacitors have lower ESR than tantalum capacitors, while polymer capacitors have the lowest ESR and high capacity. Ceramic capacitors have excellent high-frequency characteristics, but the capacity of each device (ASIC and microprocessor) is only one-half to one-quarter of tantalum or polymer capacitors.

The low-side MOSFETs should be used in parallel

A 12V to 1.2V converter requires the low-side MOSFET to be on 90% of the time; in this case, the conduction losses are much higher than the switching losses, and for this reason, two or three MOSFETs are often used in parallel. Multiple MOSFETs in parallel effectively reduce the drain-source on-resistance RDS(ON), thereby reducing conduction losses. When the MOSFET is turned off, the inductor current continues to flow through the body diode of the MOSFET. Under this condition, the drain voltage of the MOSFET is essentially zero, greatly reducing switching losses. Table 1 shows the losses of several multi-phase configurations (from 6W for a single phase to 1W for eight phases). Note that the total loss of the low-side MOSFET decreases as the number of phases increases (from 18W for a single phase to 8W for eight phases), thereby reducing the temperature rise of the MOSFET.

High-Side MOSFET Selection

At a duty cycle of 10%, the switching losses of the high-side MOSFET are much greater than the conduction losses. Because the high-side MOSFET is only on for a short period of time, the conduction losses are not significant. Thus, reducing switching losses is more important than reducing on-resistance. During the switching process (tON and tOFF), a certain voltage and transmission current must be withstood. The product of this voltage and current determines the peak power loss of the MOSFET; therefore, the shorter the switching time, the lower the power loss. When selecting a high-side MOSFET, a device with lower gate charge and lower gate-to-drain capacitance should be selected, which are more important than low on-resistance. As can be seen in Table 1, the total loss of the MOSFET decreases as the number of phases increases (from 4.4W for a single phase to 1.76W for eight phases).

Inductor selection

The inductor value determines the peak-to-peak value of the ripple current. The ripple current is usually expressed as a percentage of the maximum DC output current. For most applications, a ripple current of 20% to 40% of the maximum DC output can be selected. When the core voltage of ASICs and microprocessors is low, the inductor current does not decay as fast as it rises. When the load decreases, the output capacitor will be overcharged, causing an overvoltage phenomenon. If a smaller inductor is selected, less inductor energy is transferred, resulting in a lower surge voltage.

Cooling by design

In a forced convection cooling system providing 100LFM to 200LFM, a single-phase design requires a fairly large heat sink to achieve a thermal resistance of 0.6°C/W. In a four-phase design, the thermal resistance can be increased to 2°C/W. This thermal resistance is easily achieved without a heat sink and 100LFM to 200LFM of airflow.

Design Examples

A four-phase DC-DC converter is configured using the MAX5038 chip controller. The remote voltage detector pins (VSP and VSN) of the MAX5038 master controller are used to detect the output voltage, and its output signal (DIFF) is also used as the input of the EAN terminal of the master/slave controller to realize the parallel operation of the master/slave controller. The MAX5038 master controller output (CLKOUT) also provides a clock input (CLKIN) for another MAX5038 slave controller. Floating the PHASE pin causes the internal clock of the slave controller to produce a 90° phase shift with the CLKIN signal. By setting the appropriate gain, the error amplifier (V ERROR AM)P can also realize active voltage positioning. If the gain is set with precision resistors, the accuracy of accurate load balancing can be ensured. The output of the error amplifier (EAOUT) determines the load current of each phase. Each current loop can be compensated through the CLPl and CLP2 pins. After proper compensation, it can provide a very stable output under most input and load conditions.

VAP is the voltage signal "+", VAN is the voltage signal "-", PLL RAMP GEN is the phase-locked circuit ramp clock generator, DIFF is the differential signal, ISENSE is the current detector, IERROR AMP is the current error amplifier, VERROR AMP is the voltage error amplifier, COUT output capacitor VOUT output voltage.

The master controller performs voltage telemetry (sensing) functions and clock generation functions, and the slave controllers extend the output current and work synchronously with the master controller.

in conclusion

Multiphase synchronous DC-DC converters designed with the MAX5038 chip controller can effectively drive ASICs or processors that operate at IV to 1.5V and consume 100A or even higher current. They solve many basic problems in the power supply system, including capacitor ripple current, MOSFET power consumption, transient response, and output voltage ripple, etc. It should be said that this solution is a new and efficient technology.