Buck Converter Losses Under the Microscope

Feb 1, 2005 12:00 PM
By Alan Elbanhawy, Director, Computing and Telecommunications Segments, Advanced Power Systems Cente

The synchronous buck converter in Fig. 1 is the most popular topology in the PC market. The race between the top MOSFET companies is heating up with the rush to design new devices that are faster and better suited to meet the demands of the microprocessor makers. These customers require higher switching frequencies for their core dc-dc converters to achieve wider control-loop bandwidth and lower converter-output impedance at lower cost. It is also well known that faster switching frequencies introduce higher losses unless MOSFET parameters, such as gate-drain charge (Q_gd and Q_sw) and drain-source on-resistance R_DS(ON), are optimized for such applications.

Nevertheless, gate charge and on-resistance are not the only parameters that create dc-dc converter losses. The inductances associated with device packaging and pc-board traces also contribute to converter losses in the synchronous buck converter. To understand these loss mechanisms requires simulation as well as testing of actual prototypes. Both of these parasitic parameters play a pivotal role in determining the power loss in the converter and ultimately how fast we can switch MOSFETs. Armed with an understanding of how stray inductances contribute to converter losses, we can devise ways to overcome their effects.

The High-side MOSFET

Fig. 2 shows the switching waveforms of an actual voltage regulator module (VRM). Ch1 is the gate-ground voltage of the high-side (HS) MOSFET Q1 in Fig. 1; Ch2 is the source-ground voltage of the HS MOSFET Q1 in the same figure. The trace M1 in Fig. 2 is the calculated Ch1-Ch2 representing the gate-source voltage of the HS MOSFET, and Ch3 and Ch4 are the drain currents of Q2 and Q1.

In this particular test, we had to use a current loop in both the sources of Q1 and Q2 of Fig. 1 (partially represented by L2 and L3) to enable current measurements. This clearly alters the circuit performance by introducing large inductances in the circuit. Although this is not the real circuit, it helps us understand the loss mechanism during turn off of these MOSFETs. Later on, we will introduce a “noninvasive” way to measure the current without disturbing the circuit too much.

In Fig. 2, the trace M1 clearly shows a plateau of about 2.5 V after the gate voltage has turned off, as can be seen in Ch1. During this plateau, which lasts for about 100 ns, the drain currents of both the HS and the low-side (LS) MOSFETs change states. The HS MOSFET turns off and the drain current (Ch4) returns to zero, while the LS MOSFET turns on and the drain current (Ch3) reaches the inductor current. This is where the circuit performs in a way contrary to the current understanding of how the synchronous buck converter performs.

On closer examination, this phenomenon can be clearly explained. When the HS MOSFET gate-ground voltage falls from V_CC + V_GDrive to ground, the drain current will start falling from its current level toward the off state of 0 A. (V_CC is the input voltage and V_GDrive is the gate-drive voltage). Notice that Ch2 (source-to-ground voltage) falls to ground within 15 ns to 20 ns, giving the false impression that the switching transition is completed within this time. In reality, the drain currents take about 100 ns to complete the transitions.

Also remember that the 100 ns is large because of the source loops. In actual circuits where there are no loops, the current will take shorter time to reach its final level. We will show that in simulation results where we can probe without disturbing the circuit. Notice that the source-ground voltage goes negative.

The rate of change of the drain current

(where I_drain is the drain current and t is time) will generate a backward voltage =

pulling the source voltage in a negative direction with respect to the gate. This action, in turn, makes the gate voltage positive with respect to the source, hence allowing current to continue to flow though the gate voltage is firmly held to ground. Since the current transition is by large a linear transition then

In other words, this plateau has a fixed level with time until the current completes its transition. This negative source voltage plateau will now allow the drain current to continue to flow until the current reaches zero.

We are used to a MOSFET being fully ON when the current is flowing, but closer examination of Fig. 3 reveals that the drain current is flowing in the HS MOSFET while the voltage across it is V_CC. Why is that? It is crucial that we examine the MOSFET performance at this plateau to further understand what is happening.

The MOSFET is in a balancing situation where the voltage

causes the source to go negative enough for the MOSFET to barely supply the current for the transition, which is clear from the fact that

a mere 1 V above the gate-source threshold voltage. The channel is not fully enhanced; therefore, the MOSFET operates as a current source where the current level is maintained by the negative feedback of the voltage plateau. Having the MOSFET operating as a regulated current source, albeit during the current transition only, means the voltage across the drain source will sustain the full V_CC voltage across it causing higher losses.

To further complicate matters, the fall time of the drain current is dependant on its magnitude (i.e., the larger the current, the larger the fall time, which leads to the conclusion that the losses are not linear with current as it has been assumed so far). Fig. 4 shows this dependency clearly. In Fig. 4 (a), the trace M1 is the gate-source voltage of the HS MOSFET at no-load and in Fig. 4 (b), the trace M1 shows gate-source voltage of the HS MOSFET at 12.5-A load current. This leads to the equality

TurnOffLosses = I_drain × V_CC × fs × K × I_drain = K × I_drain² × V_CC × fs.
(1)

When

the plateau will reach zero voltage and the MOSFET turns off completely. Simulation results in Fig. 5 using the schematic in Fig. 6 confirm the lab test findings. In Fig. 5, the simulation scans the value of the source inductance between 0.1 nH and 5.1 nH for the same load current. Clearly, the current fall time is dependant on the parasitic source inductance further aggravating the total losses

Parasitic Inductance Effect on MOSFET Turn-on

Now that we have explained a new switching loss mechanism in the HS MOSFET during turn off, let's investigate what is the loss during the turn-on transition. The situation here is completely different because L2 in Fig. 1 does not carry any current at the start of the turn-on cycle. This means that the voltage

now is positive and will tend to make the source move in the positive direction opposing the gate drive's attempt to force the gate-source voltage to be equal to the gate-drive voltage. This is not a difficult situation to correct, because we can raise the gate-drive voltage to the level sufficient to execute clean and fast switching within the MOSFET V_GSS limits. In this case, the current will follow the gate drive.

The losses during turn on are much smaller than those during turn off. This is because during turn on, the total inductance in the loop will carry a substantial amount of the total input voltage V_CC. This loop comprises V_CC trace inductance, HS MOSFET with its lead inductance, the pc-board trace inductance between the HS and the LS MOSFET, the LS MOSFET with its lead inductance and the pc-board inductance between the source of the LS MOSFET and the return of V_CC.

Fig. 3 illustrates that point. Ch1 is the source-ground voltage, Ch2 is V_CC, Ch3 and Ch4 are the drain currents of MOSFETs Q1 and Q2, respectively. The trace M1 representing the drain-to-source voltage drop across the HS MOSFET will hover around an intermediate value between 0 V and V_cc while the current is switching on from zero to the output inductor current.

This intermediate voltage across the inductances may be calculated from the equation

where

is the sum of all the inductances in that loop.

Note that individual voltages within the above summing equation may be positive or negative, depending on the position in the circuit. This means the voltage drop across the MOSFET during turn on may be calculated from the equation:

and the losses may be calculated from the equation

where t_r is the current rise time. This equation represents a much lower loss because of several factors. One is that the rise time t_r is a much smaller value since the opposing effect of

may be overcome by raising the gate voltage to accommodate this turn on voltage plateau. Another difference between turn-on and turn-off performance is that in the turn-on transition, the HS MOSFET is actively being driven by the gate driver; and hence, we have fast transition controlled to a large degree by the gate driver and to a lesser degree by the source parasitic inductance. Conversely, in the turn-off transition the HS MOSFET is passively driven on by the

back-EMF effect and “regulated” to operate as a current source. In addition, the voltage drop from drain to source of the HS MOSFET is a small fraction of V_cc as can be calculated from Equation 2.

The Low-side MOSFET

The loss mechanism for the LS MOSFET agrees so far with the current understanding of their behavior. As HS MOSFET turns off, the inductor current starts commutating into the body diode of the LS MOSFET and the drain-source voltage of the LS MOSFET switches from V_CC to one diode drop, as seen on Fig. 7(b). As the gate voltage of the LS MOSFET turns on after the dead time, the drain current will continue to rise until it reaches the inductor current and the HS MOSFET turns off.

The slow rise time of the current can be directly attributed to the HS and LS source inductors because the inductors will resist any sudden change in current. Because the current flows from the source to the drain of the LS MOSFET, the voltage developed across the source parasitic inductance (lead and pc-board trace inductance) actually assists the turn-on cycle.

It appears that the parasitic inductance of the HS MOSFET determines the current transition here, as described before. When the LS MOSFET is turned off, the current will flow in its body diode until the HS MOSFET starts turning on, the voltage developed across the parasitic inductance will force the gate-source voltage into a negative value. Since the gate is held at the ground, this will help keep the LS MOSFET off during this transition.

The graphs shown in Fig. 7 present no surprises, and hence, the losses in the LS MOSFET appear to continue to be mostly conduction or ohmic losses since the drain-source voltage is always held low while the current is switching on and off. The measured currents in the HS and LS MOSFETs do not exactly match in these graphs because of the inaccuracies of the current dividers used.

Noninvasive Current Measurement

Current measurement in high-frequency dc-dc converters is difficult, because inserting a loop to measure current or a sense resistor will alter the original conditions. This led us to propose the use of the circuit in Fig. 8.

In this approach, we use two parallel loops of wire (L1 & L2 and L3 & L4). L1 and L3 are loops that are barely long enough to accommodate the scope current probe, while L2 and L4 are a very short wire (e.g., the MOSFET package lead plus the pc-board trace inductance). In this way, we have created a current divider where the majority of the current flows in the short wire (L2 and L4) without disturbing the circuit much while a much smaller portion of the source current will flow in the large loop (L1 and L3). The ratio between the two currents in the two parallel paths is

which is roughly the ratio between the two areas occupied by the two loops. Fig. 9 shows the method used to measure a sample of the current with the least disturbance of the original circuit.

The current understanding of HS MOSFET switching must include the effects of the circuit parasitic inductances and their effects on the switching performance of the synchronous buck dc-dc converter and other topologies. These parasitic inductances are namely the package lead inductance and the pc-board trace inductances.

The HS MOSFET source parasitic inductance has several effects on switching performance. First, the HS MOSFET will continue to conduct current even after the gate voltage reaches 0 V (ground level). This is due to the negative back EMF of the parasitic inductance pulling the source to a negative level with respect to the gate, allowing the current to flow freely. A second effect is that the drain current fall time is proportional to the current level. This leads to losses proportional to I_drain², and thus much larger switching losses even with the best MOSFETs. Thirdly, the drain current fall time is proportional to the parasitic inductance value. This puts a limit on the maximum power efficiency available from the dc-dc converter regardless of how fast the HS MOSFET is.

In order to reach high switching speeds, certain component characteristics must be obtained. MOSFETs must have low Q_gd and Q_sw. Packages must facilitate the connection of the gate driver directly between the gate and the source using non current-carrying leads or pc-board traces. A ball grid array (BGA) is an excellent package for this purpose. Furthermore, gate drivers must provide a true differential gate-source drive signal, low source and sink resistance, and very fast rise and fall times.