Understand and reduce DC/DC switching-converter ground noise

Technology News | November 22, 2011

By eeNews Europe

DC/DC switching-power converters are notorious for physically disrupting an otherwise carefully designed system and circuit schematic designs. These power converters drive unwanted charge onto electrical ground, causing false digital signals, flip-flop double clocking, EMI, analog-voltage errors, and damaging high voltages.

As the complexity of these designs increase and applications become more densely populated, the physical-circuit implementation plays a critical role in the electrical integrity of the system. This article illustrates two major sources of ground noise and offers suggestions on how to reduce both.

Ground noise: Problem #1

Figure 1 shows an ideal buck converter with a constant load current. Switches t₁ and t₂ toggle back and forth, chopping V_in across L_buck and C_buck. Neither inductor current nor capacitor voltage can change instantaneously, and the load current is constant. Hopefully, all switching voltages and currents successfully span L_buck or pass through C_buck respectively, since an ideal buck converter produces no ground noise.

But experienced designers know that a buck converter is a notorious noise source. This fact means that Figure 1 is missing important physical elements.

Figure 1: Buck converter circuit—inductor current cannot change instantaneously, so identifying a source of ground bounce in an ideal buck converter is not easy.

Whenever charge moves, a magnetic field develops. Current in a wire, resistor, transistor, superconductor, and even a capacitor’s plate-to-plate displacement current creates a magnetic field. Magnetic flux, Φ_B, is magnetic field, B, passing through a current loop area, A, and equals the product of the field cutting the loop surface at a right angle, Φ_B= B·A. The magnetic field at a distance, r, encircling a wire is directly proportional to the wire’s electrical current, B = μ_oI/2πr.

Electrical components have length and charge must flow from one device to the next in the various wire segments. But moving charge creates a magnetic field, so the schematic in Figure 1 can be improved. Figure 2 shows a better model of a simple buck converter.

Figure 2: Magnetic flux = (B-field) × (current loop area). Changing flux induces voltage. As a buck switches, the changing current-loop path causes a changing flux and induces ground bounce.(Click here for enlarged image.)

In Figure 2, the wire remains ideal in every way, except current must flow some distance in each segment while traveling from one electrical component to the next. As this charge flows, magnetic field wraps around the energized wires and is magnetic flux passing through the t1 and t2 switch loops.

Changing t1 and t2 current-loop areas is the first major source of switching-converter ground noise. Magnetic flux in the Vin-t1-gnd loop grows and collapses on every switch cycle. That changing flux induces voltage everywhere in that loop, including the ideal ground return line. No amount of copper, not even a superconductor, can eliminate this induced voltage. Only a reduction in the changing magnetic flux will help.

Changing magnetic flux has three factors: rate-of-change, magnetic field strength, and loop area. Since the clock frequency and maximum output current may be design requirements, minimizing loop area becomes the best solution.

Inductance is proportional to magnetic flux, so Figure 3 shows an electrical model for Figure 2 where changing current in parasitic inductor L_p1 causes ground noise, while constant current in parasitic L_p2 does not.

Figure 3: Parasitic inductance models energy stored in the magnetic field. Changing current in L_p1 induces ground bounce whereas constant current in L_p2 does not. (Click here for enlarged image.)

Although Figure 3 presents the problem in a familiar way, it makes a poor substitute for the physically enhanced model shown in Figure 2. Figure 3 shows parasitically induced voltage across L_p1 and L_p2; whereas voltage will actually be induced everywhere in a loop enclosing changing magnetic flux. However, this circuit element drawing will still serve the purpose of showing how to reduce induced ground noise.

Figure 4: Careful input-capacitor placement minimizes the changing loop area, and routes the changing current away from ground return to eliminate ground bounce. (Click here for enlarged image.)

As drawn in Figure 3, ground-return current flows and changes in L_p1, and it causes a voltage bounce problem. But a carefully placed input capacitor, as shown in Figure 4, reduces the parasitic magnetic-flux area, and routes changing buck current in a path that does not include ground return.

In this case, current in parasitic inductors L_p1 and L_p2 is constant, so the ground voltage will be stable. Additionally, the reduction in this magnetic flux area proportionally reduces EMI and all other unwanted, induced loop voltages as modeled in Figure 3.

In short, the first important source of switching converter ground noise is a result of changing magnetic flux area. Good printed circuit board (PCB) design uses both trace routing and careful bypass-capacitor placement to minimize changing current-loop areas and changing current in a ground-return path.

Ground noise: Problem #2

The second, major ground-noise problem, shown in Figure 5, is a result of parasitic-inductor capacitance.

Figure 5: Changing LX node voltage pumps charge through the parasitic buck-inductor capacitance, C_L, and into the parasitic ground-path inductors, L_p1 and L_p2 , causing ground noise. (Click here for enlarged image.)

Voltage cannot change instantaneously across a capacitor, nor can current instantaneously change through an inductor. So, voltage changes on the LX node couple directly across both the parasitic buck-inductor capacitance, C_L, and the buck-filter capacitor, C_buck, to appear across the parasitic ground inductors, L_p1 and L_p2.

Initially, no charge flows, but in the next moment, current builds in all of those components until the energy stored in the parasitic buck inductor capacitor,

E_CL = ½ C_LV_LX²,

transfers to the wiring’s parasitic magnetic field,

E_Lp = ½ L_pi²_{changing_max}

(where L_p = the sum of all parasitic loop inductors). Then like a swing, that unwanted energy passes back-and-forth from the electric to the magnetic field until it radiates or dissipates in resistive elements not modeled in Figure 5.

Both the peak voltage and the duration of a ground-noise oscillation are a problem. The peak voltage, measured at node Vgb, is a function of the LX node’s voltage change, the parasitic buck inductor capacitance, C_L, and additional parasitic trace capacitance (not shown). A large C_L stores more energy, so smaller is better. After selecting the buck inductor’s inductance and current rating, choose an inductor with the highest self-resonate frequency to limit the capacity of C_L.

An inductor’s self-resonate frequency is:

f_{self_resonates}= 1/[2π√(L_buckC_L)].

Notice that a doubling of the self-resonate frequency reduces the parasitic inductor capacitance, and therefore the ground-noise energy, by a factor of four!

In the case where performance takes priority over cost, maintain the same value of inductance by replacing the single L_buck inductor in Figure 5 with two series-connected inductors, each having ½L_buck (Figure 6). For a manufacturer’s series of inductors, the parasitic capacitance is typically proportional to the rated inductance, so one-half the inductance results in one-half the parasitic capacitance.

When inductors are series connected, their values add to increase inductance, but parasitic capacitors add as the inverse sum of inverse values, to decrease total parasitic capacitance. In the case of two series-connected one-half L_buck inductors, total inductance will be L_{buck_new} and total parasitic capacitance will drop by a factor of four to one-quarter C_L.

This reduction is parasitic inductor capacitance will, in turn, reduce ground bounce, Figure 6.

Figure 6: Two series-connected inductors have the same inductance but with one-quarter the parasitic capacitance; charge-pumping is reduced and, therefore, so is ground bounce. (Click here for enlarged image.)

By exploring the models and understanding the two sources and mechanisms of ground noise as induced by the ubiquitous DC/DC switching converter, engineers can minimize the effects in the early stages of design, component selection, and layout, and the subsequent product headaches and re-spins.

About the author

Jeff Barrow is a Senior Director of Analog IC Design at Integrated Device Technology, Inc. in Tucson, AZ. He works on the development and usability of power integrated circuits and is an active analog-IC designer. He received a bachelor’s degree in electrical engineering from the University of Arizona (Tucson), and his personal interests include geology, astronomy, physics, and electronics.