Access Official Technical PDF

Share Summary:LinkedInEmail

Managing 3L-ANPC over-voltage is an important design consideration as power converters adopt wide-bandgap devices such as silicon carbide (SiC) MOSFETs. Their fast switching can support higher switching frequencies, but rapid voltage and current transitions also make converter parasitics more influential. In a three-level active neutral point clamped (ANPC) converter, multiple commutation paths can expose a switch that is not actively switching to a substantial voltage spike.

A control adjustment can reduce this stress without adding circuit components or requiring a complex software algorithm. Understanding why the spike occurs—and how the adjustment changes device voltage conditions—helps power electronics designers evaluate its use alongside layout and switching-speed decisions.

Why non-active switches can experience over-voltage

Three-level converters are used in applications including grid-tied inverters, motor drives, electric transportation and uninterruptible power supplies. Compared with two-level converters, they can offer benefits such as lower electromagnetic interference, improved power quality and lower required device voltage ratings. ANPC designs also allow active control of the switches.

Those advantages come with a more complex switching environment. In the outer-mode modulation examined for a single ANPC phase, some switches operate at high frequency while others switch at line frequency or remain off during a portion of the line cycle. Consequently, a switching transition in one part of the phase leg can affect devices outside the main, actively commutating loop.

For example, during a transition between the O- and N-states, a high-frequency switch commutates current to another device. The associated voltage transition drives a conventional commutation loop, but an additional loop also exists through other switches in the converter. Under conventional control, a non-active switch can be effectively paralleled with the switching device. Its drain-source voltage then follows the voltage transition, despite the switch not receiving a switching command.

How parasitics and capacitance shape the voltage spike

The additional loop can be represented as a parasitic inductance interacting with the output capacitance of a non-active device. A fast voltage transition excites this LC network, producing a resonant voltage response. The resulting peak depends on factors including the transition rate (dv/dt), loop inductance and device output capacitance.

Output capacitance is not constant: it typically changes with the voltage across the device. At low voltage, capacitance can be much higher than it is at high voltage. This nonlinearity affects how energy moves through the resonant loop and can push the peak voltage beyond what a constant-capacitance model would predict. The coupling between loops can also mean that a high peak on a non-active switch coincides with increased voltage stress on the active switching device.

Conventional mitigation options have trade-offs. Adding capacitance across a non-active switch can increase output-capacitance (Coss) loss because that capacitance charges and discharges during switching. Reducing dv/dt by slowing the switching transition can increase switching loss. Lowering loop inductance through layout improvements is useful, but the additional loop may be physically large and difficult to minimize. Snubbers add components and can increase loss and circuit complexity.

A control change to limit 3L-ANPC over-voltage

The proposed approach changes the switching state so that S3H and S3L remain off when they are not switching. In the operating condition analyzed, this prevents the non-active switch from being directly paralleled with the high-frequency switching device. The additional loop then includes multiple device output capacitances rather than the same direct capacitive relationship found with conventional control.

This changes the steady-state voltage distribution before the next commutation. The non-active switch starts with a higher voltage across its output capacitance, helping avoid the low-voltage region where capacitance changes rapidly. With less influence from this strongly nonlinear region, the subsequent resonant response can produce a lower peak voltage.

The approach does not remove parasitic inductance or change the physical converter layout. Instead, it changes the electrical conditions that shape the transient. Because device capacitance is voltage-dependent, numerical analysis may require simulation using capacitance data from the selected device rather than relying on a constant-capacitance approximation.

What converter testing showed

Testing on a single-phase 3L-ANPC converter using SiC MOSFET modules compared conventional and proposed control. The test used a 500 V DC bus, a 400 A peak load current and a 60 kHz switching frequency. The busbar used a laminated structure, with estimated additional-loop inductance of 17 nH.

At a measured active-switch dv/dt of 10 V/ns, conventional control produced peak drain-source voltages of 736 V on the non-active switch and 754 V on the high-frequency switch. With the proposed control, the corresponding peaks were 560 V and 592 V. Thus, both measured peaks fell by more than 20% in this test. The reduction on the active switch reflects coupling between the devices and switching loops.

The control also redistributed steady-state voltage across devices, resulting in an additional calculated Coss loss of 124.1 μJ per switching cycle in the reported comparison. The practical assessment therefore involves balancing voltage-stress reduction against this added capacitance-related loss; the reported work notes that the additional loss was small relative to switching loss at full load in the studied case.

Design implications

For converter engineers and system designers, the results highlight that non-active devices deserve attention during transient analysis—not just the switches receiving gate commands. The relevant design checks include the complete commutation-loop topology, device output-capacitance curves, busbar parasitics and the effects of control state selection.

The proposed method offers a control-level option when layout changes, slower switching or added snubbers are undesirable. Its measured performance applies to the tested converter and operating conditions, so designs should verify voltage sharing, device stress and losses with their own topology, components and modulation strategy.

Conclusion

In a 3L-ANPC converter, over-voltage on a non-active switch can arise from an additional commutation loop and the nonlinear output capacitance of the devices within it. Keeping selected switches off when they are not switching changes the initial voltage distribution and reduces the influence of the capacitance’s strongly nonlinear region. The tested SiC converter showed lower peak voltage on both non-active and active switches without added hardware or a complex control algorithm. For practical designs, this is a useful complement to careful parasitic-loop analysis and validation of switching losses and voltage margins.

Access Official Technical PDF

About The Author