IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO.

3, MAY/JUNE 2005 855

A Comparison of Three-Level Converters Versus

Two-Level Converters for Low-Voltage Drives,
Traction, and Utility Applications
Ralph Teichmann and Steffen Bernet, Member, IEEE

Abstract—This paper evaluates three-level topologies as alterna- converters. With the emerging need for higher switching fre-
tives to two-level topologies in converters for low-voltage applica- quencies in applications such as high-speed drives, a desire for
tions. Topologies, semiconductor losses, filter aspects, part count, smaller and lighter (filter) components in grid-connected and
initial cost, and life-cycle cost are compared for a grid interface, a
conventional drive application, and a high-speed drive application. traction converters, a three-level-based low-voltage converter
family appears to be one technologically attractive solution.
Index Terms—Life-cycle cost, multilevel converters, semicon- A first indication of a commercial interest was stated in [1]
ductor losses.
and [2]. References [3] and [4] explore the utilization of
high-volume automotive components to build cost-effective
I. INTRODUCTION three-level low-voltage drive converters. Reference [5] shows
the efficiency gains of three-level converters over two-level
T HE low-voltage power conversion market (nominal
line-to-line voltage up to 690 V (IEC); 575 V
(ANSI)) is almost exclusively satisfied by the conventional
converters in wind power applications with a high share of
partial load operation. A three-level converter technology is
two-level dc-voltage link hard-switching converter topology. discussed for three typical application cases. Insulated gate
Alternatives such as matrix converters or soft-switching con- bipolar transistor (IGBT) semiconductor losses and switching
verters have failed to penetrate this market. Technological frequency boundaries are discussed for state-of-the-art IGBT
progress has been slow, relying mainly on semiconductor technology. Initial cost and life-cycle cost comparisons will
device improvements and control intelligence refinements, as conclude the evaluation.
well as on better integration and packaging techniques.
In contrast, the medium-voltage power conversion market II. TOPOLOGIES
offers some diversity with the presence of a dc-current link
converter and various dc-voltage link multilevel converter struc- DC-voltage link multilevel converters were proposed in
tures. In particular, the multilevel converters became very suc- several configurations. The diode-clamped multilevel con-
cessful in the past decade. While the development of the multi- verter (DCMLC) [6], also known as the neutral-point-clamped
level structures was mainly driven by the limited semiconductor multilevel converter, the flying-capacitor multilevel converter
voltage blocking capability in conjunction with static and dy- (FCMLC) [6], also known as the imbricated cell converter, and
namic voltage-sharing concerns, it turned out that the multilevel the series-connected/cascaded multilevel converter (SCMLC)
topology offered some additional benefits. Among those are are distinguished [6]. A comparison of these topologies was
a superior harmonic spectrum for a given gate switching fre- presented in [6] and [7]. The common roots of DCMLC and
quency, a lower overvoltage stress at cables and end windings the FCMLC topologies were shown in [8]. It should be added
of transformers/motors, a lower common-mode voltage, and that the total semiconductor losses and the terminal voltage
substantially lower semiconductor switching losses. spectrum of a three-level diode-clamped converter and a
The objective of this paper is the evaluation of the poten- three-level flying-capacitor converter are exactly the same for
tial benefits of a three-level topology in low-voltage power any given operating point in terms of modulation index and
displacement angle. Unidirectional and partially bidirectional
converters as shown in [9] also feature exactly the same total
Paper IPCSD-04-075, presented at the 2003 Industry Applications Society semiconductor losses and terminal voltage spectrum if operated
Annual Meeting, Salt Lake City, UT, October 12–16, and approved for publica- within functional boundaries. Despite using different current
tion in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Industrial
Power Converter Committee of the IEEE Industry Applications Society. Manu- paths, the device number and device types conducting during
script submitted for review March 1, 2004 and released for publication February each switching state and their duty cycles remain the same.
3, 2005. This work was supported by ABB Corporate Research, Germany. Only the distribution of the semiconductor losses among the
R. Teichmann is with GE Global Research, Niskayuna, NY 12309 USA
(e-mail: teichman@crd.ge.com). semiconductor modules is different. This assumes a comparable
S. Bernet is with the Fakultät IV Elektrotechnik und Informatik, Institut modulation technique, a neutral point clamp/auxiliary diode
für Energie- und Automatisierungstechnik, Fachgebiet Leistungselek- technology similar to that found in the inverse diodes of the
tronik, Technische Universität Berlin, D-10587 Berlin, Germany (e-mail:
steffen.bernet@tu-berlin.de). main switches, and disregards the influence of parasitic circuit
Digital Object Identifier 10.1109/TIA.2005.847285 elements.
0093-9994/$20.00 © 2005 IEEE
856 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 3, MAY/JUNE 2005

converters with the same device technology eliminated device

technology dependency. Representing the main competing
technologies, conventional and trench-gate nonpunch-through
(NPT) IGBTs (EUPEC) [11] and trench-gate punch-through
(PT) IGBT’s (MITSUBISHI) [12] were included. With a ma-
jority of three-phase low-voltage converters in a power level
between 10 kVA–1 MVA, the evaluation is based on a 100-kVA
converter. Common to both the three-level and two-level
topologies is a nominal grid/motor voltage of 400 Vrms and
a dc-bus voltage of V. The same displacement
power factor (DPF) is set in both topologies. Two-level and
the three-level converters are controlled by an asynchronous,
continuous, sine-triangle modulation scheme with added third
Fig. 1. Three-level NPC converter (DCMLC).
harmonics. Both converter topologies feature the same carrier
frequency yielding the first carrier frequency band at the same
frequency in the ac voltage spectrum.
The following converter applications were compared: 1) a
100-kVA 400-V 50-Hz grid converter; 2) an inverter for a con-
ventional 75-kW 400-V NEMA B induction motor; and 3) an
inverter for a 90-kW 400-V 1.2-kHz high-speed drive.

IV. EVALUATION OF SEMICONDUCTOR LOSSES

Semiconductor losses are a central evaluation criterion for
a topology due to their direct correlation with virtually all
other electrical and mechanical converter parameters. Apart
from gate control characteristics and commutation circuit
parameters, the semiconductor device losses in three-phase
Fig. 2. Conventional two-level converter.
voltage-source dc/ac converters depend on the device tech-
nology, the dc voltage and ac current levels ( ),
the junction temperature , the switching frequency , the
A useful implementation of the SCMLC starts with a displacement phase angle (displacement power factor), and the
five-level topology featuring four times the semiconductors/gate modulation index . The modulation index is defined as peak
drives compared to a conventional two-level converter. This is line-to-line voltage over dc voltage with a definition range of
based on the desire for the optimum utilization of the inevitable 0–1. A displacement phase angle of zero (unity displacement
isolation transformers. The DCMLC is best implemented as power factor) designates a real power flow from dc to ac, i.e.,
a three-level converter; higher order levels typically require a inverter operation.
series connection of the clamp devices. The FCMLC is subject
to a more complex startup routine and a higher number of A. Device Technology/ Switching Frequency
capacitors with their detrimental effects on reliability. Based
For a given dc-link voltage the devices in the three-level
on these considerations a three-level DCMLC (Fig. 1) is set
converter require half the blocking voltage capabilities of that
against a two-level converter (Fig. 2) for the low-voltage power
of a two-level converter, featuring superior conduction and
converter tradeoff analysis.
switching characteristics for any given current. For example,
600-V IGBTs feature on-state voltages that are roughly 10%
III. DEFINITION OF EVALUATION CRITERIA lower than that of a 1200-V IGBT at the same rated current and
Low-voltage dc/ac power conversion for drives and grid con- technology (Table I). Similarly, the switching loss energies of
verters is usually required for nominal three-phase rms voltage a 600-V IGBT with the same device technology and current
levels of 200–230 V, 380–460 V, or 575–690 V. These voltage are smaller by a factor of 3–5. This implies that, theoretically,
ranges correspond to IGBT voltage classes (two-level structure) even a series connection of two 600-V devices features only
of 600, 1200, and 1700 V, respectively. A three-level converter 40%–60% of the switching losses of one 1200-V IGBT while
for similar terminal voltages is theoretically to be fitted with increasing the conduction losses by less than twice. Similar
300-, 600-, and 850-V modules. The topological evaluation high-loss energy differences exist between the voltage classes
was based on the 1200-V/600-V IGBT technology (two/three 1200 and 1700 V.
level) due to the widespread availability of semiconductors of To compare a two-level converter and a three-level converter
the same device technology and similar packaging. IGBTs with the semiconductor losses were calculated using a MATLAB-
300- and 900-V ratings are available as discrete devices [10], based simulation tool calculating the switch states and the in-
and can easily be packaged in modules if a general technology stantaneous device voltages and currents over one period of the
transition to three-level converters is desired. Fitting both output frequency at a sampling rate of s. Conduction
TEICHMANN AND BERNET: COMPARISON OF THREE-LEVEL CONVERTERS VERSUS TWO-LEVEL CONVERTERS 857

TABLE I
IGBT LOSS DATA FOR MITSUBISHI (PT-IGBT) AND EUPEC (NPT-IGBT) DEVICES [10],[11]

TABLE II
SEMICONDUCTOR LOSS CHARACTERISTICS AND COEFFICIENTS

losses are attributed to devices being in the on-state. Switching passive device turn-off loss energy;
losses are distributed to the devices involved in a commutation instantaneous commutation voltage;
whenever a state transition takes place. The losses during the device manufacturer’s test voltage;
sampling period and the switching loss energies were calculated instantaneous device current;
by sampling period;
curve-fitting parameter.
On-state voltages and loss energies as a function of the de-
(1)
vice current for the IGBT and diode were obtained from man-
(2) ufacturer data sheets [11], [12]. The curve-fitting parameters
are summarized in Table II. For simplicity, the variation of the
(3)
semiconductor losses as a function of junction temperature is
neglected; the values specified for a junction temperature of
(4) C were assumed. Details of the semiconductor loss
evaluation tool can also be found in [13] and [14]. For the semi-
(5) conductor loss and switching frequency boundary evaluation the
converters were assumed to be operated in steady state at a con-
where stant case temperature of C and a maximum junction
active device on-state loss energy in sampling period; temperature of C.
passive device on-state loss energy in sampling Fig. 3 shows the total semiconductor losses in a three-phase
period; two- and three-level inverter as a function of device technology
active device turn-on loss energy; (PT, NPT, trench/conventional, fast/low ) and carrier fre-
active device turn-off loss energy; quency at one operating point ( V, ,
858 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 3, MAY/JUNE 2005

Fig. 3. Total semiconductor losses as a function of carrier frequency and

parameter device technology (sine-triangle modulation with added third
Fig. 5. Semiconductor loss distribution of two-level converter as a function of
harmonic, V = 400 V, V = 700 V, T = 80 C, T = 125 C,
displacement power factor (sine-triangle modulation with added third harmonic,
DPF = 0:9, I = 147 A, 200-A IGBTs). V = 400 V, V = 700 V, T = 125 C, I = 147 A, f = 10 kHz, 3 2
IGBT: CM200DU24F).

B. Load Current
The magnitude of the load current in relation to the device
current rating has a substantial influence on the switching fre-
quency above which the three-level converter becomes superior.
This crossover switching frequency is shifted to lower values as
the load current becomes smaller. This is due to the fact that
on-state losses decline faster ( quadratic dependency) than
switching losses ( linear dependency) with falling device cur-
rent and both loss categories are of roughly the same magni-
tude. Fig. 4 shows an equi-loss line (identical semiconductor
losses of two- and three-level converters) of a dc/ac inverter
( V, , , A) as a func-
tion of the load current. For any given load current, switching
frequencies above the lines are preferably implemented with a
three-level converter. Most converters are typically operated in
Fig. 4. Equi-loss line (two-level/three-level converter) as a function of load
current and parameter semiconductor technology (sine-triangle modulation with a load range between 0.4–0.7 per unit. A three-level topology
added third harmonic, m = 0:95, V = 700 V, DPF = 0:9, I = 147 A). therefore becomes a very attractive solution for switching fre-
quencies as low as – kHz.

, A). Both 1200- and 600-V modules C. Modulation Depth, Displacement Power Factor
have the same nominal current rating. A device developed for
fast switching operation (FF200R12KS4, mJ, A variation of the modulation depth and the terminal dis-
mJ, V at A) was included for com- placement angle changes the total semiconductor losses and
parison purposes. loss distribution among active and passive semiconductors.
Despite the higher conduction losses (two devices in current Figs. 5–8 depict the loss characteristics with the acronyms
path) the three-level topology quickly features lower total PconD, PconT, PonT, PoffT, and PoffD describing diode con-
semiconductor losses than conventional and fast IGBTs in duction losses, IGBT conduction losses, IGBT turn-on losses,
a two-level topology as the switching frequency increases. At IGBT turn-off losses, and diode turn-off losses, respectively.
this operating point the three-level topology features fewer total In a two-level converter feeding a three-phase balanced load
semiconductor losses at switching frequencies above 4/5 kHz the losses of all diode chips are the same and the losses of all
for trench-gate PT as well as planar NPT IGBTs. Clearly, IGBT chips are the same [15], [16]. Fig. 5 shows that the total
the recent introduction of the trench-gate technologies shifts conduction losses vary with the displacement power factor with
this crossover point to smaller switching frequencies in favor maximum conduction losses in the IGBT and the inverse diodes
of the three-level converter. Some of the loss curves are not at and , respectively. The switching losses
extended to the upper switching frequency range, indicating of each individual chip are not affected by the displacement
that the losses cannot be dissipated from the device within power factor. References [15] and [16] have shown that the con-
the thermal specifications ( C, C). duction losses of the IGBT chips increase almost linearly with
TEICHMANN AND BERNET: COMPARISON OF THREE-LEVEL CONVERTERS VERSUS TWO-LEVEL CONVERTERS 859

Fig. 8. Semiconductor loss distribution of three-level converter in rectifier

mode as a function of modulation index (sine-triangle modulation with added
0
third harmonic, V = 400 V, V = 700 V, DPF = 0:9, T = 125 C,
Fig. 6. Semiconductor loss distribution of three-level converter as a function of
displacement power factor (sine-triangle modulation with added third harmonic,
I = 147 A, f = 10 kHz, 6 2 IGBT: CM200DU12F).
V = 400 V, V = 700 V, T = 125 C, I = 147 A, f = 10 kHz, 6 2
IGBT: CM200DU12F).

Fig. 9. Common-mode voltages of two-level converter and three-level

converter (f = 10 kHz, V = V ).

Fig. 7. Semiconductor loss distribution of three-level converter in inverter decline. The influence of different continuous and discontinuous
mode as function of modulation index (sine-triangle modulation with added PWM schemes on the losses of a three-level DCMLC was dis-
third harmonic, V = 400 V, V = 700 V, DPF = 0:9, T = 125 C,
I = 147 A, f = 10 kHz, 6 2
IGBT: CM200DU12F).
cussed in [17].

V. PASSIVE COMPONENT CONSIDERATIONS

increasing modulation depth while the conduction losses of the
inverse diodes are declining. Passive components are a substantial contributor to weight,
Diode and IGBT losses exhibit a similar pattern in the three- cost, and losses in power converters. In fact, a substantial share
level converter. The sum of each type of switching losses, e.g., of the converter cabinet of industrial converters operating at
diode switching losses, is also constant. However, the distribu- switching frequencies between 2–5 kHz is taken up by passive
tion of both switching and conduction losses among the devices filter components. The design implications for the ac filter, dc
is a function of the displacement angle (Fig. 6). Inverse diodes filter, filter, and common-mode filter are discussed.
( 1, 2, 3; 1–4) are subject to the same conduction
losses irrespective of the displacement angle. The loss depen- A. Common-Mode Filter
dency on the modulation depth is displayed in Figs. 7 and 8 for Common-mode currents are considered to be the main
inverter and rectifier mode, respectively. The conduction losses reasons for bearing currents/failures and electromagnetic inter-
of the IGBTs in switches and remain constant while the ference (EMI) problems. Fig. 9 depicts the magnitudes of the
conduction losses of the NPC diodes and the conduction losses common-mode voltages, which cause the common-mode
of the IGBTs in and vary with the modulation index in in- currents, impressed on the ac system for a two- and a
verter mode. In rectifier mode the conduction losses of the main three-level converter with a sine-triangle modulation. Clearly,
switch inverse diodes increase with increasing while the con- the three-level converter has a much smaller common-mode
duction losses of the NPC diodes and the IGBTs in and voltage. Absolute values depend on the operating point and
860 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 3, MAY/JUNE 2005

modulation scheme. For the conventional sine-triangle modula- TABLE III

tion, a reduction of the switching frequency component of the COMPONENT COUNT/RATING COMPARISON
common-mode voltage of the three-level converter by roughly
25%–30% was found. Furthermore, [18] proposes a three-level
converter modulation scheme with which the common-mode
voltages can be suppressed completely.

B. Filter
Transient overvoltages caused by the pulsewidth-modu-
lation (PWM) switching operation are a major concern for
transformers/generators/motors connected with long cables.
The overvoltage phenomenon due to fast switching transients
(high ) in long cables has been discussed in several
papers [19]–[26]. The overvoltage at the machine/transformer
terminals is caused by a reflection of the voltage pulse of the
inverter due to a mismatch in load and cable surge impedance.
If the voltage pulse takes longer than one-third of the voltage
VI. EVALUATION OF PACKAGING AND CONTROLS
rise time to travel from the inverter to the machine/transformer
a full reflection will occur at the machine/transformer terminals A. Packaging
and the voltage pulse amplitude will approximately double
[22]. A 50% reduction in the pulse magnitude in the three-level The design of the packaging is highly application dependent.
converter will obviously reduce the overvoltage stress at the Until higher production volumes justify an integration of com-
load terminals in a design without additional filter. plete converter legs, the three-level topology requires at least the
The drastically reduced switching losses of the three-level additional NPC diode modules. In cases where single 1.2-kV
converter can also be used to reduce the using the IGBT modules can be replaced by a dual 600-V module (typically,
gate units. A potential avoidance of the filter at the con- when device current rating is set by thermal considerations) the
verter terminals may justify the increased switching losses. main switch arrangement must not be modified.
Specifically for an introduction of a three-level converter into
C. AC Filter a dc/ac converter with 1.2-kV IGBTs, the arrangement of the
dc-link capacitors must not be modified. Other voltage levels
The high-frequency content in the terminal voltage of the
require a re-design of the dc capacitor arrangement (Table III).
three-level converter is substantially smaller due to the three-
The total size and weight of the converter will certainly ben-
level characteristic with an effective commutation voltage of
efit from the smaller cooling unit and smaller passive compo-
50% of that of the two-level converter. The resulting harmonic
nents. In certain applications a transition of the cooling method
currents are reduced by the same factor [24]. Because the high-
to simpler convection cooling might reduce maintenance efforts
frequency losses are a function of the square of the high-fre-
and increase reliability.
quency content of the current, the application of a three-level
converter at a given inductance, i.e., motor/generator, will re-
sult in substantially smaller losses. Alternatively, the ac filter B. Sensors/Controls
inductance can be smaller/less heavy for a three-level topology The number of dc-link voltage sensors, in some cases as
to meet a given specification in terms of voltage or current har- simple as a resistive divider, is increased by a factor of two.
monics. The number of -processors remains the same. For example,
Specifically, for a three-phase system with a modulation the widely used fixed-point TMS320F24x series offers between
scheme without a phase difference in the carrier signals, the 7–16 dedicated PWM channels. An additional benefit exists
carrier switching frequency is a common-mode signal and for larger manufacturers that can reuse their medium-voltage
is not present in the line-to-line voltages. Only the typically converter control platform, producing a unified control platform
smaller side bands around the switching frequency need to be for low- and medium-voltage converters.
attenuated, additionally reducing filter size and losses. The stabilization of the center point potential in the three-
level diode-clamped converter imposes one additional control
D. DC Filter criteria. Control of the dc-link center point can be achieved un-
Two- and three-level converters show ripple current com- less the converter is operated in overmodulation [26], [27]. Un-
ponents on the dc side at side bands that are centered around doubtedly, the biggest disadvantage of the three-level converter
the switching frequency. The three-level converter additionally is the higher number of gate units with its accompanying isola-
shows a third harmonic (and other triplen harmonics) at the dc tion requirements. The number of gate drivers must be increased
side [25]. The three-level converter requires a dc-link center by a factor of two. However, since each gate unit is subject
point management. Despite the fact that these points are intrin- to only half the switching frequency on average, the total gate
sically stable, device asymmetries and variations in component drive power remains the same. Simplified gate drive units for
values will offset this balance. three-level structures were presented in [4].
TEICHMANN AND BERNET: COMPARISON OF THREE-LEVEL CONVERTERS VERSUS TWO-LEVEL CONVERTERS 861

VII. RELIABILITY EVALUATION

The three-level converter features a higher part count

(Table III), which tends to compromise reliability. However,
a closer analysis reveals other effects that partly balance the
component-count-based reliability numbers. Assuming a fully
integrated converter leg and a dedicated control processor, the
unavoidable additional components are the gate units, cabling,
and sensors.
The failure in time (FIT; FIT failure in
operating hours) rate of the semiconductors (
– ) is very small compared to other components in the
system. Cabling, control electronics, and fan failures are the
predominant causes of failure in power electronics [28].
All causes of failure are/can be positively influenced by ade-
quate design choices. In particular, the lower losses of the three-
level converter can reduce the ambient temperature of the gate Fig. 10. Total semiconductor losses during grid operation of converter (V =
units (e.g., 10–15 K) such that the number of failures is sta- 750 V, f = 4 kHz, m = 0:85, DPF = 0:98=00:98, T = 125 C, 3 2
tistically equal to that of a two-level converter, despite doubling CM200DU24F, 6 2 CM200DU12F).
the number of units. The power and number of cooling fans, a
component contributing highly to total system failures, can be
reduced, in particular, in high-frequency applications. Similarly,
the capacitor life expectancy can be increased due to the lower
ripple current stress.
A complete reliability assessment is case dependent; a
three-level converter is not automatically inferior to a two-level
converter.

VIII. EVALUATION OF APPLICATION EXAMPLES

A. Grid Converter
Grid converters operate in rectifier and inverter mode
depending on the application. Both types of operation are
characterized by a relatively constant modulation index and
power factor. Variations are to be expected in the current due
to varying load conditions. Typical mains frequencies are Fig. 11. Total semiconductor losses during grid operation of converter (V =
50/60 Hz. Certain utility interconnect requirements regarding 750 V, f = 8 kHz, m = 0:85, DPF = 0:98=00:98, T = 125 C, 3 2
the current quality and the behavior during grid irregularities
CM200DU24F, 6 2 CM200DU12F).
must be fulfilled, e.g., IEEE 519-1992. A switching frequency
between 2–4 kHz is typically chosen at this power level, which For a switching frequency of kHz, the semiconductor
has traditionally been regarded as a good compromise between losses are displayed in Fig. 11. In this case, the three-level con-
filter size/losses and converter efficiency. verter is clearly superior in both applications over the entire load
The following parameters were set for the purpose of the range. For both rectifier and inverter operation the loss savings
evaluation of a grid-connected load and a grid-connected gen- amount to roughly 50%–55% and 25% at 0.2 and 1 p.u., respec-
eration unit: kVA, V, V, tively. It should be noted that the two-level converter cannot be
, and kHz. Fig. 10 depicts the operated at a switching frequency of kHz and the thermal
semiconductor losses of the two- and three-level converter as a requirements of C and C with the
function of the load current for the PT trench-gate technology MITSUBISHI 200-A devices. In contrast, the three-level con-
and a switching frequency of kHz. While for inverter op- verter can be operated with the 200-A devices up to a switching
eration the three-level converter is superior over the entire load frequency of 28 and 10 kHz in inverter and rectifier mode, re-
range, in rectifier mode the three-level converter is superior up spectively.
to a load current of about 0.8 p.u. at this switching frequency. An LCL filter is designed to meet grid current quality criteria
The loss savings amount to roughly 30%–35% at 0.2 p.u. current set forth in the recommendations of IEEE 519-1992. The design
for both rectifier and inverter operation. During rectifier opera- rules are outlined in [16]. In order to attenuate the switching
tion at rated load the losses of the three-level converter are 8% harmonics below the limits required, the inductance values and
higher than that for the two-level converter. filter capacitor values (Y-connected) as shown in Table IV are
862 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 3, MAY/JUNE 2005

TABLE IV
FILTER INDUCTANCE VALUES

respectively. For a switching frequency of 8 kHz, 200 A, 600 V,

and 400 A, 1200-V PT trench-gate IGBTs are assumed. As dis-
cussed earlier, the 1.2-kV devices are not able to dissipate the
losses at a switching frequency of kHz and the thermal
parameters of C, C. In this case,
400-A modules are needed and the installed switch power of the
three-level converter is only 62.5% compared to the two-level
converter (see Table III).
Apart from the converter losses, the losses in the sine filter or
the harmonic losses in the motor are smaller due to the lower
harmonic voltages of the three-level converter.

C. High-Speed Drives Converter

The elimination of the mechanical gear in applications such
as compressors and small gas turbines is the main advantage of
Fig. 12. Relative semiconductor loss difference for field-oriented operation of
induction machine with two-/three-level converter (V = 700 V, f = 4 kHz,
high-speed drive technology. Both inverter and rectifier func-
T 2 2
= 125 C, 3 CMX00DU24F, 6 CMY00DU12F, IM: P = 75 kW, tionality are, thus, required. High-speed drives in the 100-kVA
V = 400 V, 50 Hz, X = X = 0:1, X = 2:7, R = 0:021, R = power class are typically implemented as permanent-magnet
0:025).
machines operating at a fundamental frequency between
0.5–3 kHz. For moderate harmonic losses in the machine,
required. At this current level the weight for a three-phase input typically featuring a low leakage inductance, the minimum
inductor amounts to roughly 1 kg/10 H. switching frequency should be substantially above 9–12 times
the fundamental frequency. A similar switching frequency
B. Conventional Drives Converter range is desired for servo drives [2].
In contrast to the grid application, the modulation depth, dis- For the purpose of an evaluation of a two- and three-level con-
placement power factor, and load current are highly variable. A verter the switching frequency shall be set to kHz.
standard 75-kW 400-V 50-Hz induction motor ( The thermal boundaries were set to C and
p.u., p.u., p.u., p.u.) C. The analysis in the previous section has re-
is assumed. The current at rated operation amounts to vealed that the two-level converter with 200-A devices cannot
A. The losses of the two- and three-level converter based achieve this switching frequency within the thermal specifica-
on PT trench-gate devices were calculated at distinct operating tions given. The chip area of a 400-A 1.2-kV module was needed
points in the speed–torque plane. Field-oriented motor control to achieve the switching frequency while safely dissipating the
is assumed. For the carrier frequency of kHz the relative semiconductor losses. The investigation will be limited to rated
loss difference operation in both rectifier and inverter mode ( V,
, A, C, C).
(6) Fig. 13 shows the semiconductor loss distribution of the two-
level and three-level converter with smallest possible IGBT cur-
is shown in Fig. 12. The loss difference of the three-level and rent rating for inverter and rectifier operation. The lower losses
two-level converter is given relative to the losses of the two-level of the three-level converter permit the application of 200-A de-
converter. The losses of the three-level converter are smaller in vices for the inverter ( kHz) and 300-A devices for
all operating conditions. Variation of torque has a significant the rectifier application ( kHz).
impact on the loss savings. The loss savings range from 3% Additionally, the cooling effort needed for the three-level
(40 W) at rated torque and to 30% (120 W) at low torque. converter is substantially lower since the losses are roughly
Variations in speed modify the relative losses only marginally. 50% of that of the two-level converter. The semiconductor loss
If the switching frequency is increased to 8 kHz, the three- savings in actual terms amount to 2000 and 2100 W for inverter
level converter features substantially fewer losses in the entire and rectifier operation, respectively. The losses of the
speed–torque plane. The loss savings amount to filter, in this case a highly damped system required by the high
W and W for and , switching frequency, will be substantially smaller.
TEICHMANN AND BERNET: COMPARISON OF THREE-LEVEL CONVERTERS VERSUS TWO-LEVEL CONVERTERS 863

kilowatthour at present value are , , and

for Japan, Germany, and the USA, respectively.
For the three-level converter technology to be attractive on
a life-cycle cost basis the present worth savings must be
higher than the sum of additional initial and development costs
of the three-level technology. Additional benefits such as cost
savings due to a weight/volume reduction in mobile applica-
tions, simpler cooling method, reduced installation cost, etc.,
are not included.
Higher initial costs of the three-level converter are mostly at-
tributed to the semiconductor cost. A market survey of semi-
conductor prices (dual modules, year 2000) indicated that the
prices for 10 000 units would roughly amount to 17 ct/A and
9 ct/A for 1200- and 600-V IGBTs, respectively. This shows that
two 600-V IGBTs are only slightly more expensive than one
1200-V IGBT. Additionally, the 600-V NPC diodes were ac-
counted for with 6 ct/A per diode. The total IGBT/diode module
costs amount to $204 and $288 for the two-level and three-level
Fig. 13. Losses of two-level and three-level converter with smallest possible converters fitted with 200-A devices, respectively. For an op-
IGBT current rating (given in brackets) for inverter and rectifier operation
(V = 700 V, f = 20 kHz, m = 0:95, DPF = 0:98= 0:98, 0 eration at 8 kHz, the two-level converter must be fitted with
T = 125 C, 3 2 2
CMX00DU24F, 6 CMY00DU12F). 400-A devices, bringing the two-level converter cost to $408.
Gate units were accounted for by $10 per channel. The com-
IX. COST EVALUATION plete controller was included at $50 independent of the number
of channels.
Initial cost analyses and life-cycle cost analysis are funda- The Austerlitz heat sinks RLS 250.16-500 and RLS 300.14 –
mentally different cost assessment methods. With increasing 500 [30] were assumed, which were quoted at 400 USD and
integration of power electronics components in overall en- at 450 USD, respectively. Based on the loss calculations
ergy supply solutions and continuing trends toward business discussed in the previous section the converters operating at
models in which sole entities are manufacturer/owner/operator a switching frequency of 4 kHz and the three-level converter
a cost analysis over the entire product lifetime becomes of operated at 8 kHz were fitted with the RLS 250.16-500 heat
higher importance [14], [29]. A detailed cost analysis for the sink. The two-level converter operated at 8 kHz was fitted with
grid-connected bidirectional converter operated at 4 and 8 kHz the larger RLS300.14 type to achieve a comparable heat-sink
is presented. temperature at the higher losses.
Before the initial costs of the two- and three-level converters For a grid converter application, the immediate effect is a
are to be assessed, the cost savings potential of 1 kW of saved reduction of the filter cost. Prices for a three-phase coil were
losses will be determined. The net cost (excluding recoverable quoted at roughly 60 ct H, and the price for three capacitors
taxes) per kilowatthour paid by an industrial client in year 2000 at this voltage level is roughly 1 USD F. At 60 ct H in this
ranges from roughly 13 ct/kWh (Japan), 6 ct/kWh (Germany), current class the inductor cost amounts to $432 and $300 for the
4 ct/kWh (USA), down to 2 ct/kWh (South Africa) [15]. The two-level and three-level converter (Table IV) operated at 4 kHz,
present worth of the future energy cost savings can be deter- respectively. The cost for the individual converter components
mined as are summarized in Table V for operation at 4 and 8 kHz. From
Table V, it is clear that the initial cost of the three-level converter
(7) is equal to or substantially lower than that of the two-level con-
verter of the same power rating despite the higher cost for semi-
conductors and gate/control units. This is mainly due to savings
where in the filter elements and the heat sink.
accumulated savings in present worth terms; For a grid interface chosen to operate at 4 kHz the losses
energy cost savings in month ; during rectifier operation close to rated load are 120 W higher
annual interest rate; in a three-level converter in comparison to a two-level con-
number of years. verter (Fig. 10). However, the losses of the three-level converter
Disregarding potential energy price escalations or additional are lower than that for the two-level converter by 40 W and
costs of debts, a converter system with 1 kW fewer losses oper- 50–120 W for inverter operation and rectifier operation at par-
ated 8000 h per annum achieves a cost saving in each individual tial load, respectively. From life-cycle cost perspective, a clear
year of roughly , , and in the understanding of the operation schedule of the grid interface is
major markets of Japan, Germany, and the USA, respectively. needed to make a decision on the optimum converter topology.
For a converter operated over eight years with 8000 h per year Fig. 11 shows that the absolute loss savings amount to
and an interest rate of 4% ( ), the energy cost saving per 400 W for a switching frequency of kHz and rated
864 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 3, MAY/JUNE 2005

TABLE V
COMPONENT COST ANALYSIS

load for both rectifier and inverter operation. This indicates that 5) Substantial reductions in filter size/weight and cost (
substantial cost savings can be achieved throughout the entire 30%) are possible due to the lower voltage harmonics
product life cycle. Even if the initial costs of the three-level in the three-level converter.
converter are slightly above those of the two-level converter 6) Reliability concerns can be mitigated by adequate design
the three-level converter is paying off in high energy price choices. The reduction of semiconductor losses will re-
markets like Japan. At a switching frequency of kHz, duce the average temperature at the components and, thus,
the application of the three-level converter is also justified decrease the failure rate.
in all markets due to the higher initial semiconductor cost 7) The three-level converter is economically viable in high-
in comparison to the two-level converter which needs 400-A energy-cost markets, even at low switching frequencies.
devices. Note that potential loss savings in the filter and the
machine are not yet included. The same statement is valid for REFERENCES
the conventional converter driving an induction machine. A [1] T. Tanaka. (2003) Yaskawa Tech. Rev. (Special Issue on Inverter Drive).
filter will, if needed, improve the cost balance in favor [Online] Available: http://www.yaskawa.co.jp/en/technology/gihou/65-
2/t01.htm
of the three-level converter. The utilization of a three-level [2] A. Buente, T. Oezgen, and T. Ortmann, “Dreipunkt-wechselrichter fuer
topology for high-speed drives and other high-switching-fre- hochgeschwindigkeitsantriebe,” in Conf. Rec. PCIM Europe, Nurem-
berg, Germany, 2003.
quency applications is recommended due to significant cost [3] B. A. Welchko, M. B. R. Correa, and T. A. Lipo, “A cost effective three-
reduction in both initial and operating costs. level MOSFET inverter for low power drives,” in Proc. IEEE IECON’02,
vol. 2, 2002, pp. 1267–1272.
[4] S. Park and T. M. Jahns, “A self-boost charge pump topology for a
gate drive high-side power supply,” in Proc. CPES Seminar, 2002, pp.
267–272.
X. CONCLUSION [5] O. Al-Naseem, R. W. Erickson, and P. Carlin, “Prediction of switching
loss variations by averaged switch modeling,” in Proc. IEEE APEC’00,
Three-level topologies are highly attractive for low-voltage vol. 1, 2000, pp. 242–248.
[6] Y. Shakweh and E. A. Lewis, “Assessment of medium voltage PWM
power converters, specifically for applications with medium to VSI topologies for multi-megawatt variable speed drive applications,”
high switching frequencies. in Proc. IEEE PESC’99, vol. 2, 1999, pp. 965–971.
[7] R. Teodorescu, R. Blaabjerg, and J. K. Pedersen, “Multi-level converters
1) The difference in switching loss energies between IGBTs – A survey,” in Conf. Rec. EPE’99, 1999, pp. 408–418.
of the same current class and commonly adjacent voltage [8] F. Z. Peng, “A generalized multilevel inverter topology with self voltage
classes is a factor of 3–5. This is the key device character- balancing,” in Conf. Rec. IEEE-IAS Annu. Meeting, vol. 3, 2000, pp.
2024–2031.
istic enabling low switching losses. [9] K. A. Corzine, J. R. Baker, and J. Yuen, “Reduced parts-count multi-
2) Due to the specific correlation of switching and conduc- level rectifiers,” in Conf. Rec. IEEE-IAS Annu. Meeting, vol. 1, 2001,
tion losses with load current, a three-level topology is su- pp. 589–596.
[10] (2000) IGBT Datasheet. IXYS Corp. [Online]. Available: www.ixys.com
perior in terms of total semiconductor losses at switching [11] (2000) IGBT Data Sheets: BSM200GB120DLC, BSM200GB06DLC,
frequencies as low as and beyond 2–3 kHz in practical FF200R12KS4, FF200R12KE3, FF200R17KE3. EUPEC Corp. [On-
applications. At switching frequencies above 5 kHz, the line]. Available: www.eupec.com
[12] (2000) IGBT Data Sheets: CM200DU24F, CM200DU12F. MIT-
three-level converter always features lower losses. SUBISHI Corp. [Online]. Available: www.mitsubishichips.com
3) Compared to the two-level converter, the three-level con- [13] F. Blaabjerg, U. Jaeger, S. Munk-Nielsen, and J. K. Pedersen, “Power
verter enables substantially higher switching frequencies losses in PWM-VSI inverter using NPT or PT IGBT devices,” IEEE
Trans. Power Electron., vol. 10, no. 3, pp. 358–367, May 1995.
( 4 ) applying devices of the same technology and cur- [14] R. Teichmann, “Auxiliary resonant commutated pole converters,” Ph.D.
rent class in both converters. dissertation, Dept. elect. Eng. Inf. Technol., Dresden Univ. Technol.,
4) The introduction of low on-state voltage trench-gate Dresden, Germany, 2002.
[15] S. Bernet and R. Teichmann, “Potential and risks of the matrix converter
IGBTs has a higher leverage effect on three-level con- for modern AC-drives,” in Conf. Rec. COBEP, Belo Horizonte, Brazil,
verters than on two-level converters. 1997, pp. A1–A10.
TEICHMANN AND BERNET: COMPARISON OF THREE-LEVEL CONVERTERS VERSUS TWO-LEVEL CONVERTERS 865

[16] S. Bernet, S. Ponnaluri, and R. Teichmann, “Design and loss compar- Ralph Teichmann received the MS. and Ph.D.
ison of matrix converters and voltage source converters for modern AC degrees in electrical engineering from Dresden
drives,” IEEE Trans. Ind. Electron., vol. 49, no. 2, pp. 304–314, Apr. University, Dresden, Germany, in 1997 and 2002,
2002. respectively.
[17] T. Brückner and D. G. Holmes, “Optimal pulse width modulation He was a Visiting Student at the University of Wis-
for three-level inverters,” in Proc. IEEE PESC’03, vol. 1, 2003, pp. consin, Madison, in 1995–1996 and a Research As-
165–170. sistant at Nagasaki University, Japan, in 1998–1999.
[18] K. R. M. N. Ratnayake and Y. Murai, “A novel PWM scheme to elim- Since 1997, he has been an independent consultant
inate common mode voltage in three-level voltage source inverter,” in for various companies, including ABB Corporate Re-
Proc. IEEE PESC’98, vol. 1, 1998, pp. 269–274. search and Philips Medical Systems. He is currently
[19] E. Persson, “Transient effects in applications of PWM inverters to in- with GE Global Research, Niskayuna, NY. His re-
duction motors,” IEEE Trans. Ind. Appl., vol. 28, no. 5, pp. 1095–1101, search interests include high-power conversion, renewable energies, distributed
Sep./Oct. 1992. generation, hard- and soft-switching converter topologies, as well as converter
[20] P. van Poucke, R. Belmans, W. Geysen, and E. Ternier, “Over-voltages in controls.
inverter fed induction machines using high frequency power electronics Dr. Teichmann has been the recipient of several awards, including two schol-
components,” in Proc. IEEE APEC’94, vol. 1, 1994, pp. 542–548. arships, one industrial fellowship, an IEEE paper award, and a German electrical
[21] A. H. Bonnett, “Analysis of the impact of pulse-width modulated in- engineering society publication award.
verter voltage waveforms on AC induction motors,” IEEE Trans. Ind.
Appl., vol. 32, no. 2, pp. 386–392, Mar./Apr. 1996.
[22] A. von Jouanne, D. Rendusara, P. Enjeti, and W. Gray, “Filtering tech-
niques to minimize the effect of long motor leads on PWM inverter fed
AC motor drive systems,” IEEE Trans. Ind. Appl., vol. 32, no. 4, pp.
919–926, Jul./Aug. 1996. Steffen Bernet (M’97) was born in Ilmenau, Ger-
[23] P. T. Finlayson, “Output filters for PWM drives with induction motors,” many, in 1963. He received the Diploma degree
IEEE Ind. Appl. Mag., vol. 4, no. 1, pp. 46–52, Jan./Feb. 1998. from Dresden University of Technology, Dresden,
[24] P. Guggenbach, “Pulse removal in sinusoidal PWM with constant Germany, in 1990, and the Ph.D. degree from the Il-
switching frequency to reduce switching frequency or current har- menau University of Technology, Ilmenau, Germany,
monics,” in Proc. IEEE PESC’98, 1998, pp. 294–301. in 1995, both in electrical engineering. The subject
[25] F. Jenni and D. Wuest, Steuerverfahren Fuer Selbstgefuehrte Strom- of his Ph.D. dissertation was the investigation of
richter. Zurich, Switzerland: vdf Hochschulverlag Zurich, 1995. power semiconductors in soft-switching converters.
[26] S. Ogasawara and H. Akagi, “Analysis of variation of neutral point po- He was a Development Engineer in the Depart-
tential in neutral point clamped voltage source PWM inverters,” in Conf. ment of Private Communication Systems at Siemens
Rec. IEEE-IAS Annu. Meeting, 1993, pp. 965–970. from 1994 to 1995. During 1995 and 1996, he was
[27] C. Newton and M. Summer, “Neutral point control for multi-level a Postdoctoral Researcher in the Electrical and Computer Engineering Depart-
inverters: Theory, design, and operational limitations,” in Conf. Rec. ment, University of Wisconsin, Madison. In 1996, he joined ABB Corporate Re-
IEEE-IAS Annu. Meeting, 1995, pp. 2451–2458. search, Heidelberg, Germany, where he led several strategic power electronics
[28] O. V. Thorsen and M. Dalva, “A survey of the reliability with an analysis and drives research projects for low-voltage and medium-voltage applications.
of faults on variable frequency drives in industry,” in Conf. Rec. EPE’95, He led the Electrical Drives Group at ABB Corporate Research from 1998 to
1995, pp. 1033–1038. 2001. From 1999 to 2000, he was responsible for ABB research worldwide in
[29] J. J. Stroker, “What is the real cost of higher efficiency,” IEEE Ind. Appl. the areas of power electronic systems, electric drives, and electric machines. In
Mag., vol. 9, no. 3, pp. 32–37, May/Jun. 2003. 2001, he joined the Berlin University of Technology, Berlin, Germany, as a Pro-
[30] (2004). Austerlitz Electronic GmbH. [Online]. Available: www.auster- fessor of Power Electronics. His main research areas are high-power converter
litz-electronic.de topologies, power semiconductors, and motor drives.