High Frequency Transformer Design for Modular

Power Conversion from Medium Voltage AC to

400V DC
Shishuo Zhao, Student Member, IEEE, Qiang Li, Member, IEEE, Fred C. Lee, Fellow, IEEE
Center for Power Electronics Systems,
The Bradley Department of Electrical and Computer Engineering
Virginia Polytechnic Institute and State University
Blacksburg, VA 24061 US
zssvt@vt.edu

Abstract the paper presents a high frequency modular MV Utility Service

medium voltage AC (4160 VAC and 13.8 VAC) to low voltage DC 480V
(400 VDC) power conditioning system block (PCSB) that are 3000A @2500kVA
scalable so that they can be used for micro grids of different scale
(several-hundred kW to multi-MW). The modular approach is
intended to result in higher-volume, lower-cost, less-loss power
MV/480V
electronics building blocks that service many applications, such as 60Hz Transformer
Server Hall 1 Server Hall 2
DC data center and electric vehicle charge station. In this paper, a
225kW, 500 kHz PCSB is demonstrated to direct converter 4160 Fig. 1. Traditional LVAC Distribution before Sever Hall
VAC to 400V DC for a DC date center. WBG power devices and
CLLC resonant converter are used to minimize switching related MV Utility Service

loss at high frequency. The high frequency transformer of CLLC 4160V

resonant converter is one of the key elements for the proposed
350A @2500kVA
modular approach. This paper will focus on high frequency
transformer design to realize high-voltage-isolation, high-
efficiency and high-density at the same time. Based on a split
winding transformer structure, transformer insulation material 4160V/480V
60Hz Transformer Server Hall 1 Server Hall 2
and dimension parameters are determined referring to insulation
standard. Transformer magnetic loss model is reviewed based on Fig. 2. MVAC Distribution before Sever Hall proposed by CCG
which loss design trade-off is carefully analyzed. Finally a 500 kHz
transformer prototype has been developed and demonstrated with
30kV isolation capability, whole CLLC resonant converter holds
98% peak efficiency and 48 W/in3 power density.

Keywordshigh frequency; medium voltage; transformer 480/277V 480/277V 12V DC

MV
insulation; magnetic loss analysis; CLLC resonant converter AC/DC DC/AC AC/DC VR
DC/DC VR
60Hz
60Hz UPS VR
PSU
I. INTRODUCTION Transformer PDU
Server Rack
Eff. 99% X 98% X 95% X 94% X 96% X 88% =73%
Due to the increasing use of cloud computing and big data,
the power consumption of the data center alone will reach 10%
of the total electrical power consumption in the world by 2020.
Considering such booming data center load development, high
copper cost and conduction loss due to low voltage (480VAC) Fig. 3. Traditional data center power architecture with low voltage AC
power distribution outside sever hall need to be solved. distribution within Sever Hall
Traditionally data centers step down from medium voltage
(4.16kV~ 35kV) utility facility to 480V through line frequency After power is distributed to server hall, usage of multi-
transformer and distribute to sever hall using 480V bus. This stage redundant power conversion stages leads to excessive
leads to bulky and costly transmission bus with large conduction losses. UPS, PDU and PSU together will bring about a 21% loss
losses shown in Fig. 1. Instead of low voltage, CCG Facility of the total system power [2, 3] shown in Fig. 3. In order to
Integration proposed to use 4160V distribution before server improve the power conversion efficiency inside data center
hall [1]. The distribution current is reduced from 3000A to only server hall, two approaches are proposed, one developing trend
350A for a 2.5 megawatts facility. This will dramatically reduce is to continues to use AC distribution within server hall with
86% copper usage as well as i2R conduction loss by factor of 9 paralleled UPS [4], another approach is that instead of AC, DC
shown in Fig. 2. distribution within server hall is employed[5]. Both of these two
methods can reduce redundant power conversion stages.

978-1-5090-5366-7/17/$31.00 2017 IEEE 2894

However in both AC and DC date center power architectures, a with an eleven-level staircase multilevel waveform can be
line frequency bulky and space consuming transformer is established. To date, the cascaded H-bridge multilevel rectifier
employed to step down MVAC to 480V AC and distribute is deemed a mature technology and can be directly applied to the
480V AC throughout the facilities. For megawatts data centers, proposed DC data center application.
presently, the 480V AC distribution lines carry thousands of
amperes of current which leads to a very bulky and costly
transmission bus and large conduction loss.

4160V AC AC/DC 400V DC 12V DC

MV AC/DC VR
350A@2500kVA DC/DC VR
PSU VR
PDU
Server Rack
Eff. x 97% x 99% x 97% x90% =84%

Fig. 5. Proposed medium voltage to 400VDC power conditioning system

block for DC data center

The most important element of PCSB system is the high

Fig. 4. Proposed DC data center power architecture with MVAC converted frequency isolated DC/DC stage, we list all the DC/DC module
directly to 400VDC survey results in the Tab. I. Traditionally mostly used devices
are high voltage IGBTs, based on three level NPC topology
Our research target to demonstrate MVAC distribution 1.7kV low cost IGBT is also employed in some application.
before server hall and converted directly to 400VDC distribution With development of wide band gap devices the DC/DC stage
within server hall as shown in Fig. 4. The crucial technology is working frequency is going higher and higher up to 50 kHz [17].
high frequency isolated AC/DC system from MVAC to 400 V Mostly used topology is DAB converter [9, 10, 12, 15], ABB is
DC. This system offers high frequency insulation between already using LLC unregulated resonant converter (DCX) [13].
MVAC and 400 V DC and totally eliminates the use of a bulky From magnetic point of view, due to the increasing working
60Hz transformer. Also, the MVAC distribution can greatly frequency, Nanocrystalline and MnZn Ferrite are the mostly
reduce the distribution current and therefore greatly save I2R used core material. Mostly used transformer structure is still UU
loss and copper cost. Total efficiency can be further improved to core shape [10,12,13, 14] and EE core shape [15,16]. All the
84%. In section II, the proposed high frequency modular transformer windings are made by litz wire to reduce high
medium voltage AC to low voltage DC power conditioning frequency current conduction loss.
system block (PCSB) is introduced for DC data center power
system. The most important DC/DC stage (CLLC resonant TABLE I. HIGH FREQUENCY DC/DC MODULE SUMMARY
converter) will also be introduced with device loss breakdown
in this section. Then, section III will focus on medium voltage Power Primary Freque Core Material
high frequency transformer design including transformer level device ncy
structure, insulation design, magnetic loss analysis. Then in ALSTON 180 6.5kV 5kHz NA
section IV transformer prototype is developed based on previous 2003[7] kW IGBT
analysis with parasitic parameters and insulation test results Bombardier 500 4.5kV 8kHz Nanocrystalline
referring to IEEE Std. C57.12.01 standard. Finally 500 kHz 2007[8] kW IGBT
CLLC resonant converter is developed and its experiment test UNIFLEX 25 kW 1.7kV 2kHz Amorphous
results are also included. 2009[9] IGBT
FREEDM 7 kW 6.5kV 3kHz Amorphous
II. HIGH FREQUENCY ISOLATED DC DISTRIBUTION DATA 2010[10] IGBT
CENTER SYSTEM
EPRI 50 kW 4.5kV 20kHz Ferrite
The proposed high frequency modular medium voltage AC 2012[11] IGBT
(4160 VAC and 13.8 VAC) to low voltage DC (400 VDC) FREEDM 20 kW 15kV SiC 20kHz Nanocrystalline
power conditioning system block (PCSB) is shown in Fig. 5. 2012[12] MOSFET
Five cascade H-bridge converters together with high frequency ABB 150 6.5kV 1.8kHz Nanocrystalline
isolated DC/DC converter are employed to convert medium 2013[13, 14] kW IGBT
voltage directly into 400V DC. The inputs of H-bridges are in ETH 2013[15, 166 1.7kV 20kHz Nanocrystalline
series and the outputs of the DC/DC stages are connected in 16]
kW IGBT &Ferrite
parallel.
NCSU 16 kW 1.2kVSiC 50kHz Ferrite N87
For the AC/DC stage, a cascaded H-bridge multilevel 2016[17] MOSFET
rectifier will be used to reach desired medium voltage [6]. For Even through mostly used topology for DC/DC stage is dual
a 4160VAC line, the phase-to-neutral input voltage is 2.4k active bridge (DAB). However this converter will lose ZVS at
VAC. With 1200V SiC MOSFET as switch, a five H-bridge light load and it turns off switches at peak working current

2895
resulting high turn off loss. We will use CLLC resonant 12uH leakage inductance which is two times larger than that of
converter to build the DC/DC module as shown in Fig. 6. It type 2 6.1uH. Type 2 winding is verified work under lower flux
employs magnetizing inductance to achieve ZVS of primary density which generate lower eddy current loss.
switches for whole load range, and synchronize rectifier (SR)
control is used for secondary to reduce conduction loss. Turn
off current is only the peak current of magnetizing current.
Converter device loss breakdown under 15kW 100C is shown
in Fig. 7 based on 1.2 kV SiC MOSFET from GE (Primary side)
and 650V GaN device from GaN System (Secondary side). Up
to 500 kHz device loss is acceptable within 1% of total power.
Our research will start from 500 kHz, power level will be
around 15 kW for each DC/DC module.
2:1:1 Fig. 8. Transformer structure comparison between type 1 EE core (left) and
type 2 UU core (right)
800V Llk1 Llk2 400V
DC DC

Lm

Llk2
Pri. : Series
Sec: Parallel

Single Core

Fig. 6. CLLC converter with two parallel output sets Fig. 9. Magnetic field simulation comparison between type 1 EE core (left)
and type 2 UU core (right)

Besides winding loss impact, leakage inductance will also

influence CLLC resonant capacitor voltage stress. According to
leakage simulation result, we can calculated corresponding
resonant capacitor value under different resonant frequency
(100kHz/ 300kHz/ 500kHz) shown in Fig. 10. Then resonant
capacitor voltage stress is measured under 15kW full power
CLLC circuit simulation shown in Fig. 10. For the worst case
under 500 kHz 15kW condition type 1 structure holds 1200V
voltage stress which is two times higher than that of type 2
Fig. 7. CLLC converter device loss breakdown at different frequency structure 618V. Finally, type 2 UU core structure is chosen as
our structure candidate.
III. HIGH FREQUENCY MEDIUM VOLTAGE TRANSFORMER
Resonant Capacitor Crp Comparison Resonant Capacitor Voltage Stress
The medium voltage high frequency transformer is the most 450 415
400 1500
important component in the whole DC/DC converter in terms of 350 1200
both insulation and power density. 300
250 211
1000 837
Volts
nF

200 618
512
A. Transformer Structure Determination 150
100
500
256
422
46.1
23.3 8.4 16.6
Frequently used transformer structures are still EE core and 50
0 0
UU core structures shown in Fig. 8. For both of these two 100kHz 300kHz 500kHz 100kHz 300kHz 500kHz
Type1 Type2 Type1 Type2
structures sectionalized winding are employed to guarantee Fig. 10. Resonant capacitor value and stress comparison between two types
insulation capability. Gap between two core parts can be used to
insert a shielding board between primary and secondary coil to B. Transformer Isulation Design
increase both clearance and creep age distance. If we make a It is obvious that only the high frequency isolated
comparison between two structures, type 1 holds two times transformer can provide insulation safety function for overall
higher magneto motive force (MMF) value comparing to type 2 system and it should be designed based on the input medium
, moreover type1 holds shorter air flux path length (ld), all of voltage level 4160V. Insulation between primary high voltage
which results in larger leakage inductance and higher eddy winding and secondary low voltage winding and transformer
current loss comparing to type 2 structure. core should all designed based on 4160V insulation
3D FEA simulation is done in Maxwell and the magnetic requirement. However insulation between secondary winding
field simulation result is shown in Fig. 9 under same scale. It is and core only need to be designed based on 400V level.
obvious that type 1 holds higher magnetic field value among the According to IEEE Std. C57.12.01 standard [18], there are
window area because higher MMF value and shorter air flux at least 3 dielectric test need to be passed: basic lightning
path length. From leakage inductance point of view type 1 holds impulse, applied voltage and partial discharge tests. Lightning

2896
 0.4
impulse test should have 30 kV crest value, and low frequency Ds T
apply voltage should be test up to 12 kV RMS value for the = ref
Ds ref T
4160V application shown in Tab II. The third insulation test is (1)
partial discharge test. TABLE V DIELECTRIC PARAMETER OF TRANSFORMER
Bobbin thickness 4 mm
TABLE II INSULATION TEST LEVEL FOR DRY TYPE TRANSFORMER OF IEEE
STD. C57.12.01 Primary insulation Tape thickness 2.2mm (14 layers)
Secondary insulation Tape thickness 0.35mm(2 layers)
Nominal Low- Basic Lightning impulse insulation Primary bobbin to core spacer 4 mm
L-L frequency levels(BIL ratings) in common use kV Secondary bobbin to core spacer 1 mm
system voltage crest(1.2*50us) Primary bobbin to secondary bobbin spacer 4 mm
voltages insulation Based on previous insulation design parameter. We build an
level
(kV) (kV rms) 10 20 30 45 60 95 110 electric field simulation in Maxwell to simulate the electric
0.25 2.5 None field distribution among all insulation material. Primary side
0.6 3 S 1 1 high voltage winding is connected to the high insulation test
1.2 4 S 1 1 voltage 30kV. Both transformer core and secondary winding is
2.5 10 S 1 1 grounded. Final simulation result is shown in Fig. 11. We can
5.0 12 S 1 1
8.7 20 S 1 1
see within the insulation solid material (bobbin, insulation tape,
15 34 S 1 1 and encapsulating material) maximum electric field is around
In order to pass IEEE Std. C57.12.01 standard, transformer 5.5kV/mm which is much smaller than that of all insulation
clearance distance and creep age distance should be designed solid material.
based on the transformer steady state operation voltage (4160V)
and transient maximum peak insulation test voltage (30 kV).
According to standard [19], for 4160 V operating voltage
minimum creep age distance should be at least 41.6 mm; for
30kV peak working voltage the minimum clearance distance
should be at least 42 mm. Besides these requirements all the
insulation solid material thickness like insulation tape and
bobbin should all designed based on maximum insulation
requirement 30kV.
Three different solid insulation material for transformer Fig. 11. Electric field simulation under highest insulation test voltage 30kV
bobbin are shown in Tab III. PTFE is chosen as insulation
bobbin material with high dielectric strength (19.7 kV/mm) and C. Core Material Selection and Magnetic Loss Modeling
high working temperature. Another three different insulation According to survey mostly used core materials are
tape candidates are compared in Tab IV. 3M 8943 is chosen as Nanocrystalline and Ferrite. For our target 500 kHz application
dielectric tape for high dielectric strength (9.8 kV/0.17mm) and Nanocrystalline is ruled out due to their high core loss density at
high thermal conductivity (0.4 W/ mK). such high working frequency range comparing to Ferrite. MnZn
TABLE III TRANSFORMER BOBBIN MATERIAL CANDIDATES COMPARISON Ferrite is our material candidate. There are plenty of MnZn
Ferrites for our 500 kHz application. We compared five material
Material Candidate PTFE PC ABS candidates in Fig. 12 to figure out the appropriate Ferrite
Dielectric Strength@25C 19.7kV/mm 15 kV/mm 16.7 kV/mm
material. 3F36 holds lower loss comparing to other competitors
Dielectric Constant 2.1 2.9 2.87
Thermal conductivity 0.25 W/mK 0.22 W/mK 0.17 W/mK and is chosen as our core material.
High working temperature 250C 130C 100C 1000

TABLE IV TRANSFORMER INSULATION TAPE CANDIDATES COMPARISON 500kHz

Material Candidate 3M 8943 Kapton 5413 Mica tape
100
Dielectric Strength 62.8kV/mm 102 kV/mm 10 kV/mm
Pv kw/m3

Thermal conductivity 0.4 W/mK 0.1 W/mK 0.2 W/mK P61 ACME
Stable Temperature -40~200C -73~260C Up to 950C PC95 TDK
10 3C98-- Ferroxcube
Note that plastic material dielectric strength reduces when 3F45 -- Ferroxcube
material thickness increases. Rule of thumb relationship is 3F36 Ferroxcube
100 C
listed in equation (1) [20]. Ds is dielectric strength under 1
10 100
specific thickness T, Dsref is reference dielectric strength under Bmax(mT)
reference thickness Tref. Considering 50% margin, 45kV is
chosen to determine insulation thickness calculated result is
Fig. 12. MnZn Ferrite candidates core loss comparison @500kHz
shown in Tab. V. Besides this we also encapsulate primary high
voltage winding with epoxy material (EPIC Resin
The state of art core loss calculation method is Steinmetz
R1055/H5083) to eliminate insulation weakness and avoid
Equation:
moisture and dust on high voltage winding.

2897
 Pv = k f Bm (2)

Where Bm is the peak flux amplitude, Pv is time average

power loss per unit volume, and f is the frequency of sinusoidal
excitation, and k, , and are constants found by curve fitting
results from core loss measurement. This method cannot be
directly used in CLLC resonant converter since it is only valid
for sinusoidal voltage excitation. For CLLC resonant converter
square voltage excitation waveform is added on the transformer
core. Rectangular Extension Steinmetz Equation (RESE) is Fig. 13. Magnetic field distribution characteristic under peak working current
proposed to exactly predict core loss under rectangular voltage Ipri=31.8A Isec=28.6A
excitation [21]. Due to squared (D=0.5) voltage excitation for
CLLC resonant converter, the core loss calculation for CLLC g:
resonant converter can be further simplified to (3).
N
02 lw d s 4 dH (t ) 2
8 (3) Peddy ( N , t ) = N s ( ) (6)
Pv _ rec = 2
kf Bm j =1 64 c dt
j

Then Transformer core loss can be calculated based on Where N is transformer turn number, Ns is litz wire strand
RESE method by multiplying core volume with core loss number, lw is length per turn, <>j is space average calculator
density under square voltage excitation shown in equation (4). over j turn winding. There is still two dependent variable N and
Where Vpri is the primary side DC voltage, N is the turn number t. From time domain point of view, the magnetic field is a quasi-
of primary side, lc is the magnetic loop length of the core. Note sinusoidal shape and here we assume it is sinusoidal shape and
that there is only one dependent variable that is transformer turn time average eddy current loss calculation is give below in (7)
number.
N
02 lw d s 4 dH m sin( wt ) 2 (7)
V pri Peddy ( N ) = N s ( )
Pcore ( N ) = Pv _ rec lc
(4) j =1 64 c dt j
4 N Bm f

Where Hm is peak magnetic field value for external field

Litz wire is used in high frequency transformer. There are under peak excitation current, is time average calculation over
two assumptions to calculate the litz wire winding loss due to
whole cycle. Based on this equation static 3D magnetic field
inherent nature of litz wire. First assumption is that the field
simulation result can be used to calculate litz wire winding eddy
remains constant inside each strand, equivalent to the
current loss. Winding working current conduction loss can also
assumption that each strand diameter is not large compared to
be calculated easily by equation (8). Finally total winding loss
skin depth. Second assumption is that there is no eddy current
can be calculated by summation of equation (7) and (8) shown
flows between different strands due to proper braided litz wire
in expression (9). Note that there is only one dependent variable
construction [22]. Based on the assumptions above, litz wire
that is transformer turn number N.
winding loss model should only consider both eddy current
flowing inside each strand due to external field and conduction
loss neglecting eddy current effect. The latter is easy to calculate 4 c Nlw 2 (8)
Pnon _ eddy ( N ) = I rms
via dc resistance of conductor, therefore this paper will focus on N s d s 2
the former. We assume that each litz strands diameter is small
enough that for each small strand the external magnetic field is N
02 lw d s 4 dH m sin( wt ) 2 4 c Nlw 2 (9)
uniform. For a cylinder conductor in a uniform flux density, The Pwinding ( N ) = N s ( ) + I rms
j =1 64 c dt N s d s 2
Squared Field Derivative (SFD) method is widely used to j

calculate instantaneous eddy current loss [22]:

2 4
ldc dH (t ) 2
0 (5)
Peddy (t ) = ( )
64 c dt
Where l is cylinder conductor length, c is resistivity of wire.
From equation (5) we know that in order to calculate each litz
strand eddy current loss, its corresponding external magnetic
field is necessary. Then a 2D Magnetic field simulation for
previous mentioned transformer structure is developed,
magnetic field distribution is shown in Fig. 13. It is obvious that
the magnetic field is non-linear and irregular.
D. Transformer Magnetic Loss Analysis and Design
Based on previous two magnetic loss calculation model,
transformer optimal design can be developed. Litz wire strand
AWG 44(diameter 51um) is chosen since it is enough small
consideration 500 kHz skin depth 92um. According to National
electrical code, litz wire strand number around Ns=825 is
chosen for 15kW 21A RMS current application [23]. The initial
design core loss density is chosen as Pv =300. Because there is
only one dependent variable for both core loss and winding loss
calculation which is transformer turn number N, then
transformer core loss and winding loss with swept transformer
turn number is plotted in Fig. 14. Core loss decrease with
increasing transformer turn number because larger turn number

2898
gives smaller transformer section area and smaller total TABLE VI TRANSFORMER DESIGN PARAMETER
transformer volume. Under certain core loss density, the core Working Frequency(fs) 500 kHz
loss will decrease. At same time larger transformer turn number Core Material 3F36
will increase total transformer winding length which resulting Core loss density(Pv) 300kW/m3
in higher winding loss. Finally summation of both core loss and Strand AWG 44
winding loss contribute to transformer total loss U shape curve Litz wire stand number(Ns) 825
can be plot shown in Fig. 14. Turn Number(N) 12
Winding loss 36 W
110
Core loss 38 W
100
Total loss Transformer volume 0.37 Liter
90
Magnetizing inductance 28uH
80

IV. TRANSFORMER PROTOTYPE AND INSUALTION TEST

Loss /W

70

60

50

Winding loss
A. Transformer Prototype and Parasatic Inductance Test
40
Core loss
30
Based on previous transformer designed parameters shown
20
6 8 10 12 14 16 18 20
in Tab VI, medium voltage high frequency transformer
N
prototype is developed shown in Fig. 17. Dimension marked is
including mechanical fixture. Transformer gap value is
Fig. 14. Transformer loss vs transformer turn number N determined by transformer magnetizing inductance. Parasitic
leakage inductance and magnetizing inductance is measured
Following the same method, considering core loss density and shown in Tab VII. At same time 3D FEA simulation results
impact, we swept Pv from 100 to 400 kW/m3 and plot of leakage inductance and magnetizing inductance is also
transformer total loss vs transformer turn number as shown in shown in Tab VII.
Fig. 15. The optimal design valley region is slightly moved to
higher turn number when core loss density increases. In order
to figure out our transformer design point. Then we plot
transformer total loss vs transformer volume relationship in Fig.
16, each mark with the curve represent one transformer turn
number design point. Transformer volume is calculated based
on minimum box volume that transformer can be put into
(Mechanical fixture volume is not included). Final design point
(N=12, Pv=300) locate around the corner region as the trade-off
between volume and total loss. Transformer design parameter
is summarized and shown in Tab VI.
120 Fig. 17. Medium voltage 500 kHz transformer prototype
Pv =400 kW/m3
110
Pv =300 kW/m3 TABLE VII TRANSFORMER DESIGN PARAMETER
100 Pv =200 kW/m3
Pv =100 kW/m3 Parasitic parameters Prototype Test 3D FEA
Loss /W

90

80
Magnetizing Inductance(Lm) 27.78uH 28uH
Primary Leakage Inductance(Lrp) 6.76uH 6.1uH
70
Secondary Leakage Inductance 1(Lrs2) 4.33uH 3.6uH
60 Secondary Leakage Inductance 2(Lrs2) 4.26uH 3.6uH
50
6 8 10 12

N
14 16 18 20
B. Transformer Medium Voltage Insulation Test
After transformer prototype is built up, since target
Fig. 15. Transformer loss vs turn number N application is 4160V medium voltage, insulation safety should
120
be first concern. In order to verify transformer insulation
110
Pv =400 kW/m3 capability for 4160V medium voltage application, the
Pv =300 kW/m3 transformer prototype is tested under previous mentioned IEEE
100
Pv =200 kW/m3 Std. C57.12.01 insulation standard. Three insulation test is
Loss /W

90
Pv =100 kW/m3 conducted including basic lightning insulation test, applied
80
voltage insulation test, partial discharge insulation test all of
70 which will be talked about in the following separately.
60 Total loss=74W According to IEEE Std. C57.12.01, for 4160V application
50
Volume=1.02Liter transformer the basic lightning insulation test should be at least
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6
Volume /L 30kV. The voltage waveform is also presented in the figure.
Within 1.2us the test voltage should be increased to 30kV peak
Fig. 16. Transformer loss vs transformer volume trade-off value and gradually go down to half of peak value at 50us.
During the test if the test voltage waveform is smooth and no
collapse happens during the waveform going down. This means
the transformer passes this lightning test. Then the transformer

2899
is tested under thus condition test waveform is also in Fig. 19.
Test waveform meets the requirement and shapes smoothly
going down. Blue and red curves represent two independent test.
Two test waveforms agree with each other and overlap very
well. Based on these test results, we conclude that our
transformer passes the basic lightning insulation test.
one more equipment is added that is partial discharge
measurement unit shown in Fig. 21. This unit can measure the
transformer dynamic partial discharge value during the whole
test process. The test voltage requirement is shown in Fig. 21.
Our partial discharge test result is 3.8 pC which is much smaller
than 50 pC of IEEE standard requirement during 5.4kV
3minutes period.

Equivalent Connection Circuit

Voltage
30 sec
180 sec
7.4kV
Pri. Sec. 5.4kV
Impulse
1.8Vr 1.3Vr
generator
Vr : Rated working voltage Time

Fig. 21. Transformer under partial discharge test and test condition

Fig. 18. Transformer under basic lightning insulation test and connection V. CLLC CONVERTER EXPERIMENT TEST
Finally 500 kHz 15kW CLLC resonant Converter is
developed based on our medium voltage high frequency
transformer shown in Fig. 22. The power density reaches 48
kW/in3. We gradually increase converter power level, the
waveform of 15kW full power load is shown in Fig. 23. The
dead time is around 80ns and turn off current of primary side
device is around 12A. Zero voltage switching is achieved with
little oscillation due to hard switch turn off.

Fig. 19. Transformer under basic lightning insulation test waveform

The second insulation test is applied voltage test. During

this test transformer low voltage coil is grounded and
transformer cores are all grounded. Primary high voltage coil is
firstly shorted and then added 12 kV RMS line frequency
medium voltage shown in Fig. 20. The test condition is referred
IEEE standard. After the test voltage is increased to 12 kV, the
test voltage is hold for 60 seconds and during this time there
should be no flash over, spark over or puncture. During our 1
minute test, the transformer insulation is stable and no any
insulation failure happens.

Equivalent Connection Circuit

Fig. 22. 500 kHz 15kW CLLC resonant converter prototype hardware
Line Pri. Sec.
Frequency
MV

Fig. 20. Transformer under applied voltage test and connection

The third insulation test is partial discharge test. Protracted

partial discharge can erode solid insulation and eventually lead
to breakdown of insulation. Partial discharge test equivalent
circuit is same as previous applied voltage test, however only Fig. 23. 500 kHz CLLC working waveform at 15kW 1us/div

2900
 In order to achieve high efficiency power conversion
secondary side GaN device need to work at synchronize
rectifier mode, we need to maximize the conduction time of the
device channel to minimize device conduction loss. After fine
tune secondary side SR the highest efficiency we can achieve is
presented in Fig. 24. Then peak efficiency at around half load
is around 98%.
99% Peak Efficiency 98.0%
98%
97%
96%
Efficiency
95%
94%
93%
92%
91%
90%
1000 3000 5000 7000 9000 11000 13000 15000
Power Level
Fig. 24. 500 kHz CLLC converter efficiency test result
VI. CONCLUSIONS
CLLC resonant converter is the key element in PCSB for
next generation DC data center. Medium voltage high
frequency transformer is the most crucial component in terms
of insulation and power density. UU core with sectionalized
winding structure is chosen to restrict leakage inductance with
enhanced insulation capability. Transformer insulation
parameter is determined based on standard requirement
.Magnetic loss model of both core loss and winding loss are
presented based on which transformer design trade-off between
total loss and volume are analyzed. 4160V 15 kW/500 kHz
transformer prototype is developed and tested to pass industry
insulation standard IEEE Std. C57.12.01. Based on transformer
prototype 15 kW 500 kHz CLLC resonant converter experiment
test is also presented with 98% peak efficiency and 48kW/in3
power density.
REFERENCES
[1] M. Mosman. (September 2012), Medium Voltage Electrical Systems for
Data Centers Presentation [Online]
[2] Pratt, Annabelle, and Pavan Kumar. "Evaluation of direct current
distribution in data centers to improve energy efficiency." INTEL
Corporate Technology Group, 2007.
[3] A. Pratt, P. Kumar, and T. V. Aldridge, "Evaluation of 400V DC
distribution in telco and data centers to improve energy efficiency," in
Telecommunications Energy Conference, 2007. INTELEC 2007. 29th
International, 2007, pp. 32-39.J. Taufiq Power electronics technologies
for railway vehicles", Proc. Int. Power Convers. Conf., pp.1388 -1393
2007.
[4] Pierluigi Sarti, Samir Kala , Zoh Ebrahimzadeh ,Open Compute Project:
V2 Power Shelf Specification 6600W @ 12.6V DC (2+1) Redundant
Facebook.
[5] G. AlLee, W. Tschudi, "Edison redux: 380 Vdc brings reliability and
efficiency to sustainable data centers," IEEE Power and Energy Mag.,
vol.10, no.6, pp.50-59, Nov. 2012
2901
[6] Zhao, T.; Wang, G.; Bhattacharya, S.; Huang, A. Q., "Voltage and Power
Balance Control for a Cascaded H-Bridge Converter-Based Solid-State
Transformer," in Power Electronics, IEEE Transactions on , vol.28, no.4,
pp.1523-1532, April 2013
[7] J. Taufiq Power electronics technologies for railway vehicles", Proc. Int.
Power Convers. Conf., pp.1388 -1393 2007.
[8] M. Steiner and H. Reinold Medium frequency topology in railway
applications", 12th Eur. Conf. on Power Electron. Appl., 2007.
[9] A. J. Watson, P. W. Wheeler, and J. C. Clare, "Field programmable gate
array based control of Dual Active Bridge DC/DC Converter for the
UNIFLEX-PM project," in Power Electronics and Applications (EPE
2011), Proceedings of the 2011-14th European Conference on, 2011, pp.
1-9.
[10] G. Y. Wang, S. Baek, J. Elliott, A. Kadavelugu, F. Wang, X. She, S.
Dutta,Y. Liu, T. F. Zhao, W. X. Yao, R. Gould, S. Bhattacharya, and A.
Q. Huang, "Design and hardware implementation of Gen-I silicon based
solid state transformer," in Proc. IEEE APEC., 2011, pp. 1344-1349 .
[11] J.S.Lai, A.Maitra, A.Mansoor, and F.Goodman, Multilevel intelligent
universal transformer for medium voltage application, in
Proc.IEEE.Ind.Appl.Conf. 2005, pp.1893-1899.
[12] Xu She; Xunwei Yu; Fei Wang; Huang, A.Q. "Design and Demonstration
of a 3.6-kV120-V/10-kVA Solid-State Transformer for Smart Grid
Application", Power Electronics, IEEE Transactions on, On page(s):
3982 - 3996 Volume: 29, Issue: 8, Aug. 2014.
[13] D. Dujic, A. Mester, "Laboratory scale prototype of a power electronic
transformer for traction applications," Power Electronics and
Applications, European Conference on, vol. 1, 2011.
[14] C. Zhao , D. Dujic, A. Mester , J. K. Steinke , M. Weiss , S. L. Schmid ,
T. Chaudhuri and P. Stefanutti "Power electronic traction transformer:
Medium Voltage Prototype", IEEE Trans. Power Electron., vol.
61, pp.3257-3268 2014.
[15] J.W.Kolar, J. Huber Solid-State Transformers (SST) Concepts,
Challenges and Opportunities ECPE Workshop "Smart Transformers for
Traction and Future Grid Application", Zurich, Switzerland, February 4-
5, 2016.
[16] J. E. Huber and J. W. Kolar, "Volume/weight/cost comparison of a 1MVA
10 kV/400 V solid-state against a conventional low-frequency distribution
transformer," in 2014 IEEE Energy Conversion Congress and Exposition
(ECCE), 2014, pp. 4545-4552.
[17] M. A. Bahmani, T. Thiringer, and M. Kharezy, "Optimization and
experimental validation of medium-frequency high power transformers in
solid-state transformer applications," in 2016 IEEE Applied Power
Electronics Conference and Exposition (APEC), 2016, pp. 3043-3050.
[18] "IEEE Standard for General Requirements for Dry-Type Distribution and
Power Transformers," IEEE Std C57.12.01-2015 (Revision of IEEE Std
C57.12.01-2005), pp. 1-52, 2015.
[19] Safety of information technology equipment IEC 60950-1:2005(E)
[20] Chanda, Manas; Roy, Salil K Plastic Technology Handbook Plastics
engineering series, 2007, 4th ed.
[21] M. Mingkai and F. C. Lee, "A new core loss model for rectangular AC
voltages," in Energy Conversion Congress and Exposition (ECCE), 2014
IEEE, 2014, pp. 5214-5220.
[22] C. R. Sullivan, "Optimal choice for number of strands in a litz-wire
transformer winding," IEEE Transactions on Power Electronics, vol. 14,
no. 2, pp. 283-291, 1999.
[23] NFPA 70 National Electrical Code 2014 Edition. Table 310.15(B)(16)
(formerly Table 310.16) page 70-161