вход по аккаунту



код для вставкиСкачать
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
A Tri-Slope Gate Driving GaN DC–DC Converter
With Spurious Noise Compression and Ringing
Suppression for Automotive Applications
Xugang Ke, Student Member, IEEE, Joseph Sankman, Member, IEEE, Yingping Chen, Student Member, IEEE,
Lenian He, Member, IEEE, and D. Brian Ma, Senior Member, IEEE
Abstract— Targeting on electromagnetic interference (EMI)
regulation and ringing suppression issues in automotive applications, this paper presents a gallium nitride (GaN)-based
dc–dc converter operating at 10 MHz. A spurious noise compression technique compresses and re-distributes spurious switching noise within a defined frequency sideband, achieving EMI
noise reduction at main switching frequency and its harmonics.
Meanwhile, a tri-slope gate driver is designed to control voltage
and current slew rates of GaN switches for effective ringing
suppression, which is adaptive to load and input voltage changes.
Tailored for high switching frequency and high-efficiency operation, the dynamic level shifters achieve about 0.8-ns propagation delay and near-zero quiescent current. Fabricated in a
0.35-µm Bipolar-CMOS-DMOS process, the converter accomplishes an EMI noise reduction of 40.5 dBµV and suppresses VSW
ringing by 79.3%. The converter retains above 60% efficiency
over 96.6% of its 6-W power range, with a peak efficiency
of 85.5% at 1.5-W load.
Index Terms— Electromagnetic interference (EMI) regulation,
gallium nitride (GaN) power converter, ringing suppression,
spurious noise compression (SNC), tri-slope gate driving (TSGD).
N RECENT years, high-frequency and high-density power
converters have seen significant increase in automotive
electronics. As power level keeps rising, size and thermal
limits have made it perpetually challenging to the silicon
technology-based solutions. Gallium nitride (GaN) technology, on the other hand, demonstrates a highly competitive
performance as an alternative [1], [2]. Fig. 1 provides a
brief comparison between state-of-art silicon [3], [4] and
GaN power switches [5]. With comparable voltage rating,
a GaN FET demands much lower figure of merit (as the
product of ON-resistance RDS,ON , and gate charge Q G ) and
shows no reverse recovery current. Due to the low Q G ,
Manuscript received April 21, 2017; revised June 11, 2017 and
August 2, 2017; accepted August 25, 2017. This paper was approved by
Guest Editor Pui-In Mak. (Corresponding author: D. Brian Ma.)
X. Ke, Y. Chen, and D. B. Ma are with the University of Texas at Dallas,
Richardson, TX 75080 USA (e-mail:
J. Sankman was with the University of Texas at Dallas, Richardson,
TX 75080 USA. He is now with Texas Instruments, Dallas, TX 75243 USA.
L. He is with Zhejiang University, Hangzhou 310027, China.
Color versions of one or more of the figures in this paper are available
online at
Digital Object Identifier 10.1109/JSSC.2017.2749041
it demands much shorter turn-ON transition times, including
di /dt transition period (shown as t p1 in Fig. 1) and dv/dt
transition period (shown as t p2 in Fig. 1). Hence, a GaN FET
is inherently far superior to its silicon counterpart in terms of
switching loss [6], allowing more efficient power conversion
at high switching frequency f SW [7]–[9].
However, there are some challenges that severely degrade
the performance and reliability of GaN power converters.
First, compared to low f SW operation in silicon-based power
converters, high- f SW operation in GaN converters incurs
much more drastic di /dt and dv/dt transitions, which cause
severe electromagnetic interference (EMI) noise. In a GaN
converter illustrated in Fig. 1, high- f SW operation incurs highspeed transitions at switching node voltage (VSW ) and drain
current (IDH ) of high side GaN FET M H . During the turn-ON
switching of M H , VSW ramps up within nanoseconds [7]–[9].
The instantaneous current Ispike , which is much higher than
the inductor current I L , is drawn from VIN , charging VSW
up and increasing EMI noise. An input filter can reduce
EMI [10], but it greatly increases system volume and cost.
Moreover, high- f SW operation causes significant voltage
spikes Vspike due to the parasitic inductance and capacitance
at the source/drain of M H shown in Fig. 1. This could lead to
GaN VDS breakdown or damage to the internal logic FETs.
To suppress Ispike and Vspike , a series resistor is typically
added at the gate [8], [11], as shown in Fig. 2. With a lower
driving current (limited by RG ), the drain current IDS in
M H takes a longer time to catch up I L , resulting in a longer
di /dt transition period from t1 to t2 . Also, a longer time is
required for VSW to rise to VIN (VDS of M H drops to zero)
in dv/dt transition from t2 to t3 . The turn-ON switching loss
PSW_MH in M H during di /dt and dv/dt transition periods is
defined as
VIN × I L × t × f SW
where t is the overlap time from t1 to t3 in Fig. 2. After
VSW reaches to VIN and VGH (VGS of M H ) exits from
Miller plateau, a much longer charging time from t3 to t4 is
required with RG , resulting in a high conduction loss in M H .
In addition, the passive metal gate resistor could cause hot
spot or thermal issues under a high driving current (with
0018-9200 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 1.
Brief comparison on silicon and GaN power switches.
Fig. 2.
Turn-ON switching loss comparison in M H w/o gate resistor RG .
on-chip RG ), or cause a false turn-ON of M L when VSW
rises (with off-chip RG ) which leads to shoot-through current
in the power switches. Second, as f SW rises to several tens
of megahertz, spurious switching noise at fundamental frequency ( f 0 ) and its harmonic frequencies (N f0 ) can no longer
be handled effectively by a passive LC network or an active
spur filter [10], [12], [13]. Interleaved converters [14] reduce
the spurs with harmonic cancellation effect, however, multiple
inductors and complex clock/phase control are involved.
Frequency hopping techniques [15] were also proposed,
however, inevitable VO transient associated with instant
changes of f SW narrows down the frequency hopping range
and reduces its effectiveness.
In this paper, an automotive-use GaN dc–dc converter is
presented, which overcomes the EMI challenges in high- f SW
operation. A spurious noise compression (SNC) scheme is
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 3.
Block diagram of the proposed high- f SW GaN-based buck converter.
proposed to compress and evenly redistribute the spurious
noise within a defined frequency sideband f m , achieving significant EMI noise reduction at main switching frequency f 0
and its harmonics. In the meantime, a tri-slope gate driver
is presented to control the slew rates (SRs) of IDH and VSW
adaptively, and thus suppress the ringing at VSW . The remainder of this paper is organized as follows. Section II first
introduces the system architecture of the proposed converter.
Then, two key operation schemes—SNC and tri-slope gate
driving (TSGD)—are introduced in due course. Section III
presents the design of key circuit modules. The functionality
and performance of the proposed work are verified by the
experimental results in Section IV. Finally, we conclude this
research work in Section V.
A. System Architecture
Fig. 3 illustrates the system architecture of the proposed
converter. Two enhancement mode GaN FETs M H and M L are
employed as power switches. To effectively drive the high side
switch M H , a nA-I Q dynamic up-level shifter is utilized to
transmit drive signals from low-voltage domain to high-voltage
bootstrap (BST) domain, and a BST circuit is implemented to
generate sufficiently high BST rail voltage (VBST ). To drive
low side switch M L , a dual slew-rate gate driver is used to
achieve a reliable turn-ON of M L , and meanwhile preventing
a false turn-ON of M L during fast dv/dt transition of VSW .
And a delay matching cell is developed to compensate additional delay introduced by the up-level shifter. A fixed dead
time (tdead ) is ensured between the turn-ON of M H and M L to
prevent shoot through. For a given VIN or I O , with an optimal
fixed tdead , M L is turned ON when VSW crosses 0 V at the
VSW falling edge. Moreover, a pulse-width modulation (PWM)
controller is used to generate a proper turn-ON time and
regulate an accurate output voltage. Rather than operating at a
fixed f SW , the converter employs an SNC scheme to generate
a clock signal Vclk , which randomly shifts the switching
frequency within a sideband of f m . Meanwhile, a tri-slope
gate driver charges VGH up at a variable current (I M ) which
helps suppress the ringing effect adaptively.
B. Spurious Noise Compression
Fig. 4 describes the proposed SNC scheme. An envelope generator in Fig. 4(a) generates a triangular envelope
signal VTR , as shown in Fig. 4(b). Meanwhile, a random
noise (Vn ) source that is acquired from avalanche noise of
the Zener diode D Z is coupled to the envelope signal. The
common level of VTR is level shifted by Vdc , achieving a
reference Vref (contains noise) as
Vref = Vn + VTR + Vdc .
As shown in Fig. 4(b), when Vramp reaches Vref , the clock signal Vclk is flipped to “1”. After a fixed delay tpd , the switch S1
is turned ON to discharge Cr . A new switching cycle then
starts. Here, a switching period TSW is defined as
f SW
Since Vref is a function of Vdc , triangular envelope signal VTR ,
random noise signal Vn , Vramp hits Vref in a randomized way.
Thus, TSW differs in each cycle. Here, Vdc and the dc voltage
level of VTR determine main frequency f 0 of the converter.
The envelope of VTR determines the frequency modulation
band f m . When VTR goes higher, a longer time is required
until Vramp reaches Vref . TSW becomes longer, and switching
frequency f SW (1/TSW) goes lower. When VTR reaches to the
highest point, Vref is at the peak value. At the same time,
f SW reaches to lowest point ( f 0 − f m /2). Similarly, when
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 4.
SNC: (a) System block diagram. (b) Key operation waveforms.
VTR goes lower, a shorter time is taken to charge Cr up to Vref ,
resulting in a smaller TSW . When Vref reaches to valley point,
f SW achieves highest value ( f0 + f m /2). To achieve aperiodic
distribution of the frequencies, the modulation period (Tmod )
VTR is defined as
Tmod = N × TSW
where N is an integer. According to Carson’s rule described
in [16] and [17], the total power of a signal is unaffected by
the frequency modulation. Take sine wave with single tone
frequency f0 as an example, the modulated signal Vmod (t) is
expressed as
Vmod (t) = A cos[2π f 0 t + β sin(2π f mod t)]
where f mod (= 1/Tmod) is the modulation frequency and β is
the modulation index
In terms of Bessel functions of nth order ( Jn ), the Fourier
transform of (5) is given by
Vmod (t) = A
Jn (β) cos[2π( f 0 + n fmod )t].
According to [18], the spur reduction or amplitude attenuation
is defined as
1 2
J (β) ≈ 2πβ .
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 5.
(a) Noise spectrum comparison w/o proposed SNC scheme. (b) FFT analysis on Vin_n without/with SNC scheme.
Thus, the extent of distribution or the power magnitude (peak
value) of the spurs depends on the modulation index β. The
higher the β value, the more evenly distributed the spectrum
of the spurs. Since the modulated spur power density is concentrated and remains quite high by solely using the periodic
triangular envelope VTR . Thus, a random noise signal Vn is at
last employed to achieve an aperiodicity of the f SW and evenly
distribution of spur energy. In this way, the generic expression
of a frequency modulated sinusoidal waveform F(t) with
VTR and Vn is given as
F(t) = A cos 2π f 0 t + kω
(VT R (t) + Vn (t))dt
where A is the amplitude of original signal, kω is a sensitivity
coefficient of the modulating signal.
Fig. 5(a) demonstrates spurious noise spectrum with and
without the proposed SNC scheme. A fast Fourier transform (FFT) analysis of input noise Vn_in with and without
SNC scheme is shown in Fig. 5(b). With an amplitude
of VTR and Vn at 50 and 1.32 mV, respectively, a nearly
20-dBμV spur reduction is achieved at 10 MHz. Since the total
power maintains constant, the spurious noise energy at f 0 and
its harmonics (N f0 ) are compressed along frequency domain,
resulting in effective EMI reduction. Spurious noise floor drops
below the EMI standard limit, eliminating the need of an input
filter. Moreover, in comparison with other frequency spread
spectrum techniques [16]–[18], the random noise acquired
from Zener diode achieves a random distribution of fSW and
evenly re-distributes spurious noise energy.
C. Tri-Slope Gate Driving
To mitigate VSW and IDH spikes during M H turn-ON transition and reduce high-frequency EMI noise, an adaptive TSGD
scheme is presented. Instead of using fixed current (IGH ) gate
driving, by sensing I O and VIN , three different driving currents
I M1 , I M2 , and I M3 are determined correspondingly to provide
a tri-slope charging profile of M H , as shown in Fig. 6. When
VPWM goes high, VDH triggers high and is level shifted up
to the BST domain by a nA-I Q dynamic up-level shifter.
As shown in Fig. 6(b), after VGL goes low, VGH starts to get
charged with a current of I M1 , which is proportional to −I O .
At high I O with a low modulated I M1 , VGH charges from Vth to
Vmp within a period of t p1 until M H enters the Miller plateau
region. VGH is charged up to Miller plateau voltage Vmp
Vmp = Vth + Vov
where Vov is the overdrive voltage for M H . At this time,
IDH reaches to the peak value of inductor current, I L ,peak .
Given a gate charging current of IGH , the required charging
time (t p1 ) is computed as
(Cgs + Cgd ) × Vov
Hence, the SR of IDH can be modulated by adjusting the gate
charging current IGH . When VGH is larger than Vmp , M2 turns
on to increase the driving current by I M2 which is proportional to −VIN . At low I O with a high I M1 , a less charging
time of t p1 ’ (<t p1 ) is modulated before VGH reaches Vmp .
t p1 =
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
to the gate driving scheme using passive gate resistor RG ,
the TSGD completes much faster charging action in t p3 to
ensure a lower conduction loss.
A. Spurious Noise Compression Clock Generator
Fig. 6.
M H turn-ON transition with adaptive TSGD. (a) System block
diagram. (b) Timing diagram.
In the second phase, VSW rises with a gate charge current of
I M2 until VSW reaches VIN . When IDH rises higher than I L ,
VSW is charged up by
IDH,avg = IDH − I L .
With a parasitic capacitance of CSW at VSW , the charging
period (t p2 ) from t2 to t3 is defined as
t p2 =
For low VIN , VSW is charged up to VIN earlier resulting
in a shorter t p2 ’ (<t p2 ). In this way, the rising slopes of
IDH and VSW are adaptively adjusted. The overshoot voltage
spikes and ringing at VSW are greatly suppressed. With a
controlled rising slope of IDH and a reduced current spike,
the instantaneous current drawn from VIN is minimized, reducing high-frequency EMI noise that injected to VIN . Compared
The circuit schematic of the SNC clock generator is
illustrated in Fig. 7. To modulate f SW within a sideband,
a triangular envelope VTR is generated as follows. A modulation capacitor, Cfm , is charged with a fixed current (Ib )
until it reaches the high reference voltage of V H . Then Cfm
is discharged with the same current of Ib till it drops to
the low reference voltage of VL , and then the new cycle
starts. In this way, a triangular envelope waveform between
V H and VL is generated. Random noise is produced from the
avalanche noise in the Zener diode D Z , which is in series with
a large resistor R Z . Compared to other on-chip noise sources,
such as thermal noise generated by resistor or MOSFET,
noise from Zener diode not only demonstrates a much higher
intensity of noise level but also shows a weak relation with
temperature [19], [20]. The random noise is similar to white
noise in nature and is coupled to the envelope signal VTR
through capacitor CC . A wide-bandwidth source follower
(A in Fig. 7) drives CC . The amplitude of Vn is small
compared to VTR . Take 5% modulation range ( f m / f 0 ) as an
example, the amplitude of Vn is only 3.1% of the amplitude of
VTR (50 mV). Then, the common mode level of VTR is level
shifted up with a dc voltage value which equals (I f × Rref ),
achieving a reference voltage Vref . The difference between
V H and VL determines modulation range ( f m ), while the
dc portion of Vref determines f 0 . In the SNC clock generator,
a ramp current that is defined by the voltage at Vb charges
the capacitor Cr , and a saw-tooth ramp voltage is produced.
Vramp hits the reference Vref , generating a clock signal Vclk
(at the output of high-speed comparator CMP1) with randomly
distributed frequencies over a sideband f m . The rising edge
of Vclk triggers the turn-ON of M H , and f SW of the converter
is modulated in the same way.
To achieve effective EMI noise reduction over different VIN
and I O conditions, the f m is adjustable by changing Ram .
To achieve a higher f m / f 0 , a smaller Ram is selected
externally. Since
+ I f × Rref ,
Vref =
with a smaller Ram and given VTR , the ac amplitude of Vref
becomes larger, achieving a larger f m . The dc level shifting
caused by VTR is compensated by an adjustable current I f ,
achieving a constant f 0 under different f m / f 0 . Typically,
Ram is set at a low value to achieve a larger f m . This helps
to compress the spurious noise over wide frequency band.
B. Adaptive Tri-Slope Gate Driver
Fig. 8 shows the circuit schematic of the adaptive tri-slope
gate driver. The load current I O and input voltage VIN are
detected by the inductor DC resistance and resistor divider,
and then converted to respective sensing currents. The sensing
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 7.
Schematic of the SNC block.
Fig. 8.
Schematic of the adaptive tri-slope gate driver.
Fig. 9.
(a) nA-I Q dynamic up-level shifter. (b) nA-I Q dynamic down-level shifter.
currents are level shifted up from low-voltage rail to BST
rail, achieving the currents of IOSNS and IVIN in high-voltage
BST rail. The detailed operation was depicted in Fig. 6(b).
During the turn-ON of M H , VGH starts to get charged initially
with a current of I M1 , which is defined by the overdrive
voltage (VOV,M1 ) in M1
VOV,M1 = R1 × (I B − IOSNS ) − Vth .
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 10.
Fig. 11.
PWM controller.
Fig. 13.
PCB test board.
Fig. 14.
Measured efficiency.
Dual slew-rate low side gate driver.
When VGH is further charged up until it reaches Vmp (Miller
plateau voltage), detected by MT 2 , the second charging branch
from M2 turns ON. The gate charging current is increased
by I M2 , which is defined by the overdrive voltage (VOV,M2 )
in M2
VOV,M2 = R2 × (I B − IV IN ) − Vth .
If gm2 is the transconductance of M2 , then I M2 is defined as
I M2 = gm2 × VOV,M2 .
Fig. 12.
Chip micrograph.
If gm1 is the transconductance of M1 , then I M1 is defined as
I M1 = gm1 × VOV,M1 .
With the charging current of I M1 , VGH first reaches to
Vth of M H . When VGH goes higher than Vth , M H starts
conducting and its drain current IDH rises up. Since I B is
fixed, with a higher I O and sensing current IOSNS , VOV,M1 is
smaller, resulting in a lower I M1 . Therefore, during the di /dt
transition period (t p1 in the waveform), VGH is charged up with
I M1 which is negatively proportional to I O (I M1 ∝ −I O ) to
realize a modulated slope of IDH .
For high VIN and thus a large IVIN , a low I M2 is generated
to reduce VSW rising slope, and suppress VSW ringing. When
VGH rises above the Miller plateau region and goes slightly
higher than 2Vth , V P3 is pulled down by MT 3 quickly to enable
a large charge current I M3 from M P3 in the third charging
branch. With a much larger charging current during t p3 period,
VGH reaches to final value quickly to reduce the conduction
loss in M H .
C. nA-I Q Dynamic Up- and Down-Level Shifters
High-speed level shifters are crucial for effective gate
driving in high- f SW converters. However, conventional level
shifters suffer from long propagation delay (>10 ns) or consume high static current from BST to ground [7], [8]. At light
load, the static I Q would discharge CBST when the converter
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 15.
Measured VSW without/with proposed SNC scheme.
Fig. 16.
(a) Noise test setup and (b) measured noise spectrum w/o proposed SNC scheme.
stays in sleep mode that involves no switching actions. When
the converter returns to active mode, M H cannot be turned ON
effectively due to a low BST voltage. M H can be damaged
with a high RDSON under a high thermal generation. Thus,
two nA-I Q dynamic up- and down-level shifters with complementary structure are proposed in Fig. 9. The proposed
up-level shifter employs a latch-cell structure, which consists
of high-voltage drain-enhanced MOS (DEMOS) Md5 , Md6 ,
and low-voltage nMOS Mn3 and Mn4 . When the input signal (IN_UP) goes high, Mn1 is OFF and Mn2 is turned ON
to reset the latch by pulling down the drain of Md6 . Since
the p-type transistor Md4 stays on with its gate connected
to VSW , the source terminal of Md4 drops down until Md5
is turned ON. The drain voltage of Md5 is pulled high and
the p-type transistor Md3 stays on with its gate tied to VSW .
Thus, the output signal (OUT_UP) goes up to achieve an
effective signal level shifting. Similarly, when IN goes low,
Mn1 turns ON to pull down the drain of Md5 , and thus
OUT_UP decreases. In the circuit, the cascaded LDMOSs
Md1 and Md2 provide high-voltage isolation, and low-voltage
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 17.
(a) Measured V O ripple and (b) load transient w/o SNC scheme.
Fig. 18.
Measured VSW ringing suppression w/o proposed TSGD scheme.
NMOSs Mn1 and Mn2 ensure fast driving and level shifting.
Moreover, to block a direct discharging from VSW to ground,
two DEMOS Md3 , Md4 are connected back-to-back with
Mn3 and Mn4 . Also, two gate resistors are added for effective
gate voltage overstress protection during a charged device
model electrostatic discharge event. The up-level shifter consumes a static current of only 4 nA.
To down-shift VGH to the low-voltage rail for tdead control,
Fig. 9(b) presents a nA-I Q dynamic down-level shifter. Two
p-type transistors M p1 and M p2 enable high-speed driving,
and DEMOSs Mde1 and Mde2 withstand high voltage for fast
level shifting. When the input signal (IN_DN) goes high,
M p1 turns ON to pull up the drain of Mde5 . Since the
nMOS Md7 stays on with its gate tied to VDRV , the output
of this down-level shifter (OUT_DN) goes to high immediately. Similar operation occurs when IN goes low. Here, the
high-voltage LDMOS Md7 and Md8 are connected back-toback with M p3 and M p4 to prevent a discharge from VBST
to VDRV . With a similar latch-cell structure of DEMOS Mde5 ,
Mde6 , and pMOS M p3 , M p4 , a static current of only 2 nA is
D. PWM Controller
To achieve an accurate output voltage (VO ) regulation,
a PWM controller is presented in Fig. 10. VO is regulated to
a voltage which is defined by VREF and the feedback resistors
(R1 and R2 ) at VFB . The PWM control scheme operates as
follows. When VO drops and VFB < VREF , the output of
error amplifier Vcomp goes up. A higher inductor sensing current (Isns ) is required to trigger the main comparator (CMP),
and thus the inductor current (I L ) ramps higher to compensate
VO drop until VO is regulated back. Similar regulation happens
when VO ramps high. Since VFB > VREF , Vcomp goes lower,
resulting in a smaller I L , and VO is regulated lower. In this
way, an accurate VO is achieved. With the fixed frequency
peak current mode control, the slope compensation (Islp ) technique is required to ensure the stability over wide operation
ranges. With the help of the RS latch, the rising edge of
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 19.
Measured VGH with proposed TSGD at different levels of I O and VIN .
Fig. 20.
Measured conductive EMI noise: (a) without SNC&TSGD, (b) only with SNC, and (c) with SNC&TSGD.
clock (Vclk ) triggers turn-ON of M H . When I L ramps up,
the sensed current Isns is combined with Islp , which is further
converted to a voltage (VS ). Until VS goes higher than Vcomp ,
the RS latch is reset, and VPWM goes low to turn OFF M H .
E. Dual-SR Low Side Gate Driver
To control the low side GaN FET M L , a dual-SR low
side gate driver is presented in Fig. 11. A higher turn-OFF
SR is implemented with a large-size (120×) transistor Mpd ,
in comparison with a lower turn-ON SR with Mph . When
M H turns ON, VSW rises up to VIN (12 V, nominal battery
voltage) in nanoseconds [7]–[9]. Due to Miller capacitance
Cgd of M L , and the parasitic inductance/resistance in the
source/gate terminal of M L , the gate (G) of M L is pulled high
at high VSW dv/dt transients. The gate voltage jumps higher
than Vth of M L , and M L turns ON again, resulting in a large
shoot-through current from VIN to ground. This either causes
a destructive damage of both GaN FETs due to an unrestricted
high current from VIN to ground, or a catastrophic collapse of
VIN bus. Two approaches are implemented to solve the issue.
First, the ON-resistance of Mpd (120×) is scaled much smaller
to reduce the overall resistance from the G to source (S)
of M L . Second, multiple bond wires are employed at the
gate of M L to reduce the total parasitic inductance/resistance.
Furthermore, with a separated source terminal (S R ) of M L ,
the gate drive loop is separated from drain–source current path
of M L which carries a high current. This ensures an effective
turn-ON of M L , and minimizes common source inductance to
reduce switching loss.
A prototype of the proposed GaN dc–dc converter is implemented in a 0.35-μm HV Bipolar-CMOS-DMOS process.
The chip micrograph is shown in Fig. 12, with an active die
area of 0.86 mm2 , including a 0.6-nF BST capacitor (CBST ).
The printed circuit board (PCB) test board is shown in Fig.
13. Two enhancement mode GaN power switches are placed
closely to minimize parasitic inductance and resistance. This
helps to achieve fast driving transitions and suppress gate
voltage overshooting. A 470-nH inductor (L) and a 1-μF
capacitor (C O ) are used in the power stage.
Fig. 14 shows the measured efficiency, which peaks at
85.5% for a 12-to-5-V conversion, and at 82.1% for a
24-to-5-V conversion. The efficiency stays above 60% over
96.6% of its 6-W power range. With the tri-slope gate driver,
VGH slope reduces as I O increases to suppress EMI at the
expense of less than 0.8% efficiency. With the SNC scheme,
the efficiency reduction is mainly caused by the variable
I L and VO ripples. With a low ESR C O (<10 m) and a
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 21.
Measured V SW with different conditions of I O , VIN , and f m / f 0 .
modulation range f m / f 0 of 5%, a peak efficiency reduction
is less than 0.2%.
Figs. 15 and 16 show the measured VSW and noise spectrum
without and with the proposed SNC scheme, respectively.
In Fig. 15, with the SNC scheme, it demonstrates a frequency
modulation range ( f m / f 0 ) of 5% and 12%. In Fig. 16(a),
a line impedance stabilization network (LISN) captures the
EMI noise generated from the converter which is then sent
to the spectrum analyzer. With the proposed SNC, peak EMI
noise is reduced from 84.84 to 64.48 dBμV with a modulation
range of 5%, and is further reduced to 33.03 dBμV with a
modulation range of 12%. Moreover, VO ripple and load transient response are measured in Fig. 17. With the SNC, a 19.1mV peak-to-peak VO ripple is measured with the f mod of VTR
at 20 kHz. In response to a load step-up/down of 0.5 A/100 ns,
a 51-mV VO droop is observed when f SW is modulated
higher (with the SNC), and a 49-mV VO overshoot is obtained
when f SW is nearly same as f 0 .
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 18 shows the measured VSW ringing suppression with
the TSGD scheme. At 1-A I O and 12-V VIN , VSW ringing
is reduced by 79.3%, compared to the case without the
TSGD. Specifically, the di /dt transition is modulated longer
to achieve a slower rising slope of IDH , and minimizing
instantaneous current drawn from VIN and bouncing voltages
in the VIN and VSW bond wires. The measured VGH with
adaptive tri-slope gate driver in different I O and VIN conditions
is shown in Fig. 19. When I O increases, the di /dt transition
period (t p1 ) is modulated from 0.9 to 2.5 ns. As VIN increases,
the dv/dt transition period (t p2 ) varies from 0.6 to 2 ns.
Fig. 20 shows the measured EMI noise spectrum in a
frequency range of 2 to 100 MHz, with a resolution bandwidth RBW of 20 kHz. Compared to the conventional
design with fixed f SW , peak EMI noise at 10 MHz is
reduced from 84.73 to 51.33 dBμV with the SNC scheme
( f m / f 0 = 9%), and is further reduced to 44.23 dBμV with
the TSGD. Similarly, peak EMI noise at 90-MHz frequency
is reduced from 78.07 to 61.07 dBμV with the SNC scheme,
and is further reduced to 55.60 dBμV with the TSGD. At the
middle frequency range from 10 to 70 MHz, the EMI noise
floor is reduced from 52.32 to 43.62 dBμV with the TSGD.
Fig. 21 gives the measured VSW at different I O , VIN and
modulation range ( f m / f 0 ). As the converter operates in a
wide VIN range of 3 to 40 V and offers a maximum I O of 1.2 A
at 10 MHz f SW , it supports different frequency modulation
ranges to achieve effective EMI reduction in different I O ,
VIN conditions.
As a summary, Table I compares this design with prior arts.
A TSGD GaN dc–dc converter for EMI-regulated automotive electronics is presented. The SNC scheme provides a
random distributed f SW within a sideband of f m , resulting
in a sizable EMI noise reduction. With the proposed adaptive
TSGD, the SRs of IDH and VSW in high side power switch
are modulated according to I O and VIN conditions, achieving
an effective VSW ringing suppression. Noise floor at the midfrequency range is also reduced, eliminating the need of
an input filter. Experimental results successfully verify the
effectiveness of the design techniques.
[7] X. Ke, J. Sankman, M. K. Song, P. Forghani, and D. B. Ma,
“A 3-to-40 V 10-to-30 MHz automotive-use GaN driver with active
BST balancing and VSW dual-edge dead-time modulation achieving
8.3% efficiency improvement and 3.4 ns constant propagation delay,”
in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers,
Feb. 2016, pp. 302–304.
[8] M. K. Song, L. Chen, J. Sankman, S. Terry, and D. Ma,
“A 20 V 8.4 W 20 MHz four-phase GaN DC-DC converter with fully
on-chip dual-SR bootstrapped GaN FET driver achieving 4 ns constant
propagation delay and 1 ns switching rise time,” in IEEE Int. Solid-State
Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2015, pp. 302–303.
[9] X. Ke, J. Sankman, Y. Chen, L. He, and D. B. Ma,
“A 10 MHz 3-to-40 V VIN tri-slope gate driving GaN DC-DC
converter with 40.5 dBμV spurious noise compression and 79.3%
ringing suppression for automotive applications,” in IEEE Int. SolidState Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2017, pp. 430–432.
[10] A. Majid, J. Saleem, H. B. Kotte, R. Ambatipudi, and K. Bertilsson,
“Design and implementation of EMI filter for high frequency (MHz)
power converters,” in Proc. Int. Symp. Electromagn. Compat., Rome,
Italy, Sep. 2012, pp. 1–4.
[11] J. Gottschlich and R. W. De Doncker, “A programmable gate driver
for power semiconductor switching loss characterization,” in Proc. 11th
IEEE Int. Conf. Power Electron. Drive Syst., Sydney, NSW, Australia,
Aug. 2015, pp. 456–461.
[12] D. Kos, F. Mihalic, and K. Jezernik, “Conductive EMI reduction
in switched-mode power converters,” in Proc. IEEE Int. Symp. Ind.
Electron., Dubrovnik, Croatia, Jun. 2005, pp. 441–446.
[13] L. E. LaWhite and M. F. Schlecht, “Active filters for 1 MHz power
circuits with strict input/output ripple requirements,” in Proc. 17th Annu.
IEEE Power Electron. Specialists Conf., Vancouver, Canada, Jun. 1986,
pp. 255–263.
[14] P. Zumel, O. Garcia, J. A. Cobos, and J. Uceda, “Exploring interleaved
converters as an EMI reduction technique in power converters,” in Proc.
28th IEEE Annu. Conf. Ind. Electron. Soc., Seville, Spain, Nov. 2002,
pp. 1219–1224.
[15] C. Tao and A. A. Fayed, “Spurious-noise-free buck regulator for direct
powering of analog/RF loads using PWM control with random frequency
hopping and random phase chopping,” in IEEE Int. Solid-State Circuits
Conf. (ISSCC) Dig. Tech. Papers, Feb. 2011, pp. 396–398.
[16] F. Lin and D. Y. Chen, “Reduction of power supply EMI emission by
switching frequency modulation,” IEEE Trans. Power Electron., vol. 9,
no. 1, pp. 132–137, Jan. 1994.
[17] J. Alfonso, “SSCG methods of EMI emissions reduction
applied to switching power converters,” Ph.D. dissertation,
Dept. Electron. Eng., Polytech. Univ. Catalonia, Barcelona,
Spain, Jun. 2004.
[18] K.-A. Tran, “A spread-spectrum clock generator using phase interpolation for EMI reduction,” Ph.D. dissertation, Dept. Elect. Eng. Comput.
Sci., Massachusetts Inst. Technol., Cambridge, MA, USA, Jun. 2014.
[19] B. Wilamowski and J. Irwin, “Noise in semiconductor devices,” in Industrial Electronics Handbook: Fundamentals of Industrial Electronics.
Boca Raton, FL, USA: CRC Press, 2011, ch. 11.
[20] W. T. Holman, J. A. Connelly, and A. B. Dowlatabadi, “An integrated
analog/digital random noise source,” IEEE Trans. Circuits Syst. I,
Fundam. Theory Appl., vol. 44, no. 6, pp. 521–528, Jun. 1997.
[1] B. J. Baliga, “Gallium nitride devices for power electronic applications,”
Semicond. Sci. Technol., vol. 28, no. 7, pp. 1–8, 2013.
[2] A. Lidow, J. Strydom, M. Rooij, and D. Reusch, GaN Transistors for
Efficient Power Conversion, 2nd ed. West Sussex, U.K.: Wiley, 2015.
[3] On Semiconductor. FDP054N10: N-Channel Power Trench MOSFET
100 V, 144 A, 5.5 m. Accessed: Nov. 2013. [Online]. Available:
[4] Infineon Technol. International Rectifier IRLB4030PbF. Accessed:
[5] Efficient Power Conversion Corp. EPC1001—Enhancement Mode
Power Transistor. Accessed: Apr. 2010. [Online]. Available:
[6] Y. Xiong, S. Sun, H. Jia, P. Shea, and Z. J. Shen, “New physical insights
on power MOSFET switching losses,” IEEE Trans. Power Electron.,
vol. 24, no. 2, pp. 525–531, Feb. 2009.
Xugang Ke (S’14) received the B.E. and M.E.
degrees in electrical engineering from Zhejiang University, Hangzhou, China, in 2007 and 2010, respectively. He is currently pursuing the Ph.D. degree with
the Integrated Power System Laboratory, University
of Texas at Dallas, Richardson, TX, USA.
In 2010, he joined Texas Instruments, China
Design Center, Shanghai, China, as an Analog/Power IC Designer. He involved in designing
eight power management ICs, including high-voltage
step-up/-down converters, backlighting LED drivers,
and LCD bias PMUs until 2014. His current research interests include highfrequency GaN drivers or GaN-based power converters, high-voltage siliconbased power converters.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Joseph Sankman (S’05–M’15) received the B.S.
degree in electrical engineering from the University
of Arizona, Tucson, AZ, USA, in 2010, the Ph.D.
degree in electrical engineering from the University
of Texas at Dallas, Richardson, TX, USA, in 2014.
He joined Texas Instruments, Dallas, TX, USA,
in 2015, involve in linear and switch-mode battery
chargers. He is currently involved in ultra-lowpower circuits. He has co-authored more than a
dozen peer-reviewed conferences and journals. His
research interests included energy harvesting circuits
and systems and high-performance switch-mode power converters.
Dr. Sankman was awarded the U.S. National Science Foundation Graduate Research Fellowship as well as the Texas Instruments/Semiconductor
Research Corporation Graduate Fellowship in 2011. He received the Best
in Session Award at SRC TECHCON in 2012 and 2013 for his work on
piezoelectric energy harvesting.
Yingping Chen (S’15) received the M.E. degree
in micro-electronics and solid-state electronics from
the University of Chinese Academy of Sciences,
Beijing, China, in 2012, and the B.E. degree in
polymer materials and engineering from the Beijing
Institute of Technology, Beijing, in 2008. He is
currently pursuing the Ph.D. degree in electrical
engineering with the University of Texas at Dallas,
Richardson, TX, USA.
Between 2012 and 2014, he was an Analog Engineer with Sigma Micro., Beijing, where involved in
high-voltage dc–dc converters, LED drivers, and constant-current/constantvoltage ac/dc chargers. In 2014, he joined Dialog Semi., Beijing, where he
involved in PMICs. His current research interests include high-frequency/low
EMI power converters, and GaN drivers.
Lenian He (M’17) received the B.Sc. degree from
the Department of Electronic Engineering, Southeast
University, Nanjing, China, in 1983, and the Ph.D.
degree in Kanazawa University, Ishikawa, Japan,
in 1996.
From 1983 to 1990, he served as a Principal Engineer/CTO with Suzhou Semiconductor Cooperation,
Suzhou, China, where he was responsible for developing CMOS sensors and ICs. From 1996 to 1999,
he was a Senior Research Scientist with the Research
and Development Center, Toyo Tanso Cooperation,
Osaka, Japan. In 1999, he joined Zhejiang University, Hangzhou, China, as an
Associate Professor, where he has been a Full Professor with the College
of Electrical Engineering, since 2002. He is currently the Professor and the
Director of the Institute of VLSI Design, and the Vice Dean of the School
of Microelectronics, Zhejiang University. He has authored or co-authored
more than 120 papers and three books on analog/power IC design, and
holds 18 patents. His current research interests include power management
ICs, drivers for LED lighting and backlighting, ac/dc controllers, and data
Dr. He received the 1991 Komatsu Green Research Award, the 1994
Yoneyama Research Award both in Japan, the 2005 Excellent Staff Award
from College of Electrical Engineering, Zhejiang University, for his teaching
and research contributions, the Scientific and Technological Achievements
Awards from Zhejiang Provincial Government, and the Teaching Achievements Awards from the State Educational Ministry of China, in 2009 and
2013, respectively.
D. Brian Ma (S’99–M’02–SM’07) received the
Ph.D. degree in electrical and electronic engineering
from the Hong Kong University of Science and
Technology, Hong Kong, in 2003. He received the
B.S. and M.S. degrees in electronic science from
NanKai University, Tianjin, China, in 1995 and
1998, respectively.
He was a Faculty Member with Louisana State
University, Baton Rouge, LA, USA, from 2003 to
2004 and University of Arizona, Tucson, AZ, USA,
from 2004 to 2010. He is currently a Distinguished
Chair in microelectronics and a Full Professor of electrical and computer
engineering with the University of Texas at Dallas, Richardson, TX, USA,
where he is a Founding Director of the Integrated Power System Laboratory.
He has authored one book, four book chapters, and over 160 peer-reviewed
journal and conference papers. He also holds five U.S. Patents. His current
research interests include integrated power electronics, with primary interests
on silicon, GaN and SiC-based power IC solutions for big data, IoTs,
automobile electronics, and battery powered portable electronics.
Prof. Ma was a recipient of the U.S. National Science Foundation CAREER
Award in 2009. He also received the University of Arizona AAFSAA
Outstanding Faculty Award in 2006, and was a finalist for University
of Arizona Accolades Outstanding Faculty Award in 2009. He received
numerous research paper or design awards from international conferences
and journals. He was awarded the Analog Devices Professorship during 2004–2008, the TxACE Chair Professorship during 2010–2012, the
Erik Jonsson Chair during 2012–2017, and the Distinguished Chair in Microelectronics since 2017.
Без категории
Размер файла
7 618 Кб
2017, jssc, 2749041
Пожаловаться на содержимое документа