A Study on Shunt Resistor-based Current

Measurements for Fast Switching GaN Devices

Thilini Wickramasinghe, Bruno Allard, Cyril Buttay, Charles Joubert,
Christian Martin, Jean-Fraçois Mogniotte, Hervé Morel, Pascal Bevilacqua,
Thanh-Long Le, Stephane Azzopardi

To cite this version:

Thilini Wickramasinghe, Bruno Allard, Cyril Buttay, Charles Joubert, Christian Martin, et al.. A
Study on Shunt Resistor-based Current Measurements for Fast Switching GaN Devices. 45th IEEE
IECON, Oct 2019, Lisbonne, Portugal. �10.1109/IECON.2019.8927490�. �hal-02405883�

HAL Id: hal-02405883

https://hal.science/hal-02405883v1
Submitted on 11 Dec 2019

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 A Study on Shunt Resistor-based Current
Measurements for Fast Switching GaN Devices
Thilini Wickramasinghe1 , Bruno Allard1 , Cyril 1 1
, Buttay , Charles Joubert , Christian Martin
1

Jean-Fraçois Mogniotte1 , Herve Morel1 , Pascal Bevilacqua1 , Thanh-Long Le2 , Stephane Azzopardi2
1
Univ Lyon, Universite Claude Bernard Lyon 1, INSA Lyon, CNRS, Ampere, F-69621, Villeurbanne, France.
2
Electrical & Electronics Division, Safran Tech - Safran Paris Saclay, France.
[email protected]

Abstract—Current viewing resistor (CVR) based measurement • Shunt resistor-based transducers [3]– [5].
of a high frequency switching cell is presented. The drain • Magnetic induction-based transducers (Rogowski coils,
current of the low-side GaN HEMT switch was measured at current transformers) [8]– [9].
100 kHz switching frequency for a maximum load current of
8 A and a dc input of 80 V. To analyze the results, both • Magnetic-field based transducers (Hall effect, flux gate,
simulations and experiments were used. The parasitic elements of magnetoresistive—MR) [10].
the PCB layout (i.e. calculated using the Ansys-Q3D Extractor In general, shunt resistor-based CSs (SR-CS) are simple,
software) were included in the simulation model. This paper
explains the consequences of mounting CVRs in the common-
highly linear and cost effective but they do not provide
source path of transistors compared to a circuit without a CVR. galvanic isolation [3]– [5]; whereas magnetic induction or
Simulation results are found to be in very good agreement field-based CSs are galvanically isolated [8]– [10]. However,
with measurements. Hence, the simulation model can be used compared to high-performance SR-CS (such as coaxial shunts)
to estimate the current waveforms when it is not desirable to the bandwidth of Rogowski coil CSs is 10–100 times smaller
resort to intrusive measurement methods.
Index Terms—CVR, GaN HEMT, high frequency, current
and they introduce delays. Furthermore, the Rogowski coils
measurements cannot sense DC or low frequency currents.
Transformer-based CSs are applicable only for AC or high
I. I NTRODUCTION frequency pulse current. Planar transformers with high power
are available for on-board integration [11], however, they are
Wide bandgap semiconductor materials enable power tran- physically larger than other CSs.
sistors to operate at high frequencies, high voltages and at high
Compared to the SR-CSs, the hall effect sensors and the
temperatures. In particular, Gallium Nitride (GaN) is attractive
fluxgate suffer from imperfections due to offset and gain drift
to build high electron-mobility transistors (HEMTs). They
under temperature. The dynamic response of a magnetoresis-
offer higher power density and lower on-resistance over silicon
tive (MR) CS is faster than a hall-based alternative sensor.
based counterparts. Therefore, lower losses are expected with
However, it requires specific layout in designing primary bus
a favourable impact on the heat dissipation strategy with a gain
bar to measure MR-based current, which is challenging in a
in mass. Due to a relatively stable threshold voltage over the
compact design [12].
range of 25o to 150o C and the positive temperature coefficient
of the on-resistance, GaN HEMTs are more desirable for
parallel configurations. However, a slight mismatch in the
threshold voltage from one device to another may degrade
the current distribution during switching. These circumstances
are especially concerning as the GaN HEMTs have very fast
switching times (with few tens of nanoseconds) and large
oscillations may arise during transients.
To apply them in practical situations, it is important to
examine the switching behaviour in order to identify and
control these high current-voltage transients. However, these
current waveforms are particularly difficult to measure without
disturbing the behaviour of the circuit.
There are many current measuring concepts based on some
basic physical principles [1], [2]. Most of the novel sensors
are incorporated with digital signal processing technologies
to achieve greater noise immunity and stringent design con-
straints of a printed circuit board (PCB).
Primarily, the current sensor (CS) techniques are based on; Fig. 1. A comparison of shunt resistor technologies (adapted from [4])
 Figure 1 illustrates some commonly used CS resistor types the device with short length, approximately 6.6 nH at high
with a comparison of the rated power against both resistor frequency range.
value and temperature coefficient [4]. As illustrated in Fig. 1,
metal strips, high power exposed element-based devices and
coaxial shunts resistors exhibit better temperature coefficient
than that of a current sense metal alloy. However, the high
power resistors are larger than the current sense metal alloy
i load
and they have parasitic inductances in the order of tens
of nanohenries. Materials with high Rcs vRcs (high thermal
density
capacity and low specific heat) reduce the size of the resistor Rcs
and are suitable for compact applications.
The coaxial shunts v supply
(also known Ls as current viewing Ls
resistors—CVR) are suitable for high-frequency measurements
despite of the constraints in their continuous
rs operations that
can cause thermal issues, and relatively large dimensions. They rs
have extremely flat frequency response from DC to 2 GHz,
present high performance in rise time and in accuracy. The
(a) (b)
CVRs are noise immune and have fast response for high
current measurements (e.g. 5 kA with a 20 ns rise time) [6]. Fig. 2. CVR: (a) simplified circuit model of CVR, (b) SDN-414-025 from
They have no offset and do not require an auxiliary power TandM Research [15].
supply.
This paper presents a study on current measurements of
a high frequency GaN device using a CVR. An experiment- 0.05
Resistance (Rs )
10 0
based analysis was conducted to investigate the impact of 0.04

Rs ( )
|Z| ( )

Impedance (Z cs )
mounting a shunt resistor in the power commutation path. 0.03
For further evaluations, a simulation model of the circuit was 0.02
used. The Section II describes about a coaxial shunt resistor- -2
10 0.01
based current sensor followed by basic configurations of a 12 10
5
10
6
10
7 8
10 100
test prototype. The experiment setup, test conditions and the frequency (Hz)

teta (deg.)
Ls (nH)

experimental results are detailed in Section III. In Section IV, 10 Phase angle ( )

an analysis of the experiment and simulation results are Inductance (L s ) 50

8
presented. Practical issues and suggestions to address those
issues are described in Section V with a conclusion. 6 0
5 6 7 8
10 10 10 10
Frequency (Hz)
II. C URRENT VIEWING SHUNT RESISTOR - BASED
MEASUREMENTS
Fig. 3. Frequency response of SDN-414-025 from TandM Research, measured
by 4294A precision impedance analyzer.
Illustrated in Fig. 2(a) is a simplified equivalent circuit of
a shunt resistor. Here, Rcs is the rated resistance while rs
and Ls are parasitic resistance and inductance respectively. A. An overview and the experimental prototype
The parasitic inductance in the CVR limits the operation The main objectives of the experiment are to obtain CVR-
bandwidth. To minimize the uncertainty of the resistor value, based current measurements at high frequency operation and
the voltage across the CVR is measured by Kelvin bridge to observe the effect of inserting a CVR in the power path. To
connections. achieve them, two test cases are considered: case 1– prototype
Selection of an appropriate power rating of a resistor is with the CVR and case 2– CVR unmounted prototype. A
made according to its joule losses (I 2 R, where I is the comparison of the voltage and current waveforms of the two
maximum continuous current requirement of the application cases can be used to observe the deviations. For the case 2,
and R is nominal resistance). To reduce power dissipation, the current measurements can be predicted by a simulation
smaller resistor values are preferred. Shunt resistors with very model. Therefore, a precise model is essential for reliability of
low inductance or CVRs are being used in high frequency the study. This can be validated by replicating the experiment
applications [5]. results.
For this study, a lead connection type CVR with 25 mΩ The basic configuration of the test circuit is a half-bridge
resistance [15] shown in Fig. 2 was used. A 2.5 W capable converter built with a GaN HEMT as the low-side switch
CVR is sufficient for a case of maximum current of 10 A. and a Schottky diode in the high-side. Figure 4 illustrates a
The impedance measurement of the CVR (i.e. SDN-414-025 schematic diagram of the prototype with the CVR. For non-
from TandM Research) in Fig. 3 depicts a low inductance in isolated probe measurements, the best position is to mount
 TABLE I
L IST OF MAIN COMPONENTS

Function Component Manufacturer Description

Low side switch GS66516T GaN Systems inc. 650 V, 60 A, 27 mΩ
High side switch IDH16G120CS Infineon 40 A, 80 V/ns
Gate driver UCC27611DRV Texas Instruments 5 Vout , 4–6 A
Iso. power supply of the gate driver MEJ1S1205SC Murata 5 V, 200 mA, 1 W, 5.2 kVdc isolation
Fiber optics receiver HFBR-2521Z1 Broadcom Links at dc-5 MBd
Bulk capacitors EEVEB2C100Q Panasonic 10µF±20%, 160Vdc, Al. Elect.
Filter capacitors CKG57K TDK single layer 2.2µF ±20% 450Vdc, Ceramic
Current sensor SDN-414-025 T&M Research 25 mΩ ± 4%, 2 W, 1200 Mhz

Its model corresponds to a 10 µH inductor in series with a

260 mΩ resistor and the bandwidth of 42.5 MHz which in
agreement with measured values as shown in Fig. 5.

TABLE II
T EST CONDITIONS

Parameters Minimum Nominal Maximum

Input voltage 10 V 25 V 80 V
Load 7A 10 A
Switching frequency 100 kHz 200 kHz 400 kHz
Pulse width 250 ns 500 ns
No. of bursts 10
Fig. 4. Schematic diagram of the test prototype
Pulse frequency 10 Hz 10 kHz

the CVR in the source path. As shown in Fig. 4 the current

10-5 104
sensor was implemented between the source terminal and 15

the common ground that reduce the complexity of having a 0

X: 4.021e+05
ground-referenced signal. 10 Y: 0.2587
Series inductance (H)

-0.5

Series resistance ( )
The output characteristics of the gate driver controls the
-1
switching behaviour of the GaN HEMT. By adjusting the 5
X: 4.021e+05
gate resistance, on-off transient speeds of the switch can be Y: 9.997e-06 -1.5
changed. The pulse width modulated (PWM) signal is supplied 0
-2
through a fiber optics receiver to provide isolation. A capacitor
Series resistance (Rs)
bank is used for decoupling during high current transient. Two -2.5
a -5 Series inductance (Ls)
10 µF electrolytic capacitors and six 2.2 µF ceramic capacitors
-3
were implemented for decoupling. The key components of the 104 105 106 107 108
experimental prototype are listed in Table I. Frequency (Hz)

III. E XPERIMENTAL SETUP AND RESULTS Fig. 5. Frequency response of the load
A. Test setup and conditions
A repetitive pulse test setup was used to evaluate the switch-
The main testing apparatus used in the setup are digital ing performance of the GaN device. In the operation of the
oscilloscope (Tektronix DPO 4034B, 350 MHz) combines circuit, a set of discrete pulse is fed to the gate at a repetition
with passive voltage probes (Tektronix P6139B, 500 Mhz), a rate of 1 kHz to avoid self heating of the GaN device. There
current probe (TCP0030A , dc to >120MHz, 30 A—maximum were 10 pulses per set, each with 250 ns pulse width and
range), a differential probe (EDITEST GE-8115, 30 MHz 10 µs pulse period. This corresponds to a 100 kHz switching
bandwidth), and an arbitrary signal generator (DG645 digital frequency with 2.5% duty cycle. The oscilloscope is triggered
delay/pulse generator for PWM generation). For the case of at the 10th pulse of the set to obtain the measurements.
CVR measurements, a BNC cable was used. The EDITEST
GE-8115 probe was used to verify the differential voltages B. Test results
obtained by two passive probes. Figure 6(a) illustrates the case 1 configuration of the CVR
In general, the prototypes were tested under the conditions mounted printed circuit board (PCB). To avoid adding more
that are summarized in Table II. The results presented in the parasitics to the circuit, the leads of the CVR connectors were
paper are based on an input voltage of 80 V and a maximum kept short. In the experiment setup, a heat sink was mounted
load of ∼8 A. The load is a solenoid made with 17 turns to prevent overheating of the Schottky diode. The measured
wound around a polyvinyl-chloride tube with 5 cm diameter. voltage across the resistor is shown in Fig. 6(b) which was
used to estimate drain current. As seen in the figure, the
estimated drain current (current via CVR) with respect to drain
voltage and load current was obtained.

(a)
8

Gate--source voltage (V)/ load current (A)

(a) 100
9 7
0.3 80
6

Load current (A)

8
Voltage (V)

0.2
5 60

Voltage (V)
0.1 7
4 Load current
0
6 Drain to source voltage 40
3
Input voltage
-0.1 CVR voltage (V)
5
2 Gate to source voltage 20
100 -2 -1 0 1 2 3 4
Load current (A) 1
Time scale (s) 10-7
-7 10 0
Drain current (A)

0
Voltage (V)

50 Drain voltage (V)

-1 -20
Load current (A) 5
Current via CVR -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5
0
0 Time scale(s) 10-7

(b)
-50
-5
-2 -1 0 1 2 3 4 Fig. 7. Case 2: (a) prototype without current sensor, (b) experimental results
Time scale (s) 10-7 of voltage-current of PCB without the CVR

(b)
100
Fig. 6. Case 1: (a) Experimental prototype with current sensing via coaxial
shunt resistor. (b) CVR-based measurements during transients. Input voltage
is at 80 V. Drain-to-source voltage
50 (without CVR)
In order to investigate the effect of adding shunt resistor
Voltage (V)

in the power rail, the CVR was removed from the PCB. A Drain voltage (CVR)
copper strip was used to connect source and the ground as
shown in Fig. 7(a) and no drain current can be measured in 0
this configuration. Figure. 7(b) illustrates experimental results
acquired from the PCB.
The drain voltages of the two cases can be compared as
-50
illustrated in Fig. 8. The results depicts that the coaxial shunt
-1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3
has a clear influence on the drain to source voltage. Indeed, Time scale(s) 10-7
the coaxial shunt resistor introduces parasitic inductance in
the circuit due to its connecting leads. Further, it explains Fig. 8. A comparison of drain voltages of the two PCBs with and without
the lower frequency of the ringing compared to the case 2. CVR mounted in the power path.
However, to compare the drain current waveforms in the two
cases, it is required to obtain the drain current waveform of
the case 2. This can be achieved using a simulation model. waveform. Further, a model with an ideal layout can be used
to compare the parasitic effect of PCB traces. A SPICE-based
IV. S IMULATION MODEL OF CIRCUIT FOR THE VALIDATION simulation meta-model was designed based on the simplified
OF EXPERIMENT RESULTS circuit diagram with the key components. The manufacturer-
The aim of the simulation is to replicate experimental supplied SPICE models of the GaN HEMT, the Schottky diode
results of the case 2 in order to predict the drain current and the gate driver were used. The load was represented by
 an inductor connected in parallel with a parasitic capacitor.
Further, the parasitic elements of the decoupling capacitors
were included to their models.
For the extraction of the PCB layout parasitics, the Ansys-
Q3D software was used. Figure 9(a) illustrates the simplified
diagram of the parasitic components considered for this case.
In this circuit, both power and the control commutation paths
are in the same side of a single layer configuration. The par-
asitics between the gate and gate-driver output is insignificant
due the compact design as seen in Fig 9(b). Therefore, the
parasitics in this path can be neglected.
The values of parasitic components corresponding to the
PCB traces were calculated using Ansis-Q3D software. They
were included in the SPICE meta-model as seen in Fig. 10.
For better fitting with the experiment curves, some of these
values were slightly adjusted. The estimated resistance of the
copper strip used in case 2 is approximately 0.15 mΩ, where
the length, thickness and the width are respectively 11 mm,
5 mm and 0.254 mm.
Fig. 10. LTSpice meta-model of double-pulse GaN HEMT circuit without
mounting a CVR
Z1

100 Input voltage (sim) 100 Off transient

Drain-source voltage (sim)
80
Z2 Z3
80

60 Drain-source 60
Z2

Voltage (V)
Voltage (V)

voltage (measured)
40 Input voltage (measured) 40
R gate 1
20 20 Drain-source
voltage (measured)
0 0 Input voltage (measured)
Input voltage (sim)
-20 -20 Drain-source voltage (sim)
On transient
Z3 R gate 2 Z6 Z4 2 4 6 8 10 2.5 3 3.5 4 4.5 5
Time (s) 10-8 Time (s) 10-7

(a) (b)
8
Gate--source voltage (V)/

Gate to source voltage (measured)

6
Z5 Load current (measured)
Z4
load current (A)

4
(a) (b)
2
Fig. 9. Consideration of parasitic components: (a) simplified circuit model
Z1 of parasitics in the layout, (b) the GaN HEMT and driver
with consideration Gate to source voltage (sim)
0 Load current (sim)
on the PCB.
Z2
The simulation results of the case 2 are compared with 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s) 10-7
its experiments as shown in Fig. 11. The inclusion of stray
capacitors of the PCB (i.e. Cx1 , Cx2 and Cx3 ) made the (c)
Z3
simulation Rresults more comparable to its experiments. The
gate 1
Fig. 11. Comparison of results of the case 2 SPICE simulation model. No
outcome of the approach indicates that the simulation results current sensing resistor is implemented in this circuit.
during transients are in fair agreement with its experiments to
predict the drain
R current waveform.
gate 2
Z4
The manufacturer provides three different models for Figures 12(a) and (b) illustrate a comparison of CVR-based
GS66516T HEMT [16]. The basic model represents the elec- current measurements in case 1 with the simulation results of
trical characteristics while the second and the third models are case 2. Both load and the drain currents are compared during
comprised with thermal characteristics, and stray inductances on and off transients. As seen in the figure, the oscillations of
respectively. All three models were studied by comparing the measured drain current during the on-transients are higher
Z
the results of simulations 5 with the experiments. For this compared to a predicted results of case 2. This confirms that
application, there was no substantial difference in the outcome measuring drain current using a CVR adds error to the read-
provided that the absent device parasitics are incorporated ings. Besides, low frequency oscillations in the measurements
externally. can lead to inaccurate estimations of power dissipation in the
 On transient Off transient results comparisons, CVR-based measurements introduce con-
10 10
siderably oscillations during the on-transients. Despite of slight
difference in the period of oscillations, the CVR can measure
Current (A)

Current (A)
5 5
the drain current during off-transients more accurately.
0 0
R EFERENCES
[1] B. Mammano, ”Current sensing solutions for power supply designers”.
-5 -5
In Proc. TI Power Supply Design Sem. 1997.
2 4 6 8 10 2.5 3 3.5 4 4.5 5
Time (s) 10 -8 Time (s)
[2] S. Ziegler, R.C. Woodward, H.H.C. Iu, and L.J. Borle, ”Current sensing
10 -7
Drain current with CVR (measured) Drain current without CVR (sim)
techniques: A review.” IEEE Sensors Journal, vol. 9, no. 4, pp.354-376,
Load current with CVR (measured) Load current without CVR (sim) 2009.
[3] B. Strzalkowski, J. Sorensen, ”Challenges and solutions of high-side cur-
(a) (b) rent and of phase current measurement”, presented at ECPE Workshop,
Hamburg, Germany, Oct 17th, 2017.
On transient Off transient
10 10
[4] J. Schiffner, ”Shunt resistors for precision current measurement, pre-
sented at ECPE Workshop”, Hamburg, Germany, Oct 17th, 2017.
[5] M. Adelmund, C. Bodeker and N. Kaminski, May. ”Optimisation of
Current (A)

Current (A)

5 5 Shunt Resistors for Fast Transients” In PCIM Europe 2016; International

Exhibition and Conference for Power Electronics, Intelligent Motion,
0 0 Renewable Energy and Energy Management, 2016, pp. 1-8.
[6] C. Xiao, L. Zhao, et.al, ”An overview of integratable current sensor
technologies”, in Industry Applications Conference, 38th IAS Annual
-5 -5
Meeting, October 2003, pp. 1251-1258.
2 4 6 8 10 2.5 3 3.5 4 4.5 5
Time (s) -8
[7] Billmann, ”Dimensioning of a proper coaxial shunt type”, presented at
10 Time (s) 10 -7
Drain current with CVR (measured) Drain current without CVR (sim)
ECPE Workshop, Hamburg, Germany, Oct 17th, 2017.
Load current with CVR (measured) Load current without CVR (sim) [8] C.Hewson and J. Aberdeen, ”Wide-Bandwith Rogowski based Measure-
ment in the Lab”, presented at ECPE Workshop, Hamburg, Germany,
(c) (d) Oct 17th, 2017.
[9] M. Denk and S. Hain, Case Study - ”High Bandwith and Minimal In-
Fig. 12. A comparison of measured drain current of case 1 and the SPICE vasive Current Measurement”, presented at ECPE Workshop, Hamburg,
simulation models: (a) and (b) compare the results with the simulation model Germany, Oct 17th, 2017.
of case 2. In the simulation, the drain-current obtained at the drain terminal. [10] J. Lenz and S. Edelstein, 2006. ”Magnetic sensors and their applications”
For the comparisons of (c) and (d), the parasitics in PCB traces are ignored IEEE Sensors Journal, vol. 6, no.3, pp.631-649.
in the simulation model. [11] Vishay, ”Medium Power Planar Transformer 1 kW to 3 kW”, PLA51
datasheet, 2017 [Revision: 8 Feb. 2017]
[12] R. Slatter and M. Kramb, ”High bandwidth, highly integrated current
device. sensors for high power density electromobility applications”, In Proc.
Power Electronics, Machines and Drives (PEMD 2016), 8th IET Inter-
For further investigations, a simulation model was designed national Conference, April 2016, pp. 1-6.
with an ideal PCB layout. In this model, the parasitic com- [13] R. Slatter, ”High accuracy, high bandwidth magnetoresistive current
ponents related to the PCB traces considered in Fig. 9(a) sensors for spacecraft power electronics”, In Proc. 2015 17th European
Conference on Power Electronics and Applications (EPE’15 ECCE-
were ignored. The results of the simulation were used to Europe), Sept.2015, pp. 1-10.
demonstrate the layout parasitic effect as seen in Fig.s 12(c) [14] A. Bouzourene, T. Bensalah, L. Cima and A. Couderette, ”Investiga-
and 12(d) in blue traces. In the ideal case of this design, the tion of Neel Effect technology for current sensors in next-generation
aeronautical applications”, In Proc. 2011-14th Power Electronics and
drain current waveform would oscillate a minimum of 60 ns Applications (EPE 2011) Conference, Aug. 2011 pp. 1-10.
during on-transients whereas in case 2 it is approximately [15] T&M Research Products, Inc., ”Series SDN-414”, SDN-414-025 spec-
100 ns (see Fig. 12(a)). A maximum of 40% oscillation period ifications.
[16] GaN Systems Inc.: ”LTspice Model User Guide” [available online]
can be reduced by eliminating the parasitic components in https://gansystems.com/wp-content/uploads/2018/05/LTspice-Model-
the PCB layout. However, compared to both designs the error User-Guide.pdf
of drain current through a CVR is high. Although it seems
to offer the best current measurement performance, the CVR
introduces unavoidable perturbations to the measurements.

V. C ONCLUSION
A study of CVR-based drain current measurement is per-
formed. By implementing a fairly precise simulation model,
the effect of adding CVR in the power path is observed.
The case 2 SPICE simulations achieved a fair agreement of
experiment results. The model was able to predict drain current
even during transients. Therefore, the PCB without CVR can
be consider as more accurate, and may use for analysis of the
converter operation. As seen in the results, more perturbations
in readings can be reduced by optimizing the PCB design
to reduce the copper-track related parasitics. According to the