From Planar to Trench - Evaluation of Ruggedness across Various Generations of Power MOSFETs and Im

From Planar to Trench - Evaluation of Ruggedness across Various Generations of Power MOSFETs and Im
From Planar to Trench - Evaluation of Ruggedness
across Various Generations of Power MOSFETs
and Implications on In-circuit Performance
Hemal Shah, Steve Oknaian, Eric Persson and Rongjun Huang
International Rectifier Corporation
El Segundo, CA 90245
[email protected]
Abstract—In this paper we compare the ruggedness of
various generations of power MOSFETs using three
criteria - single-pulse avalanche robustness, forwardbiased safe-operation-area (FBSOA), and body-diode
reverse recovery ruggedness. We observed that the
trench-gate FETs are generally inferior to planar-gate
counterparts in both FBSOA and avalanche capabilities;
however, both trench and planar FETs show
comparable performance under certain test conditions.
Moreover, the trench FETs evaluated in this study are
superior in reverse recovery ruggedness.
I.
TABLE I: Parts selected for the comparison
INTRODUCTION
This paper evaluates single-pulse avalanche capacity,
FBSOA, and body-diode reverse recovery ruggedness on
different generations of middle-voltage MOSFETs. By
pointing out the strengths and weaknesses of existing
generations, this evaluation will provide knowledge that
users need for properly selecting and better utilizing existing
MOSFETs technology.
Knowing that MOSFET ruggedness depends heavily on
test condition, device voltage, die size, and many other
factors, we choose different generations of devices that have
the same voltage-ratings, the same packages, and
approximately the same die sizes to ensure silicon design
generation is the sole factor of differences in the ruggedness
performance.
The parts chosen for the testing are listed in Table I are
all TO220 packages. Parts A1 to A5 are 100V rated with
similar die sizes varying from 15.3 to16.9 mm2; while parts
B1 to B4 are 55V-60V rated, with similar die sizes varying
from 5.3 to 6.2 mm2. Comparisons were made separately
among the devices with the same voltage ratings. The
generation information of each part is in the fourth column
with the smaller number indicating an earlier generation.
Note that, Generations G5 and G7 are planar parts, while G8
to G10 are trench parts. The fifth and sixth columns show
that a newer generation has lower RDS(on) and higher ID
rating than an older one.
Device
Voltage
(V)
Die Size
(mm2)
Generations
A1
A2
A3
A4
A5
B1
B2
B3
B4
100
100
100
100
100
55
60
55
55
16.83
15.35
16.26
16.94
15.35
5.70
5.70
5.33
6.25
G5, planar
G7, planar
G8, trench
G9 trench
G10, trench
G5, planar
G7, planar
G9, trench
G9, trench
II.
A.
Rdson
(mΩ,
10Vgs)
36
23
14
10
9
40
26
13.6
11
ID
(A)
42
57
75
96
97
26
30
51
61
SINGLE PULSE AVALANCHE PERFORMANCE
Unclamped Inductive Switching (UIS) results
Increase-to-fail Unclamped Inductive Switching (UIS)
test was used for characterizing avalanche ruggedness in our
evaluation. The test was taken under eight conditions: two
ambient temperatures (25°C and 175°C) and four inductors
(20, 50, 100 and 500 μH). The detailed circuit and operation
of the UIS tester can be found in application note [1]. A
typical increase-to-fail UIS waveform is shown in Fig. 1.
The part starts to operate at avalanche mode when the VDS
exceeds its break-down voltage. The peak inductor current
or IAS gradually increases till the part fails. The highest IAS
at 20 μH, 25°C and 500 μH, 25°C are listed in Table II.
For a rugged MOSFET without parasitic bipolar
transistor effect, some references suggested that the
avalanche failure is thermally induced. It occurs when the
cell temperature reaches a critical value, beyond which the
intrinsic carrier concentration exceeds the epi doping
concentration [2]. Based on this understanding and the fact
that the single-pulse thermal impedance curve of a
MOSFET normally presents a constant slope in certain time
ranges, we can conclude a linear relationship between IAS
and the inductance (or between IAS and TAV, the avalanche
time) in a log-log scale [3].
Vds(50V/div)
IAS(20A/div)
Failure
Vgs(20V/div)
Fig. 2a IAS vs inductance for devices A1-A5 under TA=25°C
Figure 1. Typical increase to fail UIS waveform
TABLE II: Comparison of IAS across parts
Device
VBR(DSS)
A1
A2
A3
A4
A5
B1
B2
B3
B4
100
100
100
100
100
55
60
55
55
IAS (A) @20uH,
25°C
233.22
195.36
81.16
84.62
138.33
106.30
105.58
114.46
125.06
IAS (A) @500uH,
25°C
62.86
64.16
61.62
61.44
55.96
35.16
37.38
34.82
39.14
Figs. 2a-2b and 3a-3b separately compare IAS of parts
A1-A5 and B1-B4 with respect to the inductance under two
different ambient temperatures (25°C and 175°C). Fig. 2a
and Fig. 2b suggest that 100V planar parts have much
higher IAS with smaller inductors, exhibiting more
robustness than their trench counterparts do; while with
larger inductors, the difference between planar and trench
parts is small. The dramatic fall-out of A3 and A4 (the two
old trench technologies) in comparison to A1-A2 (the newer
technologies) is most likely caused by the parasitic BJT,
which is more prone to turn-on at high currents. In contrast,
Fig.3a and Fig. 3b show that, for 55-60V parts, planar and
trench parts have very close avalanche performance under
all tested conditions.
B. Analysis of results and correlation to device physics
The test results obtained from the testing indicate that
the newer Generation 10 trench devices have a much
improved avalanche behavior compared to the older
generation 8,9 trench devices. This is attributed to the
design of the device structure that modifies the p-body
region doping concentration through the addition of a deep
P- implant. The P-body now extends beyond the bottom of
the trench. Additionally an N-implant at the base (tip) of the
trench is also added. It is higher in concentration than the Nepi and usually similar in concentration to the P- implant.
Fig. 2b IAS vs inductance for devices A1-A5 under TA=175°C
Fig. 3a IAS vs inductance for devices B1-B4 under TA=25°C
The balance of charge in these two regions, when done
correctly can provide a breakdown voltage to the device that
is higher than that of the plane-junction breakdown of the
epitaxy resistivity chosen. Additionally the optimization of
these two regions can force the breakdown of the MOSFET
to occur in the region directly under the source contact,
thereby preventing the hole current from providing a bias
voltage to the source and improving UIS performance as a
result.
minimize overall loop inductances that can cause large
voltage spikes. Additionally packages such as the
DirectFET™ offer better dual-sided heatsinking options and
a more uniform thermal distribution across the die-surface
that helps with improved avalanche performance.
III.
Even though the avalanche performance of the planar
FETs is better than that of all generations of trench FETs,
Fig. 2a and Fig. 2b clearly show the superior avalanche
performance of the Gen 10 devices compared to earlier
generation 8 and 9 trench MOSFETs. Comparing Fig. 2a, 2b
and Fig. 3a, 3b it is also evident that at lower breakdown
voltages (55V-60V) and smaller die sizes the avalanche
performance of the trench FETs is comparable to that of the
planar FETs. This is most likely because at the lower
breakdown voltage the destruction seems to be more energy
dependent as opposed to the current density dependent as
highlighted in the literature [4].
C. Choosing MOSFETs for in-circuit applications
Datasheet parameters are merely a first-pass selection
criteria and should not be used by designers prima-facie.
MOSFETs are subject to many stresses depending on the
application and the designer needs to be aware of the
parasitic inductances as well as thermal limitations of the
circuit board and environment as these can lead to
unexpected MOSFET failures. For example leakage
inductances in a switching circuit such as a ZVT boost
topology can cause FET’s to avalanche on successive
pulses, and although one pulse might not raise the
temperature of the MOSFET to cause it to fail, successive
pulses cause thermal buildup that leads to MOSFET
destruction.
Additionally package inductance, heatsinking and
ambient temperature are all factors that need to be closely
monitored as inductances can lead to voltage spikes that can
exceed the breakdown voltage and combined with higher
die temperatures due to inadequate heatsinking can make it
easier for MOSFETs to fail from avalanche destruction.
Higher temperatures increase the gain of the parasitic
bipolar in MOSFETs, which can cause latchup failures and
eventual thermal runaway leading to destruction. Using tight
board layout and low inductance packages (such as
DirectFET™ for switching applications, designers can
The forward-bias safe operating area (FBSOA) of a
MOSFET defines the region of safe and stable operation of
the device. In comparison with planar FETs, the trench
FETs have demonstrated severe thermal instability,
especially at high voltage low current regions of operation.
Thermal instability occurs when the rate of change of the
generated power exceeds the rate of change of the dissipated
power. In this paper, the Spirito approach [5] is used in
order to address the thermal instability problem, which is
more severe on trench FETs. The conventional thermal
boundary is over-estimated, because it is calculated based
on the assumption of uniform thermal distribution across the
entire die area, which is not always true for the trench FETs.
Fig. 4a plots the thermal instability boundary for A1-A5
devices in the same graph. Compared with the 100 μs curve
in Fig. 4b, the 10 ms SOA curve in Fig. 4a drops more
dramatically from the conventional boundary at larger
voltages, indicating more thermal instability problems in
longer period of operation. Both figures suggest that the
earlier generation parts have wider SOA boundaries than the
later ones.
As explained in [6], the narrow FBSOA of trench parts
is not necessarily due to cell structure of trench gates. The
lower the on-resistance of a MOSFET, the higher the zero
temperature crossover point (point of intersection of two
transfer curves at different temperatures) will be, and the
less robust a MOSFET will be for linear mode applications.
This applies to both trench and planar parts.
10ms DC Spirito SOA Plot
1000
100
ID, Drain Current [A]
Fig. 3b IAS vs inductance for devices B1-B4 under TA=175°C
STUDY OF FBSOA BEHAVIOR
A1
10
A1_100V_G5_Planar
1
A3
A2
A2_100V_G7_Planar
A3_100V_G8_Trench
0.1
A4
A4_100V_G9_Trench
A5_100V_G10_Trench
0.01
1.00
10.00
A5
100.00
VDS, Drain-to-Source Voltage [V]
Fig. 4a Spirito SOA thermal boundaries for devices A1-A5 under 10ms
loop inductance of 37 nH. This includes the parasitic lead
and wirebond inductances of both D2-Pak FETs, the 10 mΩ
coaxial current-measuring shunt, the PCB layout, and the
DC bus itself (the complete current-loop excluding the
external inductor). This value is verified experimentally by
turning-on both FETs and watching the current slew-rate.
With a 100V bus voltage, we measured a linear current
ramp of 150 amps in 55 ns across the coaxial shunt, or
2.73A/ns, thus calculating from V=Ldi/dt that L = 37 nH.
This value is also consistent with an LT Spice model of the
tester developed using both measured and estimated
parameters which shows a current slew-rate matching our
measurement when the total loop inductance is 37 nH.
100 us DC Spirito SOA Plot
ID, Drain Current [A]
1000
A4
A3
100
A2
A1_100V_G5_Planar
A2_100V_G7_Planar
10
A5
A1
A3_100V_G8_Trench
A4_100V_G9_Trench
A5_100V_G10_Trench
1
1.00
10.00
100.00
VDS, Drain-to-Source Voltage [V]
Fig. 4b Spirito SOA thermal boundaries for devices A1-A5 under 100
μs
The low current high voltage instability region reduces
the capability of the device to operate within a wider range
of the SOA curve. Care should be exercised in selecting
MOSFETs to guarantee robust operation in strictly linear
mode applications, such as hot swap controllers, inrush
current limiting circuits during turn-on of power converters,
or linear controllers for fan motors. In switching
applications, including clamped inductive, trench
MOSFETs are most suitable because of their low Rdson and
certain level of linear robustness. The SOA curves of the
55V-60V parts (B1-B4) were similar to that of the 100V
parts in Fig 4a and Fig 4b, with vanishing Spirito Effect for
shorter times (100μs), showing consistency in SOA
behavior of trench and planar parts, with planar parts having
larger SOA area compared with the trench parts.
IV.
BODY-DIODE REVERSE-RECOVERY PERFORMANCE
FET body diode reverse recovery behavior is
traditionally evaluated by the double-pulse test method [7,
8, 9]. Because this test methodology is intended to extract
Trr, Qrr and IRRM parameters, the applied di/dt is relatively
slow (100 A/μs). The slow di/dt is partly an historical
legacy, but also because as di/dt is increased, resolving the
parameters of interest becomes more difficult due to
parasitic inductance (L di/dt) masking the measurement of
the parameters of interest. Since today’s power electronics
are aimed at highest efficiency, many designs push
switching transition times lower, and it is not uncommon to
see FETs operating with current slew-rates exceeding
several thousand amps per microsecond. While this is
possible for rugged trench FETs, as will be shown, older
generation planar devices cannot survive these slew rates.
A. Circuit Layout
A simplified schematic of the double-pulse tester used
is shown in Figure 5. The circuit board layout was designed
to add minimal parasitic inductive elements into the highcurrent loop. The board design achieved a measured total
Fig.5 Double-pulse tester schematic
B. Gate Drive
Fast, low impedance gate drive is critical to obtaining
double-pulse measurements that are not distorted by the
switching FET operating slowly, spending significant time
in transition [8]. The gate driver used is a discrete design
based on reference [10] with 10-90% rise and fall times <
15 ns driving the largest gate capacitance used in this study.
Both the turn-on and turn-off RG were set to 2Ω (not
including any internal FET RG). Switching speed in the
double-pulse test benefits from somewhat lower effective
CISS and no Miller-effect during the turn-on edge of the
switching FET (the beginning of the reverse-recovery
period). This is because the Device Under Test (DUT – the
upper FET in our implementation) is still conducting during
switch turn-on, thus there is little change in VDS of the
switched FET, and therefore virtually no Miller capacitance
plateau to slow the gate voltage risetime (the drain voltage
doesn’t fall until long after the gate is fully enhanced –
when the DUT body-diode finally recovers). Moreover,
since VDS is at nearly 80% of BV(DSS) during the entire turnon process, the nonlinear junction capacitance is at its
lowest value, thereby making CISS lower than if the drain
were not effectively clamped to the bus.
D. Results
The example in Figure 6a shows a typical reverserecovery waveform example from device A1. At the
beginning of the trace, the current has been flowing in the
DUT body-diode (magenta trace) for 10 μs and the VDS of
the DUT (Green trace) is at ~0V. The switching FET is
turned-on at 100 ns and the 5A forward current immediately
slews negative, while the VDS shows the L di/dt voltage
across the FET package. The slow risetime of VDS on the
leading edge is due to the combined COSS of the DUT plus
the switch. Once the body-diode recovers, the VDS
immediately slews up to avalanche at ~110V, and the
current reverses direction and heads back to 0, dissipating
the energy stored in the parasitic inductance.
The old generation planar device shown in Figure 6a
exhibits a huge reverse-recovery characteristic, despite a
very small forward current. This is common of planar
MOSFETs from this era. This device failed when IRRM
reached 67A, at an applied forward current of only 17A, as
seen if Figure 6b.
Body-Diode Current IS
10A/div
Timebase 50ns/div
The failure in Figure 6b has the signature of a classic
second-breakdown of the parasitic NPN bipolar. This occurs
during normal sustaining avalanche, when the current
density exceeds a (temperature dependent) threshold. The
mechanism is that as the current (IRRM) is increased, the
bipolar device shifts from normal avalanche multiplication
to avalanche injection and thus second-breakdown,
characterized by the snapback of VDS from VBR(DSS) to ~1/2
that value [11] as can be clearly observed in the VDS
waveform in figure 6b.
Timebase 50ns/div
Body-Diode Current IS
20A/div
Parasitic NPN SecondBreakdown Failure
V DS
20V/div
Fig. 6b Reverse-recovery avalanche failure of device A1
Each device-type in table I was tested in this method –
incrementally increasing the test current until either the
device failed, or ISM (maximum rated source current) was
reached. This data is summarized in Figure 7.
1010
IRRM Reverse Current Density (A/mm2)
C. Experimental Procedure
The double-pulse tests were run at 25°C. The DUT and
the lower (switched) FET were the same device type, as
would be commonly employed in a real half-bridge
application circuit. The bus voltage was set to 80% of
datasheet-rated VBR(DSS) for each device. The gate drive
voltage was set to 15V, and the length of the first pulse was
adjusted to achieve the desired test current IF by the time the
reverse-recovery pulse was applied. Maximum diode current
was limited to ≤ISM. When the switching FET was turned-on
for pulse 2, the resulting reverse recovery current and
voltage of the DUT were captured on the oscilloscope. The
duration of pulse 2 was limited to 500 ns to minimize
damage to the test circuit in the event of failure (the
switching FET could still be turned-off safely even after the
DUT had failed). Also, in our experience, if a DUT was
going to fail, it would do so within <100 ns of the peak IRR,
in agreement with the findings of Blackburn [11].
A2
B2
A1
B1
A3
B3
A5
B4
Planar
Trench
1.0
1
0.1
0.1
A4
1.0
10
1
10
Applied Forward Current Density (A/mm2)
30
Fig. 7: Peak IRRM versus IS (normalized by area) for the devices tested
V DS
20V/div
Fig. 6a Typical reverse-recovery waveform from device A1
The x-axis of figure 7 shows the applied forward
current through the DUT body-diode, just prior to the
second pulse. The data are expressed as a current density by
dividing by the respective die area, resulting in amps/mm2.
The y-axis shows the magnitude of the resulting reverserecovery current (similarly expressed as a current density).
The orange “explosion” icons identify any data points where
device failure occurred.
A red dividing-line is placed on the graph to highlight
the clear separation of the data into two groups – the planar
devices are all above this line, and the trench devices are all
below the line. So, clearly planar FETs have higher reverserecovery current per unit area for a given forward currentdensity, compared to trench FETs. Because of this, it was
possible to drive all of the planar devices to a reverserecovery current-density sufficient to cause device failure,
while staying within the devices’ maximum current-rating.
Also, notice that the two different-sized Gen 5 devices (A1
and B1) failed at the same reverse-current density (3.98 and
4.03 A/mm2 respectively). Similarly, the two different-sized
Gen 7 devices (A2 and B2) failed at similar current densities
(5.54 and 5.26 A/mm2 respectively).
In contrast to the planar devices, none of the trench
devices could be driven hard enough to cause any failures
due to peak reverse-recovery current density. Even though
section II shows that trench devices have lower UIS
avalanche capability than their planar counterparts –
suggesting that they may also have a similarly lower
threshold current density for parasitic NPN secondbreakdown failure, the body-diode performance was so
good that the peak reverse current density we could achieve
was about 2.5 A/mm2. This was not sufficient to cause any
failures for any of the trench FETs tested.
V.
CONCLUSIONS
As MOSFET technologies evolve at break-neck speeds
with the need for matching parameters to specialized
applications, MOSFET ruggedness needs to be evaluated
similarly from an application-circuit perspective. In our
study using few different generations of planar and trench
gate technologies, we have demonstrated and reiterated that
Planar FETs have superior FBSOA and UIS avalanche
capabilities. However, we have also shown that reverse
recovery performance of the body diode of IR’s trench-gate
MOSFETs as characterized by the reverse-current densities
show superior performance than those of planar devices.
Our results also indicate that for a given technology there is
a reverse current density, which is the “failure” value and
exceeding this value results in destruction of the MOSFET.
When selecting MOSFETs for use in their application,
designers should be aware of the parasitics in their circuit,
the thermal environment of operation and the specific
strengths and weaknesses of the chosen MOSFET
technology from a ruggedness standpoint.
ACKNOWLEDGMENT
The authors would like to acknowledge the support of
Thanh Le, Al Diy, Eric Fernandez, Tim Henson, Bhaskar
Tetali, Aram Arzumanyan and Ling Ma for their assistance
with data collection and analysis. In addition, the authors
would like to thank Scott Bowman and Doug Carpenter from
Curtis Instruments for their assistance and insight with high
slew-rate di/dt testing.
REFERENCES
[1]
T. McDonald, M. Soldano, A. Murray, and T. Avram, “Power
MOSFET avalanche design guidelines”, Application Note AN1005,
International Rectifier.
[2] C. Blake, T. McDonald, D. Kinzer, J. Cao, A. Kwan and A.
Arzumanyan, “Evaluating the reliability of power MOSFETs”, Power
Electronics Technology, 2005, pp. 40-44.
[3] R. Stoltenburg, “Boundary of power-mos, unclamped inductiveswitching (UIS), avalanche-current capability”, IEEE Applied Power
Electronics Conference, 1989, pp. 359-364.
[4] I. Pawel, R. Siemieniec, M. Rösch, F. Hirler, and R. Herzer:
Simulating the Avalanche Behavior of Trench Power MOSFETs. In
Proc. ISPS, 233-238, Prague, 2006.
[5] Spirito, P.; Rinaldi, N.; Breglio, G.; d’Alessandro, V.; “Thermal
Instabilities in High Current Power MOS Devices: Experimental
Evidence, Electro-thermal Simulations and Analytical Modeling,”
MIEL 2002, 23rd International Conference on Microelectronics,
Volume 1, May 12-15, 2002; pp. 23-30.
[6] A. Kwan, K. Teasdale, N. Nguyen, J. Ambrus, and T. McDonald,
“Improved SOA analysis for trench MOSFETs using the Spirito
approach”, 9th Annual Automotive Electronics Reliability Workshop,
2004
[7] JEDEC Standard JESD24-10, “Test Method for Measurement of
Reverse Recovery Time trr for Power MOSFET Drain-Source
Diodes,” Electronic Industries Association, Engineering Department,
August, 1994
[8] K. Gauen, “New Test Circuit Enhances Understanding of Reverse
Recovery Stress in Power MOSFET D-S Diodes,” PCIM, September
1988, pp 36-49
[9] J. Witcher, “Methodology for Switching Characterization of Power
Devices and Modules,” Virginia Tech MSEE Thesis, January 29,
2002
[10] AN-937, www.irf.com
[11] D. L. Blackburn, “Turn-Off Failure of Power MOSFETs,” PESC ’85
Record, Proc. Power Electronics Specialists Conf. Rec., pp. 429-435,
June 1985 (Also reprinted in IEEE Transactions on Power
Electronics, Vol. PE-2, No. 2, pp. 136-142, April, 1987)
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement