Recent Advances in GaN Power HEMTs Related to Thermal

Recent Advances in GaN Power HEMTs Related to Thermal
Recent Advances in GaN Power HEMTs
Related to Thermal Problems
and Low-Cost Approaches
WW05
Kazuya Yamamoto1, Hiroshi Okazaki2,
Kenjiro Nishikawa3
1Mitsubishi
Electric Corp., 2NTT DoCoMo R&D Labs,
3Kagoshima University
Yamamoto.Kazuya@bk.MitsubishiElectric.co.jp
Slide 1
of 190
Thermal management of electronics: Measurement and the limits of GaN‐
on‐diamond electronics
Professor Martin Kuball
University of Bristol
Martin.Kuball@bristol.ac.uk
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 1
of 60
Outline





GaN electronics
Thermal management
challenges
Thermal materials and
device characterization
Ultra‐high power
electronics: GaN‐on‐
Diamond HEMTs
Conclusions
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 2
of 60
Microwave GaN Electronics
www.fujitsu.com
You can buy this already (GaN‐on‐SiC, GaN‐on‐Si) !!!
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 3
of 60
Thermal management
Electronic Warfare
Challenges: New material and device design; how to measure channel temperature of a device ?
Directed Energy Systems
Radar Systems
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 4
of 60
Performance and reliability
Impact ionization
Bulk trap generation
Interface trap generation
Killat et al., Compound Semiconductor
Jan/Feb 2013
Device degradation is temperature and electric field accelerated.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 5
of 60
Electric field & temperature
Source field plate
Passivation
Gate
Drain
AlGaN
Channel
GaN
Electric fields can be ‘shaped’ using e.g. field plates, T‐shaped gate, slanted gate i.e., electric field driven device degradation can be limited.
Nucleation layer
4H SiC
Temperature is the main factor at present limiting the reliability ie determining the maximum possible power density.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 6
of 60
GaN HEMT thermal history
Worked example 20W 10GHz solid‐state module
GaAs based
Cell size determined by operating frequency
module
PRESENTLY
4x reduction in size
5W
4.47W
0.5dB
manifold loss
GaN HEMT Vds=28V (5W/mm)
1 x 1mm transistor
Zopt=50ohm // 0.2pF
GaAs pHEMT
Vds=8V (0.9W/mm)
4 x 1.38mm transistor
Zopt=4.5ohm // 1.1pF
GaN
based module
GaAs (0.5 W/mK) thermal conductivity much lower than GaN‐on‐SiC (1.6; 4.5 W/mK).
Is this
the limit
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 7
of 60
Channel temperature
Device degradation determined buy temperature
Rate of device failure  exp(‐Ea/kBT), with Ea = activation energy
T = channel temperature (or temperature at specific location inside channel).
Gate metal diffusion
J. A. del Alamo et al., IEEE IEDM 2004
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 8
of 60
Why waste with experiment
I can do a thermal simulation … this saves me a lot of money and time
Heater width ?
Heat has to traverse interfaces
Thermal
conductivity
of material
and variation
through layer(s)
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 9
of 60
IR Thermography
Basic principle: Measures intensity of thermal IR radiation
Measured intensity:
with σ Stefan‐Boltzmann constant
Intensity
T2
T 2 > T
1
T1
10 m
Wavelength
1 m
Often 3‐5 m or 8‐10 m spectral window is used.
Fast, but diffraction‐limited spatial resolution of >3‐10 µm.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 10
of 60
IR Thermography
Limited lateral resolution:
IR: 3‐10 m
Typical ‘no’ depth resolution (for uncoated devices):
often 2‐5 m
IR measures a temperature average which is often not easy to define.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 11
of 60
Electrical Methods
Basic principle: Quantifies changes in IV curve with temperature rise,
e.g., a change in saturation current.
E.g. Kuzmik et al, IEEE Trans. Electron Dev.48 1496 (2002); McAllister et al., J. Vac. Sci. Techn. 24, 624 (2006); Simms, IEEE Trans. Electron Dev. 55, 478 (2008). Advantage: Uses electrical test equipment standard in most laboratories;
measures however average temperature over whole device.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 12
of 60
Raman thermography
Based on that vibrations of ‘atoms’ (phonons) of materials are temperature dependent
off
on
Spatial resolution  0.5‐0.7 µm. Temperature resolution < 2‐5 C.
Time resolution  10 ns.
Easy to use.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 13
of 60
The Raman method
Photons of a laser ‘hit’ device
Scattered light contains three different
wavelengths
Photon energy change
Light of same photon energy
as laser
Light of increased photon energy
(photon absorbs a vibration of ‘atoms’)
Light of reduced
photon energy
(photon created a vibration of ‘atoms’)
Raman probes typically this light
M. Kuball, Surf. Interface Anal. 31, 987 (2001).
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 14
of 60
A typical Raman spectrum
SiC
GaN
GaN
GaN
SiC
Temperature in ‘all’ different material layers in a device can be probed simultaneously:
 GaN, SiC temperature in GaN/SiC HEMT,  AlGaAs, GaAs temperature in GaAs pHEMT,
…
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 15
of 60
Raman vs IR thermography
160
120
Raman
IR
3D sim.
gate
Temperature [oC]
AlGaN/GaN HEMT on SiC
drain
source
Traditionally used IR underestimates temperature by 40%
80
M. Kuball et al., CSICS 2007
-10
-5
0
5
10
Position [m]
Spatial resolution: Raman 0.5‐0.7 μm; IR 7 μm.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 16
of 60
Importance of interfaces
180
100
80
60
40
20
0
GaN
SiC
Thermal Boundary Resistance (TBR)
GaN
0.1
SiC
1
10
100
1000
Depth (m)
CS-6
5
URI-4
4
30%
CS-2
CS-5
URI-2
3
20%
URI-3b
CS-1
1
URI-1
100
Manoi et al., IEEE Electron Dev. Lett 31, 1395 (2010).
CS-4
URI-3a
2
CS-3
150
Measurement
accuracy
200
10%
5%
Extra channel temperature rise
o
Temperature rise ( C)
120
TBR x10-8 (m2K/W)
×

140
US, Europe, Japan
6
Thermal resistance of interface
160
250
Interface temperature (oC)
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 17
of 60
How to obtain time resolution
ON
Pulsed laser source
OFF
Best ever reported IR time resolution
Time resolution: 10 ns.
Spatial resolution: 0.5 μm.
Order of magnitude faster and more accurate
than other thermography techniques.
Riedel et al., IEEE Electron Dev. Lett. 29, 416 (2008); Kuball et al. IEEE Electron Dev. Lett. 28, 86 – 89 (2007).
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 18
of 60
Time‐resolved Raman
S
100 m
Ability to trace temperature with 10ns time resolution
D
S
G
25 µm
D S
Temperature splits
Identical increase
Time‐dependent
thermal cross‐talk
Manoi et al, Solid State Elecronics 57, 14 (2011). Slide 19
of 60
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Can one improve resolution ?
(a)
Source
Gate
(b)
Drain
Source
Gate
Solid Immersion Lens (SIL)
Drain
GaN
θ
SiC n=2.6
R
SIL
air
(c)
Axial:
6
widefield
confocal
Designed
SIL
4
0.4
SiC SIL limit
0.2
2
0
0.0
1
N.A.
drain
source
o
Axial resolution [m]
8
heat
0.6
Temperature [ C]
Lateral:
widefield
confocal
Lateral resolution [m]
10
180
160
gate
Upper GaN (Exp.)
Lower GaN (Exp.)
Average GaN (Sim.)
2
2
3
Lateral position [m]
J.W. Pomeroy et al., J. Appl. Phys. 118, 144501 (2015)
4
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 20
of 60
DC vs RF temperatures
Junction temperature
256°C 225°C
1E+8
175°C
1E+7
MTTF [hrs]
1E+6
RF test
1E+5
Ea=1.92eV
Ea=1.82eV
1E+4
1E+3
DC test
1E+2
1E+1
18
19
20
21
22
23
24
25
26
27
28
1/KT [1/eV]
Activation energy determined in DC and RF lifetime test similar.
J.W. Pomeroy, Microelectronics Reliability 55, 2505 (2015).
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 21
of 60
However, for >100V
5.9 W/mm average
100V
J.W. Pomeroy et al. ROCS 2015.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 22
of 60
How to do it in real life ?
top
Measure
Raman
from
bottom
Can be performed on‐wafer or in‐package. Only condition is that
the semiconductor of the device is
optically visible.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 23
of 60
Temperature measured
For optically transparant materials – Average of temperature in small volume
Gate
2 m
Ga
0.75 m
N
C
Si
0.75 m
For optically non‐ transparant materials – Small ‘surface’ area
Temperature average over 0.75 m x 0.75 m x 50nm (for GaAs)
It is well defined over which area an average of temperature is measured and should be compared a subsequent thermal simulation !!!
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 24
of 60
Simulation to aid experiment
Pomeroy et al, IMS 2012
If there is a T‐gate or field plate, we consider this by using thermal simulation, as those ‘screen’ the hot spot.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 25
of 60
Methodology developed …
180
Laser 160
light
Step 2 – We fit in thermal simulation any resistances at internal chip interfaces.
Step 1 – We fit in thermal simulation the gradient in temperature through SiC
to extract SiC thermal conductivity.
Ungated device
Ohmic
140
contact
120
o
Temperature rise ( C)
Ohmic
contact
AlGaN
GaN
100
Depth
scan
80
60
40
20
Extract key material parameters
0
GaN
0.1
SiC
1
10 source 100
gate
1000
drain
Depth (m)
HEMT
GaN
SiC
depth
Step 3 – We measure temperature in active device region and determine Edge of Chip (EoC) temperature (temperature 500 m away from device).
Step 4 – We fit thermal simulation to HEMT and EoC temperature; EoC
temperature considers die attach.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 26
of 60
Thermal simulation
2D drift‐diffusion model
3D finite element thermal model
AlGaN
e.g., 20VDS, ‐3VGS
GaN
3.75µm
drain
source
20nm
Gate Gate F.P.
edge F.P.
edge
Joule heating map
Calibrated 3D thermal model including die and package
This approach combines the advantage of accurate Pdiss profile (drift‐diffusion) with 3D finite element, e.g. large models
J. W. Pomeroy et al., Microelectron. Reliab., (55)12, 2505 (2015).
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 27
of 60
Good repeatability
≤ ±5°C
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 28
of 60
GaN‐on‐Diamond HEMT
Qorvo & Bristol, CSICS 2013;
Pomeroy et al, Appl. Phys. Lett.
104, 083513 (2014).
A 3× power density increase was achieved by the DARPA NJTT program
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 29
of 60
Thermal conductivities
Single crystalline diamond
Enriched Diamond
Natural diamond
J.R. Olson, Phys. Rev. Lett. 70, 14 (1993).
SiC thermal conductivity: 4.8 W/cmK
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 30
of 60
Single‐crystalline diamond
Not a realistic option to use for semiconductor technology
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 31
of 60
Poly‐crystalline diamond
Grain size
Defects
phonon‐phonon
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 32
of 60
The role of interfaces
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 33
of 60
Importance of interfaces
Source
Drain
Gate
GaN
interface
Diamond
¼
model
Raman
probe
1 µm
Ungated
Transistor
2 finger transistor
Polycrystalline diamond properties as well as interfaces need to be optimized.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 34
of 60
Diamond thermal conductivity
GaN
contact
map
mesa
Diamond interface
contact
Polycrystalline
Diamond
F.E. model of ungated HEMT
¼ cross section
Fit finite element model by
adjusting two parameters:
Diamond thermal + GaN/diamond conductivity
interface TBReff
Pomeroy et al, CSICS 2014.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost 35
Approaches
Slide 35
of 60
Diamond thermal properties
TBReff: Effective thermal boundary resistance
Effective thermal conductivity:
Weighted average, influenced by grain size.
Substrate T.C.
[W/mK]
GaN
Heat flux
160 W/mK
thin dielectric Increasing thermal
conductivity along
growth direction
100 µm polycrystalline diamond
TBReff
‐8
×10 [m2K/W]
GaN‐on‐SiC
420
2‐5 (~2.5 typical)
GaN‐on‐di
1200 (effective)
2.7±0.3
O.W. Kiiding et al.Diamond Relt.Mater., 3 (1994) 1178
REMINDER:
Bulk diamond:
2000‐3000 W/mK
(Pomeroy et al., CSICS 2013) WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 36
of 60
Near nucleation diamond
Thermal conductivity impacted
by grain size
Grain evolution can be
controlled by manipulating
the chemistry of the
diamond growth.
J.Anaya et al. Acta Materiala 103, 141 (2016); Appl.
Phys. Lett. 106, 223101 (2015).
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 37
of 60
Near nucleation diamond
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 38
of 60
Diamond beyond nucleation Z
3‐heaters
X
Thermal conductivity. determined by 3‐omega technique.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 39
of 60
Diamond beyond nucleation R.Baranyai et al. APEX 9, 061302 (2016)
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 40
of 60
Modeling of diamond properties
(i) thermal resistance between grains, (ii) shortening in the phonon mean free path due the reduced size of the grains
S. Bhattacharyya et al. Appl. Phys. Lett. 79, 1441 (2001)
Grain Boundary thickness
Callaway
, ,
, ,
,
1
J.Anaya et al., Acta Materiala
103, 141 (2016).
, ,
Kapitza‐like
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 41
of 60
GaN‐diamond interface
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 42
of 60
Transient Thermoreflectance
Wafer mapping & fast wafer screening
Varied thickness
28 nm to 100 nm
Presently we develop this into a commercial equipment.
Sun et al. IEEE Electron Dev. Lett. (accepted) 2016; Appl. Phys. Lett. 106, 111906 (2015); Pomeroy et al. IEEE Electron Dev. Lett. 35, 1007 (2014).
Slide 43
of 60
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Validation of the technique
Normalized signal
Exp.
Sim. surface temp.
Sim. total refl.
0
200
400
600
800
t (ns)
Measured transient is consistent on wafers
w/ and w/o gold transducer
Precautions have been taken to ensure that the measured signal represents the
surface temperature transient.
• Different UV powers result in identical transients.
H. Sun et al, CSMantech 2015
• Thermo-optic simulation further supports data.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 44
of 60
GaN‐diamond interface
Dielectric seeding layer needs to be optimized.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 45
of 60
GaN‐diamond interface
2000
sub (W/m-K)
1500
H. Sun et al, CSMantech 2015
200 °C
MW
diamond
100
1000
50
•
HF diamond
500
60
Transistor peak
temperature rise
SiC
•
50
40
30
20
10
When substrate thermal
conductivity is low, TBReff
is not the major factor.
However, TBReff limits heat
removal for high thermal
conductivity substrates.
2
TBReff (m K/GW)
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 46
of 60
Wafer screening
Rare imperfections can be screened out
Typically good homogeneity Fast wafer‐mapping of the GaN‐on‐
Diamond thermal resistance
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 47
of 60
Thermoreflectance for devices
D
Materials:
Field Plate
Passivation S
G
•Two metallization levels:
1.Drain contact ( )
2.Source field plate ( )
GaN
•GaN ( )
SiC or Diamond
D
45 μm
6 μm
6 μm
G
Field plate
S
S Martin Horcajo et al., CS Mantech 2016
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 48
of 60
GaN‐diamond HEMT design
Decreasing thermal resistance associated with the carrier enables a further Increase in power density to ~3X
J. W. Pomeroy et al, CSICS 2014
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 49
of 60
GaN layer optimization
Source
TBReff
d
Drain
Gate
1 µm
Low TBR
A 1µm‐thick GaN buffer is optimal for the range TBReff values expected
d [m]
1.0
Measured TBR
Peak Temperature rise [Norm.]
Diamond
1.2
e.g. 4 Finger HEMT
Vary GaN buffer thickness (d) and TBReff
Heat spreading
GaN
0.8
Thin GaN
0.75
1
1.5
2
2.5
3
0.6
0
1
2
-8
3
J. W. Pomeroy et al, CSICS 2014
4
2
TBR x10 [Wm /K]
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 50
of 60
Raman experimental validation
260
GaN‐on‐SiC
240
GaN
Diamond
S G
D
GaN
220
OR decreasing interface thermal resistance
We validate all thermal simulations
with Raman thermography measurements.
180
diamond
W/mK
 710
1400
710
o
Current GaN‐on‐diamond
Increasing diamond thermal conductivity
Temperature rise [ C]
200
Measurement:
GaN
Diamond
Simulation:
diamond
Rinterface
160
o
2
Cm /W
-8
2.7x10
-8
2.7x10
0
140
120
100
80
60
40
20
0
0.1
1
Depth [m]
10
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
100
Slide 51
of 60
GaN‐diamond stability
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 52
of 60
Coefficient of thermal exansion
Diamond
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 53
of 60
Stress in GaN layer
Diamond
Determined using Raman spectroscopy
Year
2011
With nitride
transition layers
0.8
2012
2013
Without nitride
transition layers
565.0
566.0
0.2
0.0
-0.2
566.5
567.0
Stress free
Hot filament
Wafer
Microwave
E2 peak (cm-1)
0.4
565.5
Tensile
Stress (GPa)
0.6
567.5
Sun et al, CS Mantech 2016
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 54
of 60
How to test for stability ?
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 55
of 60
High mechanical stability
GaN
Cantilever
to appy
force
diamond
We try to cause fracture at the GaN-diamond interface
D. Liu et al, Appl. Phys. Lett., 107, 251902 (2015).
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 56
of 60
Estimation of interface strength
GaN layer fractures as >3 GPa; GaN/diamond interface
fracture strength is much greater.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 57
of 60
Thermomechanical stability
Stress in GaN‐on‐
diamond induced by heating (CTE mismatch)
Fracture at
> 3GPa
Good thermo-mechanical stability
in the areas studied ie no change in
the TBR after annealing.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 58
of 60
Conclusions
GaN electronics main challenge at present is its heat sinking.
Raman thermography offers the opportunity to quantify channel temperature, and to identify and optimize thermal bottlenecks such as interfaces. Raman thermography enables 0.5µm spatial and 10ns time resolution thermal imaging in 3D, which can be improved further using solid immersion lenses (SILs).
GaN‐on‐Diamond HEMTs enable 3x improvement in power density, however, require optimization in diamond thermal properties near the interface and of the interface itself.
Transient thermoreflectance can be used for wafer mapping of thermal interfaces (before device fabrication) and for device thermal analysis.
High mechanical & thermo‐mechanical stability of GaN‐diamond interface
was demonstrated.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 59
of 60
Acknowledgment
Research Professor & Research Fellows
Prof. Michael J. Uren
Dr. James W. Pomeroy
Dr. Dong Liu
Postdoctoral Researchers
Dr. Julian Anaya
Dr. Roland Baranyai
Dr. Tommaso Brazzini
Dr. Indranil Chatterjee
Dr. Sara Horcajo
Dr. Huarui Sun
Dr. Serge Karbojan
PhD students:
Peter Butler
Callum Middleton
Bahar Oner
Alexander Pooth
Maire Power
Ben Rakauskas
Will Waller
Yan Zhou
and others
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 60
of 60
•
•
•
•
—
—
—
—
—
•
•
•
•
•
•
•
5
•
•
—
—
•
•
•
•
Æ
—
ƒ
ƒ
—
—
—
•
•
•
Æ
Æ
Æ
•
•
–
–
–
–
•
°
•
•
•
–
–
•
•
°
•
•
°
•
—
•
•
•
•
°
•
—
—
•
—
—
•
–
•
–
•
•
•
—
—
Æ
•
•
•
•
—
—
•
—
•
9
9
9
9
ITEM
UNITS
Bulk Thermal Conductivity W/m-K
CTE
ppm/⁰C
Modulus @ RT
Gpa
AuSn
57
16
59
Ag Epoxy
23
38
3.0
Mfr A
50
26
6.7
Sintered Ag
Mfr B
100
28
10
Mfr C
180
22
26
•
—
•
•
•
—
—
•
•
•
•
•
•
—
—
—
—
—
—
•
•
•
•
•
•
Pout(dBm)
Drain Efficiency (%)
20
16
Gain, Return Loss (dB)
12
8
4
0
-4
-8
-12
S11
S21
S22
-16
-20
1.0
1.1
1.2
1.3
1.4
1.5
1.6
61.0
80
60.5
70
60.0
6
60
59.5
50
59.0
40
58.5
30
58.0
Output Power (dBm)
20
57.5
Drain Efficiency (%)
10
57.0
1.10
1.15
1.20
1.25
1.30
1.35
1.40
1.45
Drain Efficiency (%)
Frequency (GHz)
0
1.50
Ω
20
10
5
0
-5
-10
S(2,1)
S(1,1)
S(2,2)
-15
-20
2.5
2.7
2.9
3.1
3.3
3.5
Frequency (GHz)
800
100
700
90
600
80
500
70
400
60
300
50
Drain Efficiency (%)
2.3
Output Power (W)
Gain, Return Loss (dB)
15
200
40
Output Power
100
30
Drain Efficiency
0
20
2.6
2.7
2.8
2.9
Frequency (GHz)
3.0
3.1
3.2
20
10
5
0
-5
-10
S11
S21
-15
S22
-20
4000 4200 4400 4600 4800 5000 5200 5400 5600 5800 6000 6200 6400 6600 6800 7000
51.0
100
50.5
90
50.0
80
49.5
70
49.0
60
48.5
50
48.0
Output Power (dBm)
40
47.5
Drain Efficiency (%)
30
47.0
20
5.2
5.3
5.4
5.5
5.6
Frequency (GHz)
5.7
5.8
5.9
Drain Efficiency (%)
Frequency (MHz)
Output Power (dBm)
Gain, Return Loss (dB)
15
16
12
Gain, Return Loss (dB)
8
4
0
-4
-8
-12
S21
S11
S22
-16
-20
8.2
8.4
8.6
8.8
9.0
9.2
9.4
9.6
9.8
10.0
10.2
10.4
10.6
Output Power (dBm), Drain Efficiency (%)
Frequency (GHz)
55
50
45
40
Drain Efficiency
35
Output Power
30
8.8
8.9
9.0
9.1
9.2
9.3
9.4
9.5
9.6
9.7
Frequency (GHz)
20
10
5
0
-5
-10
S21
S11
S22
-15
-20
3.0
3.1
3.2
3.3
3.4
3.5
3.6
3.7
Output Power and Efficiency vs. Frequency
Frequency (GHz)
100
100
90
90
80
80
70
70
60
60
50
50
40
20
3.05
40
Output Power (W)
30
30
Efficiency (%)
3.10
3.15
3.20
3.25 3.30 3.35
Frequency (GHz)
3.40
3.45
3.50
20
3.55
Drain Efficiency (%)
2.9
Output Power (W)
Gain and Return Loss (dB)
15
25
20
S21
10
S11
5
S22
0
-5
-10
-15
-20
1.85
1.9
1.95
2
2.05
2.1
2.15
2.2
2.25
2.3
2.35
2.4
Frequency (GHz)
57.0
75.0
56.5
70.0
56.0
65.0
55.5
60.0
55.0
55.0
Psat (dBm)
54.5
54.0
2.10
50.0
Efficiency (%)
2.11
2.12
2.13
2.14
2.15
Frequency (GHz)
2.16
2.17
45.0
2.18
Efficiency (%)
1.8
Output Power (dBm)
Gain, Return Loss (dB)
15
•
•
•
•
•
•
Cost effective GaN HEMT developments with appropriate thermal transfer
Kazutaka Inoue
Sumitomo Electric Industries, Ltd.
inoue‐kazutaka@sei.co.jp
Wednesday, October, 5th, 2016
Slide 1
of 44
Outline
1. Fundamentals 2. Thermal Study of Substrate
3. Thermal Design (GaN for Radar)
4. Thermal Design (GaN for Base Station)
5. Summary
Slide 2
of 44
History of Sumitomo GaN HEMTs
ー2000
2003
2004
2005
Fujitsu,
Fujitsu Laboratories
Development Start
2006
2007
2008
Eudyna Devices
2009
2010 -
Sumitomo Electric
Device Innovations
Engineering Sample Release
Mass-production

#1 Volume leader
in RF high power GaN

We





strive for
Performance
Quality
Reliability
Cost
Capacity
2013 GaN Market Share
above 5W below 4GHz Source: ABI Research 2014
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 3
of 44
Microwave Device Products
【Mobile Base Station】
【Satellite】
【Automotive Radar】
GaAs MMIC
GaN HEMT
GaAs FET
【Radio Link】
GaAs MMIC
【Radar】
Gulf Control
Air Traffic Control
Marine
GaN HEMT for Radar
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 4
of 44
Why GaN ?
Johnson's Figure of Merit
Johnson’s FoM = ft x BV
= vs /2π x BF
BV = BF × Lg
Material
Si
GaAs
SiC
GaN
Bang Gap Energy
(eV)
1.1
1.4
3.2
3.4
Critical Breakdown
Field (MV/cm)
0.3
0.4
3.0
3.0
1.5
0.5
4.9
1.5
600
1500
2.0
2.7
Thermal Conductance
(W/cm/K)
Mobility
(cm2/V/s)
Saturated Velocity
(*107 cm/s)
1300 6000
1.0
1.3
GaN
GaAs,
InP
Si, SiGe
fT = vs / ( 2πLg )
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 5
of 44
Cost Effective Design, Focused on Thermal Transfer
 Chip cost is determined by ...
‐ Wafer process cost and yield
‐ Maximum channel temperature design
 Maximum channel temperature is determined by...
‐ Reliability of the device technology
‐ Thermal conductance of material (on‐SiC, on‐Si)
‐ Chip pattern layout
‐ Operating condition
(Thermal dissipation ↔ Efficiency)
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 6
of 44
Yield of GaN HEMT Wafer Process
Psat= 52.2 dBm  = 0.15 Gp= 16.6 dB  = 0.15 Rth = 1.2 C/W  = 0.02
BTS GaN HEMT 2.7GHz 160W, n = 9000 pcs, 181 lots,  = lot average
GaN HEMT wafer process has been refined, through over 10‐years mass production.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 7
of 44
Reliability of GaN HEMT
DC‐HTOL test
RF‐HTOL test
(Vds=60V , Tch=250, 275, 300, 315degC)
(Vds=55V, Tch=270, 290, 310degC, P4dB)
K.Osawa et.al, ” Over 74% Efficiency, L-Band 200 W GaN-HEMT for Space Applications ,” EuMC2016.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 8
of 44
Material vs. Thermal Design
 Parameters, related to thermal design ;
‐ Thermal conductance of substrate material (SiC:4.9 Si:1.5 [W/cm・K] ) ‐ Substrate thickness (Typical thickness is ∼100 μm )
‐ Gate to gate pitch (Wide pitch improves the degree of heat spreading)
Gate
Drain
Source
AlGaN
GaN
Gate
Drain
HEMT Layers∼1μm
Substrate Thickness
∼100 μm
Heat Spreading
Slide 9
of 44
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Thermal Design vs. Efficiency
Thermal Dissipation = Pdc ‐ Pout + Pin
= Pdc × ( 1− Power‐Added‐Efficiency )
DC supply power
(Pdc)
RF input power (Pin)
Power Amp.
Thermal dissipation
RF output power
(Pout)
Slide 10
of 44
Outline
1. Fundamentals 2. Thermal Study of Substrate
3. Thermal Design (GaN for Radar)
4. Thermal Design (GaN for Base Station)
5. Summary
Slide 11
of 44
Case Study of GaN on‐Si ∼RF Performance
S
G
RF loss(low gain)
D
Large parasitic C
AlGaN
2DEG
Low resistivity region
Silicon
Diffusion of the Ga & Al
into the silicon substrate
GaN HEMT on‐Si realize comparable performance to GaN on SiC. I. Makabe et al.,“ Improvement of RF performance of GaN‐HEMT on silicon substrate,”Proc. International Workshop on Nitride Semiconductors, No. TuEP12, Wrocław, Poland(Aug. 2014)
Lg=0.6μm, f=2GHz Vds=50V, Idq=0A WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 12
of 44
Case Study of GaN on‐Si
∼Thermal Design (Sub. Thickness)
1. Simulated Thermal resistance is not comparable, even in 30 μm thickness. ・・・・・
2. Excessive thinning causes chip warping @ GaAs MESFET
Sub. Thickness : 20 μm
Sub. Thickness : 30 μm
GaN HEMT on‐Si or on‐SiC
(Wg=36 mm) AuSn
Chip (flat)
AuSn
Cu‐based Package
Chip (warped)
Package Base
Package Base
Net improvement is limited by chip warping !
Small deviation
Large deviation
Slide 13
of 44
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Case Study of GaN on‐Si
∼Thermal Design (Gate Pitch)
Thermal Conductance SiC:4.9 Si:1.5 [W/cm・K] Chip (on‐SiC)
Thermal Resistance
Package Base
Chip (SiC)
Chip (on‐Si)
AuSn
Package Base
Chip (Si)
AuSn
+ PKG
on‐SiC
AuSn
+ PKG
on‐Si
 Thermally equivalent pitch of GaN on‐Si is 1.6 times of GaN on‐SiC.  SiC substrate cost used to be several times higher to the wafer process cost. In such cost structure, GaN on‐Si was a reasonable solution.
 It should be noted the 1.6 times chip size increases “net wafer process cost” of GaN on Si
 The power density of GaN HEMT has been increasing. It used to be ∼5 W/mm, but now reaches to ∼10 W/mm. Higher power density favors better thermal conductance material. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 14
of 44
Case Study of GaN on‐Si
∼Thermal Design (Chip Stretch)
Two options to realize thermally equivalent GaN on‐Si
Option.1 (Lateral Stretch)
Option.2 (Vertical Stretch)
on‐SiC
300μm
on‐Si
‐ Larger PKG is required for on‐Si chip.
‐ Larger unit finger increases gate
resistance, which degrades the gain by more than 1dB.
Gain
Effect
480μm
PKG for on‐SiC
≧1dB
PKG for on‐Si ?
300μm
Judgement
480μm
Unit
Length
Total cost strongly depends on PKG price. Trade‐off issue arises (vs.gain).
GaN on‐SiC is the best solution at present, judging from both cost and RF property.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 15
of 44
In addition...
∼SiC Quality Improvement
10 years ago (Surface Defects )
 SiC quality was one of the issues to be solved.
 The current SiC has realized smooth surface, and it contributes to the drastic improvement of the GaN on SiC cost structure. Current SiC (Drastically Improved )
Abnormal
epitaxial growth
Defect on substrate
GaN
SiC
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 16
of 44
Prospect of Substrate Material
Power density [W/mm]
 Power density of GaN HEMT have been increased, and will be continued.  The benchmark of GaN on Si vs. SiC indicates the availability of GaN on diamond. ?
2005
Higher power density favors better thermal conductance material.
2016
on‐Si
‐ RF‐loss elimination
required
‐ Thermal limitation
‐ No more cost effective
on‐SiC
‐ Expensive & many defects
‐ Chip selection needed
‐ Most promising
2025
‐‐
‐ Will be most promising
‐ Thermal design may be critical @ >20W/mm
‐ Competitive @>20W/mm
‐ Quality & cost required
Slide 17
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
of 44
on‐diamond ‐‐
‐ Similar to early stage SiC ? Outline
1. Fundamentals 2. Thermal Study of Substrate
3. Thermal Design (GaN for Radar)
4. Thermal Design (GaN for Base Station)
5. Summary
Slide 18
of 44
Output Power vs. Frequency
Pout
Satellite
CW operation
1KW
Case.2 BTS
Backed-off operation
100W
Case.1 Radar
Pulsed operation
Si
10W
1W
GaN
Si/GaAs
1GHz
GaAs/GaN
GaAs
10GHz
Frequency
100GHz
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 19
of 44
Satellite Application
(CW‐operation, as a reference)
K.Osawa et.al, ” Over 74% Efficiency, L-Band 200 W
GaN-HEMT for Space Applications ,” EuMC2016.
Gate width
Chip size
: 24 mm × 2 chip
: 6.0 ×0.86 mm (5.16mm2)
Design point
of 100W GaN HEMT die

Designed for CW and saturated operation . (Most severe thermal requirement)
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 20
of 44
Case.1 Rader (Pulsed operation)
Rader
System
Object
CW operation
Channel Temp.
Pulsed operation
ON
Device
OFF
Tch image
 Thermal design of pulsed operation chip differs from CW one, and the estimation of the transitional Tch is important. The following slides explain the detail. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 21
of 44
Channel Temperature Simulation using Transient Thermal Resistance
RF/Vg ON

OFF
T‐
(Rth Approximation)
Rth()
dTch peak1
-(Rth Approximation)
 During the pulse‐ON, channel temperature rises due to the heat generation.  At the pulse‐OFF, the channel temperature falls down by heat dissipation, which is expressed as reverse behavior of heat generation.  With combining these 2 lines, the channel temperature falls.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 22
of 44
Channel Temperature Simulation using Transient Thermal Resistance
RF/Vg ON
Rth(T+)-Rth(T)
OFF
T‐

Rth(2T+)-Rth(2T)
(Rth Approximation)
Rth()
dTch peak1
dTch peak 1 = Pd*Rth()
-(Rth Approximation)
dTch peak3
dTch peak2
dTch peak 2 = Pd*{Rth(T+)-Rth(T)+Rth()}
= Pd* {Rth(T+)-Rth(T)}+dTch peak 1
dTch peak 3 = Pd*{Rth(2T+)-Rth(2T)+Rth(T+)-Rth(T)+Rth()}
= Pd*{Rth(2T+)-Rth(2T)}+dTch peak 2
 The peak Tch at the 1st pulse is calculated as shown in above.
 The 2nd peak Tch is also calculated, combining these 3 values.
 The total Tch rise is summation of the delta Tch at the each on‐state condition.
Slide 23
of 44
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Channel Temperature Measurement & Analysis
150
150
Shutter Speed 1msec
140
140
= Averaging Time
130
120
120
T ch [degC]
T ch [d e g C ]
Simulated Tch
130
Simulated IR response
110
100
90
110
100
80
80
70
70
60
60
50
50
-4
-2
0
2
Time [msec]
4
6
Simulated IR response
90
Measured IR Tch
-4
-2
0
2
4
6
Time [msec]
 The analysis performed more than 10 years ago. We only had an IR system with rather slow shutter speed. But we found the measured Tch from IR agreed with the simulated Tch curve, by taking the shutter speed into consideration.  Thus, we concluded that the mentioned simulation in previous slide is reliable. And the compact kW‐class GaN device was developed by utilizing this analysis. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 24
of 44
kW‐Class GaN HEMT Pallet Amplifier for Radar VGG
VDD
RF out
RF in
VDD
VGG
Size : 58.5 mm X 40.0 mm X 8.0 mm • Input/Output matched to 50 ohm
• Includes RC Bias Circuit
• Cu base
E.Mitani et.al, ” A kW-class
AlGaN/GaN HEMT pallet amplifier
for S-band high power application ,”
EuMIC2007.
Slide 25
of 44
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
kW‐Class GaN for Radar RF Performance
IDD(DC)=2.0A, Pulse Width 200sec, Duty 10%
59.4dBm
60.0dBm
VDD=80 V
60
90
f=2.9 GHz
70
55.0%
55
60
50
40
30
50
13.8dB
20
80
Output Power [dBm]
80
Drain Efficiency [%]
Gain [dB]
Output Power [dBm]
90
70
f=3.2 GHz
55
60
49.5%
50
40
30
50
14.1dB
20
10
0
45
35
40
45
Input Power [dBm]
Output Power
Drain Efficiency
50
Gain
Drain Efficiency [%]
Gain [dB]
VDD=65 V
60
10
0
45
35
40
45
Input Power [dBm]
Output Power
Drain Efficiency
50
Gain
 Compact 1 kW class GaN HEMT for radar application has successfully demonstrated, by optimizing for pulsed operation. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 26
of 44
Up‐to‐date Achievement
X‐band 300 W GaN HEMT for Rader
K.Kikuchi et.al, ” An 8.5–10.0 GHz 310 W GaN HEMT for radar
applications”, 2014 IEEE MTT-S Int. Microwave Symposium
Digest, 2014.
:
:
:
:
55.2 dBm
14.5
55
Pout
14.0
GL
45
14.1 dB
13.0
40
35
13.5
42%
PAE
12.5
30
55.2 dBm
(333 W)
56
25
20
54
52
Gain
15
10
50
Pout
48
10.2 dB
Gain (dB)
50
58
15.0
GL (dB)
Pout (dBm), PAE (%)
60
24.0 mm × 17.4 mm
14.4 mm × 2chip (>10 watt/mm)
5.4 mm ×0.7 mm
Lg=0.4μm, Via-Hole
Output Power (dBm)
PKG size
Gate width
Chip size
Chip outline
5
0
46
12.0
32 34 36 38 40 42 44 46
35 40 45 50 55 60 65 70
Input Power (dBm)
Vds (V)
Pulse Width = 100 sec, Duty = 10%, Frequency = 9.0 GHz
Slide 27
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
of 44
30
Up‐to‐date Achievement
X‐band 300 W GaN HEMT for Radar
400
Vds = 65 V, Idq = 0.80 A
Pulse Width = 100 sec, Duty = 10%
350
Output Power (W)
This work
(Maximum)
Frequency
8.5 – 10.0
(GHz)
(9.0)
Output Power
310
(W)
(333)
Power Gain
10.0
(dB)
(10.2)
CW
300
Pulse
250
This work
333 W
200
150
100
50
0
7
8
9
10
11
Frequency (GHz)
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
12
Slide 28
of 44
Prospect of Cost Effective GaN for Rader applications  Using GaN to replace the klystron ‐ High reliability, Maintenance free ‐ Compact, Easy operation ‐> Reduction of OPEX
 GaN output power is not as large as klystron, but it allows long pulses. Pulse compression can be used.  Cost effective GaN HEMT will penetrate into the radar applications.
Filter
Linear FM Pulse
Frequency Dependent
Delay Filter
http://www.river.go.jp/xbandradar/
Compressed Pulse
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 29
of 44
Outline
1. Fundamentals 2. Thermal Study of Substrate
3. Thermal Design (GaN for Radar)
4. Thermal Design (GaN for Base Station)
5. Summary
Slide 30
of 44
Case.2 BTS Application
(Backed‐off Operation)
Drain Efficiency
Power Probability
BTS trend → Smaller → Higher efficiency
W‐CDMA
WiMAX
LTE, LTE‐adv
5G
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 year
Normalized Output Power Both efficiency improvements of device itself & efficiency enhancement circuit techniques are required, at backed‐off power region.
Slide 31
of 44
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Operation Class (F, inverse‐F)
D
Id
G
Inverse Class-F
Class-F
Vd
Ids
Ids
Vds
Vds
Id, Vd
Id, Vd
S
0
90
180
270
360
ωt
Load Impedance for Harmonics
Even : 0 (Short)
Odd : ∞ (Open)
Peak Value of Voltage, Current
Voltage: 2x Vds(DC)
Current: 3.14x Average Ids
0
90
180
270
360
ωt
Even : ∞ (Open)
Odd : 0 (Short)
Voltage: 3.14x Vds(DC)
Current: 2x Average Ids
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 32
of 44
Combination of Inv. Class‐F & GaN HEMT Property
Ids
On-R
1.6
1.4
Vgs=+2V
1.2
Ids [A]
Inverse-F
-F
1.0
0.8
Vgs=+0V
0.6
0.4
Rds
Vgs=-1V
0.2
0
0
50
Vds
100
150
200
Vds [V]
 Inverse Class‐F requirement are, ‐ Low Ron (Steep Knee) reduces RF‐loss
‐ High breakdown voltage contributes to keep good pinch off
 GaN HEMT is the best technology for these requirements at present. WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 33
of 44
Circuit of Inverse Class‐F 100W GaN HEMT
Design Concept
Circuit : Harmonic tune by simple(=low cost) LC circuit (LPF)
GaN chip : Size minimization by referring thermal loss Capacitor (r=150)
Input
100W
GaN Chip
Output
Product: EGN21C105I2D
Vds=50V, f=2.14GHz
100
50
90
48
80
46
70
44
60
42
50
40
40
38
30
36
20
Drain Efficiency [%]
52
Pout [dBm]
Inductor
• Bonding Wire
• Transmission Line
on Alumina sub.
18 20 22 24 26 28 30 32 34 36 38
Pin [dBm]
Slide 34
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
of 44
Id
Theory of Doherty Amplifier
(a) Backed‐off
Zopt
Peak Amp. OFF
/4
Power Matching
Open
Class-C
Main Amp.
/4
Zopt/2
2xZopt
Vd
Class-B or AB
2xZopt
Efficiency Matching
(b) Peak Power
Peak Amp. ON
/4
Class-C
Main Amp.
Class-B or AB
Zopt
/4
depend on both main and peak amp's eff.
depend on
main amp's eff.
Zopt/2
B
A
Efficiency
Zopt
b
a
Zopt
6dB BO
Psat
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Pout
Slide 35
of 44
GaN HEMT for Doherty PA
a
b
gm
Rds
Rs
c
Drain
Ids
Rd
Cds
a) Ron
DC IV-Characteristics
RF Signal Loss
b) Rds
Source
c) Cds
Vds
Key of High Frequency Performance
70W-class Device for 2.6GHz Base Stations
70W-class Device
Vds
Imax
Wg
Cds
Si LDMOS-FET
28V
26A
120mm
31pF
GaN HEMT (This work)
50V
11A
18mm
3.8pF
H. Deguchi et.al, "A 33W GaN HEMT Doherty Amplifier with 55% Drain Efficiency for 2.6GHz Base Stations”,
2009 IEEE MTT-S Int. Microwave Symposium Digest, pp.1273-1276, 2009.
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 36
of 44
RF Performance of Doherty PA
Pout
Pout(Pulse)
Drain Efficiency
55
100
50
45
80
40
70
35
Effi.
30
60
50
25
40
20
30
15
20
Gain
10
Cds-reduced
Device
90
Pout
Drain Efficiency [%]
D rain E fficiency [% ]
Output P ower [dB m ], Gain [dB ]
Vds = 50V, Idq-m = 200mA Idq-p : Class-C Bias
f = 2.6GHz, CW Pulse Duty = 10% (6s/60s)
Gain
Gain(Pulse)
Conventional
Device
Freq.=2.57GHz, Vds=50V
10
5
0
20 22 24 26 28 30 32 34 36 38 40 42
Input Power [dBm]
H. Deguchi et.al, "A 33W GaN HEMT Doherty Amplifier with 55% Drain Efficiency for 2.6GHz Base Stations”,
2009 IEEE MTT-S Int. Microwave Symposium Digest, pp.1273-1276, 2009.
Slide 37
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
of 44
Example of Envelope Tracking (ET)
Envelope Tracking Operation
Fixed Vds
Supply
Vds
Vds
Conventional Operation
Time
Time
Output Signal
Output Signal
F.Yamaki et.al, ” A 65 % drain efficiency GaN
HEMT with 200 W peak power for 20 V to 65
V envelope tracking base station amplifier ,”
2011 IEEE MTT-S Int. Microwave Symposium
Digest, 2011.
Modulated
Vds Supply
GaN HEMT for Envelope‐Tracking
1. Cds reduction & Vds dependence reduction
2. BVdsx ≧300 V , for 20‐65V ET‐operation
Imax =680mA/mm, Cds=0.15pF/mm, BVdsx≧300V
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 38
of 44
Optimum Thermal Design for BTS 2. Efficiency Boosting
Output Power [W] General use
Depend on efficiency
Specified for BTS
1. Thermal Design
∼2005
2006
2009
2011
Circuit
Technique
Class‐AB
Higher Class
Doherty w/Class‐F ‐1
ET
Drain Eff. @8dB‐B.O.
34%
45%
55%
68%
Thermal Dissipation
Ref.
83%
68%
48%
2
1
Chip Size [mm2] Slide 39
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
of 44
RRH* BTS‐system and GaN HEMT
* Remote Radio Head
Base Station
RRH
RRH
RRH
Optic Fiber(0.1-20km)
Less than20kg/20L
→Easy to Settle
Control Unit
CAPEX (Capital Expenditure) ‐> Cost reduction of GaN itself has been progressed, and small & light weight PA, utilizing GaN high efficiency, contributes to BTS setting cost. OPEX(Operating Expenditure) ‐> Higher efficiency PA realizes lower power consumption.
Slide 40
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
of 44
Microwave Device Market
& SEI GaN HEMT Shipment
Total $1600M (@2013)
Si‐LDMOS
GaN HEMT
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 41
of 44
Outline
1. Fundamentals 2. Thermal Study of Substrate
3. Thermal Design (GaN for Radar)
4. Thermal Design (GaN for Base Station)
5. Summary
Slide 42
of 44
Summary (1/2)
 GaN HEMTs have already realized high quality, uniformity and reliability for infrastructure RF power applications. Thus, the cost reduction has been strongly required.  The adequate thermal transfer design is one of the solutions. GaN on‐SiC has been proved the best material to satisfy both the cost and thermal requirements. The focus design on pulsed operation is effective for compact radar devices, and the efficiency boosting in backed‐off region have reduced the chip size of base station PAs. Quality
SiC crystal, Epi‐growth uniformity, Small process deviation
Reliability
Stringent qualification, Ruggedness
Cost
Adequate thermal design, utilizing SiC material property GaN HEMT to various applications
(Satellite, Radar, Base station)
Slide 43
of 44
Summary (2/2)
 GaN HEMT has already been adopted in several markets. For further market penetration, continuous cost down efforts are essential. Price
90W GaN Chip x2
(Not specified for BTS)
24 x 17.4 mm
210W GaN Chip
(Specified for BTS)
21 x 13.2 mm
4’’ Wafer
4 inch
2 inch
Year
3 inch
Slide 44
of 44
Cost effective approaches for European GaAs & GaN Power solutions
Guillaume CALLET
guillaume.callet@ums‐gaas.com
WW05 Recent Advances in GaN Power HEMTs Related to Thermal Problems and Low‐Cost Approaches
Slide 1
Outlilne
What can be the reducing cost drivers for GaN on SiC solution ?  Presentation of UMS
Overview of GaN technologies
Thermal analysis for the development of packaged GaN solution
 UMS Products & Foundry Solutions
Conclusions
Slide 2
Outlilne
 Presentation of UMS
 III-V company with 20 year experience in semiconductor (especially GaAs
HEMT)
 Overview of GaN technologies
Thermal analysis for the development of packaged GaN solution
 UMS Products & Foundry Solutions
Conclusions
Slide 3
UMS at a glance
 Founded in 1996 by gathering Thales and AIRBUS Defense and Space
GmbH activities
 European source of RF MMIC solutions, GaAs and GaN foundry services
 2 industrial facilities in Ulm (Germany) & Villebon (France)
 400 people
Villebon facility
Ulm facility
Slide 4
Outlilne
 Presentation of UMS
 Overview of GaN technologies
Thermal analysis for the development of packaged GaN solution
 UMS Products & Foundry Solutions
Conclusions
Slide 5
UMS GaN technologies
Power by die (W)
Status
GH50 Power bar
50
New Development
GH50
GH15
Released
Released
Design kit available
Design kit available
To be released 2017
Targeted
Performances
0.50µm
0.25µm
0.15µm
Power
5 W/mm
4 W/mm
3.5 W/mm
50 V
30 V
20V
>200 V
>120 V
MTF
1e6 / 200°C
1e6 / 200°C
1e6 / 200°C
Domain of frequencies
Up to 7 GHz
Up to 20 GHz
Up to 40GHz
Breakdown voltage (Vbds)
GH25 MMIC
GH25
Gate length
Operating Vds
100
GH50
GH25
GH15
10
5 10 15 20 25 30 35 40
Target
Frequency
(GHz)
Slide 6
Outlilne
 Presentation of UMS
Overview of GaN technologies
Thermal analysis for the development of packaged GaN solution
 Various Packaging solutions
Thermal management of QFN
oPCB / Glue / Die Coating
 UMS Products & Foundry Solutions
Conclusions
Slide 7
Packaging offer for power
Main constraints for GaN packaging ‐ related to costs:
‐ Frequency band
‐ Thermal management
Slide 8
Thermal management analysis
Different solutions investigated:
Plastic molded QFN
 Package solutions:
 Flange
 Ceramic metal SMD
 SMD QFN
Ceramic metal SMD
 Stack variation
 PCB variations
 Interlayer stack
Flange package
– Mo / Diamond Tab
– CuMo Coin
Slide 9
Broad band packages 1: Plastic molded QFN
2: Enhanced flange package
3: Ceramic metal SMD
4: Flange package
Package + PCB insertion losses (dB)
Package + PCB return losses (dB)
4
2
3
1
2
3
4
1
 SMD QFN package is excellent
 Flange packages are competitive up to 20GHz
 Package optimization still ongoing to achieve VSWR <1.5:1 at 20GHz
Slide 10
 Part of the matching can be integrated into the die
Package Thermal analysis
Target: Package GH50 & GH25 products
C1
R1
GaN /SiC
R4
Thermal flow
Die attach
Leadframe
C2
R2
Solder
R5
PCB
C3
R3
 Analysis must be carried out on different samples:
 Power‐bares
 MMICs
 Finite element simulations performed on different 3 mains package family – ANSYS
Slide 11
MMIC Packaging:
Modeling assumption
Analysis performed on ANSYS:

Symmetry : ¼ of the device is meshed

GaN/SiC interface  Layer with low thermal conductivity (TBR)

Joule heating  Block heater along drain edge of gate foot (≈ 1.5 µm)

Boundary condition  Fixed temperature on backside of full assembly

Meshing  Specific methodology for power bars ≈ 1 000 000 nodes
2nd stage
Pdiss@14GHz & 34dBm
≈ 3.8 W/transistor
8 transistors  30.4W
Pdiss = 3.16W/mm
Test Vehicle GH25 MMIC:
1st stage : 4x8x125µm
2nd stage : 8x8x150µm
1st stage
Pdiss@14GHz & 34dBm
≈ 2 W/transistor
8 transistors  16W
Pdiss = 2W/mm
Slide 12
Flange Package
Tc = 80°C / Pdiss : 2nd stage  3.16 W/mm (30.4 W) / 1st stage  2 W/mm (16 W)
Cases
Flange KYO
Flange KYO + Tab Mo
Flange + Tab Diamond
GaN / TBR
SiC – 100 µm
AuSn – 25 µm
GaN / TBR – 2 µm
SiC – 100 µm
AuSn – 25 µm
GaN / TBR
SiC – 100 µm
AuSn – 25 µm
Mo – 150 µm
CVD/Cu – 300 µm
AuSn – 25 µm
AuSn – 25 µm
Stack
CuMoCu – 1.4 mm
CuMoCu – 1.4 mm
Thermal paste – 50 µm
Thermal paste – 50 µm
(°C/W)
Tj _max
(°C)
Thermal paste – 50 µm
Tcase
Tcase
Rth_tot
CuMoCu – 1.4 mm
Tcase
3.75
3.84
3.24
194
196.8
178
Slide 13
Ceramic metal SMD
Tc = 80°C / Pdiss : 2nd stage  3.16 W/mm (30.4 W) / 1st stage  2 W/mm (16 W)
Cases
SMD & PCB coin
SMD & PCB vias
PCB Coin Footprint
GaN/TBR
SiC – 100 µm
AuSn – 25 µm
GaN/TBR
SiC – 100 µm
AuSn – 25 µm
CuW – 200 µm
SnPb – 100 µm
Stack
Insulator – 1.3 mm
Cu –1.3 mm
CuW – 200 µm
SnPb – 100 µm
Cu – 18 µm
Insulator – 203 µm
CF3350 ‐ 50 µm
Cu – 1 mm
Al 6061 ‐ 3 mm
Tcase
Rth_tot
(°C/W)
Tj _max
(°C)
Tcase
5.06
4.69
233.8
222.7
PCB Via Footprint
Slide 14
Plastic Package SMD QFN
Tc = 80°C / Pdiss : 2nd stage  3.16 W/mm (30.4 W) / 1st stage  2 W/mm (16 W)
Cases
SMD & PCB coin
SMD & PCB vias
GaN/TBR
GaN/TBR
SiC – 100 µm
Ag – 15 µm
C194 – 200 µm
SnPb – 100 µm
Cu – 18 µm
SiC – 100 µm
Ag – 15 µm
C194 – 200 µm
SnPb – 100 µm
Stack
Insulator – 1.3 mm
Cu –1.3 mm
Insulator – 203 µm
CF3350 ‐ 50 µm
2 different Ag based glues evaluated:
a) 20 W.m‐1.K‐1
b) 70 W.m‐1.K‐1
Cu – 1 mm
Al 6061 ‐ 3 mm
Tcase
Tcase
Rth_tot
(°C/W)
Tj _max
(°C)
4.97a
4.72b
4.56a
4.32b
231a
224b
219a
211b
Slide 15
Synthesis
Case
study
Packaging
Remark
Rth (°C/W)
Bandwidth
Cost
1
Flange KYO
Hermetic
3.75
Up to 6 GHz

2
Flange KYO + Mo TAB
Hermetic
3.84
Up to 6 GHz

3
Flange KYO + Diamond
TAB
Hermetic
3.25
Up to 6 GHz

4
SMD + PCB coin
Hermetic
5.06
Up to 14 GHz

5
SMD + PCB via
Hermetic
4.69
Up to 14 GHz

6
SMD QFN + PCB coin
Die attach : 20 W.m-1.K-1
4.97
> 20 GHz

Die attach : 70 W.m-1.K-1
4.72
> 20 GHz

7
SMD QFN + PCB via
Die attach : 20 W.m-1.K-1
4.56
> 20 GHz

Die attach : 70 W.m-1.K-1
4.32
> 20 GHz

Slide 16
Approach power bar assembly
Temperature Gradient

8x8x400µm / Tref = 75°C / P = 2W/mm/ CW
Tj_Peak=211,9°C
Slide 17
Die Attach Impact


8x8x400 / 7x7 QFN, CW, P=2W/mm, Tcase=75°C, Cond glue = 40W/m.K
T(°C)
delta T (°C)
Rth(°C/W)
GaN / TBR
substrat SiC
colle puce (40W/mK)
Leadframe
BrasureSnPb
CI/glue
Drain Al
217.9
196.5
178.3
162.7
150.0
126.3
96.1
21.4
18.2
15.6
12.7
23.7
30.3
21.1
0.418
0.355
0.304
0.248
0.463
0.591
0.411
contribution
(%)
15.0
12.7
10.9
8.9
16.6
21.2
14.7
Total
75
142.9
2.792
100.0
8x8x400 / 7x7 QFN, CW, P=2W/mm, Tcase=75°C, Cond glue = 20W/m.K
GaN / TBR
substrat SiC
colle puce (20W/mK)
Leadframe
BrasureSnPb
CI/glue
Drain Al
T(°C)
230.9
208.9
190.5
161.6
149.5
126.2
96.1
delta T (°C)
22.0
18.4
28.9
12.1
23.3
30.1
21.1
Rth(°C/W)
0.430
0.360
0.565
0.235
0.455
0.588
0.411
contribution (%)
14.1
11.8
18.6
7.7
14.9
19.3
13.5
Total
75
155.9
3.046
100.0
Slide 18
GH50 Power Bar / Transient
 8x8x400 / 7x7 QFN / Tref = 75°C / P = 2W/mm
Slide 19
Conclusion on assembly
 SMD QFN solutions are broadband and lower cost solution
 Applied to GH25 and GH50
 Analysis shows that thermal management strongly depends on the PCB
• Coin offers very good thermal dissipation / difficult to implement to series
• Thermal glue can reduce by roughly 0.2 W.m-1.K-1
 As demonstrated GaN on SiC packaged in QFN can also operate in CW
• Important care must be take to the functioning conditions
Higher Integration 
MMIC
$$$ Internally‐Matched

Q‐MMIC
Internally Matched
DISCRETE General Purpose
L
S
C
X
Ku
Freq
Slide 20
Outlilne
 Presentation of UMS
Overview of GaN technologies
Thermal analysis for the development of packaged GaN solution
UMS Products & Foundry Solutions
 Foundry Presentation
o Market addressed are not only military
o Foundry partner contribution
Quasi-MMIC solution development
o Concept: Combination of GaN & GaAs
o Cost reduce
o Solutions
Conclusions
Slide 21
UMS Foundry Solutions
Slide 22
1W High Power Amplifier 37‐ 40GHz
PPH15X‐20
Power detector inside, Gain Control
Advanced concept / PPH15X-20 / QFN / CHA6194-QXG

6194
Application
 Point to point
 Point to Multipoint
High linearity HPA

 Specific features:
GaAs pHEMT process
Power detector dynamic 30dB
Low AM/AM, AMP/PM,
QFN 5x6
RF bandwidth: 37-40GHz

Linear Gain: 20dB

Power at 1dB comp.: 30dBm

Sat. Power : 31dBm

RL>13dB

Consumption: 6V, 0.8 A
Slide 23
1W High Power Amplifier 37‐ 40GHz
PPH15X‐20
Power detector inside, Gain Control
Vd=6V Idq=0.8A
34
32
Pout (dBm), Gain(dB)





30
28
26
Pout@1dB gain comp
Pout@4dB gain comp
Linear Gain
24
22
20
18
36
37
38
39
40
41
Freq (GHz)
Slide 24
Foundry Portfolio
 QFN Assembly offer for all processes
Slide 25
GH50 & GH25 Foundry key figures
Overall repartition
Nbre of wafer / year
200
150
GH50
100
Defense
Telecom
Space
50
GH25
0
2011
2012
2013
2014
2015
R&D
2016

GH25: Since 2010 more than 130 wafers processed in the frame of
80 different projects

GH50: Production launched and transferred to 4’’

More 150 wafers to be manufactured in both GH50 & GH25 in
2016

Design Kit including EM Stack and DRC for ADS2016
Slide 26
DK available for GH25‐10
 GH25: MMIC process: Qualified & Open in Foundry
•
•
•
•
•
•
•
•
250nm gate length
On 4‐inch SiC wafer
Frequency range: DC – 20 GHz
Vds = 30V as standard Recommended Operating Value
Idss = 850 mA/mm as average value
Very High breakdown voltage: Vbds > 120V
Power density: 4W/mm @ 10GHz in CW Mode
Design Kit available for ADS2009‐2016 & MWO
–
–
–
–
NL model for Hot FET (electrothermal)
L FET model for noise
NL model for Cold‐FET
NL model for diodes
(Scalable models for passives and active elements)
• New features: DRC & Stack EM available for ADS2016
Slide 27
ANSYS/ADS Thermal analysis
 Transistor GH25 8x125 d’UMS / Vds=25V Ids=25mA/mm
ADS2013
Slide 28
TNO – HPA GH25‐10 design
THERMAL SIMULATION
In courtesy of TNO
Input stage 1x10x275um, Pdiss=8.2 W
Output stage 4x10x275, Pdiss=23.3 W
Matching networks, Pdiss=7.2W
( = Pdiss_total - (input+output stage) )
Tj Input stage
Tj Output stage
MMIC bottom
Package bottom
GaNS‐3
Slide 29
UMS product Solutions
Slide 30
Product
CHK015A‐QBA
DC ‐ 6 GHz Packaged Transistor
 Application

Radar & Communications
General Purpose Transistor






 Specific feature



Wide-band
Low parasitic Plastic Package
Low thermal resistance
RF bandwidth:
Linear Gain:
Output Power:
Drain Efficiency:
PAE:
Package:
DC – 6GHz
14dB @ 6GHz
> 15W
> 70%
50% @ 6GHz
DFN 3x4
Slide 31
25W / X-band / QFN
Sampling Now
In sampling / GH25 Based / Die product to be packaged
V‐
 Application
 Defence / Space
In
STG1
STG2
Out
Main Performances

RF bandwidth:
8.5 – 10.5GHz

Gain:
30dB

Pout :
25W

PAE_associated: 45%

Consumption:
V+
 Specific feature

High efficiency

High power

Die / 15W version available in QFN
30V, 0.8A
Slide 32
Quasi MMIC Concept
Use of UMS proprietary Passive MMIC Technology: Q-MMIC is close to MMIC size
Input Matched Circuit
+
$
+
GaN
Power +
Bar $
$
$
+
Fast design & fabrication
Output Matched Circuit
50Ω
50Ω
$
Fast design & fabrication
☺
Customize your Q‐MMIC
☺
Slide 33
Why “Quasi MMIC”?
• Integration
– Close to MMIC size
Example of 50W C‐
band HPA
6mm²
• Cost
• Flexibility
– Short dev. cycle times (passive MMICs)
 GH50/25: 14Weeks ICT
 URLC: 5 Weeks
7.6mm²
6
MMIC / Quasi‐MMIC Cost
– Close to hybrid solutions
3.6mm²
5
4
3
2
1
10
15
20
MMIC Size (mm²)
Applicable to QFN for low cost development
25
Slide 34
New Product
100W / L‐band / QFN
In development / GH50 based / Internally‐Matched / Q‐MMIC
 Application

Radar / Dual Use
Main Performances
 Specific feature @ 1.3GHz

Peak Pout=110W

With PAE=57%

Gain = 14dB

RF bandwidth:
1.2 – 1.4GHz

Linear Gain:
15dB

Output Power:
100W

Gain @ 50W
10dB

PAE :
55%

Package:
DFN 7x7
“High Performance Plastic Packaged 100W L‐Band Quasi‐MMIC HPA” D. Bouw et al. ‐ EuMC06‐05
50W / C‐band / QFN
Slide 35
New Product
In development / GH25 based / Internally‐Matched / Q‐MMIC
 Application

Radar & Communications
Main Performances
 Specific feature

High PAE

Low parasitic Power Plastic Package

Low thermal resistance

RF bandwidth:
5.2 – 5.9GHz
5.9-6.9GHz

Linear Gain:
14dB
13dB

Output Power:
50W
50W

Gain @ 50W
10dB
9dB

PAE :
47%
45%

Package:
DFN 7x7
DFN 7x7
Slide 36
C‐band DPA Description
Advanced Development
Symmetric DPA Architecture / GH25 based / Q‐MMIC / QFN packaging  Application
 Communications / 5G
Main Objectives


Configuration

1 stage => for investigations purpose

Single & dual input

Linearity / Main & Peak
synchronization …
Key Performance

Frequency band : 5.6 – 6.6GHz

Peak Output Power: 15W

PAE @ 6dB OBO: > 35%
Slide 37
C-band DPA Characterizations
PAE vs Freq / 6 & 10dB OBO / 4 Boards / T=25°C
 Biasing conditions:


Vd = 30V
Id_q_main = 60mA / Vgs0_peak = -7V
6dB OBO
10dB OBO
Slide 38
Outline
 Presentation of UMS
Overview of GaN technologies
Thermal analysis for the development of packaged GaN solution
UMS Products & Foundry Solutions
Conclusions
Slide 39
Conclusions
 QFN Solution validated for power technologies up to 40GHz – including
GaN

MMIC in QFN for GH25 validated with enhanced glue + PCB solutions
• High efficiency solutions already available

Q‐MMIC in QFN used for GH50 & GH25
• Challenging solutions evaluated

Release of PPH15X‐20 for applications up to 40GHz
 Foundry access allowing more and more QFN assembly



Tools available for ADS allow more accurate thermal analysis
Association with URLC (passive process) available in Foundry
Indicators allow to identify the cost decrease (Industrial Manufacturing cycle time
reducing, …)
 Final passivation is under development and should be available to
improve the robustness versus humidity for GH50/GH25/URLC
Slide 40
Aknowledgment
• TNO for their participation
• Colleagues for their support & works
–
–
–
–
–
–
M. Camiade
P‐F. Alleaume
L. Brunel
M. Feron
E. Leclerc
J‐P. Viaud
Slide 41
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

advertising