Kuball, M. H. H., & Pomeroy, J. W. (2016). A Review of Raman
Thermography for Electronic and Opto-Electronic Device
Measurement With Submicron Spatial and Nanosecond Temporal
Resolution.
IEEE Transactions on Device and Materials Reliability
,
16
(4), 667-684. https://doi.org/10.1109/TDMR.2016.2617458
Peer reviewed version
Link to published version (if available):
10.1109/TDMR.2016.2617458
Link to publication record in Explore Bristol Research
PDF-document
This is the author accepted manuscript (AAM). The final published version (version of record) is available online
via IEEE at http://ieeexplore.ieee.org/document/7590091. Please refer to any applicable terms of use of the
publisher.
University of Bristol - Explore Bristol Research
General rights
This document is made available in accordance with publisher policies. Please cite only the
published version using the reference above. Full terms of use are available:
http://www.bristol.ac.uk/red/research-policy/pure/user-guides/ebr-terms/
TDMR-2016
1
Abstract— We review the Raman thermography technique,
which has been developed to determine the temperature in and
around the active area of semiconductor devices with submicron
spatial and nanosecond temporal resolution. This is critical for the
qualification of device technology, including for accelerated
lifetime reliability testing and device design optimization. Its
practical use is illustrated for GaN and GaAs based high electron
mobility transistors, and opto-electronic devices. We also discuss
how Raman thermography is used to validate device thermal
models, as well as determining the thermal conductivity of
materials relevant for electronic and opto-electronic devices.
Index Terms— Thermography, thermal simulation, GaN,
GaAs, HEMT, thermal management, reliability.
I. INTRODUCTION
LECTRONIC and opto-electronic semiconductor devices
have advanced greatly in recent decades by incorporating
new materials and designs to achieve increased speed,
decreased size and improved efficiency. This has enabled
higher power densities at higher frequencies for electronic
devices, and further wavelength ranges or increased optical
output power for opto-electronic devices. The heterogeneous
integration of complimentary materials has also increased
device performance and functionality, giving a wide range of
technological design options [1]. However, an ever increasing
power dissipation density has resulted in thermal management
challenges. It is critically important to be able to accurately
determine device temperatures in order to assess the reliability
of new technologies, because temperature is one of the
dominant drivers of device degradation and influences device
performance [2-7]. Phonon transport at the nanoscale and at
interfaces impacts how waste heat is transported away from the
erials (e.g. SiC, GaN, diamond, graphene) are
This work is in part supported by the UK Engineering and Physical Sciences
Research Council (EPSRC), the European Space Agency (ESA), the European
Defense Agency (EDA), the US Office for Naval Research (ONR) and the
Defense Advanced Research Projects Agency (DARPA) under Contract
FA8650-15-C-7517 monitored by Dr. Avram Bar Cohen, supported by Dr. John
Blevins, Dr. Joseph Maurer and Dr. Abirami Sivananthan. Any opinions,
findings, and conclusions or recommendations expressed in this material are
those of the authors and do not necessarily reflect the views of DARPA.
also often not as well-known as commonly thought, depending
on the growth conditions or growth methods used. Relying on
device thermal simulations alone, without experimental
verification, can therefore potentially limit the accuracy
achieved when predicting operating temperatures in electronic
and opto-electronic devices, negatively impact device design or
reliability.
Long term reliability is a concern with any new technology
and qualification is one part of the process towards
commercialization. For example, GaN high electron mobility
transistors (HEMTs) are presently being developed and
commercialized for both microwave and power electronics
applications. A corresponding processes has happened before
for Si, GaAs and InP, and will happen again for future materials
and device systems. Part of this qualification is temperature
accelerated lifetime testing which relies on detailed knowledge
of channel temperatures which are used to predict the mean time
to failure (MTTF) at the designed operating temperature.
Current standards, such as JEDEC [7] for the determination of
channel temperature, still mostly
characterization approaches, such as IR thermography, which
often can no longer be applied with high accuracy in high power
density devices due to their intrinsically limited spatial
resolution and the resulting underestimation of the device peak
temperature [8]. For example, temperature gradients as high as
-
based RF HEMTs [9].
The question also arises - Which temperature is relevant for
device reliability assessment? This has not been discussed
extensively in the past because the spatial resolution limitations
of existing thermography techniques has made it impossible to
measure close to the device channel and the temperature profile
directly. For example, it is important to identify that not only
the peak channel temperature near the gate contact in a HEMT
is a critical parameter, but also other lower temperature regions
M. Kuball and J. W. Pomeroy are with H. H. Wills Physics Laboratory,
University of Bristol, Bristol BS8 1TL, U.K. (e-mail:
martin.kuball@bristol.ac.uk).
Copyright © 2016 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be
obtained by sending a request to pubs-permissions@ieee.org
M. Kuball, Member, IEEE, and J. W. Pomeroy.
E
TDMR-2016
2
of the device may result in device degradation, such as of the
ohmic source and drain contacts. Pulsed operation is also used
in many applications, so any generally applicable thermography
technique must combine high spatial resolution with high
temporal resolution to be able to record fast temperature
transients, which cannot be achieved using conventional IR
thermography. In this paper, we review the Raman
thermography technique which enables submicron spatial
resolution and nanosecond temporal resolution measurements
in semiconductor electronic and opto-electronic devices.
Following a discussion of existing thermography techniques,
the general principles of Raman thermography are given
together with examples of applications.
II. SEMICONDUCTOR DEVICE TEMPERATURE MEASUREMENT
TECHNIQUES
Various electrical characterization based methods have been
used extensively to assess the channel temperature in electronic
devices. The equipment needed for these temperature
assessments is available in standard electrical testing
laboratories. Different methods are used, all based on
measuring changes in temperature dependent electrical
parameters, including: Saturated drain current [10-14]; gate
leakage current and threshold voltage, among other parameters.
These techniques offer no spatial resolution as such, but instead
are sensitive to the entire periphery of the device channel [14].
As a consequence, electrical methods tend to underestimate the
peak device temperature. For example in a HEMT, temperature
may be averaged not only along the whole gate width, but also
typically between source and drain contact. Because the
temperature measurement location is not well defined, it is can
be difficult to relate the measured temperatures to a thermal
simulation in a meaningful way. For opto-electronic devices,
light emitting diodes and laser diodes, the emission wavelength
is temperature dependent, following the semiconductor
bandgap temperature dependence. Measuring the emission
wavelength is therefore a convenient method for temperature
estimation in the active region of opto-electronic devices [15].
Naturally this technique can only provide temperature
information in areas that emit light, whereas obtaining the
temperature distribution in other parts of the device structure
may also be required to assess the full benefits of the device
design. Alternatively, the temperature-induced shift in lasing
threshold voltage can be used to measure the temperature in
opto-electronic devices [16], although this technique lacks
spatial resolution, in the same way as when electrical methods
are applied to transistors.
IR thermography is the most commonly used technique for
semiconductor device and circuit technology temperature
measurement in industry [8, 17, 18]. This technique is based on
measuring the thermal radiation emitted from the surface of a
device, typically detected in the 3 10 µm spectral range,
depending on the detector used. The radiated light intensity
(radiance) scales as T
4
, following the Stephan-Boltzmann law,
where T is the surface temperature in Kelvin. Therefore,
temperature can be determined by measuring radiance, after
calibrating the surface emissivity of the device under test
(DUT) to take into account that the measured surface is not a
perfect black body (emissivity = 1). Emissivity can be
calibrated by measuring the radiance of the DUT at single or
multiple well defined temperatures. Calibration must be
performed on a pixel-by-pixel basis because the emissivity
typically varies across the surface of a device, which has
regions of exposed semiconductor, passivation and different
metal contacts. Sample movement during the emissivity
calibration is one potential source of artifacts in IR
thermography temperature maps, particularly for transistor
structures which have features of a similar or smaller length
scale than the IR camera resolution; such artifacts can be
particularly apparent at the edges of structures.
The achievable spatial resolution limit of IR thermography is
typically on the order of the wavelength measured. Therefore,
the long wavelength of the measured IR emission is a limitation
when measuring devices with micron or sub-micron scale
feature sizes. The effect of lateral spatial averaging on
measured device temperatures must be carefully considered
when interpreting results. For example, temperature is often
averaged over the whole sourcedrain gap when a HEMT is
measured. Furthermore, many semiconductor materials are
transparent at the wavelength of the emitted IR thermal
radiation (unless heavily doped) and radiation originating from
below the surface contributes to the thermal image, including
from the die attach/wafer backside. The resulting lateral and
depth temperature averaging can cause the temperature
measured using IR thermography to be significantly lower than
the actual peak temperature [8, 19], as illustrated in Figure 1 for
a GaN HEMT. Another challenge is that metal surfaces tend to
have rather low emissivity which makes temperature
determination of metal areas such as contacts challenging,
unless measurements are performed at elevated background
temperatures. Low emissivity or transparent surfaces can be
addressed by covering the device surface with an opaque high
not truly represent the surface temperature and may be
impossible to fully remove from the device after the
measurement [8]. It should be possible in theory to compare the
Fig. 1: (a) Device temperature in an AlGaN/GaN HEMT obtained by IR
thermography; (b) High-spatial-resolution Raman thermography temperature
profile measured across the source-drain region. Reprinted from [37],
Copyright 2005, with permission from Institute of Physics Publishing.
TDMR-2016
3
data obtained by IR thermography to a device thermal
simulation, considering all the factors previously described.
Attempts have been made to do this [20], although whether this
approach can be used to accurately determine the peak
temperature in all devices is controversial, particularly in
devices with very high local temperature gradients such as GaN
RF HEMTs. Although peak temperature accuracy may be
limited by spatial averaging in IR thermography, it is a very
efficient method for mapping the temperature over large surface
areas, making this technique especially suitable for comparative
screening of identical devices where exact knowledge of peak
can be identified quickly, such as those caused by local leakage
currents or power imbalances in monolithic microwave
integrated circuits (MMICs). Integrating fast low resolution IR
temperature mapping with slower high resolution serial
temperature measurement approaches can therefore be
beneficial and a combined Raman-IR thermography
measurement system has been developed for this purpose, as
described in [8].
Other optical techniques have been demonstrated for
electronic and opto-electronic device thermography. Micro-
photoluminescence (PL) measurements can be used to measure
the semiconductor surface temperature in active devices [21],
exploiting the temperature dependence of the band edge
emission. PL is a laser-based technique offering submicron
spatial resolution, although it should be noted that electron hole
pairs generated by the above bandgap laser illumination can
modify the current/field distribution in the device being
measured. This additional photo-induced current can cause a
significant additional device heating, which often dominates
even the direct laser heating induced temperature rise [22].
Reflectivity changes can also be used to optically probe the
device temperature, either considering only the change in
reflectivity [23, 24], or additionally by measuring the phase
shift of the reflected light [25, 26]. Calibration of the
thermoreflectance coefficient can be challenging as the change
in reflectivity with temperature is often small, however it is
possible by performing careful measurements of specially
prepared samples [27]. Liquid crystal thermography is another
optical technique and relies on the known phase transition
temperatures of liquid crystals [28]. However, it is less
convenient and no longer commonly used; multiple depositions
of liquid crystals with different phase transition temperatures
are needed to map each temperature contour.
Raman thermography is the focus of this review paper. The
use of Raman spectroscopy for temperature measurements was
first demonstrated on Si devices [29], but has been more fully
exploited following the advancement of GaN electronic devices
[30], and has since been demonstrated extensively by several
groups [31-35]. This technique is often combined with finite
element (FE) thermal simulation to determine the channel
temperature, as well as the measurement of material thermal
properties used for device thermal simulations [36-38]. Figure
1 shows a comparison of temperatures measured in the same
device using Raman thermography and IR thermography. The
improvement in spatial resolution is clearly apparent, with a
lateral resolution of 0.5 µm for Raman thermography and about
7 µm in this case for IR thermography. This clearly illustrates
the benefit of using a higher spatial resolution technique for the
thermal analysis of transistors in particular, where the heat
generating regions tend to be on the micrometer or smaller
length scale.
In the following sections we will discuss this technique in
detail, including its practical application with examples. This
optical technique exploits the temperature dependent properties
of quantized lattice vibrations (phonons) of the materials in
electronic or opto-electronic devices. As it is a laser based
microscopy technique, it can provide temperature mapping with
sub-micron lateral spatial resolutions as high as 0.5 µm (Figure
1), and even smaller if combined with advanced optical
focusing concepts such as solid immersion lenses (SILs). One
advantage of Raman thermography in comparison to PL is that
sub-band gap lasers can be used, avoiding light absorption in
the device which could impact its operation and also enabling
three dimensional mapping through transparent materials. In
addition to spatial mapping, the temperature transients can also
be measured by modulating the probing laser, with nanosecond
or better temporal resolution depending on the modulation
scheme used.
III. PRINCIPLES OF RAMAN THERMOGRAPHY
A. Spatial Resolution
Raman thermography is a laser-based microscopy
measurement technique. Laser light is focused through a
microscope objective lens onto the region of interest and
backscattered light is usually collected for analysis using the
same lens, as illustrated in Figure 2(a). By the scanning focused
laser spot over the DUT, temperature maps can be obtained by
measuring each point sequentially. For example, the channel
temperature line profile in a HEMT can be measured by
scanning laterally in one dimension (Figure 1), whereas by
scanning in two dimensions (2D) surface temperature maps can
be obtained which may be used to probe the temperature
distribution around defects or other features [39]. As with any
Fig. 2: Raman thermography setup with a device under test (DUT) where both
the device structure and substrate are transparent to the laser light, showing:
(a) Configuration for a top surface measurement; (b) back side measured
through the device substrate. The DUT is a typical GaN-on-SiC HEMT
structure.
TDMR-2016
4
measurement technique, it is important to take into account
spatial averaging when analyzing temperatures measured by
Raman thermography, which is determined by the optical
resolution of the measurement system. Two scenarios need to
be distinguished when assessing the axial measurement
resolution of Raman thermography: 1) When the probing laser
photon energy is below the bandgap of the materials in the
active area of the device; 2) when above-bandgap excitation is
used. We consider the former case first, when photon
absorption, heating and photo-induced current is negligible.
This is discussed for the example of an AlGaN/GaN HEMT, as
due to its bandgap of 3.4 eV, the GaN layer and substrate, with
the exception of silicon, are transparent for most of the laser
sources used in Raman systems. In this case, Raman
thermography measures a spatially averaged temperature in a
finite volume in the GaN layer and substrate, as illustrated in
Figure 2(a). The transparency of the measured materials can
also be exploited by focusing the probing laser below the
surface, allowing 3D temperature mapping.
The diffraction limited spatial resolution of a scanning
confocal laser microscope is defined as the elliptical half
intensity radius of the point spread function (PSF), which
depends on the on numerical aperture (NA) of the objective and
the laser wavelength () and is independent of the
magnification. The lateral resolution is given by
, (1)
and the axial resolution, related to the depth of field (DoF), is
given by
, (2)
where n is the refractive index of the focusing medium [40]; for
NA<0.5, nNA
2
. Although
these values can be improved slightly by using a highly
confocal measurement system with a reduced aperture size [40],
the NA of the objective lens is the main parameter that can be
increased to enhance spatial resolution [41], and especially to
improve the axial resolution. The maximum NA limit is equal
to the refractive index of the immersion medium, which is
NA
max
=n
air
=1 for a conventional air immersion microscope,
although the practically achievable NA is lower than the
theoretical limit due to practical restrictions such as the
objective lens minimum working distance. It is also important
to note that (1) and (2) are only applicable for diffraction limited
light focusing, which is the case close to the semiconductor/air
interface where optical aberration can be neglected [42].
Raman spectroscopy is material selective, so that in a
transparent layered structure, the characteristic phonons of each
material can used to simultaneously probe the temperature in
these layers. For example, a typical Raman thermography
measurement using a 488 nm laser focused near the surface of
a GaN device (n=2.4), through an 0.5 NA objective lens, will
achieve a resolution of r
lateral
=0.5 µm and r
axial
~8 µm,
calculated using (1) and (2). In this case, the depth of field is
much larger than the 1-2
measured temperature in that layer can be considered to be a
simple depth average through this layer. A separate average
through the upper portion of the SiC substrate may also be
measured by considering the SiC Raman modes of the
substrate, as illustrated in Figure 2(a), i.e., 3D temperature
information can be obtained. This also applies if other
transparent materials are being probed. Figure 3 shows the
volume of GaN averaged in a typical Raman thermography
measurement, compared with the simulated temperature
distribution in an AlGaN/GaN RF HEMT. Although it is
important to consider the effect of spatial averaging on the
measured temperature, the Raman measured average
temperature can be relatively close to the peak value.
As well as using the material specific phonon modes to
measure the temperature in different layers of a structure
separately, confocal Raman thermography can be used to map
the 3D temperature distribution inside transparent materials.
Focusing the laser through a thick transparent substrate, from
the back side of the device, can also be advantageous because
it enables the temperature to be probed underneath metal
contacts, areas which would otherwise be obscured when
measuring from the top side of the device (Figure 2(b)).
However, spatial resolution may be aberration limited owing to
refraction at the air/semiconductor interface and then (1) and
(2) do not apply. In this scenario, marginal and on axis rays are
focused at different depths, which affects both the lateral
resolution and depth of focus, but especially the latter. The
refraction limited DoF is then given by [42, 43]
, (3)
which increases linearly with depth (z). In a GaN-on-SiC
device for example, r
axial
can be as large as 30 µm when
focusing through a 300 µm thick SiC substrate. This increased
spatial averaging needs to be considered when analyzing
Raman temperature measurement results. Spherical aberration
Fig. 3: Combined drift diffusion and 3D finite element thermal modelling
result, showing a temperature cross section through a packaged two finger
AlGaN/GaN-on-SiC HEMT (half the device is shown) operated at 50V
DS
, -
1.5V
GS
and a power dissipation density of 11 W/mm, at 25 C ambient
temperature. The Joule heating profile in the 2DEG channel is shown above
for visual reference. The model is adapted from [4].
25C
162C
![Fig. 4: (a) A typical Raman spectrum of GaN-on-SiC, with the GaN E2 and A1(LO) phonon modes labelled. A phonon mode related to underlying the SiC substrate is also apparent; (b) phonon frequency as function of temperature for GaN on different substrates, which is dependent on the specific growth conditions used. Copyright 2006 IEEE. Reprinted, with permission, from [8].](/figures/fig-4-a-a-typical-raman-spectrum-of-gan-on-sic-with-the-gan-uoiczj16.webp)
![Fig. 9: Simulated channel heating distribution map of AlGaN/GaN-on-SiC HEMT with field plates at different drain bias voltages. Reprinted from [4], Copyright 2015, with permission from Elsevier.](/figures/fig-9-simulated-channel-heating-distribution-map-of-algan-13swskjo.webp)
![Fig. 16: (a) Temperatures measured using Raman thermography at the center of multi-finger AlGaN/GaN HEMTs with different geometries, compared to the temperature predicated by 3D finite difference modelling (solid line); (b) thermal resistance in ºC mm/W as a function of finger width and finger spacing, predicted by thermal simulation. Reproduced from [84], with the permission of AIP Publishing.](/figures/fig-16-a-temperatures-measured-using-raman-thermography-at-1m1ixph3.webp)
![Fig. 8: Simulated temperature rise in a packaged GaN HEMT on a 100μmthick SiC substrate, shown as a function of number of gate fingers at a constant power dissipation density of 11W/mm; using a model adapted from [4]. The ratio between the peak gate temperature rise and spatially averaged temperature rise, equivalent to a Raman thermography measurement, is shown. The temperature across each layer in the device stack is also shown, normalized to the peak gate temperature. A eutectic die attach and CuW carrier are included, representing the package.](/figures/fig-8-simulated-temperature-rise-in-a-packaged-gan-hemt-on-a-170vvmy8.webp)