This article has been accepted for publication in IEEE Photonics Journal.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2025.3552951

Thermal Performance of 940 nm AlGaAs-Based

VCSELs Grown on Germanium
J. Baker, C. P. Allford, S. Gillgrass, J. I. Davies, S. Shutts, and P. M. Smowton

Abstract—VCSEL thermal resistances are determined from GaAs substrates [3], and investigated VCSELs grown on 450
power-current-voltage-wavelength measurements performed and 225 μm thick Ge substrates [4]. Other research groups
on nominally identical epitaxial structures grown on Ge and have focussed on the assessment of crystal quality and the
GaAs substrates. We show that the effective thermal transition layers between the Ge substrate and AlGaAs
conductivity of the VCSEL is increased when grown on Ge epitaxial structure [5], [6], leading to the successful
substrates. production of 940 nm VCSELs [7], demonstration of 25 Gb/s
NRZ transmission at elevated temperature [8], and the
Keywords—VCSEL, GaAs, Ge, thermal resistance
monolithic integration of AlGaAs DBRs on Si substrates via
I. INTRODUCTION a Ge aspect ratio trapping layer [9]. Recently, a SWIR VCSEL
emitting at 1380 nm was demonstrated using AlGaAs DBRs
The production of 800-1000 nm vertical-cavity surface- grown on Ge substrates and wafer-bonded to an AlGaInAs
emitting lasers (VCSELs) is a well-established industry built active region grown on InP [10]. It has also been alluded to in
upon decades of advancements in epitaxy and fabrication in the literature, but not yet shown, that the thermal performance
the GaAs-AlGaAs material system. However, in today’s of Ge-substrate VCSELs should be improved due to the higher
environmental and geopolitical context, question marks over thermal conductivity of bulk Ge and we presented our initial
GaAs have arisen relating to supply chain vulnerability and findings relating to this at the 2024 IEEE International
ecological impact. For these reasons, germanium, with its Semiconductor Laser Conference [11]. In this paper, we
broader production, non-toxicity, and potential for recycling, expand upon those results to robustly assess the thermal
presents itself as a viable alternative growth substrate in the performance of Ge-substrate VCSELs against their GaAs
future for the optoelectronics industry. In recent years, counterparts through on-wafer electro-optical testing.
germanium has been described as an ideal drop-in
replacement for gallium arsenide as the growth substrate of II. MATERIALS & METHODS
choice for high-volume production of AlGaAs-based
VCSELs [1]. A driving motivation for using Ge is the lattice A. Epitaxial Structure & Device Fabrication
match with AlAs (and high Al-composition AlGaAs) which The epitaxial wafers used for this study are generic
eliminates the strain-induced wafer bow suffered with growth structures designed and grown by IQE plc for emission at 940
on GaAs substrates. Other advantages include higher thermal nm. This consists of a MQW active region in a λ-thick cavity,
conductivity, higher electron and hole mobility, higher sandwiched between an upper p-doped AlGaAs DBR and
fracture toughness, defect-free (zero EPD) crystal quality, and lower n-doped AlAs-AlGaAs DBR. The inclusion of AlAs in
the potential to grow on thinner substrates. Subsequently, we the bottom DBR is used to improve heat dissipation due to
have highlighted the benefits of growth on Ge for improving improved thermal conductivity of binary AlAs [12]. A high Al
on-wafer uniformity of device performance resulting from a composition layer is positioned in the top DBR which, after
reduced oxidation variation [2], we have shown comparable selective thermal oxidation, provides electrical and optical
performance of VCSELs grown on 200 mm (8-inch) Ge and confinement. Nominally identical structures, within
acceptable tolerances, are grown on n-type GaAs and n-type
The EPSRC funded Future Compound Semiconductor Manufacturing Hub Ge substrates by MOVPE. For structures on Ge, a proprietary
(EP/P006973/1) and the UKRI Strength in Places fund Project CSconnected MOVPE-grown transition layer, developed by IQE, is
(107134) provided essential resources for this study. Corresponding author: included to ensure a smooth transition from Ge to AlGaAs.
Jack Baker.
J. Baker is with the School of Physics and Astronomy, Cardiff University,
The devices characterised for this work are Quick VCSEL
Translational Research Hub, UK, CF24 4HQ (email: bakerj19@[Link]). (QuickSEL) structures, the fabrication process of which is
C. P. Allford is with the School of Physics and Astronomy, Cardiff University, described in [13].
Translational Research Hub, UK, CF24 4HQ (email:
allfordcp1@[Link]). S. Gillgrass is with the School of Physics and B. Experimental Method
Astronomy, Cardiff University, Translational Research Hub, UK, CF24 4HQ To assess and compare the VCSEL thermal performance,
(gillgrasss@[Link]). S. Shutts is with the School of Physics and
Astronomy, Cardiff University, Translational Research Hub, UK, CF24 4HQ
the thermal resistance, RTh, is measured for a range of oxide
(shuttss@[Link]). P. M. Smowton (smowtonpm@[Link]) is with apertures for devices grown on both Ge and GaAs substrates.
the School of Physics and Astronomy and the Institute for Compound We employ the measurement technique described in [14] for
Semiconductors, Cardiff University, Translational Research Hub, UK, CF24 the determination of RTh, which we outline here. The thermal
4HQ. J. I. Davies is with IQE plc, St Mellons, Cardiff, UK (email: resistance can be defined as in equation (1) by the temperature
IDavies@[Link]).
Color versions of one or more of the figures in this article are available
change for a given heat flux:
online at [Link]

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2025.3552951

RTh = ΔT/ΔPdiss, (1) carried out to assess the p-contact resistance for each sample.
Specific contact resistances were found to be on the order of
where Pdiss is the power dissipation: 10-7 Ωcm2. Additionally, sheet resistances were extracted as
Pdiss = IV – Poptical. (2) 58 and 64 Ω□ for the Ge and GaAs wafer, respectively. This
is representative of the conductivity of the uppermost layers
Here, IV is the total electrical power equal to the product of the of the top DBR but provides some indication of the doping
bias current, I, and operating voltage, V, and Poptical is the total concentrations throughout the epi-structure.
optical power emitted from the VCSEL. The quantity ΔT in
equation (1) is not directly measurable for a VCSEL, however, The threshold current, Ith, and current density, Jth, is compared
the shift of the emission wavelength can be used to indirectly for Ge and GaAs devices in Fig. 2. The GaAs substrate devices
measure the device internal temperature, due to the known have a slightly higher threshold requirement, but values are
refractive-index shift, as described in [15] and [16]. The comparable. Threshold currents are within a few hundred μA
resulting shift with temperature (in nm/°C) can be used to for all devices tested and current densities are higher by ~ 0.2
calibrate the wavelength shift with dissipated power (in kA/cm2 for large aperture GaAs devices. This variation is
nm/mW), the ratio of which gives the thermal resistance (in within the tolerance of wafer-to-wafer and run-to-tun drift.
°C/mW), as in equation (3): There are some potential drivers of this which would have
differing impacts on the thermal performance which we will
RTh = ΔT/ΔPdiss = (Δλ/ΔPdiss)/(Δλ/ΔT). (3) consider, e.g., alignment of the gain peak and cavity mode
By measuring thermal resistance for a range of oxide wavelengths, overlap of the optical mode with the carrier
apertures, the equation density profile, mirror reflectivities, and absorption/scattering
loss.
RTh = 1/(2𝜉Da) (4)
can be fit to the experimental data. This approximates the
VCSEL as a disk heat source on a thick substrate, where 𝜉
represents the thermal conductivity of the material separating
the heat source from the heat sink, and Da is the active region
diameter [17]. In our case, 𝜉 represents the thermal
conductivity of the bottom DBR + substrate (including the
transition layers for the Ge substrate devices). Hence, we use
the measured thermal resistance to determine values of 𝜉 for
the Ge and GaAs substrate structures. The effect of interface
resistances throughout the structure and between the substrate
and heatsink are also included in the value of 𝜉. Often, the
active region diameter is given by the oxide aperture diameter,
which we measure via infrared microscopy in-situ during the
thermal oxidation process. We find that our experimental data
diverges from the model with this assumption, therefore, we
include a correction to the oxide aperture diameter to allow for
non-ideality, such that equation (4) becomes: Fig. 1: Differential resistance between 2.2 and 2.4 V of ~ 2 to 18 μm
oxide aperture diameter GaAs and Ge substrate devices.
RTh = 1/[2𝜉(Dox + b)]. (5)
where Dox is the oxide aperture diameter and b is a constant.
We did not observe any differences in the oxidation behaviour
(e.g. overall oxidation rate, crystallographic axis preference)
between the Ge and GaAs substrate wafers.
III. RESULTS & DISCUSSION
Firstly, to verify that the epitaxial structures are
equivalent, the electro-optical characteristics of devices
ranging 30 to 49 μm in mesa diameter are compared. This
produces oxide aperture diameters ranging ~ 1 to 19 μm. On-
wafer testing was performed for Ge and GaAs substrate
structures at the centre of a 150 mm (6-inch) wafer.
Understanding the electrical performance of the devices is
critical for determining the thermal resistance due to the
impact on the calculated power dissipation values. This was
assessed from current-voltage characteristics for a range of
oxide apertures. The series resistance was found to be Fig. 2: Threshold current and threshold current density of ~ 2 to 18 μm
comparable for both Ge and GaAs structures, shown in Fig. 1. oxide aperture diameter GaAs and Ge substrate devices.
Circular transfer length method (CTLM) measurements were

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2025.3552951

expected to have a negligible impact on the subsequent

analysis. A difference in the overlap of the optical mode with
the carrier density profile (transverse confinement factor) is
more difficult to determine. However, from in-situ IR-
microscopy, we observe no distinct differences in the shape of
the oxide apertures between the GaAs and Ge samples, hence
we can assume that the mode shape and carrier density profiles
are comparable. The VCSEL emission wavelength is
extracted from spectral measurements at a range of pump
currents (up to a maximum of 10 mA). Raw spectra are shown
in Fig. 3 for a 3 μm aperture diameter device. The
corresponding dissipated power at each current was
determined from the power-current-voltage characteristics
with Δλ/ΔPdiss then given by the slope of the curves. A plot
showing the wavelength shift with dissipated power for ~ 8
μm oxide aperture devices is shown in Fig. 4.

Fig. 3: Raw spectra at 25 °C from 0.5 to 6.5 mA for an ~3 μm VCSEL

(top) and corresponding fundamental mode wavelengths versus
current for the same device (bottom). Fig. 4: VCSEL emission wavelength as a function of dissipated power
at 25 °C for an 8.2 and 8.6 μm GaAs and Ge VCSEL, respectively.
We observe the minima in Jth as a function of temperature to
occur at approximately 50°C at the centres of all GaAs and Ge
wafers measured, indicating that the detuning of the gain peak
and cavity mode are equivalent. We do find that the emission
wavelengths are 3-4 nm longer for the Ge wafers, which
suggests a thicker inner cavity layer. If consistent throughout
the layers, this would indicate that the Ge epi-stack is ~ 0.5%
thicker than the GaAs, which would work to increase thermal
resistance. However, this wavelength shift could also be
indicative of a slight reduction in the Al composition in the
cavity. It should be stated, then, that the low index layers in
the bottom DBR pairs are composed of AlGaAs and AlAs.
Therefore, if we assume that a reduction in Al composition is
present throughout the stack, the thermal conductivity of the
bottom DBR would decrease for the Ge wafer (due to a
reduction in the Al composition in the high-index layers [12]).
Furthermore, it may be that the threshold gain is higher for the
Ge VSCELs which could be related to a reduced optical loss
and/or increased reflectivity from the DBRs. The dominant Fig. 5: VCSEL emission wavelength at zero dissipated power as a
contribution to internal optical loss is free carrier absorption function of heatsink temperature for an 8.2 and 8.6 μm GaAs and Ge
VCSEL, respectively.
in the p-doped layers in the top DBR of the structure, and,
further, any effect of reflectivity on device performance is also
This procedure is performed at increasing heatsink
originating in the top DBR, given that the bottom DBR (+
temperatures, from which we determine the temperature
substrate) is close to 100% reflective. Therefore, these
coefficient of the emission wavelength from the intercept of
contributions to a difference in threshold requirement are
the curves, that is, the emission wavelength at zero dissipated

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]
 This article has been accepted for publication in IEEE Photonics Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JPHOT.2025.3552951

power. This is done to exclude Joule heating (as detailed in the design and growth of the epitaxial structures used in this
[14]). The temperature dependence of the extracted study.
wavelengths at zero dissipated power are shown in Fig. 5 for
the same devices as Fig. 4. For both the GaAs and Ge substrate References
devices, ∆𝜆/Δ𝑇 is 0.070 ± 0.001 nm/°C, with the error given
[1] A. Johnson et al., ‘High performance 940nm VCSELs on large
by the numerical fit. This is in agreement with the expected area germanium substrates: the ideal substrate for volume manufacture’, in
shift given by the temperature coefficient of the refractive Vertical Cavity Surface Emitting Lasers XXV, SPIE-Intl Soc Optical Eng,
index of the stack [18]. Additionally, this approach also allows Mar. 2021. doi: 10.1117/12.2583207.
us to probe the temperature dependence of the thermal [2] S. J. Gillgrass et al., ‘Impact of thermal oxidation uniformity on
resistance. In our case, we find the increase in RTh with 150 mm GaAs- and Ge-substrate VCSELs’, J Phys D Appl Phys, vol. 56, no.
15, p. 154002, Mar. 2023, doi: 10.1088/1361-6463/ACC040.
temperature to be constant for both GaAs and Ge samples [3] S.-J. Gillgrass et al., ‘Characterisation of 200 mm GaAs and Ge
(with a temperature coefficient of ~ 0.006 (°C/mW)/°C) for substrate VCSELs for high-volume manufacturing’, in SPIE Photonics West,
the devices of Fig. 4 and 5), so we only consider and present Vertical Cavity Surface Emitting Lasers XXVII, San Francisco, CA: SPIE,
data at 25°C in the subsequent discussion. Feb. 2023.
[4] S.-J. Gillgrass et al., ‘Effect of Ge-substrate thickness on 940-nm
The resulting plot of thermal resistance as a function of oxide VCSEL performance’, in Vertical-Cavity Surface-Emitting Lasers XXVIII,
aperture is shown in Fig. 6 for both GaAs and Ge devices. SPIE, Mar. 2024. doi: 10.1117/12.3001658.
From the fit of equation (5), we extract thermal conductivities [5] Y. Zhao et al., ‘Monolithic integration of 940 nm AlGaAs
distributed Bragg reflectors on bulk Ge substrates’, Opt Mater Express, vol.
of 𝜉GaAs = 0.3090 ± 0.0003 W/cm°C and 𝜉Ge = 0.3787 ± 0.0011 12, no. 3, p. 1131, Mar. 2022, doi: 10.1364/OME.452161.
W/cm°C. The uncertainties are given by the standard error of [6] J. Guo et al., ‘Study of monolithically integrated 940nm AlGaAs
the numerical fit. The correction to the active diameter, b, distributed Bragg reflectors on graded GaAsP/bulk Si substrates’, Optical
which was included as a term in the numerical fit and is given Materials Express, Vol. 13, Issue 4, pp. 1077-1091, vol. 13, no. 4, pp. 1077–
by 1.7 ± 0.2 and 1.53 ± 0.04 μ m for Ge and GaAs, 1091, Apr. 2023, doi: 10.1364/OME.484840.
[7] Z. Wan et al., ‘Monolithically integrated 940 nm VCSELs on
respectively. The extracted thermal conductivity values are bulk Ge substrates’, Opt Express, vol. 32, no. 4, p. 6609, Feb. 2024, doi:
both reduced compared to the literature values of the thermal 10.1364/OE.513997.
conductivity of bulk Ge (0.55 W/cm°C) and GaAs (0.44 [8] Y.-C. Yang et al., ‘25 Gb/s NRZ transmission at 85°C using a
W/cm°C) [19], which is expected given the presence of high-speed 940 nm AlGaAs oxide-confined VCSEL grown on a Ge
AlGaAs in the bottom DBR, as well as interface contributions, substrate’, Optics Letters, Vol. 49, Issue 3, pp. 586-589, vol. 49, no. 3, pp.
and potentially non-ideal thermal contact with the heat sink. 586–589, Feb. 2024, doi: 10.1364/OL.509988.
[9] Y. Lin et al., ‘Monolithic integration of AlGaAs distributed
We find the thermal conductivity of the Ge substrate VCSELs Bragg reflectors on virtual Ge substrates via aspect ratio trapping’, Optical
to be ~ 18% lower than that of GaAs, which is on the order of Materials Express, Vol. 7, Issue 3, pp. 726-733, vol. 7, no. 3, pp. 726–733,
the difference of the bulk values. Therefore, we show that the Mar. 2017, doi: 10.1364/OME.7.000726.
higher thermal conductivity of Ge can indeed be leveraged to [10] T. Watanabe et al., ‘1380 nm VCSELs using surface-activated
improve the thermal performance of AlGaAs-based VCSELs. bonding of GaAs-based DBRs on a Ge substrate’, Applied Physics Express,
vol. 18, no. 1, p. 016507, Jan. 2025, doi: 10.35848/1882-0786/ADAA4C.
[11] J. Baker, C. P. Allford, S. Gillgrass, J. I. Davies, S. Shutts, and P.
M. Smowton, ‘Improved Thermal Performance of VCSELs on Germanium
Substrates’, Conference Digest - IEEE International Semiconductor Laser
Conference, 2024, doi: 10.1109/ISLC57752.2024.10717337.
[12] M. A. Afromowitz, ‘Thermal conductivity of Ga1−xAlxAs
alloys’, J Appl Phys, vol. 44, no. 3, pp. 1292–1294, Mar. 1973, doi:
10.1063/1.1662342.
[13] J. Baker et al., ‘VCSEL Quick Fabrication for Assessment of
Large Diameter Epitaxial Wafers’, IEEE Photonics J, vol. 14, no. 3, 2022,
doi: 10.1109/JPHOT.2022.3169032.
[14] M. Daubenschüz, ‘Simplified Determination of the Thermal
Resistance of Vertical-Cavity Surface-Emitting Lasers’.
[15] M. H. MacDougal, J. Geske, C. K. Lin, A. E. Bond, and P. D.
Dapkus, ‘Thermal impedance of VCSEL’s with AlOx-GaAs DBR’s’, IEEE
Photonics Technology Letters, vol. 10, no. 1, pp. 15–17, Jan. 1998, doi:
10.1109/68.651085.
[16] T. Flick, K. H. Becks, J. Dopke, P. Mättig, and P. Tepel,
‘Measurement of the thermal resistance of VCSEL devices’, Journal of
Instrumentation, vol. 6, no. 01, p. C01021, Jan. 2011, doi: 10.1088/1748-
0221/6/01/C01021.
[17] W. Nakwaski and M. Osiński, ‘Thermal resistance of top-surface-
Fig. 6: Thermal resistance as a function of oxide aperture diameter for emitting vertical-cavity semiconductor lasers and monolithic two-
GaAs and Ge substrate VCSELs. dimensional arrays’, Electron Lett, vol. 28, no. 6, pp. 572–574, 1992, doi:
10.1049/EL:19920361.
[18] J. Talghader and J. S. Smith, ‘Thermal dependence of the
ACKNOWLEDGMENT refractive index of GaAs and AlAs measured using semiconductor multilayer
optical cavities’, Appl Phys Lett, vol. 66, no. 3, p. 335, Jan. 1995, doi:
Device fabrication was carried out at the Institute for 10.1063/1.114204.
Compound Semiconductors (ICS) cleanroom at Cardiff [19] P. D. Maycock, ‘Thermal conductivity of silicon, germanium,
University. The authors would like to acknowledge IQE for III–V compounds and III–V alloys’, Solid State Electron, vol. 10, no. 3, pp.
161–168, Mar. 1967, doi: 10.1016/0038-1101(67)90069-X.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see [Link]