Cryogenic computing requires an energy-efficient optical data link from 4 K to the end-user at room temperature. Laser spectra and light–current–voltage family curves over a wide temperature range from 2.6 to 295 K are reported on oxide-vertical cavity surface emitting lasers coupled with OM4 fiber. Non-linear shifting of the lasing wavelength vs junction temperatures is measured with the wavelength shifting coefficients <0.005 nm/K at 2.6 K. 12.5 Gb/s non-return-to-zero error-free data transmission at 2.6 K to the error detector at 295 K is demonstrated as a viable solution for cryo-computing.

The demand for computing power driven by the boost of the internet (IT) economy, deep learning, and IoT requires urgent development of the energy-efficient computer architecture. Cryogenic computing, such as a superconducting processor based on rapid-single-flux-quantum (RSFQ) and future quantum computing, is a promising solution to classical computer architecture.1 To take advantage of cryogenic computing technology, high-speed and energy-efficient data transfer is required from below 4 K to the end-user at room temperature. For cryogenic data transfer from 4 K to room temperature, the optical data link is significantly better than the electrical data link, which suffers from low thermal resistance, frequency-dependent conductor loss, dielectric loss, and radiation loss. In fact, 850 nm vertical cavity surface emitting lasers (VCSELs) based optical links are widely used for short-haul (<100 m) in the data center,2,3 and the cryogenic VCSEL is a promising solution to a cryogenic optical transmitter because of its high modulation bandwidth (>50 GHz) at a low bias current (∼2 mA). We reported a record direct modulated laser (DML) bandwidth for a sub-100 fJ/bit optical link.4–6 

In the previously proposed cryo-computing data link, the ultra-low-power superconducting processor electrical data output drives a Cryo-VCSEL at 77 K, converting an electrical signal to an optical signal via energy-efficient direct modulation. The electrical link between the cryo-computer chip and the Cryo-VCSEL mainly consists of a superconducting RSFQ circuit for ultra-low power data processing with the signal output level (2–4 mVpp) before transmitting to a wideband low-noise amplifier,7 and the amplifier output can drive a laser diode or an optical modulator for fiber-optic communication links. The optical signal is then transmitted through a direct fiber-coupling to the end-users at room temperature.8 

The first high-speed Cryo-VCSEL with a 16 μm diameter aperture by a proton-implanted technique was reported in 1996 with laser operation down to 6 K with laser threshold, ITH ∼2.8 mA.9 This device demonstrated an impressive modulation bandwidth of 10.8 GHz at 10 K and 2 Gb/s NRZ eye-diagram data transmission use free-space optics at 77 K.

Late in 2012, a high-speed 850 nm Oxide-VCSEL with a 5.5 μm diameter aperture was reported with laser operation down to 88 K and delivered a low laser threshold, ITH = 0.38 mA at 120 K.10 This device demonstrated 10 Gb/s NRZ eye diagram using free-space optics at a low biased current, I = 1 mA operated at 145 K.

In this work, a high-speed packaged oxide-VCSEL optically coupled to OM4 fiber is developed for the optical data link operated from room temperature down to 2.6 K in the cryostat provided by the National Institute of Standards and Technology (NIST). The measured laser spectrum and light–current–voltage (L–I–V) family curves characterize a VCSEL with laser threshold, ITH = 0.177 mA at 2.6 K. The fiber-coupled Cryo-VCSEL demonstrated 12.5 Gb/s error-free data transmission from the cryostat at 2.6 K to the error detector (ED) at room temperature. In addition, the packaged Cryo-VCSEL shows no performance degradation after more than 20 rapid thermal cycles from 300 to 2.6 K.

The Cryo-VCSEL epitaxial layer design optimizes the threshold current and the differential gain at cryogenic temperature by tuning the cavity resonance to gain spectrum peak offset, as illustrated in previous works.4,5

The Cryo-VCSEL device fabrication begins with forming an Ohmic contact on top p-doped distributed Bragg reflector (DBR) pairs. After that, low pressure inductively coupled plasma dry etching is used to form a 15 μm diameter DBR mesa. With the deposited top metal, the sample is sent to the furnace filled with high-temperature water vapor to form a 6.8 μm oxide-confined aperture. Subsequently, the AuGe/Ni/Au metal is deposited and annealed to achieve an n-type Ohmic contact. Moreover, to reduce the parasitic capacitance of VCSELs, benzocyclobutene (BCB) having a low dielectric constant are used for device planarization.11,12 Finally, the high-speed device is finished by via etching and Ti/Au metal evaporation to finish the coplanar waveguide (CPW) and interconnect. The fabricated Cryo-VCSELs with 6.8 μm aperture are tested and qualified using an on-wafer fiber probing technique before the wafer is diced into an individual Cryo-VCSEL die.

The qualified Cryo-VCSEL die is transferred to a gold-plated oxygen-free copper package, as shown in Fig. 1. After aligning the Cryo-VCSEL die with the coplanar waveguide (CPW) on the gold plated ceramic substrate, cryogenic epoxy is used to secure the Cryo-VCSEL die on the copper substrate. Subsequently, 1-mil gold wires are used to connect both the signal pad and the ground pad of the Cryo-VCSEL die to the CPW. The Cryo-VCSEL can be directly modulated through an RF cable using a cryogenic SMPM connector. Our previous work demonstrated the measured light emission spectrum and L–I–V curves of fabricated Cryo-VCSELs using a free-space coupling through the optical window installed on the cryostat for the temperate range down to 5 K.4–6,8 The radiation loss of a cryostat optical window limits the further study of Cryo-VCSELs below 5 K in the free-space optical coupling cryostat. Thus, a direct fiber coupling to the Cryo-VCSEL method is adopted to reduce the radiation loss and permit us to study Cryo-VCSEL characteristics and high-speed modulation below 5 K.

FIG. 1.

The VCSEL chip is assembled with a high-speed PCB package and coupled with OM4 fiber for optical output. The SMPM connector and a microwave low-loss cable were used for the RF electrical input. The inset is an SEM picture of the fabricated high-speed cryogenics oxide-VCSEL.

FIG. 1.

The VCSEL chip is assembled with a high-speed PCB package and coupled with OM4 fiber for optical output. The SMPM connector and a microwave low-loss cable were used for the RF electrical input. The inset is an SEM picture of the fabricated high-speed cryogenics oxide-VCSEL.

Close modal

To couple the light emitting from the Cryo-VCSEL aperture, the OM4 multimode fiber needs to precisely align to the Cryo-VCSEL aperture (<1 μm tolerance) to achieve 70% coupling efficiency at room temperature. MasterBond cryogenic epoxy is used to secure the fiber to the Cryo-VCSEL. After the Cryo-VCSEL optical package is completed, it can be placed in the cryogenic chamber (from NIST) for the laser spectrum, L–I–V, and high-speed data rate measurements for temperatures from 300 down to 2.6 K. To guarantee the optical coupling efficiency in a wide temperature range, the thermal expansion coefficient (from 300 to 2.6 K) of all different parts used in the Cryo-VCSEL package shown in Fig. 1 needs to be precisely compensated.

The temperature-dependent DC characterization of the Cryo-VCSEL from 295 to 2.6 K is performed in a close-loop cryostat using the SHI Cryogenics RDK-101DL cryocooler. The DC bias current is supplied to the Cryo-VCSEL package through semi-rigid SMA cables with proper thermal lagging. For the optical path inside the cryostat, a 1-m OM4 multimode fiber glued on the top mesa of the Cryo-VCSEL directly connects to the customized vacuum fiber feedthrough. For the optical path outside the cryostat, an OM4 fiber installed on the vacuum fiber feedthrough can couple laser light into an optical spectrum analyzer or a power meter for the temperature-dependent measurements of the laser spectrum and L–I–V curves.

Figure 2(a) displays the Cryo-VCSEL laser spectrum operated at 295 K with a relatively high ITH = 4.97 mA and a bias I = 10 mA for I/ITH 2. The fundamental and first order modes are 893.88 and 893.25 nm as labeled with mode spacing Δλ = 0.63 nm. The mode spacing confirmed that the effective optical aperture diameter is around 6.2 μm. Figure 2(a) also shows the laser spectrum at 2.6 K with ITH = 0.177 and I = 0.4 mA. The fundamental and first order modes of the Cryo-VCSEL are 879.81 and 879.41 nm, respectively.

FIG. 2.

(a) Laser spectrum for the Cryo-VCSEL package operating at 295 and 2.6 K. For all cases, the spectrum analyzer reference level is down to −77 dBm, resolution bandwidth = 0.08 nm, and span = 10 nm. (b) Non-linear shifting of the Cryo-VCSEL lasing wavelength vs extracted junction temperatures.

FIG. 2.

(a) Laser spectrum for the Cryo-VCSEL package operating at 295 and 2.6 K. For all cases, the spectrum analyzer reference level is down to −77 dBm, resolution bandwidth = 0.08 nm, and span = 10 nm. (b) Non-linear shifting of the Cryo-VCSEL lasing wavelength vs extracted junction temperatures.

Close modal

Moreover, from 295 to 130 K, the fundamental mode wavelength shifts from 893.880 (295 K) to 882.938 nm (130 K). The laser cavity wavelength coefficient (dλ/dT) is calculated to be −0.0663 nm/K (based on the ambient temperature). As the ambient temperature drops from 130 to 2.6 K, the laser cavity wavelength coefficient is estimated to be −0.0245 nm/K, accordingly. These results indicate that the laser cavity wavelength coefficient is non-linearly dependent on the ambient temperature. Similar results are observed in earlier reported 16 μm aperture proton implanted Cryo-VCSEL (−0.0333 nm/K).9 To better interpret the temperature-dependent laser cavity wavelength shifting at cryogenic temperature, the self-heating of Cryo-VCSELs is be excluded in the detailed analysis based on the junction temperature.

The junction temperature of the Cryo-VCSEL is estimated using accurate bias current dependent optical spectra at different ambient temperatures. First, the Cryo-VCSEL self-heating power at each bias current can be easily calculated using measured L–I–V curves. Then the relation between the self-heating power and VCSEL fundamental mode at each ambient temperature is obtained. By extrapolate each curve to the zero self-heat power point, where the junction temperature equals to the ambient temperature, the fundamental mode wavelength at each junction temperature is extracted. Figure 2(b) shows the non-linear shifting of the Cryo-VCSEL lasing wavelength at various extracted junction temperatures. From 130 to 2.6 K, the laser cavity wavelength coefficient continuously decreases from −0.040 at 130 to −0.005 nm/K at 2.6 K, as shown in Fig. 2(b). The non-linear shifting of the cavity resonance mode can be explained by the effective complex relative dielectric constant (polarization vector and resonant polarization) of the wave propagation medium inside the VCSEL cavity at cryogenic temperature. The reduction in laser cavity size due to thermal contraction also contributes to the non-linear cavity resonance mode shifting, because laser cavity size reaches “constant” as the temperature approaches 4 K.13 

When the VCSEL is operated at cryogenic temperatures, the wavelength of the cavity resonance mode and the gain peak shift with the temperature at different rates, resulting in radical lasing threshold increase and output optical power decline. To alleviate this issue, the VCSEL gain profile is detuned relative to the cavity resonance at 300 K so that the gain peak can align to the cavity resonance at cryogenic temperature.4–6,8–10 The L–I–V (light–current–voltage) curves of the packaged 6.8 μm aperture Cryo-VCSEL are characterized from 295 down to 2.6 K, as shown in Fig. 3. At room temperature, the L–I curve slope slowly declines as the current increases. This can be attributed to the photon generation rate that is thermally limited via the e–h recombination process of the junction temperature rising above ambient temperature caused by current inject self-heating.

FIG. 3.

Temperature-dependent L–I–V curves for the Cryo-VCSEL package operating at 295 and 2.6 K. The inset shows the measured VCSEL threshold current decreases from 4.97 to 0.177 mA as the temperature drops from 295 to 2.6 K.

FIG. 3.

Temperature-dependent L–I–V curves for the Cryo-VCSEL package operating at 295 and 2.6 K. The inset shows the measured VCSEL threshold current decreases from 4.97 to 0.177 mA as the temperature drops from 295 to 2.6 K.

Close modal

In contrast, the L–I curve laser power output improved by 3.3 times, and the slope is nearly constant as the current increases up to 8 mA at 2.6 K. This implies e–h recombination lifetimes are reduced by 3.3 times and limited by the carrier injection rate. Also, the VCSEL self-heating has little effect at 2.6 K than at room temperature. A linear light power to the current relation is essential to VCSELs for a wide dynamic range signal modulation.

As the ambient temperature drops, the Cryo-VCSEL threshold current first drastically decreases from 4.97 at 295 K (room temperature) to 0.45 at 77 K, and finally drops to 177 μA at 2.6 K, as shown in the inset of Fig. 3. The temperature-dependent Cryo-VCSEL threshold current is nearly constant below 20 K, confirming that the gain peak and the Fabry–Pérot dip (cavity resonance) of the Cryo-VCSEL are adequately aligned at cryogenic temperatures.

For I–V characteristics, as the temperature decreases from 295 to 2.6 K, the turn-on voltage of Cryo-VCSELs shifts upwards and differential resistance increases. At 2.6 K, the carrier injection reduces at the heterojunction barrier because of the carrier thermal energy (proportional to kT) reduction at cryogenic temperature. The large voltage offset between 295 and 2.6 K exceeds 1.6 V when the bias current is 8 mA.

The block diagram of the Cryo-VCSEL package bit-error-rate (BER) testing setup is shown in Fig. 4. The bit-pattern-generator (BPG) and the error detector (ED) module on the Agilent N4901B serial BER tester are synchronized by an external clock source (Agilent E8257D). The Cryo-VCSEL DC bias current and the RF voltage swing from the BPG are combined via a bias-tee. The RF path inside the cryostat consists of multiple cryogenic semi-rigid segments and flexible SMA cables with proper thermal lagging. The Cryo-VCSEL converts the input RF signal from the BPG into the optical signal transmitting through OM4 fiber. The receiver side consists of Newport 1484-A-50 22 GHz high-gain GaAs photodetector and an SHF 804M broadband amplifier.

FIG. 4.

The block diagram of the Cryo-VCSEL package BER testing setup from 295 down to 2.6 K. A calibrated DT-670 Thermocouple from Lakeshore measures the package temperature.

FIG. 4.

The block diagram of the Cryo-VCSEL package BER testing setup from 295 down to 2.6 K. A calibrated DT-670 Thermocouple from Lakeshore measures the package temperature.

Close modal

For the eye diagram measurement, the amplified output signal from the photodetector-amplifier link is directly analyzed by the Agilent 86100C oscilloscope with an Agilent 86117A module. For the BER test, the photodetector-amplifier link directly connects to the ED module. In addition, for all the BER and eye diagram measurements, the test bit sequence is a non-return-to-zero (NRZ) pseudo-random-binary-sequence (PRBS7) with a peak-to-peak voltage swing from 0.5 to 0.6 V.

The 12.5 Gb/s eye-diagrams for different Cryo-VCSEL bias currents from 1.0 to 2.5 mA are measured as shown in Figs. 5(a)–5(d). The peak–peak amplitude of the electrical modulation signal is 500 mV when the DC bias current of the VCSEL is 1 mA. Ideally, the Cryo-VCSEL can be directly modulated by a very small input signal, because the lasing threshold current is only 0.177 mA. However, a non-ideal receiver, i.e., a low responsivity photodetector with high thermal noise, will significantly degrade the signal-to-noise ratio (SNR) of the received eye diagram. As the VCSEL DC bias current increases from 1 to 2.5 mA, the RF modulation is also increased to 600 mV peak–peak to obtain a higher signal-to-noise ratio (SNR). The voltage swing of 0.5–0.6 Vpp is available in a typical use case by adding a cryogenic amplifier between the cryo-computer chip and the Cryo-VCSEL. The impedance mismatch at a low bias current leads to a large reflection loss, resulting in optical modulation amplitude (OMA) degradation. When the VCSEL bias current is 1 mA, as shown in Fig. 5(a), the eye-opening is relatively small due to higher differential impedance (146 Ω), and the rising and falling crossing of the eye diagram is at nearly “0” level because the laser is operated near the threshold.

FIG. 5.

12.5 Gb/s PRBS7 NRZ eye diagrams at 2.6 K. On the top of each diagram, the electrical current specifies the VCSEL bias current. In contrast, the Vpp specifies the peak-to-peak RF modulation amplitude for each measured eye diagram. Each figure shows different VCSEL bias currents: (a) 1.0, (b) 1.5, (c) 2, and (d) 2.5 mA. The plot log (BER) vs received optical power for 12.5 Gb/s bit-error-rate-testing (BERT) in (e) indicates error-free transmission (1 × 10−12) is obtained at bias I = 2 mA.

FIG. 5.

12.5 Gb/s PRBS7 NRZ eye diagrams at 2.6 K. On the top of each diagram, the electrical current specifies the VCSEL bias current. In contrast, the Vpp specifies the peak-to-peak RF modulation amplitude for each measured eye diagram. Each figure shows different VCSEL bias currents: (a) 1.0, (b) 1.5, (c) 2, and (d) 2.5 mA. The plot log (BER) vs received optical power for 12.5 Gb/s bit-error-rate-testing (BERT) in (e) indicates error-free transmission (1 × 10−12) is obtained at bias I = 2 mA.

Close modal

As the VCSEL bias current increases from 1 to 2.5 mA, the eye-opening (OMA) increases because the Cryo-VCSEL input impedance improved toward 88 Ω. The rising and falling crossing is gradually approaching the middle point as the device bias current is well above the threshold, as shown in Fig. 5(d). Moreover, the mixed effect of intrinsic large-signal optical response of Cryo-VCSELs and frequency-dependent RF signal attenuation from the cryogenic RF path led to observable overshoot and relaxation oscillation in eye diagrams.11,12,14

In Fig. 5(e), the measured BER testing result for 12.5 Gbps is 1 × 10−6 at bias I = 1.1 mA and 1 × 10−9 at bias I = 1.5 mA. As the Cryo-VCSEL bias current exceeds 2 mA or above, error-free (BER <1 × 10−12) data transmission at 12.5 Gbps can be achieved with a confidence level of 95% (no error in 3 Tb). According to the parasitic parameter extraction based on S-parameter measured from the VCSEL on-wafer probing test,15 the overall optical responses are severely limited by the parasitic effects of the Cryo-VCSEL package and bonding wire inductance. Better light coupling efficiency and more sophisticated wire-bonding tools, such as a wedge bonder, will significantly improve the eye-opening and BER performance.

In summary, we report on the development of a high-speed direct fiber-coupled Cryo-VCSEL operated down to 2.6 K. The operation of oxide confined VCSELs and semiconductor lasers at 2.6 K are first demonstrated. Moreover, 12.5 Gbps error-free data transmission from the optical transmitter at 2.6 K to error detection (ED) at room temperature is also demonstrated. Thus, the packaged oxide-VCSEL with OM4 fiber could be a viable data link between 4 K and room temperature for cryogenic computing.

This work was supported by sdPhotonics, LLC under Dr. William Harrod (IARPA), Dr. Deppe, and Dr. T. R. Govindan (Army Research Office) on the IARPA project of Super Cables-VCSEL based Link under Grant No. W911NF-19-2-0165. The high-speed VCSEL BER test is partially supported by Dr. Mike Gerhold for ARO grant to UIUC under No. W911NF-17-P-0112. Additional work by sdPhotonics, LLC related to these results was supported by, or in part by, the U.S. Army Research Office under SBIR Phase II Contract No. W911NF-17-P-0048.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

1.
D. S.
Holmes
,
A. L.
Ripple
, and
M. A.
Manheimer
, “
Energy-efficient superconducting computing—Power budgets and requirements
,”
IEEE Trans. Appl. Supercond.
23
,
1701610
(
2013
).
2.
D. L.
Huffaker
,
D. G.
Deppe
, and
K.
Kuma
, “
Native-oxide defined ring contacts for low threshold vertical-cavity lasers
,”
Appl. Phys. Lett.
65
(
1
),
97
99
(
1994
).
3.
M.
Feng
,
C.
Wu
, and
N.
Holonyak
, Jr.
, “
Oxide-confined VCSELs for high-speed optical interconnects
,”
IEEE J. Quantum Electron.
56
(
2
),
2400115
(
2018
).
4.
D.
Deppe
,
A.
Srinivasa
,
C.
Kuznia
,
J.
Ahadian
,
M.
Feng
,
W.
Fu
,
H. L.
Wang
, and
H.
Wu
, “
Cryogenic VCSEL-link for low bit energy
,” in GOMACTech-2020 (
2020
).
5.
M.
Bayat
and
D. G.
Deppe
, “
Laser characteristics for VCSELs for 77 K and 4 K optical data application
,”
IEEE J. Quantum Electron.
56
(
3
),
2400206
(
2020
).
6.
W.
Fu
,
H.
Wang
,
H.
Wu
,
A.
Srinivasa
,
S.
Srinivasa
,
M.
Feng
, and
D.
Deppe
, “
Cryogenic 50 GHz VCSEL for sub-100 fJ/bit optical link
,” in IEEE Photonics Conference (IPC), Vancouver, BC, Canada (
2020
).
7.
O. A.
Mukhanov
,
S. V.
Rylov
,
D. V.
Gaidarenko
,
N. B.
Dubash
, and
V. V.
Borzenets
, “
Josephson output interfaces for RSPQ circuits
,”
IEEE Trans. Appl. Supercond.
7
(
2
),
2826
(
1997
).
8.
W.
Fu
,
H.
Wu
,
D.
Wu
,
M.
Feng
, and
D.
Deppe
, “
Cryogenic oxide-VCSELs with bandwidth over 50 GHz at 82 K for next-gen high-speed computing
,” in
Optical Fiber Communications Conference (OFC)
2021, June 6–10 (
2021
).
9.
B.
Lu
,
Y. C.
Lu
,
J.
Cheng
,
R. P.
Schneider
,
J. C.
Zolper
, and
G.
Goncher
, “
Gigabit-per-second cryogenic optical link using optimized low-temperature AlGaAs–GaAs vertical-cavity surface-emitting lasers
,”
IEEE J. Quantum Electron.
32
,
1347
1358
(
1996
).
10.
D. K.
Serkland
,
K. M.
Geib
,
G. M.
Peake
,
G. A.
Keeler
, and
A. Y.
Hsu
, “
850-nm VCSELs optimized for cryogenic data transmission
,”
Proc. SPIE
8276
,
82760S
(
2012
).
11.
C. Y.
Wang
,
M.
Liu
,
M.
Feng
, and
N.
Holonyak
, “
Temperature-dependent analysis of 50 Gb/s oxide-confined VCSELs
,” in
Optical Fiber Communications Conference (OFC), Los Angeles, CA
(
OSA
,
2017
), pp.
1
3
.
12.
T. Y.
Huang
,
J.
Qiu
,
C.-H.
Wu
,
H.-T.
Cheng
,
M.
Feng
, and
H.-C.
Kuo
, “
A NRZ-OK modulated 850-nm VCSEL with 54 Gb/s error-free data transmission
,” in
2019 Conference on Lasers and Electro-Optics Europe and European Quantum Electronics Conference (CLEO/Europe-EQEC 2019)
(
IEEE
,
2019
).
13.
S. I.
Novikova
,
Sov. Phys. Solid State
3
,
129
(
1961
).
14.
E. P.
Haglund
,
P.
Westbergh
,
J. S.
Gustavsson
, and
A.
Larsson
, “
Impact of damping on high-speed large signal VCSEL dynamics
,”
J. Lightwave Technol.
33
(
4
),
795
801
(
2015
).
15.
C. Y.
Wang
,
M.
Liu
,
M.
Feng
, and
N.
Holonyak
, Jr.
, “
Microwave extraction method of radiative recombination and photon lifetimes up to 85 °C on 50 Gb/s oxide-vertical cavity surface emitting laser
,”
J. Appl. Phys.
120
(
22
),
223103
(
2016
).