Characterization of heterogeneous InP-on-Si optical modulators operating between 77 K and room temperature

The silicon photonics platform is an attractive technology for very-large-scale integration of optical circuits in terms of cost, compactness, and compatibility with complementary metal–oxide–semiconductor (CMOS) microelectronics.¹ This platform provides small-footprint passive optical components at telecom wavelengths since it exhibits a high refractive index contrast and a moderately broad transparency window in the near- and mid-infrared. On the other hand, active optical components such as lasers, electro-optical (EO) modulators, and photodetectors have been effectively fabricated by heterogeneous integration with germanium and III–V compound semiconductors.^2,3

While most of the pursuit application areas for silicon photonics have centered around room temperature,^4,5 there are a large number of applications that demand operation at low temperatures including cryogenic image sensors and next-generation quantum computing systems. Those applications are characterized by generating a massive amount of electrical data beyond the bandwidth of copper cables. Optical interconnect can alleviate this limitation, providing increased bandwidth at low power consumption to transmit the data over the fiber to room temperature.⁶ Another noteworthy application is outer space exploration, where extreme temperatures and radiation fields can seriously compromise the performance of optical devices.^7–10 Although silicon passive components work efficiently at low temperatures,¹¹ active devices might suffer from several detrimental effects, such as cryogenic incomplete ionization¹² and lattice constant mismatch when different materials are integrated.² To expand the application of photonic integrated circuits (PICs) for such cutting-edge applications, optimized designs and performance evaluations in such unconventional conditions are required.

In this letter, electro-optic (EO) modulators based on III–V materials bonded on silicon are fabricated and characterized between 77 K and room temperature. At such low temperatures, the electro-optical refractive index change in III–V is mainly due to three phenomena: the Pockels effect, the Franz-Keldysh (FK) effect, and the quantum confined Stark effect (QCSE).^13–16 The other contributions (i.e., band-filling, band-shrinkage, and plasma dispersion effects) become negligible since the free carriers freeze out.¹⁷ 

The two kinds of modulators investigated in this work are named bulk modulator and quantum-well (QW) modulator, which are based on the Franz-Keldysh effect and the quantum confined Stark effect, respectively. Due to the rather wavelength insensitivity of the Franz-Keldysh effect, the bulk modulators have been demonstrated to operate with smaller variation of V_π over a wider range of temperatures compared to QW modulator. Both devices have shown an electro-optic bandwidth as large as 2 GHz over the range of temperatures between 77 K and 295 K. Such modulation bandwidth is limited by the series resistance and junction capacitance of the modulators, and it is expected to be enlarged by future design optimization. In addition, these modulators require lower doping concentration compared to silicon modulators based on free-carrier plasma effect, and as a result, they suffer less from free-carrier induced insertion losses.¹⁸ 

The cross section of the heterogeneous silicon photonics modulators is schematically shown in Fig. 1.

The device consists of a heterogeneous waveguide fabricated from bonding a III–V stack to a silicon-on-insulator (SOI) wafer. The optical mode in this heterogeneous cross section extends in both silicon and III–V region, and the confinement is controlled by varying the silicon waveguide width and etch depth. The bandgap, doping, and confinement factor in the EO layer are properly optimized to achieve large electro-optic effect. In the device under investigation, the III–V mesa width is 3 μm wide, while the silicon waveguides are 500 nm thick and 600 nm wide, with a shallow etch of 231 nm.

Since the interband absorption strongly reduces the optical signal, the EO layer is designed to have a photoluminescence peak at least 100 meV larger than the photon energy at the operating wavelength.¹⁹ As a result, for λ = 1550 nm (0.8 eV), the interband absorption is set to be around λ_g = 1370 nm (0.9 eV).

The QW modulator investigated here was previously optimized to exploit carrier depletion and the quantum confined Stark effect at room temperature.^20,21 The epitaxial layers are listed in Table I, where the EO layer consists of 15 compressive (+0.3%) QW with 16 tensile (−0.41%) barriers. Top and bottom separate confinement heterostructures (SCH) are shaped to achieve strain balance and large mode confinement in the QW layer. Both the SCH layer and multi-QWs are slightly n-doped to improve the high speed performance.^21,22 The measured photoluminescence peak of the fabricated stack is at 1364 nm.

As a comparison, a bulk modulator based on the Franz-Keldysh effect is designed to operate at room temperature. The epitaxial layers are listed in Table II, where the multi-QW layers in Table I have been replaced with a p-n junction. The composition of InAlGaAs is set to have a bandgap of 1376 nm, while the thickness and the doping of the p and the n layers are chosen to maximize the electro-optic index variation. To reach large field confinement, InP is used to replace the top and bottom SCH in the previous stack, since it holds a smaller optical refractive index.

The energy bandgaps and the refractive index variations are computed as a function of the In, Al, and Ga concentration, and the temperature by using the fitting formulas reported in the literature.^19,23 For the materials in Tables I and II, the optical refractive indices do not change significantly in the range 77 K–295 K (e.g., Δn_QW = 4.4 × 10⁻² and Δn_bulk = 3.2 × 10⁻² and similarly for the other alloys), resulting in an almost constant mode confinement over temperature (i.e., ≃27% in the QW and ≃62% in the bulk). On the other hand, as shown in Fig. 2, the energy-gap increases by about 68 meV at 77 K in both bulk and QW. This causes a weaker electro-optic effect at lower temperature, as shown later by the experimental results.

The modulators are fabricated in a Si/InP heterogeneous integration wafer run at the University of California, Santa Barbara.^2,24 The process starts on 4″ silicon-on-insulator (SOI) wafers where the passive waveguides are defined by 248 nm deep ultraviolet (DUV) photolithography and dry-etched in fluorine based pseudo-Bosch etching processes. InP chiplets, diced out from 2″ wafers grown by metal-organic chemical vapor deposition (MOCVD), are directly bonded to a prepatterned SOI wafers by oxygen plasma-assisted bonding process. After the InP substrate is removed mechanically/chemically, the InP structures are formed by CH₄/H₂/Ar based dry-etch and HCl or H₃PO₄ based wet etches. A sputtered oxide layer of 1 μm thickness is deposited, acting as a dual purpose passivation layer for InP structures and optical cladding for the passive silicon waveguides. Optimized metal stacks of Pd/Ti/Pd/Au and Pd/Ge/Pd/Au are, respectively, used for p and n metal contact to achieve a specific contact resistance in order of 10⁻⁶ Ω cm². Finally, the chips are diced out and the facets are mechanically polished for efficient fiber coupling in testing.

Figure 3(a) shows the layout of the heterogeneous InP on silicon photonic modulator chip used for the characterization. The test structures are grouped in 3 subsets where the length (L) of the heterogeneous section is varied (i.e., L = 600 μm, 1 mm, and 2 mm). Each subset consists of a waveguide modulator, a passive silicon waveguide and a Mach-Zehnder interferometer (MZI) in push-pull configuration [from top to bottom in Fig. 3(a)]. The waveguide modulator is used to extract the electro-optic absorption, the passive silicon waveguide is considered as a reference, while the value of V_π · L is extracted from transmission spectra of the Mach-Zehnder interferometer at different applied voltages.

One key aspect of these heterogenously integrated devices is the manufacturing of the mode converter between the InP modulator and the passive silicon waveguide. Figure 3(b) shows the scanning electron microscope image of such a transition confirming the high-quality of the fabrication.²⁴ 

The devices were tested in a cryogenic probe station (Lakeshore TTPX) equipped with two lensed fibers for chip butt-coupling, two direct-current (DC) electrical probes, and one radio-frequency (RF) probe. The optical transmission spectrum of the device is characterized with a linearly tunable continuous wave single-mode external cavity diode laser between 1460 nm and 1580 nm. A polarization controller, external to the cryogenic probe station, is used to maximized the transverse-electric (TE) mode launched in the device. The output power is finally measured with a photodiode.

To perform the characterization at different cryogenic temperatures, the chamber is first vacuum pumped to a pressure below 1 mTorr. Then, a heat bath is cooled down to 77 K by injecting liquid nitrogen into the refrigerator of the cryostat. The temperature is then increased with a step of 50 K up to room temperature. For each measurement, a high precision PID controller (Lakeshore 336 Temperature Controller) is used to maintain a constant temperature.

Through several thermal cycles, no degradation of device performance is observed that proves the durability of III–V bonding on silicon wafer at low temperature.

The current-voltage characteristics for both QW and bulk modulators is performed in the cryogenic probe station at different temperatures with a precision source measure unit (Keysight B2902A). When the temperature decreases, a significant drop is observed in the electrical current for a specific applied voltage, as shown in Fig. 4. This is due to the minor ionized dopants at lower temperatures and, consequently, fewer free carriers.¹² 

The voltage required for inducing a phase change of π (i.e., V_π) in reverse bias is measured from the output spectrum of the MZI. When the applied voltage is swept, the optical refractive index in the heterogeneous cross section changes, resulting in a modulated output signal. The value of V_π is the difference between the voltage values that cause a constructive (maximum) and destructive interference (minimum) in the output spectrum. The results of this measurement are shown in Fig. 5 for different temperatures when the voltage is applied to one arm of the MZI only. When the voltage is very large, the electro-optic absorption becomes dominant and the light can propagate only in the unmodulated arm, resulting in a constant output power for large reverse bias voltage. The noise at low temperatures is caused by the vacuum pump vibrations that affect the fiber-to-chip coupling, while the pump is kept off at room temperature. To avoid potential damages in the modulators, the applied reverse voltage is limited to −7 V.

The negligible power dissipated in reverse bias allows us to exclude any contribution of the thermal effect to the refractive variation, and so to the measured V_π. Indeed, the thermal impedence of the modulators is expected to be about 40 °C/W, since their layout is analogous to the hybrid silicon evanescent lasers,²⁵ while the thermo-optic coefficient²⁶ of III–V is about 2 × 10⁻⁴/°C at room temperature. As a result, a dissipated power of 1 μW might cause an index variation of 8 × 10⁻⁹/°C. At lower temperature, the dissipated power in the modulators decreases by more than 2 order of magnitudes and the thermo-optic coefficients of InAlGaAs is expected to drop similarly to the one of silicon.²⁷ 

The extracted values of V_π are multiply by the modulator length to be independent of the geometry and are shown in Fig. 6 for different laser wavelengths as a function of the temperature. Since the longest test structure is 2 mm-long and the minimum safe reverse bias is −7 V, values of V_π · L larger than 14 V mm cannot be estimated. For this reason, few data points are missing for QW modulator at wavelengths 1480 nm and 1500 nm.

At room temperature, the bulk modulator slightly outperforms the QW modulator because of its smaller energy-gap. On the other hand, at lower temperatures the V_π · L increases more rapidly for the QW-based modulator, which is explained by the stronger wavelength dependance of the QCSE.^28,29 Specifically as shown in Fig. 2, when the temperature drops the divergence between the photon energy at the operating wavelength and the bandgap increases. As a result, although the QW modulator is known to be more efficient, the bulk-modulator is proven to be more suitable for working over a wide range of temperatures.

The electro-optic absorption is measured from the transmission spectrum of the waveguide modulator test structures by varying the applied voltage from −7 V up to +1 V with a step of 1 V. This measurement is repeated at different temperatures. The results are summarized in Fig. 7 for a reverse bias voltage of −3 V. Similarly to what observed for the values of V_π · L, the absorption decreases at lower temperature due to the larger energy-gap.

From the previous experiments, the electro-absorption per unit length (i.e., Δα) is derived and plotted in Fig. 8 for different temperatures and different reverse bias. The quadratic behavior of Δα with respect to the applied voltage (dashed lines in Fig. 8 for eye guidance) suggests that the Franz-Keldysh effect (bulk) and the quantum confined Stark effect (quantum-well) are still the dominant electro-optic effects at low temperature despite of the energy band edge blue-shift.³⁰ 

The larger electro-absorption observed in bulk-modulators is due to the higher confinement factor in the EO layer and the less sensitivity to the absorption edge. Indeed, although the QW is known to have a larger electro-optic effect than bulk, the QCSE also decays more rapidly with temperature than Franz-Keldysh effect when the energy of the photons are far from the photoluminescence peak.³⁰ 

Figure 9 sketches the bandwidth measurement setup used in our tests. A vector network analyzer (VNA), Agilent E8361A generates the RF signal sweep. The RF signal is mixed by means of a broadband bias-tee with the DC bias. A broadband RF probe is then used to electrically probe the device. A benchtop tunable laser source generates the 1550 nm optical carrier to be modulated. The optical signal, at the output of the modulator, is received by a lightwave component analyzer (LCA) N4373A, which is then connected to the receiving port of the VNA. The device has been tested in the frequency range from 50 MHz to 20 GHz. The experimental results at 77 K and at room temperature are shown in Fig. 10 for both the QW and bulk modulators. In the figure, the electro-optic response (i.e., the scattering coefficient S₂₁ in Fig. 10) is normalized to its maximum value. As a comparison, the frequency response of S₂₁ is fitted with the single pole response A/(1 + j2πfτ), where A is the amplitude, f is the frequency, and τ is the time constant. The 3 dB bandwidth is estimated from the fitted curve as BW_3dB = 1/(2πτ) and its value is reported in Table III for different temperatures and two reverse bias voltages. The bandwidth of the present modulators is limited by the serial resistance and the junction capacitance. Comparing the experimental results, not a significant variation is observed in the range of temperatures.

In this work, we have demonstrated that heterogeneous silicon photonic modulators, fabricated with oxygen plasma-assisted wafer bonding, can successfully operate at temperatures as low as 77 K with a dissipated power below 20 pW. A 3dB-bandwidth of about 1.5 GHz is measured at 77 K, which is limited by the serial resistance and junction capacitance of the modulators. As a result, higher modulation bandwidth is expected by optimizing the cross section and probe pad design. The experimental results highlighted that bulk-modulators can operate over a wider range of temperature with smaller change of V_π compared to QW modulators since they are less sensitive to energy-gap variations, which is a consequence of the higher sensitivity of the QCSE to wavelength variation with respect to the FK effect (i.e., bulk modulator). While the modulator experimental data shown in this letter are for temperatures as low as 77 K due to the limitation of the measurement equipment, we believe such modulators can operate at lower temperature. The results in this letter can pave the way to new applications of PICs beyond the data center and the telecommunication fields.

The authors would like to thank Lin Chang, Mihir Pendharkar, and Songtao Liu from the University of California Santa Barbara; Duanni Huang from Intel Labs; Tin Komlienovic from Nexus Photonics; and Martin Gustafsson and Richard Lazarus from Raytheon BBN Technologies for the useful discussion and insightful advice.

This material is based upon work supported by the Army Research Office under Contract No. W911NF-19-C-0060. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the Army Research Office and IARPA.