With the explosive growth in global data traffic driven by bandwidth-hungry applications such as artificial intelligence and cloud computing^[1–4], the demand for large bandwidth and power-efficient electro-optic (EO) modulators remains relentless. EO polymers have long been considered as a paradigm-shifting material due to the high EO coefficient, fast response, and low dispersion^[3–9]. Silicon-organic hybrid (SOH) modulators, which combine the advantages of both organic and inorganic materials, have demonstrated impressive performance, with low driving voltage and compact size^[10–15]. To enhance the interaction between the optical field, electric field, and the EO polymer, slot waveguide is a prevalent solution^[16–21]. It provides significantly larger optical confinement compared to conventional waveguides, e.g.,  strip waveguides. However, the narrow and deep slot is susceptible to fabrication errors, which complicates the infiltration and poling of the polymer.

Thin-film silicon waveguides offer a competitive alternative^[22–25]. With a waveguide core thickness less than one-third of the conventional thickness commonly used in commercial foundries (e.g.,  220 nm), the modes in thin-film waveguides extend significantly into the cladding, achieving confinement factors even higher than slot waveguides. In the meantime, the interaction between guided modes and rough sidewalls is substantially reduced, leading to a significantly lower propagation loss in thin-film SOH waveguides compared to slot waveguides^[23]. A recent demonstration of a thin-film silicon Mach–Zehnder interferometer (MZI) modulator achieved a $V\pi L$ of 1.4 V·cm and a 70 GHz bandwidth using 40 nm-thick silicon waveguides^[24]. Instead of the widely adopted transverse electric (TE) modes, the modulator operates with the fundamental transverse magnetic (TM) mode, which requires a vertical electric field for polling the EO polymer and driving the modulator. Although the smooth electrode surface formed by deposition could increase the poling efficiency, the vertical electrode configuration is incompatible with the current silicon-on-insulator (SOI)-based silicon photonics integration process. An attempt to use the conventional fundamental TE mode achieved a $V_{\pi }L$ of 1.9 V·cm on SOI with a 50 nm-thick single-crystal silicon^[25]. However, the resistivity of the EO polymer in the work is relatively small. Thus, side cladding with suitable resistivity was required to enhance mode confinement and form an electric field with adequate intensity for effective poling.

In this study, we experimentally demonstrate a TE thin-film SOH modulator on SOI with 70 nm-thick single-crystal silicon cores. This thickness is compatible with the standard silicon fabrication process in commercial foundries and can be integrated with other silicon photonic devices. Benefiting from the high-volume resistivity of the EO polymer SEO100^[26], high poling efficiency can be achieved without side claddings, which simplifies the device design and fabrication. The device shows a bandwidth over 33 GHz, which is limited by the bandwidth of the testing equipment. The stability of the unpackaged modulator is experimentally verified in a long-term damp heat environment.

Figure 1(a) illustrates the structure of the thin-film SOH EO modulator. The modulator employs the classic asymmetric MZI configuration with a 45 µm length difference between the two arms. The coplanar ground–signal–ground (GSG) traveling wave electrode of 600 nm thick is leveraged for poling and modulation. The single-crystal silicon core layer is 70 nm thick. The inset in Fig. 1(a) shows the field distribution of a typical fundamental TE mode along with the cross-section of the thin-film waveguide. The refractive index of the EO polymer is set as 1.70 to emulate the SEO100 polymer. Figure 1(b) shows the molecular structure of SEO100, which is composed of the chromophore molecule AJLZ53 and the host polymer amorphous polycarbonate (APC)^[8]. AJLZ53 chromophore is a highly efficient and photochemically stable dipolar nonlinear optical chromophore developed based on the perturbational molecular orbital theory of Dewar’s rule. It can be doped into APC at a relatively high loading level. The polymer composite shows a high-volume resistivity of $1.3\times {10}^9\mathrm{\Omega }·\mathrm{m}$ at a glass transition temperature ($T_g$) around 140°C and exhibits an exceptional combination of large EO coefficients, good thermal stability, and low optical loss^[9]. Notably, a significant portion of the optical field extends outside the waveguide, which enhances the effective integration of the EO polymer onto SOI. The width of the thin-silicon waveguide is simulated to ensure the lossless propagation of a single fundamental TE mode. Figure 1(c) depicts the field interaction factors of thin-silicon waveguides as a function of waveguide width. For comparison, the dashed line in the same figure shows the interaction factor of a 70 nm wide slot waveguide for comparison^[16,20]. Although a narrower waveguide can yield a higher interaction factor, the electrodes must be put farther apart to avoid the absorption loss induced by metals. Figure 1(d) shows the variation of optical loss as a function of waveguide width and the gap between the core and electrode. Considering the trade-off between optical loss and modulation efficiency, the gap and waveguide width are selected to be 1.2 and 1.1 µm, respectively, corresponding to a field interaction factor ($\mathrm{\Gamma }$) of 43.5 % according to the definition of $\mathrm{\Gamma }$ in Ref. [27]. The 50 ohmic GSG electrodes are designed accordingly. Figure 1(d) also shows that a narrower gap can further improve the modulation efficiency without increasing the propagation loss of the waveguide. However, considering the misalignment of the in-house fabrication, a more conservative gap value is used.

The primary fabrication steps of the modulator are summarized in Fig. 2(a), with optical microscopy images of the device after dry etching, lift-off, spin casting of the EO polymer, and poling shown in the inset. The patterning of the thin-film silicon waveguide is accomplished using electron-beam lithography (EBL, nb5, Nanobeam) and inductively coupled plasma (ICP) etching (GSE200Plus). The traveling wave electrodes are formed through a conventional lift-off process. The EO polymer with a weight ratio of 35% (mass fraction) is spin-coated on the device, forming a 1.5 µm thick film, and is subsequently dried at 85°C in a vacuum oven overnight to remove the residual solvent. The EO polymer is polled at 120°C by applying a voltage of 350 V across the two ground electrodes, corresponding to a poling electric field of 100 V/µm^[28]. Figure 2(b) shows the poling voltage (black), leakage current (red), and poling temperature (blue) curves collected from the modulator poling process. Once the poling voltage of 350 V is reached, the temperature of the hot plate starts to increase until it reaches 120°C. The temperature is kept slightly lower than the $T_g$ to avoid breakdown. As soon as the temperature reaches 120°C, the hot plate is switched off while the poling voltage is kept at 350 V. Finally, the poling voltage is turned off when the temperature reaches room temperature. Figure 2(c) shows the scanning electron microscope (SEM) image of the fabricated waveguide, along with a cross-sectional view of the waveguide. After applying the EO polymer, the insertion loss of the modulator with a 0.5 mm phase shifter is measured to be approximately 4.2 dB, which consists of 0.2 dB from the phase shifters and 4 dB loss from passive waveguide components, including two multimode interferometers (MMIs) and the routing waveguide. The phase shifter sections contribute to an estimated propagation loss of 0.4 dB/mm.

The frequency response of the modulator is characterized by the setup shown in Fig. 3(a). The radio frequency (RF) electrical signal from port 1 of the vector network analyzers (VNAs) is applied to the high-frequency electrode of the modulator through a high-frequency probe. A tunable laser operating at 1550 nm provides the input light, which is adjusted to the TE mode using a polarization controller before being launched into the device under test. The modulated output light is amplified using an erbium-doped fiber amplifier (EDFA).

After amplification, the modulated light is filtered and detected using a high-speed photodetector. The optical signal is converted into a high-frequency electrical signal and measured at port S2 of the VNA. To minimize the drive voltage, a 1 mm-long device is used for the measurement. Figure 2(b) shows the transmission spectra of the device before and after poling. The static extinction ratio of the modulator exceeds 30 dB. The shift of the interference spectrum is attributed to the change of the refractive index of the EO polymer during the poling, indicating that the alignment of the EO polymer occurs.

To determine the modulation efficiency, the half-wave voltage ($V_{\pi }$) of the modulator is measured by applying different DC voltages between the ground and signal electrodes. Figure 3(c) shows the transmission spectra of the thin-film SOH modulator under varying applied voltages. The free spectral region ($\mathrm{\Delta }\lambda \mathrm{FSR}$) is 18.3 nm. The EO response of the modulator, $R=\mathrm{\Delta }\lambda /\mathrm{\Delta }V$, is measured to be 387.1 pm/V, which is obtained by tracing and linearly fitting the shift of the minimum in the spectra under different DC voltages. According to the equation $V_{\pi }=(\mathrm{\Delta }\lambda \mathrm{FSR})/(2R)$^[29], $V_{\pi }$ is calculated to be 23.6 V, leading to a $V_{\pi }L$ of 1.18 V·cm, which is approximately half of the value of thin-film lithium niobate modulators^[30]. The effective EO coefficient of the EO polymer is 215.1 pm/V. As shown in Fig. 3(e), the measured EO $S_{21}$ does not degrade within the bandwidth of the measurement setup. The 3 dB bandwidth of the modulator is much larger than 33 GHz, which is limited by the bandwidth of the equipment we have access to.

The EO polymer used in this demonstration has been proven to be able to retain $>90\%$ of its original $r_{33}$ value (169 pm/V at 1310 nm by simple reflection technique) after being annealed at 85°C for 500 h^[31]. Previously, Koos et al. demonstrated the first high-speed operation of an SOH modulator with SEO100 polymer at an elevated temperature of 80°C in a regular oxygen-rich ambient atmosphere^[32]. Thus, in this work, we mainly concentrate on demonstrating long-term stability in a damp heat environment. Figure 4 illustrates the variation of the EO response over time when the device is kept under storage conditions of 85°C and 85% relative humidity (RH) for 1000 h without protective packaging. The EO responses of the device at 0, 240, 456, 768, and 1000 h are measured. The EO response only decreases by 7.3 % after 1000 h in the 85°C and 85% RH environment.

In this paper, we experimentally demonstrated a thin-film SOH EO modulator on SOI with a 70 nm silicon layer. The measured half-wave-voltage length product is 1.18 V·cm. The frequency response of the modulator extends beyond 33 GHz without significant degradation. Furthermore, the long-term stability of the modulator is confirmed after long-term (1000 h) exposure to an 85°C/85% RH storage environment without packaging. The EO activity and stability of SEO100 as a benchmark guest-host EO polymer can be further improved by chemical crosslinking and supramolecular engineering of organic EO materials. These results demonstrate that the SOH modulator can meet the thermal reliability requirements for optoelectronic components in future energy-efficient data centers and harsh environment applications.