The Pockels effect is a significant electro-optic (EO) effect in photonic integrated circuits, which is widely implemented in on-chip high-speed EO tuning, switching, and modulation^[1]. Novel Pockels EO materials such as thin-film ${\mathrm{LiNbO}}_3$, ${\mathrm{LiTaNO}}_3$, ${\mathrm{BaTiO}}_3$, and ${\mathrm{PbZrTiO}}_3$ have been demonstrated in low-voltage and wide-bandwidth EO modulations^[2–8]. Among those EO materials, electro-optic polymers (EOPs), also called poled polymers, are organic materials that exhibit the strong and ultra-fast Pockels EO effect^[9–11]. Since the EOP material is solution-processible and CMOS-compatible in fabrication, it is particularly suitable for hybrid integration in silicon photonic platforms (Si, ${\mathrm{Si}}_3{\mathrm{N}}_4$), thereby imparting EO tuning capabilities to these devices^[12–15]. Silicon-organic hybrid and plasmonic-organic hybrid EO waveguides have been demonstrated with ultra-wide bandwidth and low half-wavelength voltage ($V_{\pi }$) EO modulation for optical transceiver in data centers^[16–19].

Typically, EOPs consist of a polymer matrix and electro-active molecules (chromophores with large dipole moments). Electric field poling is a necessary but highly challenging process for generating the Pockels EO effect^[11,20]. The process involves heating the device under a fixed strong direct current (DC) electric field ($E_{\text{poling}}$) until the temperature reaches the glass transition temperature of the EOP film (softening state). At this temperature, dipolar molecules in films quickly align along the direction of the $E_{\text{poling}}$. Afterward, a rapid cooling or withdrawing thermal field under the $E_{\text{poling}}$ is carried out to lock the molecular orientation. Finally, the $E_{\text{poling}}$ is removed at room temperature, resulting in a device exhibiting the EO effect. This multiphysics process requires harmonic cooperation in precise control of the heating temperature, $E_{\text{poling}}$ strength, and cooling timing to optimize chromophore alignment and prevent dielectric breakdown. State-of-the-art progress has demonstrated a record-high EO coefficient of 1100 pm/V in bulk EOP films sandwiched by electrodes with a dielectric protecting layer^[21,22]. Such high performance requires $E_{\text{poling}}$ higher than 100 V/µm^[23].

However, the $E_{\text{poling}}$ poling in hybrid waveguides is more challenging than in bulk EOP films. The discontinuities in the $E_{\text{poling}}$ are very pronounced due to the discrepancies in the dielectric constant and refractive index in the hybrid waveguide. The inevitable sidewall roughness of the electrode and waveguide in an imperfect fabrication can significantly amplify the discontinuity in the electric field^[15,24]. Consequently, hybrid EO waveguides with EOP in the $E_{\text{poling}}$ poling are fragile and easily produce dielectric breakdown. As such, the $E_{\text{poling}}$ is weaker than 100 V/µm, typically in the range of 50–70 V/µm^[25,26]. This compromise results in inefficient poling and EO coefficient reduction by about 50% in comparison with poled bulk film ($E_{\text{poling}}>100\mathrm{V}/\mathrm{\mu m}$). Further, the challenges in poling control are exacerbated by differences in the interface roughness and discontinuities in dielectric constants. Recent innovative strategies have been proposed, such as employing transparent graphene electrodes or dielectric protective layers to optimize the poling process and modulation efficiency^[27–29].

In this Letter, we propose a segmented electric field poling strategy that employs Step-1 weak $E_{\text{poling}}$ and then Step-2 strong $E_{\text{poling}}$ in a ${\mathrm{Si}}_3{\mathrm{N}}_4/\mathrm{EOP}$ hybrid waveguide for efficient chromophore alignment. It effectively circumvents dielectric breakdown caused by interface roughness and electric field discontinuity. The segmented poling approach yields the highest tuning efficiency $\phi$ of 52.5 to 58.3 pm/V in a Mach–Zehnder interferometer [MZI, $\text{free spectral range (FSR)}=4\mathrm{nm}$], which is significantly improved compared to 31.7 pm/V. The tuning efficiency $\phi$ is defined as the amount of wavelength drift per unit voltage. The $V_{\pi }$ of an MZI with a tuning length of 1 cm is 8.6 V, based on evanescent-wave EO tuning. Those results indicated that this novel segmented poling strategy is promising for enabling efficient EO tuning on ${\mathrm{Si}}_3{\mathrm{N}}_4$ waveguides.

Our proposed segmented electric field poling strategy is verified in the ${\mathrm{Si}}_3{\mathrm{N}}_4/\mathrm{EOP}$ hybrid MZI [Fig. 1(a)] EO filter with ground-signal-ground (GSG) coplanar Al electrodes. A electro-active dipolar chromophore YLD124 with a ellipsoidal structure [$\sim 2\mathrm{nm}$ in length, Fig. 1(b)] is doped in a PMMA-co-PS (polymethyl methacrylate-co-polystyrene) polymer (mass fraction 20%) to form the guest-host EOP film. As shown in Fig. 1(c), the passive ${\mathrm{Si}}_3{\mathrm{N}}_4$ waveguide is cladded by the EOP. The evanescent field in the EOP cladding is implemented for evanescent-wave EO tuning. COMSOL multi-physics simulation is carried out to theoretically evaluate the evanescent-wave EO interaction. The optical field and poling electric field distributions are shown in Figs. 1(d) and 1(e). The optical confinement factor, which is defined as the relative light intensity of the EOP cladding in the guided mode, is 30% when the width/height of the ${\mathrm{Si}}_3{\mathrm{N}}_4$ is 2 µm/300 nm. In accordance, the EO overlapping factor is 22%, and the $V_{\pi }L$ of the MZI is $10.6\mathrm{V}·\mathrm{cm}$, supposing that the electrode distance ($d$) is 6 µm and the $r_{33}$ is 50 pm/V. Such EO overlap is not strong enough because the most intensified evanescent field is at the interface between the ${\mathrm{Si}}_3{\mathrm{N}}_4$ and the EOP cladding, where the $E_{\text{poling}}$ is weak [Fig. 1(e), 42 V/µm]. The interface $E_{\text{poling}}$ between the electrode and the EOP is the strongest [Fig. 1(e), 473 V/µm], where the EO interaction is negligible because the optical field intensity is nearly zero. Therefore, such strong interfacial $E_{\text{poling}}$ is helpless but often leads to dielectric breakdown in poling. Moreover, overcrowded chromophore density ($\sim 1\text{molecule}/{\mathrm{nm}}^3$ in the EOP film with 20% YLD124) is a factor for accelerating dielectric breakdown. When a strong poling field is applied, the morphology change of the EOP film due to molecular rotation and aggregation at the glass transition temperature exacerbates the dielectric breakdown. To prevent dielectric breakdown and also to achieve high poling efficiency, the selection of $E_{\text{poling}}$ is contradictory. Conventionally, in order to achieve a high EO coefficient, a high DC voltage is applied to induce chromophore alignment from isotropic [Fig. 2(a)] to anisotropic [Fig. 2(b)]. However, dielectric breakdown frequently occurs due to a strong interfacial electric field, high chromophore loading density, and imperfection in waveguide fabrication.

We propose the segmented electric field poling of the ${\mathrm{Si}}_3{\mathrm{N}}_4/\mathrm{EOP}$ hybrid waveguide to avoid unwanted dielectric breakdown at the interface of the electrode and the EOP. As shown in Fig. 2(c), the segmented electric field poling is a two-step process: Step-1 is the low-voltage poling, as shown in Fig. 2(c) (the predominant poling area is near the electrode interface with a strong electric field), and then Step-2 poling with high voltage (the predominant poling area is between the EOP and the SN with strong EO overlap). Although the electric field in Step-1 poling is relatively weak, the electric field near the electrode interface is strong enough to effectively drive the chromophore orientation. So in Step-2 poling with a strong electric field, chromophore alignment near the electrode interface is not drastic because most chromophores are aligned during Step-1 poling. The chromophores between the EOP and the ${\mathrm{Si}}_3{\mathrm{N}}_4$, most of which are not aligned due to weak $E_{\text{poling}}$ in Step-1, can be efficiently oriented in Step-2 poling with stronger $E_{\text{poling}}$.

To verify our estimation, we fabricate four MZIs with an electrode length of 2 mm for segmented poling. The ${\mathrm{Si}}_3{\mathrm{N}}_4$ waveguide is fabricated by standard UV photolithography and inductively coupled plasma (ICP) etching, and the Al electrode is facilitated by the lift-off technique. The SEM images of a typical ${\mathrm{Si}}_3{\mathrm{N}}_4$ waveguide and electrode before integration of the EOP cladding are shown in Figs. 3(a)–3(c). The EOP is deposited on the ${\mathrm{Si}}_3{\mathrm{N}}_4$ waveguide by spin-coating. Figure 3(d) illustrates a hybrid waveguide with in-part EOP cladding, and the electron dispersive spectra clearly display the metal element [Al, Fig. 3(e)] and carbon [C, Fig. 3(f)] from the organic chromophore YLD124 and polymer. Note that for poling and EO tuning, the waveguide should be fully covered by the EOP cladding.

From AMZI-1 to AMZI-4, the Step-1 $E_{\text{poling}}$ are 52, 56, 60, and 65 V/µm, respectively, and the Step-2 $E_{\text{poling}}$ are the same of 65 V/µm. The poling temperature is 114°C for both steps of all the asymmetrical Mach–Zehnder interferometers (AMZIs). The parameters of the electrode gap ($d$) and $E_{\text{poling}}$, along with the tuning efficiency of the AMZI filter, are summarized in Table 1. Note that the tuning efficiency is defined as the slope of the wavelength shift as a function of the voltage.

For the AMZI-1 with the electrode gap of 7.7 µm after Step-1 poling (52 V/µm), most chromophores at the electrode interface aligns since the relative $E_{\text{poling}}$ at the electrode interface is six-fold stronger than that at the ${\mathrm{Si}}_3{\mathrm{N}}_4$ interface, and partial chromophores at the interface of ${\mathrm{Si}}_3{\mathrm{N}}_4$ align. The EO tuning of AMZI-1 with the DC voltage is shown in Fig. 4(a), and its linear relation of wavelength-voltage is plotted in Fig. 4(c). The extrapolated tuning efficiency is 31.7 pm/V for AMZI-1 with an FSR of 4 nm, and the calculated in-device $r_{33}$ is 53 pm/V. Then, in Step-2, a stronger $E_{\text{poling}}$ of 65 V/µm is applied to efficiently induce the chromophore alignment on the ${\mathrm{Si}}_3{\mathrm{N}}_4$ interface. Since chromophores at the interface of the electrode have efficiently aligned in the Step-1 poling with a weak $E_{\text{poling}}$ of 52 V/µm, the morphology change at the electrode interface is effectively mitigated during Step-2 poling with stronger $E_{\text{poling}}$ of 65 V/µm. Therefore, dielectric breakdown at the electrode interface can be avoided, resulting in reliable electric field poling. The EO tuning test [Figs. 4(b) and 4(d)] of AMZI-1 after Step-2 poling shows a much higher tuning efficiency of 58.3 pm/V, and the in-device $r_{33}$ is as high as 114 pm/V. To the best of our knowledge, this is the highest EO coefficient in ${\mathrm{Si}}_3{\mathrm{N}}_4/\mathrm{polymer}$ hybrid EO waveguides. It should be noted that 114 pm/V is very close to the $r_{33}$ value of bulk film with the poling voltage of 100 V/µm, suggesting that our segmented poling can significantly improve the poling efficiency.

For AMZI-2, Step-1 $E_{\text{poling}}$ of 56 V/µm results in a higher tuning efficiency of 42.5 pm/V than that of AMZI-1 after Step-1 poling (52 V/µm). This is due to a higher $E_{\text{poling}}$. Its tuning efficiency is also higher than that of AMZI-3 after a higher Step-1 $E_{\text{poling}}$ of 60 V/µm, which is ascribed to the narrower electrode gap. However, after Step-2 poling with an $E_{\text{poling}}$ of 65 V/µm, the improvement in tuning efficiency of AMZI-2 and AMZI-3 is not as pronounced as that of AMZI-1. As shown in Table 1, the tuning efficiencies of AMZI-1, AMZI-2, and AMZI-3 are similar and higher than 50 pm/V, indicating that the ultimate tuning efficiency of those AMZIs with such EOP recipe and waveguide parameters is about 55 pm/V. This consistency also implies that the proposed segmented poling is highly reliable.

Conventional one-step poling is carried out with a high $E_{\text{poling}}$ of 65 V/µm in AMZI-4 and a poling temperature of 114°C. Since such strong $E_{\text{poling}}$ is very risky, the timing of poling termination should be carefully controlled to prevent the dielectric breakdown. Usually, poling under strong electric fields terminates more early. This conservative operation is adopted to avoid dielectric breakdown, but this raises a question of whether it is efficiently poled. Indeed, our EO measurement shows that the tuning efficiency of AMZI-4 after a one-step strong $E_{\text{poling}}$ of 65 V/µm is only 32.8 pm/V, much lower than that after segmented poling. The second poling with the same $E_{\text{poling}}$ of 65 V/µm is carried out again, and the improvement in tuning efficiency is negligible (33.3 pm/V). The extrapolated in-device $r_{33}$ is 47 pm/V. Strong $E_{\text{poling}}$ is relatively uncontrollable, and it leads to a reduction in the poling efficiency. On the contrary, segmented poling is much more reliable and more efficient in chromophore alignment.

For the AMZI with an electrode length of 2 mm, the required voltage for an FSR ($2\pi$) is still too high. The high tuning efficiency of 58.3 pm/V is equal to $V_{\pi }$ of 36 V. To further reduce the tuning voltage, we fabricated an AMZI with an electrode length of 10 mm (AMZI-5) and implemented segmented poling (Step-1 50 V/µm and Step-2 65 V/µm). It should be noted that the longer tuning length will definitely increase the risk of dielectric breakdown because there are more defects in the waveguide and the electrode fabrication. Therefore, the segmented poling in the hybrid waveguide with a long tuning length is especially necessary. The final EO tuning is shown in Fig. 5. The DC voltages of −5, 0, 5, and 10 V correspond to the central wavelengths of 1559.47, 1558.28, 1557.09, and 1555.8 nm, respectively. The linear EO tuning efficiency is as high as 241 pm/V, and the extrapolated $V_{\pi }$ is significantly reduced to 8.6 V ($V_{\pi }L:8.6\mathrm{V}·\mathrm{cm}$).

Furthermore, a new AMZI (AMZI-6, electrode length 5 mm) with a length difference of 3.24 mm between two phase shifters is designed and fabricated to verify the effect of segmented poling on the EO tuning efficiency. The output transmission in Fig. 6(a) shows that the FSR of AMZI-6 is 0.4 nm, corresponding to the frequency spacing of 50 GHz for dense wavelength division multiplexing (DWDM). The same segmented poling is implemented (Step-1 50 V/µm and Step-2 65 V/µm) in AMZI-6. The EO tuning of AMZI-6 after Step-1 poling is illustrated in Fig. 6(a). The linear tuning efficiency is 8 pm/V, and the extrapolated $V_{\pi }L$ is $12.5\mathrm{V}·\mathrm{cm}$. The Step-2 poling produces the improved tuning efficiency of 11.3 pm/V and the lower $V_{\pi }L$ of $8.8\mathrm{V}·\mathrm{cm}$.

In summary, we proposed an efficient segmented poling in a ${\mathrm{Si}}_3{\mathrm{N}}_4/\mathrm{EOP}$ hybrid waveguide, involving initial poling with a weak electric field followed by a strong electric field. This approach was successfully demonstrated on MZI-type filters with different tuning lengths and FSRs. It effectively avoided dielectric breakdown during the poling process and significantly enhanced the tuning efficiency of EO-tunable filters from $\sim 30$ to $>50\mathrm{pm}/\mathrm{V}$. The extrapolated largest in-device Pockels EO coefficient is as high as 114 pm/V, and the lowest half-wave voltage of the MZI with an electrode length of 1 cm is 8.6 V. Although the tuning efficiency of the evanescent-field EO effect remains moderate, considering the simplicity of the hybrid integration process for the ${\mathrm{Si}}_3{\mathrm{N}}_4$ and EOPs, combined with the efficient and reliable segmented poling method, our work paves the way for the development of future large-scale, low-cost, and highly efficient ${\mathrm{Si}}_3{\mathrm{N}}_4/\mathrm{EOP}$ hybrid EO platforms.