Ferroelectric Silicon Photonics

Updated 14 January 2026

Ferroelectric silicon photonics refers to the integration of nonlinear ferroelectric materials on silicon platforms for efficient electro-optic modulation and non-volatile phase control.
Key approaches include using materials like BaTiO₃, PZT, and Hf₀.₅Zr₀.₅O₂ with CMOS-compatible processes to achieve high EO coefficients and low voltage-length products.
The field drives scalable, high-speed photonic circuits for programmable signal processing, optical memory, and emerging quantum and neuromorphic applications.

Ferroelectric silicon photonics is the field concerned with the integration and exploitation of ferroelectric materials and their associated nonlinear and remanent polarization phenomena within silicon photonic platforms. The core motivation is to transcend the intrinsic limitations of conventional electro-optic and thermo-optic materials, enabling high-efficiency, high-speed, and non-volatile optical phase and amplitude control functions directly on chip. Ferroelectric materials—including perovskite oxides (BaTiO₃, Pb(Zr,Ti)O₃), ferroelectric nematic liquid crystals, hafnium-based thin films, and recently, ferroionic 2D van der Waals crystals—offer strong second- or third-order nonlinearities, spontaneous polarization, and domain reorientation/hysteresis effects that enable both volatile (Pockels) and non-volatile (remnant polarization) operation modes. CMOS compatibility, high EO efficiency, and scalable process flows distinguish recent research directions in this area.

1. Physical Principles and Materials Systems

Ferroelectric materials exhibit spontaneous polarization due to the lack of inversion symmetry in their crystal structure. This endows them with a nonzero second-order nonlinear susceptibility (χ²), enabling the linear electro-optic (Pockels) effect:

$\Delta n = -\frac{1}{2} n^3 r_{ij} E_j$

where Δn is the change in refractive index, n is the refractive index, r_{ij} are the electro-optic coefficients, and E_j is the electric field component. In addition, remanent polarization—switchable by external field—enables non-volatile programming.

Key ferroelectric systems integrated with silicon photonics:

Barium Titanate (BaTiO₃, BTO): Perovskite oxide, r₃₃ in thin film devices ranging from ~100 pm/V up to ~900 pm/V depending on process and orientation. Non-volatile polarization switching demonstrated (Xiong et al., 2014, Ortmann et al., 2019, Catalá-Lahoz et al., 12 Jan 2026).
Lead Zirconate Titanate (PZT): Strong Pockels effect (r_eff ~61–67 pm/V), used in SiN and Si platforms (Alexander et al., 2018).
Ferroelectric Nematic Liquid Crystals (FNLCs): Spontaneous polar order, r₃₃ >100 pm/V possible, typically 24–30 pm/V extracted in device; GHz-bandwidth hybrid silicon implementations (Taghavi et al., 2024, Chiang et al., 19 Jul 2025).
Ferroelectric Hf₀.₅Zr₀.₅O₂ (HZO): CMOS-compatible, remanent polarization up to ~20 μC/cm², primarily quadratic EO response at moderate fields, linear Pockels observed in future slot configurations (Taki et al., 2023).
Ferroionic 2D Materials (e.g., CuCrP₂S₆): Layered van der Waals crystals, electric field-driven cation migration modulates refractive index linearly (Δn ~–2.8×10⁻³), V_πL ~0.25 V·cm achieved (Dushaq et al., 2023).
Lithium Niobate (LiNbO₃): High-performance but typically requires hybrid bonding rather than native growth on Si (Witmer et al., 2016).

2. Device Architectures and Integration Strategies

Architectures capitalize on strong light-matter interaction and engineered overlap between the optical mode and the ferroelectric or ferroionic active region. Process compatibility with commercial CMOS foundries, process steps, and poling/alignment protocols are critical for scaling.

Key Device Topologies and Fabrication Workflows

Platform	Waveguide Geometry	Ferroelectric Integration
BaTiO₃ on SOI	Horizontal slot (Si/BTO/Si); 1–3 μm	MBE growth, wafer bonding, epitaxial STO buffer; poling with E >50 kV/cm (Xiong et al., 2014, Catalá-Lahoz et al., 12 Jan 2026)
BaTiO₃ on SiN/SOI	Si₃N₄ ridge (1.1 μm); BTO layer below	MBE + direct wafer bonding; planarization by CMP; side/top electrodes (Ortmann et al., 2019)
PZT on SiN	Ridge waveguide; 330 nm SiN, 150 nm PZT	Sol-gel chemical solution deposition; anneal + poling E ≈ 150 kV/cm (Alexander et al., 2018)
FNLC-Si Hybrid	Slot/strip Si waveguides; slot fn ≈ 125 nm	Capillary infiltration of LC; monodomain alignment (few V, room T); no poling (Taghavi et al., 2024, Chiang et al., 19 Jul 2025)
HZO-SiN	30 nm HZO on 330 nm SiN; transverse E	ALD on SiN, 1 nm Al₂O₃ spacers; 400°C anneal (Taki et al., 2023)
2D Ferroionic (CCPS)	Si MRR (220 × 460 nm), 30–100 nm CCPS	Mechanical transfer; Au/Cr electrodes patterned under CCPS (Dushaq et al., 2023)

Significant emphasis has been placed on single-lithography or single-mask processes, planarization (CMP), and low-thermal-budget back-end steps to maximize foundry compatibility, scalability, and to enable integration with existing Si photonics process flows (Taghavi et al., 2024, Catalá-Lahoz et al., 12 Jan 2026).

3. Electro-Optic Performance and Non-Volatile Control

Ferroelectric silicon photonic devices exhibit both volatile (instantaneous field-driven, e.g., Pockels effect) and non-volatile (remnant-polarization-driven) phase shift modalities.

Performance Benchmarks

Material/Mechanism	r_eff (pm/V)	V_πL (V·cm)	Bandwidth	Mode (Volatile/Non-Volatile)
BaTiO₃-SiO₂–Si	213–923	0.17–1.5	GHz, 80 ns switching	Both (domain & Pockels) (Xiong et al., 2014, Ortmann et al., 2019, Catalá-Lahoz et al., 12 Jan 2026)
FNLC-Si	24–30	0.25	4.18 GHz (RC-limited); up to >67 GHz demonstrated	Volatile (GHz Pockels), Non-volatile via director (Taghavi et al., 2024, Chiang et al., 19 Jul 2025)
PZT-SiN	61–67	3.2	> 33 GHz	Volatile (poled) (Alexander et al., 2018)
HZO-SiN	–	–	π-shift in 4.5 mm, >10⁴ s retention	Non-volatile; quadratic EO (Taki et al., 2023)
CCPS/SiPh	–	0.25	n.r. (seconds scale, faster possible)	Both (ionotronic ≫ optoelectronic) (Dushaq et al., 2023)

In programmable meshes (e.g., BaTiO₃-based FPPGA), nanosecond-to-millisecond domain switching enables set-and-forget circuits with zero static power requirements (Catalá-Lahoz et al., 12 Jan 2026). Studies with HZO and ferroionic 2D materials have demonstrated non-volatile, multi-level index states with multisecond to hour-scale retention, and negligible optical loss, making them highly attractive for addressable optical crossbars and static interferometric calibration (Taki et al., 2023, Dushaq et al., 2023).

4. Light–Matter Interaction Engineering

Maximized interaction strength is achieved through sub-wavelength slot waveguides (FNLC-Si, slot width down to 125 nm), ridge/slot hybrid configurations (BaTiO₃), and mode-overlap design (COMSOL/Lumerical simulations) resulting in large confinement factors (up to 18–22% in BTO-SiN) and effective overlap integrals (Γ) exceeding 0.25 in optimized topologies (Ortmann et al., 2019, Taghavi et al., 2024).

Electrode geometries are tailored for strong field localization—e.g., FLS in FNLC-Si achieves deff ~200 nm and >1 V/μm field at 1 V bias, enabling sub-volt operation (Taghavi et al., 2024). Trimming of buffer thickness, electrode gap, and mode converter design further reduces insertion loss and boosts tuning efficiency (Catalá-Lahoz et al., 12 Jan 2026, Alexander et al., 2018).

5. Applications in Photonic Integrated Circuits

Ferroelectric silicon photonics underpins a spectrum of applications demanding high-efficiency, low-power, and scalable optical phase and amplitude control:

Programmable photonic circuits: Mesh architectures leveraging non-volatile ferroelectric phase shifters (BaTiO₃ FPPGA, HZO-FeFET arrays) for reconfigurable unitaries, tunable filters, and low-power routing (Catalá-Lahoz et al., 12 Jan 2026, Taki et al., 2023, Tang et al., 2022).
High-speed electro-optic modulation: FNLC-Si and PZT-SiN deliver >100 GHz bandwidths and V_πL down to 0.25 V·cm, enabling direct-drive, low-voltage, high-baudrate transmitters (e.g., 102 Gbit/s PAM-4) (Chiang et al., 19 Jul 2025, Alexander et al., 2018).
Nonvolatile memory and synaptic photonics: Ferroionic 2D materials (CCPS) and remanent BTO/HZO support multi-state, zero-hold-power phase tuning, essential for neuromorphic and quantum photonics (Dushaq et al., 2023, Tang et al., 2022).
Thermal stabilization and compensation: BTO-SiN devices allow electrical compensation of the thermo-optic shift with sub-nW static power, crucial for large-scale WDM circuits (Ortmann et al., 2019).
Quantum technologies: High-Q Si/LiNbO₃ resonators combine large optical nonlinearity with compatible microwave-to-optical coupling, relevant for quantum transduction nodes (Witmer et al., 2016).

6. Advantages, Limitations, and Prospective Directions

Advantages:

Voltage-length products (V_πL) at or below 0.3 V·cm in slot geometries (Chiang et al., 19 Jul 2025, Dushaq et al., 2023).
Static power dissipation approaches zero for non-volatile phase shifters, orders of magnitude lower than thermo-optic methods (Catalá-Lahoz et al., 12 Jan 2026, Ortmann et al., 2019).
Bandwidths limited only by device RC constants or intrinsic material response, routinely above tens of GHz (Taghavi et al., 2024, Alexander et al., 2018).
Scalable fabrication with fully CMOS-foundry-compatible processes for several ferroelectrics (BTO, HZO) (Catalá-Lahoz et al., 12 Jan 2026, Taki et al., 2023).

Limitations:

Integration challenges for epitaxial perovskites (BaTiO₃)—high losses (α up to 44 dB/cm in early work), requiring improved growth and interface engineering (Xiong et al., 2014).
FNLC and PZT: long-term material stability, and further optimization needed for insertion loss and device variability (Taghavi et al., 2024, Alexander et al., 2018).
HZO and CCPS: Pockels effect not always realized at moderate fields; quadratic or ionotronic response may dominate unless engineered for strong field overlap (Taki et al., 2023, Dushaq et al., 2023).

Future Directions:

Material developments: dopant/alloy engineering (BTO, HZO, FNLC), 2D ferroelectrics for enhanced r_eff and reduced drive voltages.
Advanced integration: site-selective ferroelectric placement, monolithic active-passive integration (modulators and detectors/lasers).
Multi-level and analog tuning: realization of dense, crossbar-addressed phase grids for programmable optical signal processors (Tang et al., 2022).
Scaling: expansion to 10⁴–10⁵ actuators per die for large MIMO photonic neural networks (Catalá-Lahoz et al., 12 Jan 2026).
Bandwidth and energy minimization: improved RF-electrode design, traveling-wave drive, and further scaling of phase-shifter length.

7. Comparative Assessment with Alternative Platforms

Ferroelectric silicon photonics offers distinct performance and scaling benefits relative to alternative active platforms (summarized below):

Platform	r_eff (pm/V)	V_πL (V·cm)	Power/FSR (μW)	Static Power	Bandwidth
BaTiO₃–Si/SOI	213–923	0.17–1.5	≈0.1 nW	≈0 nW	GHz
FNLC–Si	24–30	0.25	n.r.	n.r.	>67 GHz
PZT–SiN	61–67	3.2	n.r.	Bias-free	>33 GHz
HZO–SiN	–	–	≈0 (non-vol.)	≈0 nW	n.r.
CCPS–SOI	–	0.25	n.r.	≈0 nW	up to µs
LiNbO₃ (thin film)	~30	2.0	n.r.	n.r.	>60 GHz
Thermo-optic (Si)	–	–	2.4 mW	mW	kHz–MHz
Plasma dispersion (Si)	–	1–2	>0.5 mW	mW	>40 GHz

Ferroelectric phase shifters eliminate the need for hold power, reduce latency due to fast electric-field switching, and maintain compact footprints due to high EO coefficients, clearly distinguishing them from both traditional Si-based thermal-carrier modulation and passive platforms.

Ferroelectric silicon photonics provides a pathway to highly efficient, scalable, and low-power integrated photonic circuits, leveraging both volatile and non-volatile ferroelectric phenomena to enable advanced modulation formats, programmable meshes, and memory-embedded optical systems. Continued materials, process, and device engineering are expected to further reduce losses, enhance actuation efficiency, and expand the application space toward large-scale integrated optical information processing (Taghavi et al., 2024, Catalá-Lahoz et al., 12 Jan 2026, Xiong et al., 2014, Ortmann et al., 2019, Alexander et al., 2018, Taki et al., 2023, Dushaq et al., 2023, Chiang et al., 19 Jul 2025, Tang et al., 2022, Witmer et al., 2016).