Silicon-Based Photonic Integrated Circuits

• Silicon-based photonic integrated circuits are CMOS-compatible platforms that integrate passive (waveguides, resonators) and active (modulators, detectors) devices onto a single chip.
• They leverage high refractive index contrast and heterogeneous integration with III–V materials to overcome silicon’s indirect bandgap for efficient modulation and amplification.
• Scalable fabrication using advanced lithography, bonding, and stress-relief techniques ensures low-loss, high-density circuits for telecommunications, quantum photonics, and AI acceleration.

Silicon-based photonic integrated circuits (PICs) are monolithic or hybrid optoelectronic platforms fabricated using CMOS-compatible silicon technology, supporting a dense integration of passive and active photonic functions—such as waveguides, resonators, modulators, detectors, and light sources—on a single chip. They leverage the high refractive index of silicon and advanced foundry processes to enable compact, low-loss and scalable photonic systems for telecommunications, information processing, sensing, and emerging quantum technologies. The prevalence of silicon as a material arises from its maturity in the semiconductor industry and the viability of heterogeneous integration with III–V and other functional materials to compensate for silicon’s indirect bandgap and limited spectral response.

1. Material Platforms and Waveguide Architectures

Silicon-based PICs utilize a variety of materials and layer stacks for different functionality domains. The core silicon-on-insulator (SOI) platform employs a thin high-index crystalline silicon device layer (~220–250 nm), isolated from the substrate by a low-index buried oxide (BOX, ~2 µm SiO₂), supporting single-mode rib or strip waveguides with tight modal confinement (n_Si ≈ 3.48 at 1550 nm) and minimal propagation loss (<1–3 dB/cm for optimized geometries) (Tan et al., 2024). The strong index contrast (Δn >2) allows for micron-scale bending radii and compact integration densities.

For applications outside silicon’s transparency window (sub-1.1 µm), silicon nitride (Si₃N₄) waveguides are used, typically deposited by LPCVD and clad in SiO₂. Si₃N₄'s broad transparency (400 nm to 2.3 µm), low loss (0.1–2 dB/m in foundry processes), and wide bandgap (E_g ≈ 5 eV) make it the platform of choice for linear, nonlinear, and quantum photonics in the visible and near-infrared (Ye et al., 2023, Tran et al., 2021). Silicon-based PICs may incorporate additional materials: III–V gain epitaxy for native or hybrid light sources, lead zirconate titanate (PZT) for high-speed modulators, and amorphous silicon for hardmask etching to enable thick, crack-free Si₃N₄ films with large mode volumes and engineered dispersion (Liu et al., 2024, Snijders et al., 3 Sep 2025).

2. Integration of Active Devices: Heterogeneous and Hybrid Light Sources

As crystalline silicon is an indirect-bandgap material, silicon PICs inherently lack efficient on-chip light sources. Integration of direct-bandgap III–V materials—such as InP- or GaAs-based quantum wells or quantum dots—is critical for lasers and optical amplifiers (Tan et al., 2024).

Three classes of integration are established:

1. Monolithic (heteroepitaxial) growth of III–V or GeSn alloys on Si promises ultimate CMOS-compatibility but has struggled with defectivity and high thermal budgets.
2. Heterogeneous wafer- or die-level bonding places III–V dies on processed Si waveguides by molecular, adhesive, or micro-transfer printing methods, enabling in-plane or evanescent coupling with sub-μm placement accuracy (Mauthe et al., 2020, Katsumi et al., 2018, Tan et al., 2024).
3. Hybrid (flip-chip) assembly: pre-tested III–V dies are aligned and solder- or epoxy-bonded to silicon edge coupler regions, enabling high yield and relaxed III–V process constraints (Tan et al., 2024).

Template-assisted selective epitaxy (TASE) permits in-plane growth of III–V gain regions, self-aligned and embedded in Si photonic-crystal cavities, achieving localized light sources fully integrated with standard SOI waveguides (Mauthe et al., 2020). Transfer printing allows precise post-CMOS placement of single-photon sources (InAs/GaAs QDs) or strongly coupled QED systems on pre-fabricated Si chips (Katsumi et al., 2018, Osada et al., 2018). Hybrid integration with advanced spot-size converters and mirrors (e.g., buried Al) enables low-loss coupling, high SMSR, and telecom-grade performance (Ding et al., 2016, Tan et al., 2024).

3. Passive and Nonlinear Functionalities

Passive elements in silicon-based PICs include waveguides, directional couplers (MMIs), Mach-Zehnder interferometers (MZIs), microring and photonic-crystal resonators, and multiplexers/demultiplexers. Design of these components leverages the mature CMOS process design kit (PDK) approach, leveraging grating and edge couplers for fiber interfacing and employing inverse-designed, topology-optimized (adjoint-method) devices for low-loss, broadband, and compact functions (Soref et al., 2022).

Nonlinear functionalities, such as stimulated Raman scattering, third-harmonic generation, and Kerr combs, exploit the tight confinement and transparency of silicon and silicon-nitride waveguides. Silicon SOI nanowires (strip/slot/PhC) provide high Raman gain (g_R = 10–15 cm/GW) and are engineered to balance TPA, FCA, and the effective area for on-chip amplification and lasing (Shaikh et al., 2024). Si₃N₄ platforms, with low nonlinear absorption, support the generation of dissipative Kerr solitons and frequency combs, leveraging engineered dispersion and high Q (>10⁷ intrinsic) in thick (≥800 nm) films (Ye et al., 2023, Liu et al., 2024).

4. Active Modulation, Detection, and Programmability

Silicon-based PICs integrate a range of active components:

• Modulators: Carrier-depletion p–n junctions in Si waveguides provide high-speed electro-optic modulation (>20 GHz), while PZT on Si₃N₄ enables high-extinction, fast amplitude modulation, albeit at higher drive voltages (Snijders et al., 3 Sep 2025). Thermo-optic and phase-change tuning is used for programmability, crossbar switching, and tuning of resonance conditions (Soref et al., 2022, Zhu et al., 2 Apr 2025).
• Detectors: Telecom-band photodetection is historically accomplished with Ge epitaxially grown on Si, but recent developments in deep-level tellurium-doped Si enable all-silicon detectors with 0.56 A/W responsivity, 5.9 GHz bandwidth, and NEP = 4.2 × 10⁻¹⁰ W/Hz¹/², preserving full CMOS process compatibility (Shaikh et al., 2024). Si₃N₄ platforms, transparent in the telecom band, typically require heterogeneous integration of Ge or III–V detectors (Tran et al., 2021).
• Programmability and Reconfigurability: Large-scale mesh architectures based on cascaded MZIs or GMZI-DFT circuits, integrated with thousands of phase actuators, enable arbitrary linear transformations (unitary and non-unitary) for photonic computing, matrix operations, AI acceleration, and switching, with self-calibrating hardware–software stacks (Zhu et al., 2 Apr 2025, Hasan et al., 2022).

5. Advanced Functions: Quantum PICs and Spectrally Extended Platforms

Silicon-based PICs are a leading platform for quantum photonics due to their device density, cryogenic compatibility, and scalability. Deterministic integration of single-photon sources via QD transfer-printing or G-center nanopillars provides indistinguishable photon emitting arrays for on-chip networks (Katsumi et al., 2018, Hollenbach et al., 2021). Strongly coupled QD–cavity QED systems on silicon furnish deterministic single-photon nonlinearities vital for future quantum information processors (Osada et al., 2018).

Extension into the visible/NIR is achieved via Si₃N₄ waveguides and III–V/SiN heterogeneous integration, supporting narrow-linewidth lasers, photodetectors, and modulators at sub-µm wavelengths, crucial for biosensing, quantum metrology, and cold-atom control (Tran et al., 2021, Witzens et al., 2020, Snijders et al., 3 Sep 2025). These platforms demonstrate on-chip trapping and manipulation of >10⁷ atoms, MHz-class switching, and ultra-low loss (0.1–2 dB/m), showing distinct advantages over SOI in the visible (Snijders et al., 3 Sep 2025).

6. System-level Integration and Applications

Silicon-based PICs enables complex photonic circuits for telecom/datacom (multi-terabit WDM systems), coherent transceiver modules (hybrid laser, modulator, MZI, detector), RF signal processing (universal RF PICs, on-chip RF spectrum analyzers with 10 MHz resolution), biophotonics (compact multi-color laser engines for microscopy and cytometry), and AI/ML accelerators (programmable photonic processor meshes supporting matrix multiplication, neural inference, switching, and PUFs) (Hasan et al., 2022, Redding et al., 5 Nov 2025, Witzens et al., 2020, Zhu et al., 2 Apr 2025).

Edge and grating coupler designs, multicore fiber coupling using buried Al mirrors, and modular frameworks for plug-and-play integration of topology-optimized devices (3dB splitters, crossovers, add-drop resonators) support heterogeneous system construction, extendible to large-scale circuits (Ding et al., 2016, Soref et al., 2022). Silicon-based PIC reliability is matched to data-center standards (thermal cycling, high-temperature aging), with high-yield foundry flows directly convertible to high-volume CMOS lines (Tan et al., 2024, Liu et al., 2024).

7. Fabrication, Scalability, and Outlook

Silicon-based PIC fabrication leverages established deep-UV (DUV) and electron-beam lithography, advanced hardmask and subtractive etch processes, and wafer-scale bonding for heterogeneous integration. Techniques for stress relief (crack-isolation trenches, double-layer deposition), surface roughness minimization, and thermal annealing enable large-scale (>6-inch wafer) production of low-loss, dispersion-engineered Si₃N₄ circuits with uniform properties (Ye et al., 2023, Liu et al., 2024). Flip-chip and transfer techniques for device integration, standardized alignment marks, and under-bump metallurgy facilitate robust hybrid assemblies (Tan et al., 2024).

The trajectory for silicon-based PICs is toward monolithic platform integration—combining passive photonics, active modulation/detection, native and hybrid light sources, quantum emitters, and high-speed electronics in a fully CMOS-compatible ecosystem spanning visible to mid-infrared wavelengths (Tran et al., 2021, Snijders et al., 3 Sep 2025). This convergence is pivotal for next-generation optical network nodes, integrated sensors, quantum processors, and AI/neuromorphic photonic computing systems.