3D Integrated Photonics

• Three-dimensional integrated photonics is a field that designs and fabricates volumetric photonic circuits with multi-layer architectures enabling ultrahigh component density and advanced optical routing.
• It employs diverse techniques such as monolithic multilayer platforms, heterogeneous bonding, and two-photon polymerization to achieve low-loss, high-performance optical interconnects.
• Applications span solid-state LIDAR, quantum photonics, and neuromorphic computing, effectively overcoming the limitations inherent to traditional planar photonic systems.

Three-dimensional integrated photonics is the field concerned with the design, fabrication, and application of photonic circuits and devices that leverage spatial structuring in all three dimensions, rather than being restricted to planar or single-layer configurations. Moving beyond the architectural and connectivity limits of two-dimensional (2D) photonic integrated circuits (PICs), 3D integration enables ultrahigh component and waveguide density, advanced optical routing, multi-plane interconnects, and new functionalities relevant for communications, sensing, LIDAR, quantum photonics, neuromorphic computing, and electronic–photonic convergence.

1. Fundamental Architectures and Materials Systems

Multi-layer and 3D photonic integration is realized through a variety of substrates and stacking paradigms, each with distinct processing and performance characteristics.

• Monolithic multilayer platforms: Alternating waveguide (e.g., Si₃N₄) and cladding (e.g., SiO₂) layers enable up to eight independent photonic layers with suppressed vertical crosstalk (<–30 dB), as exemplified in the multi-layer Si₃N₄/SiO₂ PIC for optical phased arrays (OPAs) (Wu et al., 2022). Typical layer thicknesses for these platforms are ~300 nm for Si₃N₄ with ~1.3 μm SiO₂ isolation, enabling stack heights compatible with standard fiber mode-field diameters.
• Heterogeneous 3D integration: Layer stacks of disparate materials are realized using wafer-scale direct bonding, e.g., the SuMMIT platform, which employs four bonded layers (CMOS electronics, redistribution wafer, silicon-photonics, and backside multi-material integration) for dense 3D photonic-CMOS circuits with no custom foundry changes (Ranno et al., 2023). III-V gain layers, ultra-low-loss (ULL) Si₃N₄, silicon, and additional functional films are similarly vertically integrated for high-performance, isolator-free lasers and frequency synthesizers (Xiang et al., 2023).
• Additive and direct-write 3D fabrication: Two-photon polymerization (TPP) and hybrid TPP/OPP ("flash-TPP") yield sub-micron resolution for freeform 3D photonic circuits on arbitrary substrates, including polymer, silicon, or glass. Flash-TPP achieves drastic write-time reductions (80–90%), with typical core indices n₁~1.51 and numerical aperture up to 0.16 (Grabulosa et al., 2021, Grabulosa et al., 2023).
• Hybrid 2D/3D overpasses: For existing planar PICs, two-photon polymerization enables freeform 3D printed waveguide overpasses (WOPs) that eliminate waveguide crossings in non-planar topologies—scaling as O(n²) for n×n switch-and-select arrays, as opposed to O(n⁴) for purely planar crossings (Nesic et al., 2019).
• Novel vertical elements and interconnects: Rolled-up nanomembrane devices (e.g., TiO₂ vertical microtube ring resonators) are monolithically integrated atop SOI waveguides, yielding vertical, out-of-plane coupling and add/drop functionality (Madani et al., 2015). Heterogeneous integration of wide-bandgap materials (e.g., SiN/AlN on sapphire) supports ultra-broadband 3D PICs spanning the UV to near-IR (Zhang et al., 28 Mar 2025).

2. Fabrication Strategies and Key Process Steps

3D photonic integration imposes stringent demands on alignment, planarization, and process compatibility.

• CMOS-compatible multilayer processes: Successive LPCVD or PECVD deposition of waveguide and cladding layers, combined with single-step lithography and deep RIE to define vertically aligned waveguides through the full stack, is requisite for achieving monolithic multi-layer platforms. Subsequent PECVD SiO₂ cladding and, if required, CMP ensure low-loss edge facets for efficient coupling (Wu et al., 2022).
• Low-temperature hybrid bonding: For wafer-scale 3D stacks, planarized Cu-Cu and oxide-oxide direct-bonding (DBI) provides top-to-bottom electrical and optical interconnectivity. After bonding, handle and BOX removal exposes flat surfaces for backside passive/active material integration, independent of original process flows (Ranno et al., 2023).
• Direct laser writing and two-photon polymerization: 3D waveguides, couplers, and overpasses are realized via focused femtosecond laser exposure in negative-tone resists (e.g., IP-S, SU-8), with features defined by the writing voxel size (~200–400 nm lateral, ~600–800 nm axial). Blanket UV OPP is used to cure passive cladding volumes, decoupling the total fabrication time from device volume ("flash-TPP") (Grabulosa et al., 2021, Grabulosa et al., 2023).
• Vertical assembly and in-chip modification: Micro-transfer printing (μTP) with PDMS stamps enables 3D co-integration of photonic and CMOS chiplets, achieving sub-μm alignment and minimal parasitics (Gu et al., 28 Nov 2025). Sub-surface nonlinear laser lithography using nanosecond IR pulses enables embedded photonic devices or microfluidic elements in bulk silicon with index modifications ~10⁻³–10⁻² (Tokel et al., 2014).

3. Device Designs and 3D Photonic Functions

Three-dimensional integration enables a diversity of passive, active, and hybrid photonic devices:

• 3D Optical Phased Arrays (OPAs): Stacked waveguide layers form independent emitter rows; wavelength-tuned delay lines provide horizontal beam steering, while vertical stacking controls the vertical beam profile. Experimental 4-layer arrays achieve vertical FWHM ~17.4°, steering sensitivity up to 0.58°/nm, and input coupling efficiency of 35.1% (uncoated facets) (Wu et al., 2022).
• Interlayer and vertical couplers: Two-stage adiabatic interplane couplers (IPCs) with compact footprints (<40 μm) realize <0.05 dB loss per transition and <3×10⁻⁴ dB per out-of-plane crossing over wide passbands (Chiles et al., 2017). Additive 3D tapers and couplers via TPP or flash-TPP achieve adiabatic 1-to-M splitting with splitting losses as low as 0.06 dB per branch (Grabulosa et al., 2021, Grabulosa et al., 2023).
• Multi-level and mixed-material devices: Backside or interlayer integration of phase-change memory (PCM), magneto-optical or Pockels-active films on photonic–CMOS stacks enables memories, sub-volt modulators, and integrated isolators/circulators (Ranno et al., 2023). Stackable AlN and SiN layers on sapphire permit efficient nonlinear elements (e.g., SHG microcavities, SPDC sources) and low-loss, high-Q linear optical circuits in the UV–IR (Zhang et al., 28 Mar 2025).
• Advanced routing and overpasses: Freeform 3D-printed WOPs eliminate exponential scaling of waveguide crossings in complex switch-and-select fabrics. For n×n fabrics, required overpass count scales as either (n–2)² or ⌈(n–2)²/2⌉, and each optical path requires only a single overpass (Nesic et al., 2019).
• Quantum and neuromorphic elements: 3D femtosecond-laser direct-written circuits realize multi-path, multi-arm interferometers with high tunable quantum visibilities and phase sensitivity. Fractal 3D photonic interconnects enable O(N) scaling of I/O density for photonic neural networks, supporting parallel convolutions and vector–matrix multiplications in compact form factors (Moughames et al., 2019, Chaboyer et al., 2014).

4. Theoretical Modeling and Performance Analysis

Distinct modeling approaches are needed to analyze 3D photonic systems:

• Mode and coupling analysis: Overlap integrals directly quantify coupling efficiency between fibers and multilayer supermode couplers, or between free-space and waveguide–polymer transitions. The overlap integral also quantifies mode-size conversion efficiency (e.g., 3D-printed parabolic micro-reflectors with 5:2 size conversion and <0.5 dB excess loss) (Wu et al., 2022, Huang et al., 2024).
• Adiabaticity and crosstalk suppression: Adiabatic conditions for interplane transitions require |dΔβ/dz| ≪ κ(z)²; phase-velocity mapping—staggering waveguide widths—effectively detunes intra- and interlayer crosstalk (<-35 dB for Δw=80 nm, offset) (Chiles et al., 2017).
• Far-field and phased array models: Kirchhoff–Fraunhofer summations encapsulate the array farfield, with pitches and phase profiles dictating lobes and beamwidths.
• Complex mechanical behavior: For flexible 3D architectures, mechanics with multiple neutral axes are required to predict strain-optical coupling and structural robustness; these models enable devices to survive repeated bending to R~0.5 mm with no Q degradation (Li et al., 2013).

5. Quantitative Benchmarks and Comparative Metrics

Performance metrics derived from both simulation and experiment demonstrate the practical viability of 3D integrated photonics.

Metric	State-of-the-Art Example	Value / Range	Reference
Input coupling efficiency (4-layer OPA)	Multi-layer Si₃N₄/SiO₂ OPA	71.2% (sim), 35.1% (exp, no AR)	(Wu et al., 2022)
Propagation loss (ULL Si₃N₄)	3D Si/SiN with ultra-high-Q	0.5 dB/m	(Xiang et al., 2023)
Interplane coupler loss (3-plane a-Si)	0.032 dB (sim), 0.05±0.02 dB (exp)	(Chiles et al., 2017)
Vertical coupler transmission	Flexible glass photonics, SU-8/ChG	1.1 dB (single), 2.0 dB (double)	(Li et al., 2013)
3D-printed overpass insertion loss	Hybrid 2D/3D PIC (WOP), 1 μm×1 μm core	1.6–1.9 dB	(Nesic et al., 2019)
Q factor (flexible glass resonator)	ChG on SU-8/polyimide/silicone	4.6×10⁵	(Li et al., 2013)
Eye-sensitivity at 224 Gb/s PAM-4	μTP BiCMOS–SiPh receiver	–5.2 dBm OMA @ KP4-FEC	(Gu et al., 28 Nov 2025)
Bandwidth (grating coupler, Si)	3D-CMOS co-integrated device	1-dB BW ≃ 32 nm, η >90%	(Ranno et al., 2023)

Enhanced device density, reduced crossing/coupler loss, improved crosstalk immunity, and expanded functionality (e.g., combined nonvolatile memory, nonlinear elements, and quantum sources) are consistent and recurring findings.

6. Applications, System-Level Impacts, and Scaling

The unique features of 3D integrated photonics directly address long-standing challenges in multiple regimes:

• Solid-state LIDAR OPAs: Multi-layer OPAs support high-efficiency, wide-angle, solid-state beam steering—enabling longer range and finer angular resolution than planar-only arrays (Wu et al., 2022).
• Quantum photonics: Integrated multi-path circuits with high quantum visibilities and tunable phase settings outperform planar analogues in sensitivity and system stability (Chaboyer et al., 2014).
• Heterogeneous and chiplet integration: 3D-nanoprinted interposers and hardwire-configurable photonic circuits allow mixing of platforms (Si, InP, LiNbO₃, polymer) without changing process flows, supporting modular assembly, rapid prototyping, and economic deployment for the fragmented photonics market (Huang et al., 2024, Hoose et al., 2019).
• Neuromorphic and AI photonics: Fractal 3D couplers and interconnects break the O(N²) planar bottleneck, implementing parallel vector-matrix products, Haar spatial filters, and dense splitting networks essential for energy-efficient, scalable photonic neural network hardware (Moughames et al., 2019).
• Flexible and conformal photonics: Monolithic, neutral-axis aligned glass/polymer 3D photonics enables mechanically robust, high-Q devices for biosensing, interconnects, and rolled architectures (Li et al., 2013).

7. Prospects, Limitations, and Open Challenges

While 3D photonic integration delivers clear advances in density, functionality, and system-level performance, several technical challenges remain:

• Device loss and uniformity, especially in printed or direct-written structures, require further reduction for parity with the best planar Si or Si₃N₄ circuits.
• Alignment and overlay precision for multi-material, multi-wafer, and additive techniques is a critical bottleneck; however, closed-loop, machine-vision-guided alignment and in-situ metrology ameliorate this constraint.
• Mechanical stability, thermal cycling, and long-term reliability, particularly for polymeric and flexible photonic platforms, are areas of active research.
• Integration of active elements (high-speed modulators, detectors, nonlinear gain) within fully 3D architectures is in early-stage demonstration and represents a central focus for future work (Tokel et al., 2014, Grabulosa et al., 2021).
• Scaling from chip-level to wafer-scale, high-throughput manufacturing with arbitrary 3D topologies remains to be fully developed, despite promising throughput improvements in flash-TPP and μTP.

The field thus continues to evolve toward realizing high-functionality, heterogeneous, and application-adaptive photonic integration in volumetric, 3D form factors—delivering performance and flexibility well beyond what is possible with planar approaches alone.