Papers
Topics
Authors
Recent
Search
2000 character limit reached

Heterogeneous back-end-of-line integration of thin-film lithium niobate on active silicon photonics for single-chip optical transceivers

Published 8 Dec 2025 in physics.optics | (2512.07196v1)

Abstract: The explosive growth of artificial intelligence, cloud computing, and large-scale machine learning is driving an urgent demand for short-reach optical interconnects featuring large bandwidth, low power consumption, high integration density, and low cost preferably adopting complementary metal-oxide-semiconductor (CMOS) processes. Heterogeneous integration of silicon photonics and thin-film lithium niobate (TFLN) combines the advantages of both platforms, and enables co-integration of high-performance modulators, photodetectors, and passive photonic components, offering an ideal route to meet these requirements. However, process incompatibilities have constrained the direct integration of TFLN with only passive silicon photonics. Here, we demonstrate the first heterogeneous back-end-of-line integration of TFLN with a full-functional and active silicon photonics platform via trench-based die-to-wafer bonding. This technology introduces TFLN after completing the full CMOS compatible processes for silicon photonics. Si/SiN passive components including low-loss fiber interfaces, 56-GHz Ge photodetectors, 100-GHz TFLN modulators, and multilayer metallization are integrated on a single silicon chip with efficient inter-layer and inter-material optical coupling. The integrated on-chip optical links exhibit greater than 60 GHz electrical-to-electrical bandwidth and support 128-GBaud OOK and 100-GBaud PAM4 transmission below forward error-correction thresholds, establishing a scalable platform for energy-efficient, high-capacity photonic systems.

Summary

  • The paper presents a novel BEOL integration process enabling CMOS-compatible co-integration of thin-film lithium niobate modulators and high-performance Ge photodetectors on silicon.
  • It achieves a half-wave voltage of 4.4 V for modulators with a 3-dB EO bandwidth of 100 GHz and low insertion loss, ensuring efficient optical coupling and energy efficiency.
  • On-chip data links demonstrated 128 GBaud OOK and 100 GBaud PAM4 transmission with error rates below FEC thresholds, confirming high-speed performance.

Heterogeneous Back-End-of-Line Integration of Thin-Film Lithium Niobate on Active Silicon Photonics for Single-Chip Optical Transceivers

Introduction and Motivation

The architecture presented in "Heterogeneous back-end-of-line integration of thin-film lithium niobate on active silicon photonics for single-chip optical transceivers" (2512.07196) addresses critical challenges in photonic integrated circuits (PICs), targeting increasing demands for bandwidth, integration density, and energy efficiency in data-center and AI/ML interconnects. The solution leverages the distinct advantages of thin-film lithium niobate (TFLN) for high-bandwidth EO modulation and active silicon photonics for integration of Ge photodetectors and passive elements, while retaining compatibility with standard CMOS fabrication flows. Existing integration strategies have been constrained by material and process incompatibilities, restricting TFLN deployment to passive Si platforms. The demonstrated back-end-of-line (BEOL) integration via trench-based die-to-wafer bonding fundamentally removes this bottleneck, enabling full functional co-integration of high-performance TFLN modulators and broadband Ge photodetectors.

BEOL Integration Workflow and Platform Architecture

The proposed BEOL process integrates diced x-cut LNOI dies onto a pre-fabricated active silicon photonics wafer, incorporating passive Si/SiN routing, grating couplers, SiN edge couplers, TO phase shifters, epitaxial Ge photodetectors, and multilayer metallization. Trenches are etched in the thick SiO2 overcladding above the modulation regions to allow sub-micron proximity between the Si and TFLN waveguides, essential for efficient optical coupling. Benzocyclobutene (BCB) adhesive ensures robust bonding and surface planarity across these interfaces.

The modulation sections utilize TFLN ridge waveguides with coplanar traveling-wave electrodes. Vertical adiabatic couplers (VACs), based on inverse tapers in silicon, enable nearly lossless transition (measured coupling efficiency >97%, loss ~0.11 dB) from Si to TFLN. Similar strategies yield Si-SiN coupling losses of 0.06 dB. Integration of a horizontal PIN Si structure underneath Ge yields high photodetector responsivity and minimal dark current.

Device Performance Characterization

Modulator

The Si-TFLN unbalanced Mach-Zehnder interferometer (MZI) modulator achieved a half-wave voltage (VπV_\pi) of 4.4 V for a 6.4 mm device (VπL = 2.8 V·cm), extinction ratio >25 dB, and insertion loss of 4 dB (primarily from TFLN waveguide propagation and contact photolithography limits). The 3-dB EO bandwidth reached 100 GHz, exceeding conventional Si MZI modulators at comparable lengths and voltages. The integration of Si-based TO phase shifters offers stable, drift-free quadrature biasing; a π phase shift required only 17.1 mW. The EO response shows minor low-frequency gain due to resistor/electrode impedance mismatches, with further improvement possible via capacitively loaded traveling-wave electrodes and substrate engineering.

Germanium Photodetectors

Integrated Ge photodetectors exhibited robust performance pre- and post-bonding, maintaining dark current as low as 55 nA at −1 V, responsivity >0.8 A/W across C-band, and 3-dB bandwidths of 47 GHz at −1 V and 56 GHz at −2 V bias. There was no measurable degradation from the bonding step, confirming process compatibility and integrity.

On-chip data links demonstrated uniform EE bandwidths >60 GHz across four parallel channels, limited by the Ge photodetector response. High-speed transmission experiments achieved 128 GBaud OOK with BER <2.4×10⁻⁴ (well below KP4-FEC threshold) and 100 GBaud PAM4 with BER <3.8×10⁻³ (satisfying HD-FEC limits), without optical amplification. Clear eye diagrams at high received optical powers confirm signal fidelity and link robustness.

Implications, Comparisons, and Future Directions

This integration paradigm circumvents the ultimate limitations of Si EO modulation, notably the weak index modulation and high power/voltage requirements, while capitalizing on TFLN's intrinsic Pockels effect, high linearity, low insertion loss, and high bandwidth. The process retains full CMOS compatibility for all Si photonics elements, opening pathways for scalable and cost-effective manufacturing—critical for data center and optical NoC deployment.

Potential improvements include optimization of traveling-wave electrodes (e.g., capacitively loaded, differential drive) and substrate engineering (undercut or backside-etched Si), which may push modulation FOMs (BW/Vπ2\text{BW}/V_\pi^2) into the 100–243 GHz/V² regime, matching top-tier Ge photodetector bandwidths with sub-volt drive levels. Compact modulator architectures (resonant, slow-light) can further miniaturize footprints to sub-millimeter scales.

The multilevel integration (Si, SiN, TFLN) affords unprecedented design flexibility for advanced system-level functions: coherent/multi-dimensional multiplexed transceivers, microwave/quantum photonics, and integrated photonic computing. The proven coupling efficiency and fabrication tolerance for multi-material vertical transitions (e.g., ±300 nm lateral offset, ±20 nm BCB thickness) support reliable yield and high-volume production.

Conclusion

This work demonstrates a scalable BEOL integration strategy for co-packaged TFLN, active Si photonics, and Ge photodetectors, culminating in a fully functional single-chip optical transceiver platform with high modulation bandwidth, low transmission error rates, and full CMOS compatibility. The architecture addresses persistent bottlenecks in photonic interconnects for AI/datacenter applications, and provides a concrete foundation for future advances in wafer-scale computing and complex on-chip photonic systems. Further device and process optimizations are likely to elevate modulation efficiency, integration density, and data rates, underpinning a new class of heterogeneously integrated photonic circuits.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Sign Up to Generate All Videos Subscribe on YouTube

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Sign Up to Generate

Explain it Like I'm 14

Overview

This paper shows a new way to build super-fast “optical transceivers” on a single computer chip. An optical transceiver sends and receives data using light, which can be much faster and use less energy than electricity alone. The researchers combined two technologies: active silicon photonics (light-based parts made on silicon, like in normal computer chips) and thin-film lithium niobate (a special crystal that is excellent at quickly changing light using electricity). Their main achievement is getting these two materials to work together on one chip using a factory-friendly process, so the result can be made at large scale and low cost.

Key Questions

The paper asks and answers simple but important questions:

  • How can we put lithium niobate modulators (fast light switches) directly on a finished silicon photonics chip without breaking the standard chip-making process?
  • Can we keep all the useful silicon parts (like high-speed photodetectors, waveguides, and metal wiring) and still connect light efficiently between silicon and lithium niobate?
  • Will the combined chip be fast and reliable enough for today’s huge data needs (like AI and cloud computing)?

Methods and Approach

What did they build?

  • A single chip that includes:
    • Silicon and silicon nitride waveguides (tiny roads that guide light)
    • Fiber couplers (to get light in and out of the chip)
    • Germanium photodetectors (light sensors that turn light into electrical signals)
    • Lithium niobate modulators (fast “light switches” that turn electrical signals into light changes)
    • Metal layers (to carry electrical signals around the chip)

How did they make different materials work together?

  • Think of the chip like a layered cake. Normally, after baking the cake (making all the silicon parts), you don’t want to rebake or mess with the layers too much. The team used a “back-end-of-line” method, which means they added the lithium niobate on top after the silicon chip was already finished.
  • They carved “trenches” (like shallow grooves) in the top protective layer over the silicon waveguides to bring the lithium niobate close enough for light to move between the two layers.
  • They bonded small pieces (“dies”) of lithium niobate into these trenches using a thin adhesive called BCB. You can think of BCB like super-flat double-sided tape that helps stick two clean surfaces together.
  • To move light smoothly between silicon and lithium niobate, they used vertical adiabatic couplers. Imagine merging lanes on a highway: the coupler gently changes the lane shape so cars (light) can switch layers without crashing (losing power).

How did they test it?

  • They measured how fast the modulators and photodetectors could work (their bandwidth, in GHz; 1 GHz is one billion times per second).
  • They sent high-speed data through the on-chip link and checked how accurately it was received (bit error rate, or BER).
  • They also checked that bonding lithium niobate didn’t hurt the silicon photodetectors.

Main Findings

Here are the key results and why they matter:

  • Very efficient light transfer between layers:
    • The silicon-to-lithium niobate coupler lost only about 0.11 dB, meaning very little light was lost during the layer switch. That’s like keeping almost all your signal strength when changing tracks.
  • Fast, low-voltage modulators on lithium niobate:
    • The modulator reached about 100 GHz bandwidth, which is extremely fast.
    • It needed around 4.4 V to shift light by half a full cycle over a 6.4 mm length (a good efficiency for this kind of device).
  • Reliable silicon control:
    • A small, stable silicon heater kept the modulator’s “bias” point steady (so it stayed in the best operating spot), without drift. This improves long-term reliability.
  • High-quality photodetectors:
    • Germanium photodetectors had bandwidths up to 56 GHz, high sensitivity (>0.8 A/W), and very low dark current (noise). Importantly, their performance didn’t change after bonding the lithium niobate, showing the process is gentle on existing parts.
  • End-to-end on-chip speed:
    • The full link from modulator to photodetector had more than 60 GHz bandwidth, limited mainly by the photodetector speed.
  • Real data transmission:
    • The chip sent 128-GBaud OOK signals and 100-GBaud PAM4 signals with error rates below standard correction limits. In simpler terms, it carried extremely fast data with few errors, good enough for practical use.
    • “GBaud” means billions of symbols per second; OOK is a simple on/off light signal; PAM4 uses four levels instead of two, packing more bits per symbol.

Why This Is Important

  • Speed and energy: As AI and cloud services grow, we need faster, more energy-efficient ways to move data. Light-based links on chips help do that.
  • Manufacturing-friendly: The method fits standard CMOS factory processes, which means it can be scaled up and made cheaper, like today’s electronics.
  • Complete on one chip: Putting both the “send” side (modulator) and “receive” side (photodetector) together, with low-loss connections, makes a compact, powerful optical transceiver. This reduces complexity and improves performance.
  • Path to even better performance: The paper outlines clear ways to make the modulators faster and more efficient, and to shrink their size further. That means future versions could be even better.
  • Opens doors for advanced systems: Because silicon, silicon nitride, and lithium niobate all work together, the platform can support complex features—like coherent communications, microwave photonics, quantum photonics, and photonic computing—and can help build optical networks that connect many parts of giant “wafer-scale” computers.

Simple Takeaway

The researchers found a practical, factory-friendly way to “mix-and-match” the best materials for light-based chips—using silicon for integration and detectors, and lithium niobate for fast light switching—without breaking standard chip-making rules. Their single chip can send and receive very high-speed data with low loss and low power, pointing the way toward faster, greener data centers and future optical computing systems.

View Paper Prompt  View All Prompts 

Knowledge Gaps

Unresolved gaps, limitations, and open questions

Below is a single, actionable list of what remains missing, uncertain, or unexplored in the presented BEOL heterogeneous TFLN-on-active-silicon photonics platform.

  • Quantify the current TFLN ridge waveguide propagation loss (not reported) and demonstrate loss reduction with stepper lithography to the cited <0.27 dB/cm target; link these measurements to modulator insertion-loss improvements.
  • Provide wafer-scale statistics for vertical adiabatic coupler (VAC) performance: measured coupling-loss distributions, lateral misalignment distributions, and BCB thickness uniformity across dies and wafers, to define robust process windows.
  • Establish long-term reliability of BCB-bonded LNOI dies under industry-standard stress (e.g., 85°C/85% RH, thermal cycling, mechanical shock, high optical and RF power) including delamination risk, refractive-index drift, and electrode adhesion.
  • Clarify BEOL thermal budget and compatibility: maximum allowable post-bond process temperatures, impact of the 300°C BCB cure on Ge PDs and CMOS BEOL stacks, and constraints for future metallization and passivation steps.
  • Characterize the RF/microwave properties of SU8 and BCB (dielectric constant, loss tangent, dispersion) and quantify their impact on traveling-wave electrode loss, velocity matching, and EO bandwidth.
  • Demonstrate and validate advanced electrode designs (capacitively loaded TWEs, differential drive, substrate/handle removal) within this active platform, confirming compatibility with trenches, Ge PDs, and existing metallization.
  • Report actual modulator drive conditions at high speed (Vpp at the pads and across the electrodes) and measure full-link energy per bit; show operation without external RF amplifiers (driverless) or with co-integrated CMOS drivers.
  • Assess bias stability over extended durations (hours–weeks) and temperature excursions; quantify heater power for long-term quadrature lock and map thermal cross-talk among adjacent channels.
  • Measure modulator linearity (SFDR), chirp parameter, and large-signal EO transfer characteristics under OOK/PAM4 to validate applicability to analog/microwave photonics.
  • Remove the photodetector bandwidth bottleneck by integrating >100 GHz PDs (or alternatives) to reveal modulator-limited link performance and quantify EE bandwidth gains.
  • Provide a complete optical link budget (grating/edge coupling, Si–SiN and Si–LN VACs, LN propagation, Si MMIs, PD responsivity) with per-channel variability to guide system design and scaling.
  • Quantify RF, optical, and thermal cross-talk in densely packed trench regions; derive spacing/design rules and shielding strategies for high-density channel integration.
  • Report die-to-wafer bonding yield, die placement accuracy, throughput, and cost metrics to assess manufacturability and scalability of the trench-based approach.
  • Measure reflection/return loss at all interlayer transitions (Si–LN, Si–SiN, Si–Ge) and evaluate their impact on coherent systems; implement and test anti-reflection strategies.
  • Demonstrate polarization-diverse operation (TE/TM) and polarization-handling components; characterize VAC performance for TM and quantify polarization-dependent loss and stability.
  • Validate system-level functions enabled by SiN/TFLN/Si tri-layer integration (WDM filtering on SiN, IQ modulators on LN, LO routing) with coherent and multi-dimensional multiplexed transceivers.
  • Co-integrate TIAs and modulator drivers in CMOS BEOL, quantify total link energy/latency/area, and characterize electromagnetic parasitics between electronics and photonics in the trench geometry.
  • Provide environmental/reliability data for BEOL metals (Au electrodes, Ti resistors) including electromigration, corrosion, radiation hardness, and ESD robustness, and assess compatibility with standard Cu BEOL and barrier stacks.
  • Model and measure thermal management under high data rates (electrode ohmic heating, PD self-heating, heat spreading through trenches) and its impact on Vπ, insertion loss, and wavelength drift; propose heat-sinking strategies.
  • Deliver PDK-ready parameterized cells and design rules for trenches, VACs, LN waveguides, and electrodes; validate overlay tolerances with stepper lithography and metrology for foundry adoption.
  • Scale beyond four channels and report channel-to-channel uniformity and yield across large arrays (hundreds–thousands); include fiber-array coupling tolerances and packaging considerations.
  • Investigate compatibility with on-chip laser integration (III–V bonding or heterogeneous approaches) within the same BEOL stack; analyze thermal/material interactions with LN and BCB.
  • Measure noise/EVM for 100 GBaud PAM4, quantify DSP requirements (e.g., FFE length, taps), and study sensitivity to laser RIN and phase noise in the presence of interlayer reflections.
  • Analyze the impact of LN cut (x-cut) and electrode orientation on the effective EO coefficient (e.g., r33r_{33} vs r31r_{31}) in this geometry; explore alternative cuts/orientations for improved efficiency and lower VπL.
  • Determine acceptable ranges for trench topography (residual oxide thickness, sidewall angle, planarity) and correlate with coupling efficiency and scattering to set CMP/etch specifications.
  • Characterize PD saturation power, linearity, timing jitter, and noise-equivalent power post-bonding, and provide statistical variations across devices to inform receiver design.
  • Demonstrate multilayer metallization beyond the single BEOL layer used here; validate via reliability and interconnect integrity in the presence of trenches and bonded LN.
  • Validate packaging-level robustness (fiber attach, temperature cycling, mechanical shock) and alignment tolerances for SiN edge couplers and grating couplers in realistic transceiver assemblies.
View Paper Prompt  View All Prompts 

Practical Applications

Immediate Applications

The following applications can be deployed with the demonstrated platform as-is or with minimal engineering, leveraging the reported 100 GHz Si–TFLN modulators, 56 GHz Ge photodetectors, Si/SiN passives, and BEOL die-to-wafer integration that preserves standard CMOS photonics PDK flows. Each bullet lists sector(s), potential tools/products/workflows, and key assumptions/dependencies.

  • Single-chip short-reach optical transceivers for AI/ML clusters and data centers
    • Sectors: Cloud/data center, AI hardware, networking
    • Tools/products/workflows: 100 GBaud PAM4 / 128 GBaud OOK lanes realized on a single Si chip; co-packaged optics engines and on-board optics; SiN edge couplers (≈1.6 dB loss) for fiber-array attach; PDK-driven design flow with TFLN building blocks; use of existing 50–70 GHz driver/receiver electronics
    • Assumptions/dependencies: Yield and reliability of BCB-bonded TFLN dies; packaging of 60+ GHz RF I/O; reducing modulator insertion loss toward <1.5 dB (stepper lithography); thermal management and co-integration with CMOS drivers; cost and supply of LNOI at scale
  • Analog photonic links and microwave photonics modules up to mmWave
    • Sectors: Wireless infrastructure (5G/6G), test & measurement, defense/aerospace
    • Tools/products/workflows: Analog RF-over-fiber remoting; LO distribution; spectrum/analyzer front-ends; push–pull traveling-wave electrodes; integrated PDs for end-to-end links
    • Assumptions/dependencies: Linearity and noise budgeting for analog links; PD bandwidth currently ~56 GHz limits end-to-end bandwidth; packaging for low-RF-loss interconnects
  • Optical chiplet I/O for heterogeneous compute packages
    • Sectors: Semiconductors, HPC
    • Tools/products/workflows: Photonic chiplet/interposer carrying Si–TFLN transceivers to reduce pin count and PCB losses; 2.5D/3D assembly with micro-bumps; BEOL-friendly LN processing decoupled from front-end Si photonics fab
    • Assumptions/dependencies: Co-design of electronics and photonics; thermal co-management; alignment tolerances across chiplets; supply chain for multi-die assembly
  • Rapid prototyping and education using a CMOS-compatible heterogeneous photonics platform
    • Sectors: Academia, R&D labs, foundry MPW services
    • Tools/products/workflows: PDK-based design kits with VACs, SiN filters, Ge PDs, and TFLN modulators; MPW runs to explore optical NoCs, coherent links, Kerr-nonlinear functions
    • Assumptions/dependencies: Foundry access to a BEOL TFLN module; reference cells and design rules for trenches/bonding; characterization infrastructure (VNA/OCA/real-time scopes)
  • Low-loss fiber attach and manufacturing-friendly packaging
    • Sectors: Photonics manufacturing, ODM/OEM transceiver assembly
    • Tools/products/workflows: SiN edge couplers for broadband, alignment-tolerant fiber attachment; passive V-groove arrays; UV-curing workflows (e.g., SU-8 overcladding, epoxies)
    • Assumptions/dependencies: Environmental stability (temperature/humidity) of BCB and polymer overcladdings; fiber-array tolerances and long-term reliability
  • Energy-per-bit reduction in hyperscale interconnects
    • Sectors: Energy management, sustainability policy, cloud economics
    • Tools/products/workflows: Adoption of high-efficiency TFLN modulators (path to sub-2 V Vπ and low IL) to lower driver swing and power; reduced lane count at higher baud rates
    • Assumptions/dependencies: Realization of lower IL via finer lithography; driver/DSP optimization; data center-level TCO models validating energy savings
  • Standards, qualification, and benchmarking of heterogeneous BEOL photonics
    • Sectors: Standards bodies, policy, industry consortia
    • Tools/products/workflows: Reliability and qualification protocols (thermal cycling, humidity, HTOL) for TFLN-on-Si with BCB; lane-level benchmarks for 100+ GBaud PAM4; inclusion of LN modules in PDKs
    • Assumptions/dependencies: Coordinated efforts with JEDEC/IPC/IEEE; multi-vendor interoperability tests; access to long-term reliability data
  • High-bandwidth reference PICs for test and measurement
    • Sectors: Instrumentation (VNAs, oscilloscopes), calibration services
    • Tools/products/workflows: On-chip 60+ GHz links as calibration artifacts; repeatable EO S21 references; packaged dies with RF probes and fiber pigtails
    • Assumptions/dependencies: Stable packaging; consistent fiber and RF coupling; traceable calibration chains

Long-Term Applications

The following opportunities are enabled by the platform’s architectural advances (BEOL TFLN-on-active-Si, efficient VACs, SiN co-integration) but require further performance scaling, reliability data, or ecosystem maturation.

  • Optical network-on-chip (NoC) and wafer-scale photonic interconnects
    • Sectors: Semiconductors, HPC, AI accelerators
    • Tools/products/workflows: Dense arrays of on-chip modulators/PDs for multi-terabit intra/inter-chip communication; WDM using SiN filters; optical routers/switch fabrics
    • Assumptions/dependencies: Integration of laser sources (co-packaged or on-wafer), thermal stabilization, thousands of lanes with low IL and crosstalk, robust bias control and monitoring, wafer-scale yield
  • Single-chip coherent transceivers (integrated I/Q modulators and receivers)
    • Sectors: Telecom/datacom (400ZR+), metro/DCI
    • Tools/products/workflows: TFLN-based I/Q modulators, Ge or coherent receivers, SiN filters and couplers for polarization multiplexing; compact DSP-ready coherent engines
    • Assumptions/dependencies: Integration or hybrid co-packaging of narrow-linewidth LO lasers; high-linearity, low-drift modulators; polarization handling and thermal control
  • Quantum photonics subsystems
    • Sectors: Quantum communications/sensing/computing
    • Tools/products/workflows: Fast EO control and quantum frequency conversion in LN with low-loss SiN routing; entanglement/distribution circuits; interfacing with SNSPDs or Si/Ge APDs
    • Assumptions/dependencies: Cryogenic compatibility and low-noise operation; integration of quantum light sources; ultra-low-loss waveguides and low phase noise
  • Photonic computing and neuromorphic processors
    • Sectors: AI hardware, edge computing
    • Tools/products/workflows: Arrays of low-voltage, high-speed TFLN modulators for multiply–accumulate or activation functions; SiN delay lines/resonators for signal processing
    • Assumptions/dependencies: Dense driver integration and calibration; footprint reduction (resonant/slow-light LN designs); stable, low-drift biasing; co-design with electronic control loops
  • Integrated microwave photonics for 6G/THz systems
    • Sectors: Wireless, radar, satellite
    • Tools/products/workflows: Signal generation, tunable filtering, beamforming, and photonic-assisted frequency conversion; SiN microcombs combined with EO modulation
    • Assumptions/dependencies: On-chip comb sources (high-Q SiN), low-RIN lasers, improved PD bandwidth (>100 GHz), package-level RF performance
  • FMCW LiDAR and advanced sensing PICs
    • Sectors: Automotive, robotics, industrial sensing
    • Tools/products/workflows: LN modulators for linear high-speed chirps; SiN low-loss routing to Tx/Rx apertures; integrated PDs for coherent detection
    • Assumptions/dependencies: Eye-safe power handling, narrow-linewidth lasers, ruggedized packaging, detector enhancements (e.g., APD/SNSPD), temperature stability
  • Time/frequency metrology and clock distribution on chip
    • Sectors: Precision timing, finance, scientific instrumentation
    • Tools/products/workflows: EO comb generation, low-jitter timing distribution across packages/boards; integrated photonic timing links for large compute fabrics
    • Assumptions/dependencies: Ultra-low-noise lasers, very low-loss SiN (ultra-high-Q), environmental isolation, long-term phase stability
  • Foundry ecosystem and supply chain evolution for heterogeneous BEOL photonics
    • Sectors: Manufacturing, policy, standards
    • Tools/products/workflows: Standardized BEOL LN process modules in mainstream Si photonics PDKs; wafer- or die-level bonding at high throughput; MPW programs with LN options
    • Assumptions/dependencies: Cost/yield competitiveness; secure supply of LNOI; certification and export controls where applicable; cross-foundry interoperability
  • Consumer-level benefits via cloud and network back-ends
    • Sectors: Consumer internet, gaming/AR/VR streaming
    • Tools/products/workflows: Higher throughput and lower latency from data-center back-ends; more energy-efficient AI inference services
    • Assumptions/dependencies: Broad deployment in hyperscale networks; compatibility with existing fiber plants and pluggable form factors; cost-per-bit reductions realized in practice

Notes on cross-cutting dependencies:

  • Reliability of BCB adhesive bonding and long-term thermal/humidity stability across operating profiles
  • Alignment, overlay, and CD control in trenches and VACs; transition from contact lithography to stepper to cut TFLN waveguide loss
  • Packaging at 60–100+ GHz (RF launches, SI/PI co-design), thermal design, and ESD robustness
  • PD bandwidth as current system bottleneck; roadmap to ≥100 GHz PDs to fully exploit TFLN modulators
  • Access to integrated or co-packaged lasers and low-loss SiN building blocks for advanced WDM/coherent/quantum functions
View Paper Prompt  View All Prompts 

Glossary

  • Arbitrary waveform generator (AWG): A high-speed instrument that generates programmable electrical waveforms for testing. "Electrical data signals (219 pseudorandom bit sequences) were generated by a 256 GS/s arbitrary waveform generator (AWG)"
  • Back-end-of-line (BEOL): The part of semiconductor fabrication that includes metal interconnects and processes after device formation. "we introduce a back-end-of-line (BEOL) integration platform via trench-based die-to-wafer bonding of TFLN."
  • Benzocyclobutene (BCB): A polymer used as an adhesive and planarization layer in photonic/semiconductor bonding. "using a thin benzocyclobutene (BCB) adhesive layer."
  • Bias drift: Unwanted change in the operating point of a modulator over time due to material effects. "Unlike EO phase shifters on LN, which suffers from a bias drift caused by free-carrier migration"
  • Bias-tee: A three-port component that combines or separates DC bias and RF signals. "The photodetectors were reverse-biased at -2 V using a bias-tee connected to another microwave probe."
  • Bit error rate (BER): The fraction of bits received incorrectly in a digital communication link. "For 128. Gbaud OOK signals, bit error rates (BERs) below 2.4×10-4 were achieved"
  • Capacitively loaded traveling-wave electrode: An electrode design that uses capacitive loading to enhance modulation bandwidth/efficiency. "Further improvements are possible with capacitively loaded traveling-wave electrodes and undercut or backside-etched Si substrates"
  • Chemical mechanical polishing: A wafer planarization process combining chemical and mechanical removal. "and chemical mechanical polishing technologies."
  • Complementary metal-oxide-semiconductor (CMOS): A standard semiconductor process technology for electronics compatible with photonic integration. "preferably adopting complementary metal- oxide-semiconductor (CMOS) processes."
  • Critical dimension (CD): The smallest controllable feature size in fabrication, dictating pattern resolution. "with only a moderate critical dimension (CD) requirement."
  • Dark current: The current through a photodetector with no incident light, indicating noise/leakage. "The dark current is as low as 55 nA at -1 V"
  • Die-to-wafer bonding: Attaching individual dies onto a processed wafer to enable heterogeneous integration. "via trench-based die-to-wafer bonding."
  • Edge coupler: A structure that couples light between a chip waveguide and an optical fiber at the chip facet. "SiN tapers for edge couplers were also fabricated"
  • Electrical-to-electrical (EE) bandwidth: The frequency range over which an electrical input is transmitted to an electrical output through an opto-electronic link. "3-dB electrical- to-electrical (EE) bandwidths over 60GHz"
  • Electro-optic (EO): Referring to modulation of optical properties by an applied electric field. "electro-optic (EO) modulation"
  • Evanescent coupling: Optical power transfer between waveguides via overlapping evanescent fields. "Evanescent coupling is generally adopted."
  • Etching stop layer: A layer used to precisely stop an etch process at a desired depth. "TiN etching stop layer"
  • Epitaxy: The growth of a crystalline layer on a substrate with a defined orientation, used for photodiodes. "Fabrication of the Si waveguides, Ge epitaxy layer, and Si doping."
  • Feed-forward equalization: A digital signal processing technique to compensate channel distortions. "Offline digital signal processing, including resampling and feed-forward equalization"
  • Forward error correction (FEC): Coding schemes that enable error correction at the receiver to reduce BER. "below forward error correction thresholds"
  • Germanium (Ge) photodetector: A high-speed optical detector using germanium absorption integrated on silicon. "56-GHz Ge photodetectors"
  • Grating couplers: Diffractive structures that couple light between fibers and on-chip waveguides. "comprise grating couplers, 3-dB MMI couplers, SiN edge couplers, Si TO phase shifters, and Ge photodetectors."
  • Ground-signal-ground (GSG): A coplanar transmission line/probe configuration for high-frequency signals. "standard coplanar ground-signal-ground (GSG) push-pull traveling-wave electrodes."
  • Half-wave voltage: The drive voltage required to achieve a π phase shift in an interferometric modulator. "yet with a half-wave voltage over 54 V"
  • Heterogeneous integration: Combining different materials/platforms (e.g., Si and LN) in a single chip. "Heterogeneous integration of silicon photonics and thin-film lithium niobate (TFLN)"
  • Inverse taper: A narrowing of a waveguide to expand the optical mode for low-loss coupling. "These VACs rely solely on inverse tapers in the Si waveguides"
  • Lithium niobate on insulator (LNOI): A substrate with thin-film lithium niobate bonded to an oxide layer for photonics. "x- cut LN-on-insulator (LNOI) dies"
  • Mach-Zehnder interferometer (MZI): An interferometric structure used for phase modulation and intensity control. "Mach-Zehnder interferometer (MZI) based Si modulators"
  • Micro-transfer printing: A method to place micro-scale devices onto a target substrate using a stamp. "or micro-transfer printing30-32"
  • Microwave photonics: Using photonic techniques for generation, processing, and distribution of microwave/THz signals. "optical transceivers for data-center interconnects and microwave photonics37"
  • Mode field diameter: The effective width of the optical mode in a fiber or waveguide. "a 7.5 um mode field diameter"
  • Multilayer metallization: Multiple stacked metal interconnect layers for routing high-density signals. "and multilayer metallization are integrated on a single silicon chip"
  • Multimode interference (MMI) couplers: Passive devices that split/combine light using multimode interference. "3-dB multimode interference (MMI) couplers"
  • On-off keying (OOK): A binary modulation format where light is turned on/off to encode data. "supports 128-GBaud OOK"
  • Optical component analyzer: An instrument for high-frequency characterization of optical modulators/detectors. "a 110 GHz optical component analyzer (Newkey GOCA-110)"
  • Optical network-on-chips (NoCs): On-chip optical interconnect architectures akin to electronic NoCs. "optical network- on-chips (NoCs)"
  • Overlay accuracy: The precision of aligning one lithographic layer to another during fabrication. "as well as a higher CD and overlay accuracy control"
  • PIN structure: A diode with p-type, intrinsic, and n-type regions used for photodetection. "For photodetection, a horizontal PIN structure in Si is adopted"
  • Plasma enhanced chemical vapor deposition: A thin-film deposition technique using plasma to enhance reactions. "plasma enhanced chemical vapor deposition"
  • Phase matching: Designing waveguides so that modes have matched effective indices for efficient coupling/absorption. "The Ge waveguide is also designed in phase matching with the Si waveguide beneath"
  • Pockels effect: A linear electro-optic effect in certain crystals enabling high-speed modulation. "with an intrinsic Pockels effect."
  • Polarization controller (PC): A device that adjusts the state of polarization of light. "A polarization controller (PC) was used to ensure the transverse-electric polarization"
  • Pulse amplitude modulation (PAM4): A four-level amplitude modulation format increasing data throughput per symbol. "100- GBaud PAM4"
  • Quadrature (bias point): The operating point of an interferometer at half the peak transmission for linear modulation. "the modulator was biased at quadrature"
  • Resonant modulators: Modulators that use resonant cavities (e.g., rings) to enhance modulation with compact footprint. "Advanced designs such as resonant modulators or slow-light structures"
  • Responsivity: The photocurrent generated per unit of incident optical power. "a responsivity exceeding 0.8 A/W across the whole C band"
  • Ring resonator: A compact resonant cavity used for filtering or modulation on a chip. "Using a ring resonator structure, the drive voltage of a Si modulator can be decreased."
  • Scattering parameter S21: A measure of forward transmission (gain) versus frequency in RF/EO systems. "EO frequency response (S21) of a 6.4-mm-long modulation section"
  • Silicon nitride (SiN): A low-loss photonic waveguide material used for passive components and fiber interfaces. "Si/SiN passive components including low-loss fiber interfaces"
  • Slow-light structures: Photonic designs that reduce group velocity to enhance modulation efficiency or nonlinear effects. "Advanced designs such as resonant modulators or slow-light structures"
  • SU8: An epoxy-based negative photoresist used for thick overcladding or microstructures. "including SU8 over-cladding"
  • Termination resistor: A resistor used to match transmission line impedance and absorb signal energy. "50 W termination resistors made of Ti"
  • Thermal budget: The cumulative thermal exposure allowable during processing without degrading devices. "would inevitably impose constraints on the subsequent silicon photonics processing concerning material compatibility and thermal budget"
  • Thermo-optic (TO) coefficient: The rate at which a material’s refractive index changes with temperature. "owing to the strong TO coefficient of Si."
  • Thermo-optic (TO) phase shifter: A phase control device that uses temperature-induced index changes. "thermo-optic (TO) phase shifters"
  • Thin-film lithium niobate (TFLN): A thin crystalline LN layer enabling high-efficiency, high-speed modulators. "Heterogeneous integration of silicon photonics and thin-film lithium niobate (TFLN)"
  • Traveling-wave electrodes: Electrodes designed to propagate RF signals alongside optical modes for high-speed modulation. "push-pull traveling-wave electrodes"
  • Undercut or backside-etched Si substrates: Mechanical modifications to the substrate to improve RF/EO performance. "undercut or backside-etched Si substrates"
  • Vector network analyzer (VNA): An instrument that measures complex RF transmission/reflection (S-parameters) versus frequency. "using a vector network analyzer (VNA, Keysight N5227B)"
  • Vertical adiabatic couplers (VACs): Structures that transfer light between layers via gradual vertical mode transformation. "Vertical adiabatic couplers (VACs) within the trenches enable nearly lossless mode transitions"
  • Wafer-to-wafer bonding: Bonding entire wafers to integrate disparate material platforms. "wafer-to-wafer bonding26-29"
  • Wavelength division multiplexed (WDM) links: Systems that carry multiple wavelengths (channels) simultaneously. "80×10. Gb/s wavelength division multiplexed links"
  • x-cut: A specific crystallographic orientation of lithium niobate used to set electro-optic axis alignment. "One or more x-cut LNOI dies (commercially available from NanoLN, China) were then flipped and bonded"
View Paper Prompt  View All Prompts 

Open Problems

We found no open problems mentioned in this paper.

Collections

Sign up for free to add this paper to one or more collections.

Sign Up

Tweets

Sign up for free to view the 2 tweets with 65 likes about this paper.

Sign Up for Free