# Silicon Photonics Topic radar · 9,505 documents · 2021-01-01–2026-09-22 · updated 2026-09-23 Canonical: https://slate.greyb.com/radar/silicon-photonics/innovations ## In one minute ### The 3D-PIC platform is decoupling from standard Silicon-on-Insulator (SOI) to bypass the thermal and scaling limits of planar silicon A decisive pattern across the portfolio reveals that silicon is being relegated to a passive routing substrate, while active functions are moving into heterogeneous 3D stacks. In Waveguide Design, the center of gravity has shifted to "diagonal directional coupler elevators" for vertical energy transfer [1], while the Next-Gen Compute node is deploying PECVD-deposited isolation layers and vertical photonic vias to overcome planar footprint limits [2]. This architectural shift is not merely a design preference but a thermal necessity: Fabrication records show that waveguide smoothing requires budgets near 1000°C, which would destroy temperature-sensitive active components like germanium detectors if integrated on the same plane [3]. **Why this is worth reading** The convergence of vertical coupling and multi-layer Si-SiN stacks across three distinct nodes indicates that the industry is moving toward a chiplet-based 3D architecture. This transition will likely force a bifurcation in the foundry market, where standard CMOS-compatible facilities are relegated to low-cost interconnects while specialized "super-foundries" control the high-capacity, multi-material switch market. 1. [Waveguide & Component Design](/radar/silicon-photonics/innovations?bucket=strategic_value_chain_position-waveguide_component_design) 2. [Next-Gen Compute & Sensing](/radar/silicon-photonics/innovations?bucket=strategic_value_chain_position-next_gen_compute_sensing) 3. [Fabrication & Foundry Scaling](/radar/silicon-photonics/innovations?bucket=strategic_value_chain_position-fabrication_foundry_scaling) ### The foundry is absorbing the system integrator's role through heterogeneous material bonding A decisive shift is visible where foundries are moving from substrate providers to system-level integrators by internalizing III-V gain layers and lithium niobate thin films directly onto silicon wafers. This transition is evidenced by GlobalFoundries representing 13% of activity in high-power coupling structures and TSMC appearing in 22% of device fabrication records. The record shows this consolidation is driven by the need to bypass 2D lithographic constraints using "diagonal directional coupler elevators" and multi-layer stacks. ### Sovereign foundry scaling is accelerating to secure the AI interconnect supply chain Fabrication and foundry scaling represents 31.2% of the total activity (2,964 documents), signaling a transition from component research to industrial-scale production. Department of Commerce. TSMC appears in 22% of device fabrication records, suggesting that the path to market is consolidating around a few high-volume foundries in Taiwan and China. ### A "research lock-in" exists where academic entities hold the fundamental IP for active modulation while commercial foundries focus on passive infrastructure There is a stark misalignment between commercial narratives and the underlying patent baseline. While trade coverage and foundry announcements from GlobalFoundries and TSMC focus on high-volume Co-Packaged Optics (CPO) and 1.6T transceivers, the actual inventions for next-generation active modulation—such as non-volatile phase-change materials (PCM) and hybrid lithium niobate integration—are dominated by Chinese academic institutions like Zhejiang and Jilin Universities. Specifically, the Active Modulation node shows a high patent survival rate of 83% but a total absence of commercial media coverage across 879 documents, suggesting these critical functional layers are being perfected in a pre-commercial vacuum. ### Commercial narratives for AI interconnects are outrunning the underlying patent baseline There is a measurable gap between public market enthusiasm and the actual volume of core inventions for finished photonic processors. While items of commercial coverage (362) signal urgent demand for 1.6T transceivers in AI data centers, the patent record remains heavily weighted toward underlying material assembly and testing infrastructure rather than realized logic. Furthermore, despite the $8.1 billion raised by sector participants in mid-2026, the technical record in areas like active modulation shows a total absence of trade coverage across 879 inventions, suggesting these critical functional layers remain in a "research lock-in" phase dominated by academic institutions. ### Active modulation R&D is decoupling from commercial media narratives Active Modulation & Switching represents 9.2% of the corpus (879 documents) and maintains a high survival rate of 83%, yet it shows a total absence of commercial media or trade coverage. The technical record is dominated by Chinese academic institutions—with Jilin University and Zhejiang University accounting for 8% of the node—focusing on "zero-static-power" hardware using phase-change materials (PCMs) like GeSbTe (GST). This suggests a "research lock-in" phase where fundamental IP for next-generation optical neural networks is being secured by academic entities ahead of industrial adoption. ### The competitive moat is shifting from optical design to specialized back-end-of-line (BEOL) materials handling and vertical alignment precision Across the landscape, the technical engine is moving away from lithographic waveguide geometry toward the mastery of material interfaces. The portfolio favors expensive structural refinements, such as specialized etching grooves and silver-glue bonded TEC structures, over commodity manufacturing. This is mirrored in the Packaging node, where R&D is concentrating on proprietary material gradients, such as varying oxygen-to-nitrogen ratios in silicon oxynitride layers, to manage optical coupling loss. ### 3D vertical stacking is replacing planar silicon-on-insulator as the dominant R&D architecture The portfolio shows a transition from monolithic Silicon-on-Insulator (SOI) toward 3D photonic integrated circuits (3D-PICs) to solve signal latency and thermal bottlenecks. Published activity in packaging surged from 34 documents in 2025Q1 to 60 in 2026Q2, with focus concentrated on vertical photonic vias and PECVD-deposited isolation layers. This move toward verticality is further supported by the use of "trident" structures for dissimilar material bridging and multi-layer Si-SiN stacks to eliminate substrate leakage. ## Overview Silicon Photonics is transitioning from a planar, silicon-only discipline to a complex 3D platform play where the integration of heterogeneous materials and vertical interlayer coupling is being used to bypass the physical scaling limits of traditional architectures. This shift moves the industry's center of gravity from optical design toward mastery of vertical etching precision and multi-material bonding, effectively transferring the burden of reliability and yield management from the assembly stage to the foundry level. The technical record is moving decisively toward 3D architectures to solve the physical interface bottleneck. This shift is supported by a rise in technical disclosures in the Packaging & System Integration area, which peaked at 60 documents in 2026Q2, signaling a strategic move away from monolithic Silicon-on-Insulator (SOI) architectures toward multi-material stacks that utilize engineered refractive index gradients. A sharp contradiction exists between technical requirements and commercial narratives. This gap suggests that while commercial demand for AI-scale interconnects is surging, the underlying patent baseline for the necessary high-temperature manufacturing processes is still maturing and fragmented. Intellectual property for active modulation is concentrated in academic institutions, creating a potential commercial bottleneck. The total absence of commercial media or trade coverage across these 879 inventions suggests a "research lock-in" phase where fundamental patents for 3D signal routing and non-volatile weights are held by non-industrial players, potentially necessitating aggressive cross-licensing as these technologies move toward production. Foundries are evolving into system integrators as they internalize complex material libraries. This transition indicates that foundries are moving beyond substrate provision to handle complex materials like lithium niobate and tellurium oxide, which will likely marginalize smaller Outsourced Semiconductor Assembly and Test (OSAT) providers who cannot afford the capital expenditure for the required MEMS-driven micro-positioners and laser stealth cutting. ## Areas - Waveguide & Component Design · 4,046 documents - Fabrication & Foundry Scaling · 2,964 documents - Packaging & System Integration · 940 documents - Active Modulation & Switching · 879 documents - Next-Gen Compute & Sensing · 627 documents