---
title: Glass-Substrate Photonic Interposer Overview
url: https://www.emergentmind.com/topics/glass-substrate-photonic-interposer
type: topic
---

# Glass-Substrate Photonic Interposer Overview

A glass-substrate photonic interposer can be understood as a glass-supported optical redistribution layer that performs routing, fanout, pitch conversion, wavelength handling, vertical redirection, or chiplet interfacing between fibers, photonic chips, and heterogeneous electronic-photonic assemblies. In the present literature, the term spans direct fused-silica optical interposers with buried waveguides, embedded mirrors, and self-alignment sockets; glass-supported resonant nanophotonic blocks whose cladding geometry is part of the optical design; deposited-glass waveguide platforms for low-loss routing and resonant filtering; glass-carried printed chiplet couplers; and panel-scale proposals in which a single glass package substrate hosts a reconfigurable WDM optical network [2512.01972], [2311.00935], [1404.5610], [2402.11988], [2508.06079].

## 1. Architectural scope and typology

The most direct realization of a glass-substrate photonic interposer is a monolithic fused-silica optical routing chip inserted between external fibers and a silicon photonic die. In “Laser Structured Optical Interposer for Ultra-dense Vertical Coupling of Multi-core Fibers to Silicon Photonic Chip” the entire optical redistribution function is implemented inside a **1 mm thick polished fused-silica wafer**: six **7-core MCFs** are converted into a **4 × 10** silicon-photonic grating-coupler region by buried waveguides, embedded TIR micro-mirrors, and FLICE-opened fiber sockets, yielding **40 optical channels** with an **average single-pass insertion loss of 5.0 dB for the fully packaged system**, and a **minimum loss of 3.2 dB** [2512.01972].

A second architectural class uses glass primarily as a host or carrier for integrated photonic functions rather than as a fully monolithic routing medium. The clearest examples are the **GaAs 2D photonic crystal slab nanocavity on glass with an upper glass thin film**, where the glass environment determines cavity loss channels, and the **3D-nanoprinted interposer** on a **silica/glass carrier**, where printed free-space reflective couplers, chip-coupling frames, and fiber-guiding funnels bridge silicon and InP chiplets without changing the foundry process flow of the constituent dies [2311.00935], [2402.11988].

A third class is architectural and prospective: glass is treated as a package-scale substrate that can host a photonic backplane. The panel-scale proposal defines two forms, a **PIC-embedded glass interposer** and a **monolithic-photonics-integrated glass interposer**, both aimed at a **single large glass package/interposer substrate** carrying chiplets, HBM stacks, WDM waveguides, optical crossbar switches, and comb sources over areas **up to 500 mm × 500 mm or larger** [2508.06079].

Adjacent work broadens, but also sharpens, the definition. The polymer-on-glass photonic-crystal spectrometer is best treated as **substrate-guided in-plane light distribution with lithographically defined, spatially localized wavelength-selective outcouplers**, not as a core interposer; the Hydex papers present a **high-index doped silica glass CMOS compatible platform** with low-loss routing and high-Q resonators, but not a complete interposer; the multilayer SiN microcomb interposer and the dual-layer SiN 3D interposer are not glass-native, yet they formalize the passive-backplane and multilayer-routing roles that a glass substrate may assume [1001.0580], [1505.05953], [2010.09479], [2604.12566].

## 2. Materials systems and structural implementations

Several distinct material stacks recur in the literature, and they imply different interposer roles.

| Platform | Stack | Reported figures |
|---|---|---|
| Fused-silica optical interposer | **Corning 7980 polished fused silica**, **1 mm thick**, with laser-written waveguides, FLICE mirrors, and sockets | **40 optical channels**, **six 7-core MCFs**, **5.0 dB** average single-pass packaged loss |
| Glass-supported PhC resonator | **GaAs** slab, **\(h = 130\ \mathrm{nm}\)**, on **glass \(n=1.45\)** with upper glass film | **\(Q \sim 10^4\)** bare on glass; **\(Q = 4.2\times10^5\)** at **\(t \approx 260\ \mathrm{nm}\)** |
| Hydex deposited-glass PIC platform | **High-index doped silica glass** core **\(n = 1.7\)** in **SiO\(_2\)** | **17%** contrast; **\(0.04\ \mathrm{dB/cm}\)** propagation loss; **\(Q = 1.2\times10^6\)** ring |
| Polymer-on-glass PhC slab guide | **396 nm PMMA** on **22 mm × 22 mm**, **0.96 mm** glass coverslip | **450–700 nm** operation; **\(3\times3\)** array of PhC patterns |

The fused-silica interposer uses femtosecond-laser-induced refractive-index modification for buried waveguides and **femtosecond laser irradiation followed by chemical etching (FLICE)** for hollow micro-optical structures. Its mirrors are **elliptical disks**, about **30 \(\mu\)m tall** and **\(\sim 54~\mu\)m wide**, tilted **50° with respect to the waveguides** to induce TIR, while the waveguides occupy four depth layers centered nominally at **50, 140, 230, and 320 \(\mu\)m** [2512.01972].

The glass-supported nanocavity is structurally much smaller and much more sensitive to vertical refractive-index symmetry. It consists of a **GaAs 2D photonic crystal slab nanocavity on glass** with an additional **upper glass thin film**, using a **W0.9 line-defect waveguide** and a multistep heterostructure created by local elongation of the lattice constant in the \(x\)-direction by **+2.5%** and **+1.25%** for designated holes. In air-clad conditions the same cavity supports **\(Q = 4.9\times 10^7\)** and **mode volume \(V = 1.7(\lambda/n)^3\)** near **\(\lambda = 918\ \mathrm{nm}\)** [2311.00935].

Hydex represents a different strategy: the interposer-like function is realized in a deposited-glass waveguide circuit rather than in bulk glass machining. One paper reports a **core refractive index \(n = 1.7\)**, **17%** core-cladding contrast, **\(0.04\ \mathrm{dB/cm}\)** propagation loss, **\(\sim 1.5\ \mathrm{dB/facet}\)** fiber coupling, and ring resonators with **\(Q = 1.2 \times 10^6\)** and **200 GHz FSR** [1404.5610]. A related Hydex review reports **propagation loss as low as \(0.06\ \mathrm{dB/cm}\)**, **fiber pigtail / coupling loss about \(1.5\ \mathrm{dB}\)**, and **negligible bending losses for radii down to \(30\ \mu\mathrm{m}\)**, emphasizing long routing paths and compact resonant functions in a silica-derived platform [1505.05953].

Glass can also act as a low-index slab guide supporting local patterned functions. In the polymer-on-glass spectrometer, the main waveguiding body is a **glass coverslip** with **\(n \approx 1.52\)**, while a **PMMA** overlayer with **\(n = 1.499\text{--}1.512\)** is patterned into **square-lattice** PhC patches whose lattice constants span **300 to 420 nm** in the final device. This is a weak-index-contrast substrate-guided system rather than a channelized interconnect, but it demonstrates that local wavelength-selective extraction can be integrated directly on glass [1001.0580].

## 3. Optical physics specific to glass-supported interposers

A central optical problem for high-confinement devices on glass is vertical asymmetry. The GaAs PhC study makes this explicit: on glass with air above, the refractive-index environment is asymmetric, so the intended **TE-like cavity mode** hybridizes with **TM-like slab modes**. The paper identifies the loss chain as **TE–TM mode conversion**, followed by lateral leakage because the **TM-polarized light does not feel any photonic bandgap effect**. In the multistep heterostructure cavity this collapses the cavity from **\(Q = 4.9\times 10^7\)** in air to **\(Q \sim 10^4\)** for the bare on-glass case. Adding a top glass film suppresses this asymmetry-driven conversion, and the optimized thickness **\(t \approx 260\ \mathrm{nm}\)** yields **\(Q = 4.2\times 10^5\)**, which is **1.7×** higher than the thick symmetric-glass case that saturates near **\(2.5\times 10^5\)** [2311.00935].

That result also corrects a common simplification: a perfectly thick symmetric top cladding is not always the optimum on glass. The same paper attributes the improvement at intermediate \(t\) to **mode mismatch between the cavity’s radiation and the optical modes supported by the upper glass film**. In the authors’ decomposition, **lateral leakage** dominates for **\(t \lesssim 200\ \mathrm{nm}\)**, **bottom leakage** is relatively insensitive to \(t\), and **top leakage** exhibits a distinct dip around **\(t \approx 250\text{--}260\ \mathrm{nm}\)**. This establishes that in glass-supported nanophotonics the overglass is part of the optical design, not merely encapsulation [2311.00935].

A second, very different, glass-specific optical regime is multimode slab transport with local extraction. In the photonic-crystal spectrometer, light is edge-coupled into a **0.96 mm** thick glass slab and scattered upward by reciprocal lattice vectors \(\mathbf{G}\) from small PhC patches. The strongest extraction occurs when **\(|\mathbf{k}_{\parallel}| \approx |\mathbf{G}|\)**, giving the design rule **\(\lambda/n \approx a\)**. The architecture therefore uses different lattice constants to realize different spectral response functions and reconstructs a spectrum through the matrix equation **\(\mathbf{A}\mathbf{x}=\mathbf{b}\)** and a Moore–Penrose pseudoinverse [1001.0580]. This suggests that a glass interposer need not be restricted to point-to-point channels; it can also host spatially distributed monitor taps or wavelength-selective pickoff sites.

A third optical regime is low-loss, moderate-contrast deposited-glass routing. Hydex is explicitly framed as promising for **telecommunications and on-chip WDM optical interconnects for computing**, with **negligible nonlinear absorption in the telecom band**, **\(\gamma = 233 \ \mathrm{W^{-1} km^{-1}}\)** in one report, and **\(0.04 \ \mathrm{dB/cm}\)** propagation loss [1404.5610]. The related review gives **\(\gamma = 220\ \mathrm{W^{-1} km^{-1}}\)**, **\(n_2 = 1.1 \times 10^{-19}\ \mathrm{m^2/W}\)**, **\(A_{\mathrm{eff}} = 2.0\ \mu\mathrm{m}^2\)**, and operation with no detectable multiphoton absorption up to **\(25\ \mathrm{GW/cm^2}\)** [1505.05953]. For interposer use, a plausible implication is that the dominant value of such glass platforms lies less in their nonlinear demonstrations than in their passive routing, WDM filtering, and fiber-compatible I/O.

## 4. Routing, fanout, multilayer redistribution, and wavelength handling

At the package level, the fused-silica optical interposer is the clearest demonstration of what a glass-substrate photonic interposer does operationally. It accepts **six 7-core MCFs**, routes **40 channels** through a **4 to 6 mm** fanout, uses **four optical depth layers**, redirects light with embedded TIR mirrors, and launches toward SiP gratings at **10° off vertical**. The design space includes three routing topologies, with **Design C** emerging as the best compactness-loss compromise because crossings at **80°** contribute **< 0.01 dB/intersect** and produced no measurable crosstalk above a **40 dB** noise floor, even with up to **16** crossings. The final socketed interposer showed **-2.75 dB** average channel loss with **\(\sigma = 0.96\) dB**, and the packaged loop response had **48 nm 3 dB bandwidth** with a peak near **1504 nm**, the bandwidth limit being dominated by the SiP grating couplers rather than the glass interposer itself [2512.01972].

Broadband passive-backplane logic appears in the microcomb interposer architecture, even though that work is implemented in **stoichiometric silicon nitride** rather than on glass. Its interposer collects broadband light from chiplets, routes it, splits it into the **\(\sim 2\,\mu\mathrm{m}\)**, **\(\sim 1.55\,\mu\mathrm{m}\)**, and **\(\sim 1\,\mu\mathrm{m}\)** bands with two cascaded dichroic directional couplers, suppresses the pump with thermally tunable microrings, and interfaces to SHG, detectors, and photonic chiplets. The reported extinction ratios are typically in the **15–25 dB** class, the MMIs have **< 0.5 dB** excess loss, and the thick-thin SiN bilayer taper is simulated to provide **< 1 dB across an octave** [2010.09479]. This suggests that the passive-backplane role of an interposer is largely substrate-agnostic even when its compact implementations are not.

The scaling problem of dense optical redistribution is formalized in the 3D SiN interposer. For a fully connected planar network of \(k\) nodes, the paper gives **\(C_{\mathrm{tot}} = \binom{k}{4}\)**, and for the demonstrated **12-node** case the all-planar routing baseline contains **495** intralayer crossings. By moving to a dual-layer 3D routing scheme, the fabricated interposer reduces this to **150**, below the cited all-planar lower bound of **153**, and achieves a **45.8% reduction experimentally in the average loss per waveguide**. The measured component losses are **\(0.123\,\mathrm{dB}\)** and **\(0.169\,\mathrm{dB}\)** per intralayer crossing for the two layers, **\(0.001\,\mathrm{dB}\)** per interlayer crossing, **\(0.360\,\mathrm{dB}\)** per interlayer taper, and **\(0.038\,\mathrm{dB/cm}\)** propagation loss [2604.12566]. Although not a glass implementation, this is directly relevant to any glass interposer that uses multilayer dielectric routing.

At the largest proposed scale, the panel-scale glass interposer defines a WDM optical network-on-package. Its per-waveguide bandwidth is given by **\(B_{\mathrm{WG}} = N_{\lambda}\cdot R_b\)** with **\(N_{\lambda}=32\)** and **\(R_b=32\ \mathrm{Gb/s}\)**, producing **\(1.024\ \mathrm{Tb/s}\)** per waveguide. In the unit-interposer comparison, the photonic system supports **832** links and **26.624 Tb/s** per XPU using **26 waveguides**, whereas the electrical silicon comparison uses **736 transmission lines** for **13.312 Tb/s**. The path-loss model assigns **0.25 dB** to each straight bypass and **0.5 dB** to each **90°** turn, and for a worst-case **8 × 8** tiled route the paper budgets **twelve 0° bypass switches**, **two 90° turn switches**, and an additional **3 dB** for edge couplers and waveguides, leading to the stated **\(-3\ \mathrm{dBm}\)** per-carrier launch requirement for **\(-10\ \mathrm{dBm}\)** receiver sensitivity and the associated **>500 mm** repeater-free reach claim [2508.06079].

## 5. Heterogeneous integration, alignment, and packaging

The strongest packaging result on a glass carrier is the 3D-nanoprinted interposer. Its **chip-coupling frame** uses printed fixers and clamps, with stated positioning accuracy of **±1 µm** from fixers and **sub-micrometer** accuracy with clamps. Optical coupling is performed by pairs of off-chip parabolic reflectors at polymer/air interfaces, governed by the mode-conversion relation **\(d_2 = d_1 \frac{f_2}{f_1}\)** and the ratio **\(q = \frac{f_1}{f_2}\)**. The demonstrated **mode-field-dimension conversion ratio of 5:2 from fiber to chip** is described as the largest mode size conversion using non-waveguided components, with **1.7 dB inherent coupling loss** and measured fiber-to-chip loss around **2.4 dB/facet I/O** over **1480–1620 nm**. The system further demonstrates **2.5 dB die-to-die coupling loss between silicon and InP chips** over the same **140 nm** wavelength range [2402.11988].

The fused-silica interposer emphasizes another packaging mode: passive fiber registration combined with active chip alignment. On the fiber side it uses FLICE-opened **self-aligning sockets** with length increased to **250 \(\mu\)m**, and assembly occurs by a **plug-and-twist** procedure with **Norland NOA81** UV-curable optical adhesive. On the chip side the interposer-to-SiP assembly uses **active alignment** and a **thin index-matched epoxy layer**. The reported packaging-induced penalty is about **0.3 dB**, of which **0.2 dB** is linked to random **\(\pm 1~\mu\mathrm{m}\)** waveguide-core position misalignment along the MCF length, and re-measurement after **two weeks** showed **< 0.1 dB** degradation [2512.01972].

Mechanical stack design becomes more critical when the photonic layer is thin, laminated, or stressed. The flexible chalcogenide-glass platform shows that conventional multilayer bending theory fails when **PI** and **SU-8** have moduli of **\(2.5\ \mathrm{GPa}\)** and **\(2\ \mathrm{GPa}\)** while the intermediate **silicone adhesive** is only **\(1.5\ \mathrm{MPa}\)**. The resulting **multiple neutral axes** permit placement of the neutral axis inside the SU-8 photonic region, and the paper derives **\(h_{3n} = \frac{h_3}{2} + d_1\frac{\bar{E}_1 h_1}{\bar{E}_3 h_3}\)** with the condition **\(h_3 > 8.28\ \mu\text{m}\)** for the SU-8 neutral axis to remain inside SU-8. Optically, the same platform reports **\(3.3 \pm 0.6\ \mathrm{dB/cm}\)** waveguide loss, intrinsic resonator \(Q\) up to **460,000**, and repeated bending to **0.5 mm radius** without measurable degradation [1307.5937]. A plausible implication is that laminated glass-photonic redistribution layers may require mechanics-aware co-design even when the primary interposer is nominally passive.

The GaAs-on-glass nanocavity paper adds a heterogeneous-integration detail of direct relevance: it explicitly notes that such structures can be implemented via **pick-and-place integration techniques including transfer printing** [2311.00935]. That statement is narrow, but important. It places glass-supported high-Q resonant blocks within a chiplet-assembly ecosystem rather than treating them as purely monolithic laboratory objects.

## 6. Limits, misconceptions, and research trajectory

One recurring misconception is that a glass-substrate photonic interposer is necessarily a single technology. The literature instead spans at least five distinct modalities: bulk fused-silica optical redistribution, deposited-glass waveguide PICs, glass-supported nanophotonic resonators, glass-carried printed free-space couplers, and proposed panel-scale glass photonic fabrics. Another misconception is that glass only supports low-functionality passive fanout. The cited work includes **high-Q** nanocavities on glass, **200 GHz FSR** Hydex rings, distributed spectral taps on glass slabs, **octave-spanning** passive interposer logic, and **reconfigurable** WDM crossbar concepts [2311.00935], [1404.5610], [1001.0580], [2010.09479], [2508.06079].

A second misconception is that a fully symmetric or fully thick dielectric cladding is always the best route to high performance on glass. The GaAs PhC study explicitly reports the opposite for its cavity family: **\(Q = 4.2\times 10^5\)** at **\(t \approx 260\ \mathrm{nm}\)** exceeds the thick symmetric-glass result of **\(2.5\times 10^5\)** because thin-film modal mismatch suppresses an additional radiation channel [2311.00935]. Similarly, a plausible implication of the rhombic top-glass analysis is that local overglass geometry can become a secondary loss-engineering parameter rather than a purely protective layer.

The limitations are equally clear. Several of the most informative papers are not complete interposer demonstrations. The PhC-on-glass work is **entirely numerical**, does not treat waveguide-cavity coupling, loaded versus unloaded \(Q\), insertion loss, thermal resistance, packaging reliability, or fabrication tolerances in a systematic way [2311.00935]. The Hydex papers demonstrate an excellent passive and nonlinear platform, but not an interposer with active devices, package-qualified integration, or electrical redistribution [1404.5610], [1505.05953]. The spectrometer is explicitly **useful background, but not a core interposer paper** [1001.0580]. The microcomb interposer and the 3D SiN interposer are architecturally relevant but not glass-native [2010.09479], [2604.12566]. The panel-scale glass interposer remains a proposal whose **>500 mm** repeater-free claim is supported by link-budget accounting rather than a measured panel-scale system, and whose open issues include thermal control, route contention, multilayer process maturity, large-area packaging, and defect tolerance [2508.06079].

The present state of the field therefore supports a precise conclusion. Glass-substrate photonic interposers are no longer limited to the idea of a transparent carrier with a few passive waveguides. The literature already contains a direct fused-silica interposer with dense vertical coupling and packaged multichannel performance, glass-supported resonant blocks whose loss physics is now quantitatively understood, deposited-glass platforms with low-loss WDM-relevant circuitry, glass-carried chiplet couplers with passive assembly, and a credible panel-scale architectural proposal. What remains unresolved is their unification into a single package technology that simultaneously delivers low-loss optical redistribution, high-density multilayer routing, robust heterogeneous assembly, electrical co-integration, manufacturable panel-scale process control, and experimentally validated system-level operation on glass.

Source: https://www.emergentmind.com/topics/glass-substrate-photonic-interposer