Glass-Substrate IPD: The RF and Integration Advantages

Glass-substrate IPD: why through-glass and thin-film change RF integration

Integrated passive devices (IPDs) put the passives that surround a radio — matching networks, filters, couplers, baluns, and decoupling — onto a single thin-film substrate instead of placing them as discrete components. For two decades the default carrier was high-resistivity silicon or a ceramic such as alumina and LTCC. Glass is now taking share in mmWave, Wi-Fi front-end, and optical/RF co-packaged modules, and the reasons are electrical rather than fashion.

What a glass IPD actually is

A glass IPD is a planar passive circuit built in thin metal layers on a glass wafer or panel, typically with:

• electroless/electroplated Cu or evaporated Au conductors several micrometres thick;
• through-glass vias (TGV) for vertical ground and signal transitions;
• redistribution layers on both sides for grounded structures;
• panel-scale processing borrowed from display manufacturing, which keeps cost down at large format.

The result sits electrically between an LTCC module (3-D, many layers, lossy at high frequency) and a planar alumina thin-film circuit (excellent RF, small panel, expensive).

Advantage 1 — dielectric loss that is nearly invisible

RF-grade borosilicate, fused silica, and alkali-free glasses show loss tangents around 0.001 or lower at mmWave frequencies, with dielectric constants near 5 that are stable across frequency and temperature. Alumina has even lower loss but εr near 9–10 forces tighter, harder-to-fabricate lines; silicon substrates need heavy doping or suspended structures to suppress substrate-mode loss. On glass, a 50 Ω line is wide enough to fabricate repeatably and the dielectric contributes almost nothing to the link budget.

That matters directly at 28 GHz, 39 GHz, and in 112 Gbps/224 Gbps electrical lanes, where a few tenths of a dB per centimetre decide whether a channel closes.

Advantage 2 — dimensional stability and surface quality

Glass is amorphous, so there is no grain structure, no anisotropic shrinkage, and near-zero warpage through thin-film temperature cycles. Typical total thickness variation on qualified wafers is a few micrometres across 200/300 mm, and surface roughness below 1 nm is routine after polishing. For thin-film microstrip and grounded coplanar lines, smooth surfaces mean conductor loss stays close to the theoretical value instead of being dominated by surface roughness at mmWave frequencies.

The same dimensional stability is why TGVs can be placed on tight pitches and why impedance stays repeatable from panel to panel.

Advantage 3 — TGVs beat TSVs for passive RF

Through-silicon vias need a conductive liner plus a dielectric isolation layer inside the via, and the silicon around them supports parasitic modes and crosstalk. A through-glass via is an insulator with metal in the hole: no junction capacitance, no need for a depletion region, and far better isolation between adjacent vias. For a grounded coplanar transition, a dense TGV fence provides an almost ideal sidewall ground, suppressing parallel-plate and cavity modes.

Practical consequences:

• shorter, lower-inductance ground returns for PA matching;
• cleaner launch from package to board or antenna;
• isolation between transmit and receive sections on one die.

Advantage 4 — panel economics

Glass IPDs are increasingly made on 510 × 515 mm panels on adapted display lines. The area per dollar is dramatically better than 200 mm wafers, and the process steps are the familiar thin-film sequence: seed layer, lithography, plating, and etch. For low-complexity passive dies with large footprint — antenna-in-package substrates, distribution networks, large baluns — panel glass is the first IPD technology where the substrate cost stops being the dominant line item.

Advantage 5 — co-packaging with photonics and silicon

Silicon photonics interposers already use glass and silicon carriers, and CPO (co-packaged optics) assemblies need dense, low-loss RF feeds from ASIC serializers to optical engines on the same substrate. A glass IPD handles the impedance matching, AC coupling, and ESD structures with almost no dielectric penalty, and its coefficient of thermal expansion can be chosen to sit close to silicon and the lid material. The same substrate carries optical alignment features, which ceramic and organic packages do poorly.

Where glass is not the answer

Be direct about the limits:

• Thermal conductivity is low (~1 W/m·K). Glass is a passive carrier, not a heat spreader; high-power PA dies still need Cu coin, diamond, or nitride paths.
• Fracture mechanics. Glass is strong in compression but sensitive to edge defects; singulation and handling need process discipline.
• Ecosystem younger than ceramic. Qualified TGV suppliers and design kits are fewer than for alumina or LTCC, though the list is growing fast.
• Low-ε geometry. At low GHz, εr≈5 means larger passives than on εr≈9 ceramic or high-ε LTCC; miniaturization is less aggressive.

Glass vs. the alternatives at a glance

How to evaluate a glass IPD supplier

Five questions separate qualified vendors from slideware:

1. Measured insertion loss per millimetre of grounded coplanar line at your target frequency, not simulations.
2. TGV pitch, aspect ratio, and measured crosstalk between adjacent vias.
3. Warpage and thickness variation on a full panel after metallization.
4. Reliability qualification: thermal cycle, humidity, and drop data on the actual stack.
5. Design kit quality — de-embedded models, not just S-parameter sweeps at one bias.

Bottom line

Glass-substrate IPDs combine near-ideal dielectric behavior, smooth dimensional control, low-parasitic through-substrate grounding, and panel-scale economics. They are the natural carrier for mmWave passive networks, RF feeds in co-packaged optics, and the matching/filter blocks of reconfigurable front ends — everywhere the alternative ceramic or silicon solution is either too lossy, too noisy parasitically, or too expensive at the area the circuit needs.