The Compound Semiconductor Layer Inside Your 5G Phone

When people talk about a 5G phone, they talk about the processor and the camera. The part that decides whether the phone can actually hold a call at the edge of a cell sits elsewhere: a small module between the modem and the antenna, built on a material that is not silicon. I want to walk through that layer, because it is a good example of a pattern that runs through the whole electronics supply chain. The parts that matter most to the finished product are often the ones nobody puts on the box.

This piece follows the compound semiconductor layer, meaning indium phosphide (InP), gallium arsenide (GaAs) and gallium nitride (GaN), from the crystal to the handset. It relies on the compound-semiconductor and RF front-end material in my ESDM research book. Where that material stops, I say so. It stops earlier than I would like on lasers, and I will be direct about that gap rather than fill it with numbers I cannot source.

Why silicon is the wrong tool for two jobs

Silicon is unmatched for dense digital logic. Its crystal growth is mature and almost defect-free, which is what lets a fab pattern billions of transistors on one die. But two jobs sit outside what silicon does well. One is emitting light efficiently. The other is handling high-frequency power. Compound semiconductors, formed from group III and group V elements such as gallium and arsenic, exist to cover those two jobs.

The light-emission problem comes down to band structure. Silicon has an indirect bandgap, so an electron recombining across the gap rarely gives up its energy as a photon. Indium phosphide, and the related indium gallium arsenide phosphide alloys grown on it, has a direct bandgap. Recombination there can emit a photon efficiently. That single physical difference is why the fiber-optic telecom industry runs on III-V materials and not on the silicon that dominates every other layer of the stack.

The RF-power problem is about breakdown behaviour and electron mobility. Gallium arsenide and gallium nitride tolerate higher voltages and move electrons faster than silicon, which is what a power amplifier needs when it pushes signal out of an antenna. Silicon carbide is the other famous compound semiconductor, but its home is traction inverters and industrial power conversion, and I leave it out of this piece.

Three material systems, three different jobs

The three systems share a family name but not much else. Each pairs a specific crystal growth method with a specific application, and each has a narrower foundry base than digital logic.

Indium phosphide and the telecom laser

InP is the substrate for lasers at the 1310 nanometre and 1550 nanometre wavelengths that define long-haul and metro fiber links. It also supports ultra-high-frequency RF devices above 100 GHz. Those wavelengths matter because they sit where optical fibre loses the least light, so a laser that emits there can carry data a long way.

This is the thinnest part of my source material. The book identifies InP for these two roles and stops. It does not document the laser structures used in commercial optical transceivers, such as distributed feedback lasers, externally modulated lasers or vertical-cavity surface-emitting lasers. It does not name InP wafer suppliers, give yields, or describe packaging and pluggable module formats. An earlier draft of that chapter filled the gap with an invented market figure and a made-up citation, and it was removed for exactly that reason. I am not going to reinvent it here. If you want the photonics story in depth, it needs new primary research, and I would treat it as an open question, not a settled one.

Gallium arsenide and the handset amplifier

GaAs is the substrate for the power amplifiers in 4G and 5G mobile phones. It is produced at commercial scale on 150 millimetre (6-inch) wafers grown by the Liquid Encapsulated Czochralski method, a relative of the crystal-pulling process used for silicon. The foundry base is small. Win Semiconductors is the specialist foundry named in my source, and RF module designers such as Qorvo and Skyworks depend on that kind of capacity instead of growing their own GaAs substrates.

Gallium nitride and the wider power range

GaN is grown by metal-organic chemical vapour deposition as an AlGaN/GaN high-electron-mobility transistor structure, on silicon, sapphire or silicon carbide substrates, at 150mm and 200mm. It covers a wide voltage span, roughly 40V to 650V. At the low end that means USB Power Delivery fast chargers and AI server power supplies. At the high end it means 5G base-station power amplifiers, frequently grown on silicon carbide. The pairing is deliberate: GaN brings high electron mobility, and SiC brings better thermal conductivity for pulling heat out under sustained transmit power.

The pattern is worth stating plainly. Silicon logic has one dominant growth method feeding a large foundry base, backed by roughly seven decades of process learning. Each III-V system has its own method, and controlling MOCVD layers at atomic precision or growing GaAs boules is harder. The supplier base has stayed narrow while unit demand has grown.

Inside the handset: the RF front-end as a signal chain

The clearest documented example of how these materials become a working product is the Qualcomm Snapdragon 5G modem-RF system. Qualcomm is described in my source as the leading supplier of 5G modems, baseband processors and integrated RF front-end modules, holding an estimated 65% of the premium 5G smartphone modem-RF market. That figure is attributed to Counterpoint Research and TechInsights tracking, so it is an estimate and I treat it as one.

The chain runs in a specific order. A 5G baseband processor, the Snapdragon X75 fabricated on TSMC's 4nm FinFET process, hands off to an intermediate-frequency transceiver, the SDR735 on TSMC's 14nm process. That transceiver feeds two parallel RF paths. One is a sub-6GHz front-end module combining a power amplifier, a bulk acoustic wave filter and a low-noise amplifier. The other is a millimetre-wave antenna-in-package module, the QTM565, for the higher and shorter-range bands.

The green box is the one that relies on compound semiconductors. The power amplifier module, called a PAMiD (power amplifier module with integrated duplexer), bundles a GaAs amplifier die together with the duplexer that separates transmit and receive paths sharing one antenna. My source names the die as GaAs HBT capacity from Win Semiconductors or licensed to GlobalFoundries. The design logic is practical. A modern phone has to support dozens of bands across 4G, 5G sub-6GHz and 5G millimetre-wave in a device a few millimetres thick, and placing discrete parts at that band count no longer works.

The acoustic filter is the other strategic part

The second design choice worth noting is bulk acoustic wave filtering. In the documented case it comes from Qualcomm's own RF360 filter business in Munich. BAW filters hold their sharp bandpass shape at the higher frequencies and tighter band spacing that 5G New Radio requires, where the surface acoustic wave filters common in lower-frequency 4G bands fall short. Along with the power amplifier, the filter is one of the two RF components that manufacturers treat as differentiated rather than commodity.

The headline spec on a phone is the 10 Gbps downlink figure, which belongs to the digital baseband. That number still has to survive the analog physics of the front end before it reaches a base station. The amplifier, filter and antenna module are where the RF bill of materials, and the import dependency behind it, actually concentrate.

Demand: handsets at scale and base stations at high power

The scale is what keeps this layer growing. A flagship 5G smartphone integrates over 1,500 individual components across its main board and sub-boards, built on 10-to-14-layer HDI any-layer PCBs. IDC and Counterpoint Research estimate more than one billion smartphones ship globally each year, and each carries some version of the RF front-end stack. RF module demand therefore tracks global handset volume, not any single flagship cycle.

The same materials appear at far higher power in infrastructure. Massive MIMO active antenna units for 5G, and emerging 6G, base stations integrate 64 transmit and 64 receive channels, built around GaN-on-SiC power amplifiers. An outdoor unit has to dissipate over 500 watts of heat through die-cast aluminium heat sinks and embedded vapour chambers. My reading is that it is the handset PAMiD logic repeated at fifty to a hundred times the power level and packaged for the outdoors, which is the comparison the book itself draws.

Where India sits in this layer

India's documented position is assembly and export, not RF-module fabrication. Under the Production Linked Incentive scheme for large-scale electronics manufacturing, India became the world's second-largest mobile phone manufacturer. Final assembly is anchored by Foxconn (Tamil Nadu, Karnataka), Pegatron (Tamil Nadu), Tata Electronics (through its Wistron acquisition in Karnataka), Dixon Technologies (Noida) and Samsung, whose Noida plant my source describes as the world's largest single smartphone facility. Mobile exports crossed $15 billion in FY 2023-24, among the country's top five exported commodities, against annual domestic mobile manufacturing value above $45 billion.

Here is the important boundary. Those phones are populated with boards that already carry RF modules sourced from GaAs foundries and module suppliers such as Qorvo, Skyworks and Qualcomm's RF360 business. The book does not document any GaAs power amplifier fabrication, BAW filter production or antenna-in-package assembly for these devices on Indian soil, and I would not claim it does. That is an evidence gap, and it is a different claim from the wafer-fab, packaging and PCB investments that appear elsewhere in India's electronics story, which belong to other layers.

Why the foundry base stays narrow

It is worth asking why the supplier list in this corner is so short, because the answer explains most of the strategic risk. Silicon crystal growth has had roughly seven decades of process learning, and a single dominant method, Czochralski boules, feeds a large and fairly undifferentiated foundry base. The III-V systems do not have that history. Each pairs a distinct growth method with a distinct application. Metal-organic chemical vapour deposition has to lay down AlGaN/GaN layers with atomic precision. Liquid Encapsulated Czochralski growth has to produce GaAs boules that behave. The reactors and the recipes are less forgiving, and the people who know how to run them are fewer.

The result is a supply base that has stayed concentrated even as unit demand grew. I think this is the right way to read the names in the table. Win Semiconductors is not one supplier among fifty. It is one of a small number of specialist compound semiconductor foundries, and module designers such as Qorvo and Skyworks depend on that kind of capacity instead of growing their own GaAs substrates. Qualcomm's PAMiD amplifier die appears in my source as fabricated by Win Semi or licensed to GlobalFoundries, which tells me the amplifier stage rests on a very small number of fabs.

Design companies and foundries are different risks

A point that is easy to miss is that the names on the module (Qorvo, Skyworks, Qualcomm) are designers and integrators. The manufacturing risk sits one layer down, at the foundry and at the substrate. A buyer who checks only the module vendor sees a competitive market. A buyer who follows the die to its foundry sees something narrower. That is the same pattern I see in other parts of the electronics chain, where a visible brand sits on top of a much more concentrated manufacturing step.

Reading the RF module as a bill of materials

The RF front end is now a line item of its own. In a flagship 5G phone with over 1,500 components, the PAMiD modules, BAW and SAW filters and antenna tuners sit next to a 3nm or 4nm application processor and modem as some of the most complex and highest-value subsystems on the board. The board itself is a 10-to-14-layer any-layer HDI PCB with microvias filled by electroplated copper. All of that is packed into a device a few millimetres thick that must handle dozens of cellular bands at once.

Why does this matter for supply risk? Because RF module demand scales with global handset volume, and a billion-plus phones a year is a lot of amplifier die. When a single component family is embedded in every handset, a shortage or an export restriction at the substrate level travels straight up to the end product. The link to gallium is direct: GaAs and GaN both start from gallium, and I treat that supply in a separate piece on gallium and germanium.

What the photonics gap tells us

I said earlier that the laser half of this topic is thinly sourced. It is worth being specific about what is missing, because the missing list is itself a research agenda. My source does not document laser diode device physics, including the distributed feedback, externally modulated and vertical-cavity surface-emitting architectures that dominate commercial optical transceivers. It does not cover silicon photonics, meaning optical waveguides, modulators and germanium photodetectors integrated on a silicon platform, which is an increasingly important alternative to discrete III-V optics for data-centre interconnects. It does not cover pluggable optical transceiver formats or the coherent signal processing behind long-haul links, and it does not substantiate co-packaged optics.

I mention this list so that you know where to be sceptical. When an article, including a confident one, quotes an InP market size or an optical transceiver share, ask what the primary source is. My book removed exactly that kind of unsupported figure from an earlier draft. I would rather publish a narrower piece than one that reads well and cannot be traced.

What would close the gap

The fix is not better prose. It is primary research: silicon photonics platform references from bodies such as the IEEE Photonics Society, technical references from transceiver module vendors, and laser-diode fabrication references from the compound semiconductor foundries that specialise in InP. Until that work exists, the safest statement about InP is the one the source supports. It is the substrate for 1310nm and 1550nm telecom lasers and for RF devices above 100 GHz, and it works because of its direct bandgap.

What to take away

I would take four things from this. First, compound semiconductors exist because of physics that silicon cannot change: a direct bandgap for light, and breakdown and mobility advantages for RF power. Second, the handset RF front end is a supply chain of its own, and only the amplifier stage depends on the III-V layer. Third, India's strength is assembly. The step that would change its exposure is a GaAs or GaN device supply base, and on the evidence I have, nothing at that level is documented. Fourth, the photonics half of this topic is under-sourced, so any confident claim about lasers, silicon photonics or co-packaged optics should be treated as unsupported until someone does the primary research.