III-V Compound Semiconductor Fabs

III-V compound semiconductors — gallium arsenide (GaAs), indium phosphide (InP), indium gallium arsenide (InGaAs), and related alloys — enable specific capabilities that silicon cannot deliver regardless of node scaling. Direct bandgap compound semiconductors emit and absorb light efficiently, which silicon (an indirect bandgap semiconductor) cannot; they operate at RF frequencies with lower loss and higher efficiency than silicon at equivalent power levels; they respond at specific wavelengths (particularly short-wave infrared at 1.3–1.55µm) that silicon detectors are blind to. These capabilities make III-V semiconductors essential for three distinct industrial application clusters: RF front-ends in mobile devices and cellular infrastructure, optical telecommunications and datacom, and LiDAR / photonic sensing. Each cluster has its own operator landscape, its own qualification regime, and its own supply chain structure, though they share the upstream III-V substrate supply that is the defining chokepoint of the archetype.

The III-V archetype is small by silicon standards — global III-V wafer production measures in the low millions of wafer-equivalents per year across all substrate types, compared to hundreds of millions of 300mm silicon wafer equivalents. The small market size combined with specialty substrate supply concentrated at a handful of Japanese, Chinese, German, and UK producers makes the archetype particularly supply-shock-sensitive. A single LiDAR program at automotive volumes can absorb a meaningful fraction of global InGaAs wafer supply; a substrate disruption produces cascading effects across RF, telecom, and photonic sensing markets simultaneously.

Three Application Clusters

III-V semiconductor applications cluster into three families with distinct device types, customer bases, and operator landscapes. Understanding the cluster structure is prerequisite because "III-V fab" means different things in each cluster.

The clusters differ not just in device types but in production economics. RF front-ends are the largest cluster by wafer volume — billions of mobile handsets shipped annually each carry multiple GaAs RF die. Optical telecom and datacom has smaller unit volumes but higher value per device. LiDAR and photonic sensing is the emerging high-growth cluster with significant scaling ahead as automotive LiDAR transitions from demonstration programs to series production.

The Wafer Diameter Constraint

III-V substrate diameters are substantially smaller than silicon substrates, which directly constrains die throughput per wafer and production economics. Understanding the diameter progression matters for capacity and cost analysis.

The diameter constraint has direct economic consequence. A 100mm GaAs wafer produces a fraction of the dies that a 300mm silicon wafer produces at equivalent die size — roughly 1/9 the die count from area alone, before accounting for edge utilization and defect density. Die cost on 100mm GaAs is correspondingly higher than silicon die cost at equivalent node. The III-V industry's productivity growth depends primarily on yield improvement, not on diameter scaling, because diameter scaling faces fundamental substrate crystal-growth challenges that silicon substrate has largely overcome.

Substrate Supply Concentration

III-V substrate production is concentrated at a small set of specialty suppliers globally, with distinct concentration patterns by substrate type. This concentration is the archetype's primary structural constraint — the downstream device fabs depend on substrate supply that can be disrupted by trade policy, geopolitical events, or specialty supplier operational issues.

Gallium Export Controls and Upstream Metal Exposure

China's August 2023 gallium and germanium export controls introduced a specific upstream risk into the III-V substrate supply chain. China controls approximately 80% of global gallium refining capacity. The export control regime requires Chinese gallium exporters to obtain licenses for international sales, which has introduced both price volatility and supply uncertainty. Downstream gallium consumers — including GaAs and GaN substrate producers — have responded with inventory buildup, alternative supply development (limited in the near term given concentration), and advocacy for Western gallium refining capacity expansion.

The gallium exposure is structurally consequential because gallium is the Group III element in both GaAs (the largest-volume III-V substrate) and GaN (power and RF substrates, covered separately at GaN Power & RF). A sustained gallium supply disruption would affect both categories simultaneously, concentrating the risk. Indium is the Group III element for InP — China's indium position is less dominant than its gallium position but is still meaningful, and indium has been subject to parallel policy attention.

The broader structural observation is that III-V semiconductor supply chains are thin — few substrate producers, few merchant epi producers, concentrated upstream metal refining. The supply chain can support current industry scale but has limited buffer for demand shocks or disruption. This is particularly relevant as automotive LiDAR scales toward series production volumes that would add substantial incremental demand to already-thin supply chains.

The InGaAs APD Chokepoint

Long-range automotive LiDAR at 1550nm wavelength — the eye-safe wavelength that permits higher laser power and therefore longer detection range than the 905nm wavelength used in shorter-range LiDAR — requires InGaAs avalanche photodiodes (APDs) for detection. No practical substitute exists at scale for this wavelength and performance combination. InGaAs APDs are a specialty photonic device produced historically at low volumes for scientific, telecom, and defense applications; scaling InGaAs APD production to automotive LiDAR volumes represents a genuine manufacturing chokepoint.

The device physics is demanding. An APD works by accelerating photo-generated carriers through an internal avalanche multiplication region, producing gain that lets the detector respond to weak optical signals. The multiplication region must be produced with extremely tight thickness and doping uniformity — small variations produce large gain variations across the device. At scientific quality and low volume, this uniformity is achievable. At automotive volumes with AEC-Q101 qualification (the discrete semiconductor automotive qualification standard), uniformity must be maintained across production runs of millions of units. The transition from scientific-grade APD production to automotive-grade volume production is non-trivial.

The InGaAs APD supplier base is small: Hamamatsu Photonics (Japan, historical scientific and industrial APD leader), ON Semiconductor (through SensL acquisition), Excelitas Technologies, Broadcom / Avago (specialty photonic portfolio), and a handful of specialty operators. No single supplier has demonstrated clear capability to serve automotive LiDAR at mass-market BEV production volumes; industry capacity expansion is underway but trails the automotive LiDAR demand curve projected for the late 2020s. If automotive LiDAR adoption scales as forecasted by autonomous vehicle programs, InGaAs APD supply will be a binding constraint on how many vehicles can include long-range LiDAR.

This is one of the specific chokepoints that the "III-V supply chain is thin" observation manifests as in practice. The 1550nm LiDAR detection requirement cannot be substituted with silicon (wrong wavelength response), cannot be substituted with InGaAs PIN photodiodes (insufficient sensitivity for long-range), and cannot be substituted with GaAs devices (wrong wavelength). The physics of the application pins the supply chain to a narrow specialty category.

VCSEL Scaling

Vertical-cavity surface-emitting lasers (VCSELs) are the other high-volume III-V growth category. VCSELs emit laser light perpendicular to the substrate surface (unlike edge-emitting lasers), which enables dense 2D arrays of small lasers and low-cost wafer-scale testing. VCSELs at 850nm and 940nm wavelengths are mature at volume — the Apple Face ID camera uses VCSELs for 3D face recognition, and VCSEL arrays for industrial and short-range automotive sensing have scaled to high volumes.

Automotive LiDAR is driving a new generation of VCSEL scaling. Short-range automotive LiDAR at 850nm/940nm wavelengths can use VCSEL-based emitters, and several LiDAR programs have adopted this approach for cost advantages over edge-emitter diode lasers or fiber lasers. Longer-range automotive LiDAR at 1380nm (a different eye-safe wavelength option) has driven development of 1380nm VCSELs, an emerging product category.

Lumentum is the dominant commercial VCSEL supplier globally, with Apple Face ID as an anchor customer account. II-VI / Coherent is the second major VCSEL supplier with competitive position. TriLumina (Lumentum-acquired) provided specialty addressable VCSEL arrays. Automotive-qualified VCSELs require separate qualification from mobile-device VCSELs given automotive operating environment and lifetime requirements — operators must maintain parallel product streams.

Operator Landscape

III-V fab operators split into RF-focused and photonic-focused tiers, with some consolidation as Coherent (post-II-VI merger) and others operate across clusters. The operator base is substantially smaller than silicon fab operators.

Long Qualification Cycles

III-V device qualification for automotive, telecom, and defense applications runs substantially longer than commercial silicon qualification. Automotive discrete qualification under AEC-Q101 requires 3–5 years from initial product submission through qualification completion. Telecom qualification under Telcordia GR-468 (photonic device reliability) requires 2–3 years. Defense qualification under MIL-STD-883 runs multi-year. The practical consequence is that substrate and epi supply decisions made in 2023 define production capacity in 2026–2028 at the earliest — qualification lag structurally constrains how fast the III-V industry can respond to demand shifts.

For automotive LiDAR specifically, the qualification lag means that LiDAR programs launching commercial vehicles in the late 2020s depend on substrate, epi, and device qualifications initiated several years prior. Programs that deferred qualification investment during the 2022–2023 slowdown in autonomous driving timelines will face capacity constraints when demand reaccelerates. Programs that invested through the slowdown have qualification advantage as demand recovers.

Cross-Network Connections

III-V semiconductors connect to multiple other pillars of the SiliconPlans network. LiDAR is an EV/autonomous vehicle component — InGaAs APDs and VCSELs both feed into the ElectronsX autonomy and sensor coverage. Silicon photonics uses InP lasers as integrated light sources — the SX Silicon Photonics archetype is adjacent, and co-packaged optics for AI cluster interconnect represent emerging SX-DX interface applications. RF front-ends touch 5G/6G infrastructure and mobile handset supply chains. Telecom optics underpin global communications infrastructure and high-speed datacenter interconnect.

The III-V archetype is therefore structurally a bridge archetype — smaller in wafer volume than leading-edge logic but enabling capabilities that flow into multiple downstream markets. Its supply chain thinness makes it a particularly informative case study in concentration risk.

Fabs in This Archetype

Notable III-V fabs include: WIN Semiconductors Taoyuan (largest pure-play GaAs foundry); Qorvo Richardson TX (GaAs RF); Skyworks Newbury Park CA (GaAs RF); Lumentum San Jose and international sites; Coherent consolidated operations; IQE Cardiff UK and international sites (merchant epi); Sumitomo Electric Japanese operations; Mitsubishi Electric Japanese operations; MACOM Lowell and Ithaca; UMS Villebon France; specialty Chinese III-V fabs. See Fab Facilities for the full inventory.

Related Coverage

Parent: Wafer Fabs

Peer archetype pages: Leading-Edge Logic · Mature Logic · DRAM · 3D NAND · SiC Power · GaN Power & RF · Analog & Mixed-Signal · CMOS Image Sensor · MEMS · Silicon Photonics · Rad-Hard & Rad-Tolerant

Cross-pillar dependencies: RF Front-Ends · LiDAR Semiconductors · Optical Transceivers>

Cross-network supply chain: ElectronsX · Autonomous Vehicles · LiDAR Sensors

Critical materials: Gallium Supply · Indium Supply · Critical Elements & Geopolitics