5G, 6G & Wireless Semiconductors

Wireless chips depend on compound semiconductors more than any other sector. Compute sectors run on leading-edge CMOS at TSMC N3 and N5, while radio frequency (RF) front ends and wireless infrastructure rely on gallium arsenide (GaAs), gallium nitride (GaN), silicon germanium (SiGe) BiCMOS and indium phosphide (InP). Specialty foundries make these on supply chains largely separate from mainstream silicon logic, and they hold the sector's longest-lasting bottlenecks.

The sector has three sub-markets with different chip profiles:

• Infrastructure: 5G base stations (gNodeB), remote radio units (RRUs), massive multiple-input multiple-output (MIMO) antenna arrays and baseband units. Ericsson, Nokia and Huawei build the systems; Qualcomm, Marvell, NXP and TI supply ASICs and digital signal processors (DSPs).
• Devices: RF front-end (RFFE) modules in smartphones, fixed wireless access (FWA) routers, private 5G systems and satellite terminals. Qualcomm leads baseband SoCs; Qorvo, Skyworks and Broadcom lead RFFE modules.
• Private 5G: industrial, logistics and robotics networks, where 5G links autonomous systems and coordinates robot fleets.

5G, 6G and Wireless Semiconductor Device Map

Compound Semiconductor Supply

GaAs, GaN, SiGe BiCMOS and InP cannot be made at standard silicon CMOS foundries, and process scaling will not change that because the advantage comes from the materials:

• GaAs: higher electron mobility than silicon, for efficient power amplifiers at cellular microwave frequencies.
• GaN on silicon carbide (SiC): higher breakdown voltage and power density, for base station PAs delivering 10W to 100W of RF output.
• SiGe BiCMOS: low-noise amplifiers and transceivers at millimeter-wave (mmWave) frequencies of 24GHz to 100GHz, with noise figures silicon CMOS cannot match.
• InP: the highest-frequency transistors, needed for 6G sub-terahertz bands.

The supply base is narrow, with no near-term silicon foundry equivalent:

• WIN Semiconductors in Taiwan is the largest GaAs foundry and makes the pHEMT and HBT wafers in most of the world's handset power amplifiers.
• Wolfspeed in Durham, North Carolina grows SiC substrates and runs GaN-on-SiC device fabs that supply a large share of Western base station PAs. See the Wolfspeed Supply Chain Spotlight.
• GlobalFoundries Fab 9 in Malta, New York is one of a few fabs running SiGe BiCMOS at the generations mmWave circuits need.

Gallium, the base material for GaAs and GaN, comes mostly as a byproduct of aluminum smelting from bauxite, and China produces about 80% of refined gallium. China's 2023 export controls on gallium and germanium have not yet caused acute shortages in Western wireless supply chains, but they give Beijing leverage in the U.S. and China technology contest that has no silicon equivalent. Building Western gallium supply from smelting through GaAs and GaN wafers is a multi-decade investment. Fab and wafer detail is on III-V Compound Semiconductor Fabs, GaAs and InP Wafers and Semiconductor Critical Elements.

BAW Filter Manufacturing

BAW filters use piezoelectric thin films, usually aluminum nitride (AlN) or scandium-doped AlN, deposited on silicon and patterned to resonate at set RF frequencies. Film thickness sets the resonant frequency and must be controlled to atomic-level precision. The deposition equipment, materials and process control are unlike CMOS transistor fabrication.

Qorvo, Broadcom (Avago heritage), TDK and Murata each built their BAW processes over decades as proprietary IP, and none has been standardized or licensed. TSMC, Samsung Foundry and GlobalFoundries offer no BAW process design kit. Adding BAW to a CMOS foundry would take years of development with uncertain yield, so no new entrant has emerged despite the concentration.

Huawei Kirin Chips at SMIC

Huawei's HiSilicon unit designed Kirin SoCs that matched Qualcomm Snapdragon on application processing and 5G modem integration, including the Kirin 990 5G (TSMC N7+, 2019) and Kirin 9000 (TSMC N5, 2020). In May 2020 the U.S. Bureau of Industry and Security (BIS) required TSMC to obtain a license before making chips for Huawei, which ended TSMC production of Kirin. Huawei used up its stockpile by late 2021 and relied on older Kirin parts and MediaTek SoCs through 2022 to 2023.

The Kirin 9000S, revealed in the Mate 60 Pro in August 2023, and its successor the Kirin 9010 are built at SMIC on N+1, a deep ultraviolet (DUV) multi-patterning process at about 7nm-equivalent density, without extreme ultraviolet (EUV) lithography. They trail Qualcomm's Snapdragon 8 Gen 2 (TSMC N4) by about one to two generations in compute density and power efficiency, roughly two process generations behind TSMC's leading edge at the time. SMIC cannot buy ASML EUV scanners, a limit likely to last through at least 2030 under current export controls.

The chips are commercially viable in China, where buyers have no access to TSMC-dependent alternatives. They show that export controls slow Chinese capability without stopping it when engineering talent, state funding and a large domestic market can absorb the performance penalty. That result supports continued state investment in SMIC capacity and domestic EDA tools. Company detail is on the Huawei and HiSilicon Supply Chain Spotlight and China Semiconductor Bifurcation.

5G RAN Infrastructure Vendors

Ericsson, Nokia and Huawei supply most of the world's 5G radio access network (RAN) equipment. Samsung has meaningful share in South Korea, Japan and the U.S., where it gained as a Huawei alternative, and ZTE is active in China and some emerging markets. Ericsson and Nokia design proprietary ASICs with Western EDA tools and foundries and buy components from Western suppliers. Huawei sources from Chinese suppliers where it can and is cutting Western content from new RAN designs.

Restrictions on Huawei, including the U.S. Clean Network initiative, the UK ban effective 2027 and European operator decisions, have split 5G infrastructure into two supply chains. Ericsson and Nokia compete in Western and allied markets with similar chip architectures, while Huawei and ZTE lead in China with domestic silicon. The two RAN technology roadmaps will diverge further in capability and supply chain through 2030.

Open RAN (O-RAN) splits the RAN into standardized radio units (RUs), distributed units (DUs) and centralized units (CUs) that can come from different vendors. It is the main policy tool for creating Huawei alternatives beyond Ericsson and Nokia, since any vendor meeting O-RAN Alliance specifications can compete. It also opens a merchant market for baseband ASICs such as Qualcomm FSM and Marvell OCTEON Fusion, which would otherwise stay inside Ericsson and Nokia. Open RAN has not yet matched proprietary RAN on performance and cost at macro cell scale, and the transition will take until 2027 to 2028 to settle.

Private 5G for Industry and Robotics

Private 5G networks are standalone deployments in factories, warehouses, ports and industrial campuses, and they are the fastest-growing new source of 5G chip demand. They provide deterministic, low-latency links that Wi-Fi cannot guarantee at industrial scale. A warehouse with 500 autonomous mobile robots (AMRs) needs latency under 10ms and reliability of 99.9999% or higher, which Wi-Fi 7 cannot hold consistently amid multipath reflections and moving metal.

The private 5G chip stack differs from macro cellular:

• Small cell RAN: Qualcomm FSM or Marvell OCTEON Fusion small cells replace macro base stations.
• Shared spectrum: Citizens Broadband Radio Service (CBRS) in the U.S. and similar bands in Europe and Japan allow deployment without buying licensed spectrum.
• Endpoints: AMRs, automated guided vehicles (AGVs), humanoid robots, industrial sensors and worker wearables each carry a 5G modem, typically Qualcomm Snapdragon X series or MediaTek T750, plus a GaAs or GaN front-end module.

A humanoid robot on private 5G carries both the robotics stack (GaN joint drives, position encoders, inference SoC) and the 5G stack (modem, GaAs PA, BAW filters), so its chip content exceeds either a standalone robot or a standalone 5G device. Neither supply chain was sized for that combined demand. Robot-side chips are covered on Robotics and IoT Semiconductors.

6G Sub-Terahertz Semiconductors

6G standards (3GPP Release 20 and later) target commercial deployment in 2030 to 2033. The highest-throughput links are expected to use sub-terahertz bands of 100GHz to 300GHz, which need transistors with transition frequencies (ft) above 300GHz. Only InP high electron mobility transistors (HEMTs) and advanced SiGe BiCMOS below 130nm reach that today, and both are in research-grade production.

Research programs developing sub-terahertz devices alongside new air interfaces include DARPA's 6G program, the European Commission's Hexa-X, Japan's NICT Beyond 5G, Samsung and LG Electronics in South Korea, and China's IMT-2030 promotion group. Prototype systems will need more InP and SiGe BiCMOS foundry capacity from 2026 to 2028, ahead of commercial demand after 2030, while 5G deployment continues in emerging markets:

• InP wafers: mainly from Sumitomo Electric and AXT in the U.S.
• InP devices: IQE, Qorvo and WIN foundry capacity.
• SiGe BiCMOS: GlobalFoundries Fab 9 and IHP, for the sub-100GHz portions of 6G.

None of these expansions appear in the foundries' published capital plans at the scale commercial 6G will require.

Wireless Supply Chain Bottlenecks

Key 5G, 6G and Wireless Semiconductor Suppliers

Cross-Sector Supply Overlaps

• Gallium: the common upstream material for GaAs handset PAs, GaN base station PAs, EV onboard chargers and robot joint drives, and InP for 6G and datacenter optical transceivers. Tighter Chinese export controls would hit wireless, automotive, robotics and datacenter supply at the same time, with no fast Western response. Datacenter optics are covered on Datacenter and HPC Semiconductors.
• Private 5G and robotics: robot fleets on private 5G need small cell RAN hardware (open RAN baseband ASICs, GaN small cell PAs) sized to each facility, demand that standard cellular infrastructure forecasts do not include.
• Satellite and automotive V2X: 3GPP Release 17 NTN uses the standard 5G NR air interface, so automotive telematics SoCs with NTN support reach satellites through the same modem used for terrestrial 5G. Qualcomm's Snapdragon Auto 5G Modem-RF Gen 2 is the main device, and it must pass AEC-Q100 automotive qualification on top of 5G modem qualification. Satellite chip content is on the Satellites and Space Systems Spotlight.

Cross-Network: ElectronsX Demand Side

ElectronsX covers wireless demand from autonomous vehicle V2X connectivity and robot fleet communications.

EX: AV Platforms Directory | EX: Humanoid Robots | EX: Supply Chain Convergence Map

Related Coverage

Mobile & Consumer Semiconductors | Robotics & IoT Semiconductors | Automotive & Mobility Semiconductors | Datacenter & HPC Semiconductors | Space & Defense Semiconductors | AI & Machine Learning Semiconductors | AI Inference Semiconductors | RF & Networking Semiconductors | Compound & Specialty Wafers | Semiconductor Critical Elements