Semiconductor devices is the discipline that designs the transistor itself: the CMOS, FinFET and gate-all-around (GAA) architectures of logic, the power semiconductors and RF semiconductors of energy and connectivity, and the MEMS, sensors, photonic devices and quantum devices that extend silicon beyond computation. Imec's roadmap frames the logic transition: gate-all-around nanosheet transistors replace FinFETs at the 2nm node, and the path continues through forksheets toward complementary FETs in later generations . AI accelerators concentrate demand at the leading edge while mature nodes keep the device physics honest everywhere else.
Challenges in Semiconductor Devices Recruiting
Gate-all-around (GAA) transistors turn nanosheet transistors into a stacking discipline
Gate-all-around (GAA) transistors turn nanosheet transistors into a stacking discipline. The channel is a vertical stack of silicon nanosheets fully wrapped by the gate, and every advantage over FinFET, better electrostatic control at short channel lengths, is paid for in integration complexity. Imec positions gate-all-around as the FinFET successor at 3nm and beyond, enabling 5.5-track standard cells down to the A10 generation . ASM's published roadmap extends the arc: first-generation GAA at 2nm in 2025, second generation at 1.4nm in 2027, third at 1.0nm in 2029, then complementary FETs at 0.7nm in 2031 . The device engineers who own this work live inside channel release etches, inner spacers, multi-threshold gate stacks and channel stress management. Each generation resets the optimization problem, which is precisely why the small population that has taken one node to volume is bid on before the next node is announced.
FinFET specialists still hold the high-volume CMOS floor
FinFET specialists still hold the high-volume CMOS floor. Foundries accepted the transition from FinFET to nanosheet architectures gradually from 2022 and 2023 , but the installed base of FinFET fabs, and the node roadmaps that feed them, will run for another decade. Those lines need exactly the skills the leading edge is moving past: fin profile control, leakage at volume, drive current versus variability, and the reliability signatures that accumulate over billions of device-hours. The population is enormous by device-engineering standards and almost invisible to recruiters who search only for the newest architecture. A 14nm or 7nm FinFET engineer holds volume-production judgment that no nanosheet newcomer has yet earned, and mature-node owners know it.
Power semiconductors split SiC, GaN and silicon superjunction careers
Power semiconductors split into three careers that share a label. Yole Group projects the power GaN market growing at a 35% CAGR to $3.5 billion by 2031, moving past consumer fast chargers into AI data centers, electric vehicles and industrial systems . Silicon carbide MOSFETs serve the higher-voltage traction and grid work; silicon IGBTs and superjunction devices still carry the largest volume. The device physics differs across the three: GaN HEMTs are lateral, depletion-mode-adjacent devices with dynamic on-resistance concerns; SiC MOSFETs fight channel mobility and threshold drift; superjunction silicon balances drift region charge to the percent. Each carries its own qualification culture as well, automotive for SiC, datacenter for GaN, industrial for silicon. A GaN device engineer and a SiC device engineer share wide-bandgap physics and almost no process flow. The market timing splits hiring too: SiC rode an EV wave that has since cooled, while GaN's second act in AI power delivery is just beginning, which means employers are poaching from two populations whose fortunes are moving in opposite directions.
RF semiconductors depend on III-V substrates CMOS engineers never touch
RF semiconductors run on materials the CMOS world abandoned. GaAs HBTs, GaN HEMTs and InP HBTs and HEMTs populate base stations, radar, satellite links and the front ends of phones, and they are built in compound fabs with their own design kits. IQE's H1 2026 update credits strength in wireless and aerospace applications for part of its momentum . The device engineer's vocabulary is different in kind: ft and fmax, power-added efficiency, linearity, breakdown voltage and thermal resistance, judged against load-pull data rather than Ids-Vgs families. A silicon RFIC designer and a III-V RF device engineer meet at the matching network and nowhere else. Sourcing the latter means mapping compound fabs and epi suppliers, because the population rarely advertises with the words semiconductor devices at all.
MEMS work lives between release etches, ASIC co-design and packaging stress
MEMS is where the mechanical and the electrical meet inside one product. Yole Group sizes the MEMS market at $17.1 billion today, growing toward roughly $24 billion by 2031, with Bosch among the global leaders across automotive sensors, robotics and industrial monitoring . The work splits across design, process and packaging, and the deep practitioners own release etches, stiction budgets, hermeticity and the stress that packaging injects into a suspended structure. Sensors are the volume: accelerometers, gyroscopes, microphones, pressure devices, each with its own readout coupling. The scarce profile pairs the mechanical structure with the ASIC it feeds, because noise floor, calibration and package shift all cross that boundary. A MEMS person who has never argued with an analog designer has delivered half a device.
Photonic devices run on insertion loss budgets the CMOS crowd never carries
Photonic devices run on insertion loss budgets the CMOS crowd never carries. Modulators, detectors and lasers are designed against loss, bandwidth and modulation efficiency rather than drive current, and the platforms are their own: InP lasers for datacom, silicon photonics for volume, functional oxides for next-generation modulators. RIBER's ROSIE platform grows barium titanate-class oxides on 300 mm silicon for integrated photonics, and its second unit shipped in July 2026 to a leading US quantum computing player . Quantum devices extend the pattern further: superconducting qubits and spin qubits hire physicists from national laboratories who have never touched a design kit. These populations barely overlap the CMOS pipeline, yet the same employers now need all of them in one organization, and the two searches run through entirely different networks.
Transfer curve questions expose inflated gate-all-around (GAA) transistors claims
Device CVs converge on architectures: FinFET, GAA, power, RF, MEMS. The separation is in the data the candidate can reconstruct. Which device did you own, at which node, and what did the transfer characteristics look like at the corners? What moved when you shifted a process parameter, and how did you trade off short-channel effect against drive current? Which reliability signature did you chase, hot carrier, bias temperature instability, time-dependent dielectric breakdown, and what did the model say? Owners quote numbers and trade-offs. Witnesses describe plots. The cost of a miss is a yield signature nobody can explain: device problems surface weeks after process changes, the fab keeps spending wafers while the diagnosis stalls, and the senior device staff gets pulled from the next node to referee a conversation the new hire was supposed to lead.
References
- Outer Wall Forksheet: Bridging Nanosheet and CFET Device Architectures — imec. (accessed 2026-09-28)
- Advancing the CFET-Based Device Roadmap, Part 1 — imec. (accessed 2026-09-28)
- ASM Q4 and FY 2025 Results Investor Presentation — ASM International. (accessed 2026-09-28)
- Power GaN Device Market Growing at 35% CAGR to $3.5bn in 2031 — Semiconductor Today, on Yole Group data. (accessed 2026-09-28)
- IQE Trading Update, H1 2026 — IQE plc. (accessed 2026-09-28)
- Yole Group's MEMS Industry 2026 Report Confirms Bosch as a Top Leader — Yole Group. (accessed 2026-09-28)
- RIBER First-Half 2026 Business Activity — RIBER. (accessed 2026-09-28)
