Physical process tuning
90Adjusting a real, physically installed fab tool requires hands-on presence and tacit, tool-specific knowledge that does not transfer cleanly from training data.
Semiconductor engineering sits in an unusual position: the same AI boom reshaping the profession's demand is also starting to automate pieces of its own daily work, from chip layout optimization to defect classification. Unlike a purely software job, though, a large share of the work happens on a physical factory floor, running machines that cost tens of millions of dollars each and that someone has to be legally and financially accountable for.
That physical, capital-intensive core is what keeps this profession's automation risk more moderate than software engineering's. The sections below separate the tasks already being handed to algorithms from the ones that, so far, still require a person standing in a cleanroom or signing off on a decision worth millions of dollars.
Moderate
Share of the work a machine could do
Specific, well-defined tasks — classifying wafer defects from images, optimizing a chip's physical layout, generating first-draft verification tests — are already comparably fast for AI tools to handle. What remains hard to automate is physically tuning a fab tool that behaves slightly differently from its twin on the next line, diagnosing a defect with no precedent, and being accountable when a decision affects a batch of wafers worth millions of dollars.
Scored from the tasks, not the job title. Lower is safer.
Jobs AI cannot take →What machines cannot take
Accountability for a yield excursion
85Someone has to be answerable when a fab-wide batch of wafers is scrapped or a safety incident occurs — a model cannot hold that responsibility.
Hazardous-materials and safety judgment
82Fabs handle toxic and pyrophoric gases like arsine and phosphine alongside high-voltage equipment; judgment calls here carry real physical risk that cannot be delegated.
Novel defect root-causing
80Diagnosing a failure mode with no matching precedent — a new material stack, a first-of-its-kind excursion — still needs a person forming and testing hypotheses under pressure.
Cross-functional negotiation
75Deciding whether design or process absorbs the cost of a workaround, or which of two conflicting deadlines slips, is a political and technical judgment call at once.
What they already take
Wafer defect image classification
78Computer-vision models now flag defects in wafer and microscope images faster, and often more consistently, than manual visual inspection.
Chip floorplanning and routing
60Advanced EDA algorithms and reinforcement-learning research, including published work from Google in 2021, increasingly automate placement and routing decisions that once needed iterative manual tuning, though results and reproducibility remain debated.
Verification test-pattern generation
65AI tools can generate a plausible starting set of test vectors and coverage scenarios for design verification, though engineers still judge whether they test the right corner cases.
Routine SPC anomaly flagging
55Statistical process control software already auto-flags out-of-control fab parameters, a task that used to require an engineer scanning charts by eye.
How the work is changing
From manual layout to reviewing machine-proposed designs
A growing share of design work involves checking and adjusting an AI-assisted layout or floorplan rather than placing every block by hand from scratch.
Verification becomes the bottleneck
As design complexity grows and more of it is machine-generated, catching subtle correctness errors before an irreversible tape-out is becoming the field's scarcest and most valued skill.
Geopolitics reshapes where the work happens
Export controls and national subsidy programs, including the US CHIPS Act, the EU Chips Act and Japan's Rapidus project, are actively relocating fab construction and engineering jobs to new regions.
AI accelerator design becomes its own specialty
A growing share of new hires design chips specifically for machine-learning training and inference, a discipline that barely existed as a distinct specialty before the 2010s.
New jobs branching off
AI accelerator / ASIC design engineer
Designs custom silicon optimized for machine-learning workloads rather than general-purpose computing, a fast-growing specialization pulling talent from traditional chip design.
Silicon photonics engineer
Integrates optical interconnects onto or near a chip to move data faster as pure transistor scaling slows, an emerging discipline bridging electrical and optical engineering.
Export-control and chip supply-chain compliance specialist
Screens customers, tracks end-use and keeps up with rapidly changing US, EU and Chinese trade rules — a role that barely existed as a dedicated job before the 2020s.
Advanced packaging engineer
As shrinking individual transistors delivers diminishing returns, more performance gains come from stacking and connecting multiple chiplets in one package, such as TSMC's CoWoS or Intel's Foveros.
Outlook
Demand for semiconductor engineers looks durable for the next decade, driven by AI accelerator demand and a deliberate, government-funded push to build manufacturing capacity outside the current handful of dominant sites. But the job is not immune to disruption: routine layout and inspection tasks are already shrinking, and the profession's extreme capital costs mean job growth stays concentrated in a small number of countries able to fund multi-billion-dollar fabs at all.
The advantage will keep sitting with engineers who can work at the boundary between the physical and the digital — reading a wafer map, trusting a simulation only as far as measured silicon confirms it, and taking responsibility when an automated tool's suggestion turns out to be subtly wrong.
Keep exploring
More in Engineering & Technology
💻
Software Engineer
Writes, tests and maintains the code that runs modern life — and is one of the first professions watching AI automate its own daily work.
AI-resistant 35 🤖AI Researcher
Designs and tests the algorithms behind machine intelligence, in a field now racing to automate a growing share of its own research process.
AI-resistant 50 🛰️Aerospace Engineer
Designs, analyzes and certifies the aircraft, rockets and spacecraft that leave the ground, working to safety margins that leave no room for guessing.
AI-resistant 74