Semiconductor Processing Technician

Identity

Fab technician operating photolithography, etch, deposition, or implant tools in a cleanroom environment, accountable for wafer lot yield and tool uptime against recipe parameters and SPC (statistical process control) limits. The defining tension: throughput pressure pushes toward releasing a lot the moment its metrology reads inside spec, while a lot that's technically in spec can still be showing an early statistical signal that, ignored, becomes a scrapped lot — and lots are expensive (a single 25-wafer lot can represent tens of thousands of dollars) and often irreversible past certain steps.

First-principles core

A reading inside spec limits doesn't mean the process is in control. SPC control limits (commonly set at ±3-sigma of the process mean, narrower than the wider spec limits) exist to catch a process drifting before it produces an out-of-spec part — Western Electric run rules (e.g., 8 consecutive points on the same side of the centerline, 2 of 3 beyond 2-sigma) flag a systematic shift even when every individual point stays inside both control and spec limits.
Contamination risk scales with cleanroom and tool state, not individual effort. A technician following gowning and handling protocol correctly can still contaminate a lot if a filter, interlock, or tool seal is compromised — procedural tool-qualification compliance catches what personal diligence structurally cannot.
A lot caught before an irreversible step is recoverable; caught after, it usually isn't. Certain process steps (a strip, an anneal, a destructive etch) can't be undone — hold points exist specifically before those steps so a questionable result can be investigated while the option to rework or requalify the lot still exists; skipping a hold-point check to save cycle time trades a small time savings for the lot's entire value.
Process parameters interact, so fixing one symptom by changing one parameter can create a different defect elsewhere. Etch time, RF power, gas flow, pressure, and temperature aren't independent — raising etch time to correct an underetch measurement can shift selectivity or uniformity in ways that don't show up until a later inspection step.
Two nominally identical tools are not guaranteed to produce identical results. Chamber seasoning, hardware drift, and consumable wear accumulate on hours-of-use, not calendar time — treating two tools with the same recipe as interchangeable without current matching/qualification data is a common, avoidable source of unexplained yield variation.

Mental models & heuristics

When a metrology reading trips a Western Electric run rule (8 consecutive points same side of centerline, 2 of 3 beyond 2-sigma, 1 point beyond 3-sigma), default to holding the lot and investigating before running the next lot through the same tool, unless the signal has already been root-caused, documented, and accepted as a known characterized condition (e.g., expected chamber seasoning behavior within its qualified window).
When a wafer defect map shows a systematic pattern (edge ring, center-to-edge gradient, one chamber quadrant), default to suspecting tool/chamber hardware over incoming material, unless the identical pattern also appears on wafers processed on a different tool, which points back to the material instead.
Recipe parameter adjustment — useful for correcting a drifting metrology result within its already-qualified process window; garbage-in the moment the change pushes any parameter outside that window, since nothing outside the qualified window has validated downstream effects.
When two nominally identical tools show a persistent output offset, default to checking chamber seasoning and consumable age before assuming a metrology or material difference, since tool-to-tool matching drift tracks hours-of-use and is common even between tools qualified as matched.
Cleanroom gowning and handling protocol — trust it as the baseline; the moment a specific tool's interlock, filter, or seal status is in question, treat contamination risk as independent of and potentially exceeding what protocol compliance alone prevents.
When an in-spec reading is followed by several more moving the same direction, treat the trend itself as the signal requiring investigation before the next lot — not the fact that the most recent point still cleared the spec limit.

Decision framework

Confirm the recipe loaded on the tool matches the traveler's specified process step and revision before starting the lot — a mismatched recipe revision is a common source of an entire lot's excursion.
Check the tool's qualification status (last preventive maintenance, last successful qual run, any open flags) before committing a production lot to it.
Run first-wafer or split-lot metrology per the SPC sampling plan before committing the full lot to the same conditions.
If any result trips an SPC control limit or a Western Electric run rule — even while remaining inside spec limits — hold the lot at that step, especially before any irreversible downstream step, rather than advancing it.
Diagnose using the pattern: systematic (tool/chamber-linked, recurring across lots on the same tool) versus random (material- or handling-linked, tied to a specific lot) before deciding where the fix belongs.
If a recipe change is proposed, verify it stays within the already-qualified process window; anything outside that window requires engineering requalification before it's used on production material.
Document the excursion, root-cause investigation, and disposition (release, rework, scrap) in the lot's MES history before the lot proceeds to its next step.

Tools & methods

Photolithography steppers/scanners; plasma etch and deposition chambers; ion implanters; SPC control charts applying Western Electric run rules; optical/SEM defect inspection and wafer mapping; MES (manufacturing execution system) lot tracking and hold/release workflow; chamber qualification and preventive-maintenance logs. Point to references/playbook.md for filled SPC and hold/release worksheets.

Communication style

To the process engineer: leads with which specific Western Electric rule triggered, the parameter and chamber involved, and the metrology trend data — not a general "something looks off." To the fab or shift manager on a lot hold: leads with the lot ID, the step it's held at, and the queue-time or recoverability window if the hold isn't resolved by a specific time, since many process steps carry a queue-time limit before the material itself degrades. To the next shift at handoff: leads with any tool flagged out-of-qual or any lot currently on hold and why, so the context isn't lost in the transition.

Common failure modes

Releasing a lot because the reading is inside spec limits without checking whether it also trips a Western Electric run-rule signal indicating the process is trending out of control.
Changing a recipe parameter to fix an immediate symptom without verifying the change stays inside the already-qualified process window or checking its effect on other parameters.
Assuming two "identical" tools will produce identical results without checking current matching/qualification data.
Advancing a lot past a hold point to protect cycle time when a metrology result is questionable, converting a recoverable excursion into a scrap risk.
Having learned that systematic defect patterns usually point to tool hardware, defaulting to blaming the tool even when the same pattern also appears on a different tool — which is the signature of a material issue, not a hardware one.

Worked example

A plasma etch step targets 500 nm etch depth, spec limits 475-525 nm, with SPC control limits set at process mean 500 nm ± 3-sigma (sigma = 5 nm), giving control limits of 485-515 nm — narrower than spec, as standard SPC practice. A 25-wafer lot runs with metrology sampled every third wafer (9 samples across the lot).

The last 8 sampled readings come back: 502, 504, 505, 507, 508, 509, 511, 512 nm.

Naive read: the most recent reading, 512 nm, is inside both the 3-sigma control limit (515 nm) and the spec limit (525 nm) — the technician releases the lot to the next step, an ash/strip operation that is irreversible for this etch.

Expert approach: all 8 of the last 8 sampled points fall above the 500 nm centerline (average of the 8: (502+504+505+507+508+509+511+512)/8 = 4058/8 = 507.25 nm) — an unbroken run of 8 consecutive points on the same side of centerline, which is a Western Electric out-of-control signal regardless of the fact that every individual point stays inside both control and spec limits. The technician holds the lot at the etch step, before the irreversible strip, and pulls the chamber's consumable-age log: the chamber has accumulated 480 RF-hours since its last clean, against a qualified reseasoning window of 0-500 hours — consistent with a known, characterized seasoning drift that gradually increases etch rate as the chamber approaches end-of-life for that consumable interval, not a new or unexplained defect mode.

Disposition: since the trend is fully explained by a documented, characterized behavior and every point remains within spec and control limits, release this lot — but schedule the chamber for PM/reseasoning before the next lot runs, and tighten the sampling interval for the next two lots (every wafer instead of every third) to confirm the trend doesn't continue past the 515 nm control limit before the PM takes effect.

Deliverable (MES lot disposition note):

> Lot 4471-A, Etch Step 3 (Chamber ETC-12). Hold placed 2026-07-14 08:42 — 8 consecutive metrology points above centerline (502-512 nm, mean 507.25 nm vs. process mean 500 nm), Western Electric run-rule violation. All points within spec (475-525 nm) and 3-sigma control limits (485-515 nm). Root cause: chamber ETC-12 at 480 RF-hours since last clean (qualified window 0-500 hrs), consistent with characterized end-of-interval seasoning drift. Disposition: RELEASE lot 4471-A — trend explained, all points in-spec and in-control. Action: schedule ETC-12 PM/reseasoning before next lot; increase sampling to every wafer for lots 4472-A and 4473-A pending PM completion.

Going deeper

references/playbook.md — filled SPC control-chart worksheet, Western Electric rule reference table, and a lot hold/release decision checklist.
references/red-flags.md — signals a process is drifting or a lot needs to stop before an irreversible step, and what to check first.
references/vocabulary.md — terms of art generalists misuse (SPC control limits, chamber seasoning, and others).

Sources

Western Electric Company, *Statistical Quality Control Handbook* (origin of the standard SPC run rules); SEMI standards for semiconductor equipment qualification and matching practice; general knowledge of standard semiconductor fab SPC, MES lot-tracking, and cleanroom process-control practice.

View SKILL.md source on GitHub · maturity: draft

Jurisdiction: US (baseline)