GEM (SEMI E30) gets a tool talking. GEM300 is the set of standards layered on top that make a tool participate in an automated 300mm factory rather than simply describe itself to one.
The names are opaque and the specifications are long. Here is what each actually does, in the order you tend to encounter them.
E87 — Carrier Management
What it does. Manages the FOUP or carrier as an object: it arrives at a load port, it is identified, its contents are mapped, it is associated with material, it leaves.
The model. Load ports have states. Carriers have states. The host and the tool agree on what is where. When a carrier arrives, the tool reports it; the host decides whether to accept it, and the two sides converge on a shared view of the carrier ID, the slot map, and whether the contents match what was expected.
What breaks without it. Material tracking becomes manual or inferred. You know a lot is somewhere in the tool but not which carrier is on which port, and a slot map mismatch — a wafer in a slot the system thinks is empty — becomes something an operator notices rather than something the system catches.
E87 is usually the first GEM300 standard a plant actually needs, because carrier identity is the foundation everything else attaches to.
E90 — Substrate Tracking
What it does. Tracks individual wafers, not just carriers. Where each substrate is inside the tool, and what has happened to it.
The model. Every substrate is an object with a location and a state, moving through positions inside the equipment. The tool reports transitions as they occur.
What breaks without it. You lose per-wafer resolution. Yield analysis that wants to correlate a defect signature with a specific chamber, at a specific time, on a specific wafer, cannot do it — the data is aggregated to lot or carrier level and the correlation is gone.
For a plant doing serious yield engineering, E90 is where the value is. For a plant that is not yet, it is often deferred.
E40 — Process Job Management
What it does. A process job is the instruction "run this recipe against these specific substrates". E40 defines how the host creates one, how it is queued, started, paused and stopped, and how the tool reports its progress.
The model. The host creates a job against a carrier and a slot selection, specifying the recipe. The job moves through a defined lifecycle. The tool reports state changes; the host can intervene.
What breaks without it. The host cannot express "run these wafers, this way" as a single controlled unit. You fall back on remote commands plus recipe selection plus hope, and error recovery gets harder because there is no job object to reason about when something stops halfway.
E94 — Control Job Management
What it does. Groups process jobs. A control job is the factory-level unit of work that spans one or more process jobs, typically corresponding to what an operator or the MES thinks of as "this lot, through this tool".
The model. Control jobs sit above process jobs and coordinate them, including ordering and the disposition of material when things go wrong.
What breaks without it. For a single-chamber tool running one recipe against one carrier, not much — E40 alone is often enough. For a cluster tool where a lot is split across chambers and recipes, the absence of E94 means your MES is doing the coordination itself, which is exactly the work the standard exists to avoid duplicating per tool.
E116 — Equipment Performance Tracking
What it does. Reports what state the equipment is in over time, in categories that map to the SEMI E10 model: productive, standby, engineering, scheduled downtime, unscheduled downtime, non-scheduled.
Why it matters. This is the raw material for OEE. Availability, performance and quality metrics all depend on knowing, accurately and automatically, how long the tool spent in each state and why it changed. Without E116 that data comes from operator logs, and operator logs are a genuinely poor input to a metric that drives capital decisions.
What breaks without it. Your OEE numbers are an estimate assembled after the fact rather than a measurement. Most plants discover this when someone asks why two systems report different availability for the same tool.
Which ones do you actually need?
It depends on the plant and honestly on the tool.
- A 200mm line or an OSAT with mostly standalone tools — E87 if there is carrier handling worth tracking, E116 if you care about OEE. E40, E90 and E94 are often more machinery than the situation needs
- A 300mm fab with automated material handling — all five, realistically. The automation assumes them
- A modern multi-chamber cluster tool — all five, and expect it to present multiple logical entities
The mistake to avoid is treating GEM300 as a single checkbox. It is five separate standards with five separate implementation states, and a tool can have any combination.
Where FabConnect stands on these
We implement all five — E40, E87, E90, E94 and E116 — alongside E30, E5 and E37, and they are covered by an automated test suite that exercises the message layouts through encoded bytes rather than through a shared in-process shortcut.
The honest boundary, which we state in writing everywhere else and will state here too: this has been verified against our own equipment simulator, not against physical semiconductor equipment. Our simulator models a generic GEM/GEM300 tool. It does not model any specific vendor optional fields, and the message layouts still need a field-by-field check against a given tool's SECS manual.
That check is normal integration work and it is part of what a first engagement covers. We would rather you heard it from us now than found it out in a workshop.