Industries
Excursions show up in the data long before they show up on the wafer.
We work on the unglamorous half of semiconductor AI: getting tool, metrology and MES data into one place it can be reasoned over, then putting agents on the investigations that currently take an engineer a fortnight.
Yield analytics · Equipment data · Root cause · Tool health · Fab search
01 — Where the time goes
The hard part is rarely the model.
It is the twelve systems the answer lives across.
A yield question touches tool traces, recipe history, metrology, defect inspection, maintenance records and the notes an engineer left in a spreadsheet. Each has a different owner, a different clock and a different idea of what a lot is. Most fab analytics projects spend their budget here and never reach the question they were funded to answer.
We start with the plumbing, because it is what makes everything after it reproducible — equipment interfaces, trace collection, context joins and a data model an engineer recognises. The models and the agents go on top of that.
The people who build it have supported the equipment. That matters: a correlation a data scientist finds interesting and a process engineer finds obvious is a wasted cycle, and knowing the difference is not something a model does for you.
Talk to Zealogics02 — What you get back
One investigation, end to end, instead of twelve exports.
TRACES, METROLOGY, RECIPE HISTORY AND MAINTENANCE — JOINED.
03 — What the work covers
Where we work in a fab.
Where this shows up
Deposition, etch, litho, inspection and metrology tools
Wafer probe, test and back-end assembly
Equipment makers supporting an installed base
Fab engineering and yield enhancement groups
Field service and spares planning organisations
Typical stack
Equipment and process data engineering
Trace collection over SECS/GEM and OPC-UA, contextualisation against lot, recipe and chamber, and a time-series store an engineer can query without asking IT for an export.
Yield and excursion analytics
Commonality analysis across tools, chambers and steps, drift detection on the traces themselves, and correlation against metrology and inspection results rather than against a summary of them.
Root-cause and 8D support
Agents that assemble the evidence pack for an investigation — candidate tools, matching historical events, recipe changes in window — and draft the report an engineer then argues with.
Tool health and spares intelligence
Predictive signals on subsystem health, PM effectiveness and spares consumption, joined to the maintenance record so a recommendation arrives with the work order it implies.
04 — What ships with it
What a semiconductor engagement leaves behind.
A demonstration is not a deliverable. These are the artefacts an engagement leaves with your team — owned by you, runnable without us, and auditable by whoever has to sign for them.
01
A contextualised trace store
Tool data joined to lot, recipe, chamber and step, with the join logic written down rather than living in one engineer's notebook.
02
Excursion and drift detection
Monitors on the signals that actually move before a wafer does, tuned against your own history rather than a vendor default.
03
An investigation workspace
The evidence for a yield question gathered in one place, with every chart traceable back to the raw trace behind it.
04
Fab knowledge retrieval
Specifications, procedures, 8Ds and equipment manuals searchable in plain language, answering with the document and page it came from.
05
Tool health signals
Subsystem-level health indicators and the maintenance actions they map to, delivered into the system that schedules the work.
06
The handover
Runbooks, retraining procedure and the drift monitoring that tells you when a model has stopped describing the tool it was fitted to.
05 — Agentic AI, in this sector
Agentic AI in a fab.
Where agents start
- —Equipment log summaries
- —Recipe change tracking
- —Metrology report drafting
- —Spare parts lookup agent
- —Fab documentation search
- —Alarm triage assistant
Where they go next
- —Yield investigation crew
- —Multi-tool trace correlation
- —Hypothesis ranking agents
- —Predictive tool health
- —Automated 8D report authoring
- —Fab knowledge graph agents
Delivery note — Built by engineers who support the equipment — the models sit next to the fab data rather than a sanitised copy of it.
06 — In your words
What clients say before they call us.
“Every excursion investigation starts with three days of pulling data.”
Contextualised trace store
“We know the tool is drifting. We find out when metrology tells us.”
Trace-level drift detection
“The answer is in an 8D from two years ago and nobody can find it.”
Fab knowledge retrieval
07 — The rest of the map
Eleven more sectors we work in.
A problem is rarely unique to its industry. Most of what we build here has been built next door as well, which is usually why it arrives faster the second time.
Plant
Manufacturing
Shop-floor copilots, predictive maintenance and the plant reporting that currently happens in a spreadsheet at six in the morning.
Explore
Heavy
Steel
Process optimisation, quality prediction and the reporting heavy industry still assembles by hand.
Explore
Assets
Energy & utilities
Asset performance, inspection data and the field workflows that still move on paper between three systems.
Explore
TIC
Testing, inspection & certification
Digitised inspection, generated reports and compliance search across standards that change under you.
Explore
Network
Telecom
Network operations assistants, incident response and service assurance analytics that keep up with the alarm volume.
Explore
Finance
Financial services
Reconciliation, close, risk narrative and reporting automation with an approval gate on everything that moves a number.
Explore
Assurance
Audit & assurance
Evidence review copilots, parallel testing agents and document analysis that cites the page it read.
Explore
Clinical
Healthcare
Documentation support, coding assistance and operational insight, with every clinical output approved by a person.
Explore
Public
Government
Citizen services, records intelligence and deployment models that stay inside the boundary the mandate requires.
Explore
Controlled
Defense
Engineering support and controlled AI environments — local models, no external calls, clearance-aware retrieval.
Explore
Talent
Staffing
Engineering, IT and AI talent for Fortune 500 programmes across the USA, UAE, Taiwan and India.
Explore
Have a problem worth solving?
Bring us a yield question that keeps coming back.
Send us one investigation your team has run more than once. We will show you what it looks like when the evidence assembles itself.