A closed control loop around every Tool ID.
Archer turns scattered transactional data into a single governed workflow — ingest, baseline, monitor, raise exceptions, order ahead of need, confirm, and record. Below: operate it yourself, see the architecture, and understand exactly what is forecast versus what is right-sized stocking.
Operate the control loop.
Toggle Without Archer / With Archer, press play, or click any stage to jump to it and read what happens there.
Set your parts and your assumptions.
This is the real model, in your browser. Enter your own parts — service life, usage rate, lead time — and set the assumptions an engineer argues over: the lead-time and life variability, the σ safety factor, and (for random-failure parts) the Weibull shape β. Archer computes the reorder date, the explicit safety buffer, the must-stock verdict, and the status live. Click any row for the full derivation. Nothing is sent anywhere.
This is the same model that runs across your whole fleet — the arithmetic is closed-form and checkable by hand, but the wear date is an estimate with an explicit buffer, not a deterministic guarantee. See it across a fleet ▸
Sits alongside your systems. Never in the way.
Archer reads from the systems you already run and returns recommendations and alerts. Your ERP stays the system of record and executes every order. Click a layer to read its role.
Three regimes. Two are forecasts; one is stocking.
Credibility means being precise about what is a genuine forecast and what is right-sized stocking. Archer sorts every part into one of three regimes — and never claims to predict the moment of a random failure.
① Usage-wear
Usage-driven · highest ROI
Part wears with process load — RF-hours, wafer count, cycles. We estimate the replacement date from accrued usage, then carry an explicit safety buffer for the variance in wear, rate, and lead time. An estimate with a stated confidence band — not a claim of exact prediction.
reorder = EOL − lead − √(σ-buffer)
e.g. etch focus/edge rings, CMP pads, ion-implant filaments.
② Time / cycle PM
Calendar · planned
Rebuild or replace on accrued run-hours or cycle count per the OEM PM interval. Stage the kit and book the window before the clock runs out.
due = interval − hours_since_PM
e.g. turbo/cryo pump rebuilds, seals, MFC recalibration.
③ Stochastic failure
Reliability · service level
No wear curve. Archer does not predict the exact failure; it sizes stock to a target service level over the lead time. Wear-out parts (Weibull β>1) carry an over-dispersion buffer above Poisson — a disclosed heuristic, set by you.
stock = f(MTBF, fleet, lead, β, SL%)
e.g. RF generators, electrostatic chucks, robots.
Etch focus ring, end to end.
The inputs
- 250 RF-hours service life (OEM / observed)
- ~4 RF-hr/day tool load (from the tool log)
- ~62 days life at that rate (250 ÷ 4)
- 5 weeks (35 days) part lead time (procurement)
With Archer vs without
With: at ~4 RF-hr/day the ring reaches end-of-life around day 62. Working back one 35-day lead time, Archer flags the order around day 27 — early enough to have the part on the shelf, and schedules the swap into a planned window instead of risking a mid-run failure. Run those numbers yourself in the calculator above.
Without: the ring degrades mid-run — chamber down, wafers at risk, the ring air-freighted at premium, and unplanned downtime billed at whatever an hour costs your line.
The safety buffer is a planning margin (≈2σ of the combined wear, rate, and lead-time variance) — conservative by intent, not a guaranteed service percentage; wear-out and supply tails run long, so it is sized to absorb that, not to promise a number.