AI-assisted development for ATE engineers
ATE·IQ is a development platform for semiconductor test engineers. It takes the four artefacts every test program lives across — the device spec, the IG-XL test program, the load board, and the production STDF data — and joins them into one traceable, machine-checked loop. AI drafts and explains; deterministic engines verify. Everything runs on the engineer's own machine.
A production test program is never one file. The device limits live in a PDF, the program lives in an Excel workbook, the load board lives at a PCB vendor, and the proof it all worked lives in STDF files nobody opens twice. Every question that matters — does the program match the spec? does the board match the program? which component do I probe when test 1100 fails? — crosses at least two of those silos, and today crosses them by hand.
ATE·IQ keeps all four artefacts in one workspace and joins them mechanically:
requirements to tests on test_number, tests to board nets on the pin/channel
axis, schematic nets to routed copper through a layout binding, program limits to
the limits each lot actually ran. The joins are pure functions — same inputs, same
answer, every time. AI is used where AI is good: drafting programs, authoring
boards, explaining failures — and every AI draft goes back through the
deterministic checkers before an engineer sees it.
Those checkers now include the device itself. A generated program is replayed against a model of the DUT — derived automatically from a vendor CMSIS-SVD file or the project's own register map, or your own RTL/SystemC model connected as-is behind a small wire protocol — so semantic errors that compile perfectly, like the wrong register or a compare the silicon can never satisfy, are caught offline instead of on the test floor. Every surface states who judged: your model, or the derived stand-in. What the model cannot judge, it declines out loud.
The platform is grounded in real tester knowledge: a typed registry of UltraFLEX instrument behaviour — envelopes, settle times, prohibitions — authored in our own words with a cited source and a confidence tag on every entry, alongside a working reference programme and 400+ cited ATE load-board design rules. When ATE·IQ makes a claim about an instrument, a settle time or a DIB rule, the claim carries its source.
Each page below explains one capability end to end — what it does, how it works underneath, and what it looks like on the TMP126 demonstration project.
Datasheet PDF or Jama Connect in; structured requirements with limits, forced conditions and test numbers out.
Read →Spec to complete IG-XL project — sheets, VBT modules and digital patterns — generated per requirement, validated, repaired, gated.
Read →The generated program replayed against a model of the device — derived from a vendor SVD file, or your own RTL connected as-is — catching semantic errors that compile cleanly, with DV waveforms, a dual-scope simulator cross-check, and honest declines.
Read →Configure SPI or I2C and a transaction sequence; get cyclized IG-XL digital patterns and a live, semantically-labelled waveform.
Read →ATE·IQ inside the open IG-XL workbook: live sheets and VBA as context, quality analysis, edits applied in place.
Read →Dependency map across VBA, patterns and sheets — active-function coverage, quality scans, and a pattern-to-test graph.
Read →The logical board, authored in ATE·IQ — a multi-sheet ATE schematic engine with DUT-first sheets, drawn terminations and Kelvin legs, channel flags from the saved plan, and validator-gated live edits with history.
Read →The physical copper, imported verbatim from your CAD flow — IPC-2581, ODB++, Specctra, KiCad, SES — then DRC-checked, bound to the schematic, requirement-traced, and exported to fab.
Read →Nine analyzers over every design: power domains, pin match, test impact, geometric and electrical DRC, 166-rule DIB check.
Read →The DIB definition as a first-class artefact — pins to tester channels, checked against the board and the program.
Read →Describe the load board; ATE·IQ authors it against your spec, compiles it, and repairs its own findings before you see it.
Read →Native STDF V4 parsing — yield, Cpk, outlier screens, drift, wafer maps — and statistics-backed limit proposals fed back to the program.
Read →One dashboard for the whole chain: requirement to test to net to copper to lot, every hop reconciled and attributed.
Read →A co-pilot that sees your project — curated context, callable tester-knowledge tools, and answers that cite their sources.
Read →See and shape exactly what the model receives — scored artefacts, honest budgets, one-click handoff to chat.
Read →Local-first by design: loopback-only services, your own LLM keys, encrypted credentials, nothing leaves the host.
Read →Every page in this guide references the same live project: a Texas Instruments TMP126 — a ±0.25 °C digital temperature sensor with a 3-wire SPI interface — taken through the entire loop inside ATE·IQ.
The TMP126 datasheet becomes 34 structured requirements — DC, SPI protocol, register and temperature-accuracy tests — each with limits, forced conditions and a test number.
ATE·IQ generates the full IG-XL project: Flow, TestInst, Limits, Binning and Pin Levels sheets, a VBT module implementing all 34 tests, and 24 deterministic SPI pattern files. 100% requirement coverage, all pins attributed.
A five-sheet UltraFLEX-class DIB schematic — DUT, signal conditioning, power, DCVI and digital I/O — with series terminations, Kelvin force/sense, ESD clamps, and routed copper that passes the electrical DRC clean.
Wafer-sort and final-test STDF lots parse natively. One test trends toward marginal (Cpk 0.27) and the dashboard names the board components to probe first.
The traceability dashboard shows 34 of 34 tests traced end to end — requirement → test → net → copper → lot — with every reconcile green and one documented measurement-path compensation disclosed, not hidden.
Three deliberate boundaries, because a verification tool that overstates itself is worse than none. First: offline checks are labelled as offline — final compile authority for a generated program remains IG-XL itself, and ATE·IQ says so rather than claiming tool-correctness it cannot prove. Second: statistics-backed limit proposals are proposals; the tool never silently edits a production limit. Third: when data is missing — no channel plan saved, no STDF ingested, no scan run — the surface says awaiting data and shows the step that unlocks it, instead of rendering an empty chart as if it were a clean one.
The useful first conversation is a walkthrough on your device and your tester — a spec you already ship, a workbook you already maintain. Tell us what you are working on and we will reply within one business day.