Skip to main content

4.4 Programming & Tuning

ProgrammingInspection programming translates raw component and solder data into actionable pass/fail judgments. The primary operational goal of tuning inspection is whereto rawminimize machinethe capabilityFalse turnsAlarm intoRate reliablewhile judgment.ensuring zero escapes (False Negatives). This requires a structured approach based on Golden samplesSamples, disciplined Library Management, and disciplinedcontinuous libraries create a stable foundation, while risk-based limits ensure the system focuses on what truly matters—catching escapes without drowning in cosmetic noise. Iterative feedback,iteration driven by Pareto analysis,Analysis keepsof false callscalls. under control and makes tuning a structured process instead of a guessing game. Most importantly, inspectionInspection programs are not substitutesa substitute for process control; they actfunction as a highly sensitive indicatorsprocess monitor that point back toflags upstream drift in printing andor reflow when patterns emerge.placement.

4.4.1 WhatThe “programmingInspection &System tuning”Programming really meansLoop

You’reProgramming turningand tuning transform static design data into a robust, repeatable inspection judgment. This judgment must be consistent across shifts and operators.

  1. Data Import: CAD/Gerber data (centroid location, polarity, component shape) is imported and mapped to the machine's library.
  2. Teaching and Calibration: The machine is physically taught using the Golden Board into(Chapter a2.5) to establish optimal repeatablelighting judgment. That’s:

    1. good lighting/slices, 2) lean, reusable library items, 3) sane limits per risk class, and 4) a feedback loop that trims false calls without letting escapes slip through.



    4.4.2 Build the right golden set (don’t teach from randoms)

    • Golden boardprofiles (doAOI) not ship): known-good after AOI/AXI/ICT/FCT.
    • Near-limit board: same build, but with “barely OK” features (smallest fillets, highest allowed voids). Teaches the tool whatand stillslice passesplanes (AXI/Laminography).
    • Bad examples packTuning:: croppedThresholds imagesand inspection criteria are iteratively adjusted to balance the detection capability against the operational cost of realmanual verification.

Mandate: The inspection program's success is measured not only by the defects (bridges,it tombstones,catches HIP,but void clusters) labeledalso by type.

  • Store top/bottomthe AOIlow lightingnumber profilesof false calls it generates.

    4.4.2 Library and Data Standardization

    A standardized library is essential for rapid changeovers and high throughput across multiple products.

    • One Package, One Library: A single library item should be created for each unique component package (e.g., 0402, 0.5 mm QFN). This item defines the Presence Window, AXIPolarity slice heights/ROIs, and a one-pager of acceptance limits with each product rev.



    4.4.3 Library discipline (the stuff that scales)

    • One package = one library item, reused across products; keep per-finish lighting variants if needed (e.g., ENIG vs OSP).
    • For AOI items: define presence window, pin-1/polarity regionRegion, and padSolder ROIsPad ROI (solderRegion checks).of AvoidInterest) usingfor logos/lot text as truth.inspection.
    • Reusability: Library items must be reused across all product programs to ensure consistency and minimize programming time for new products (NPI).
    • AXI Specifics: For AXIAXI, items:the library item must lock parameters such as ball count/pitchpitch, component height, slice plane locations (Laminography),, slice plane(s), and void %voiding rules (per ball vsvs. total)total area).
    • NamingGolden Set:: The program must be built using data derived from the Golden Board<Pkg>_<Pitch/Size>_<Finish if(known-good special>_v#after FA) and, ideally, supplemented with Near-Limit Boards (assemblies with barely acceptable features) to teach the system what still passes. No “temp_final_new2”.



    4.4.43 Risk classesClassification &and Limit Setting

    Not all defects are equal. Inspection limits must be tailored to the risk level to prevent cosmetic issues from blocking critical high-reliability production (soWIP).

    tuning

    Risk hasClass

    Defect guardrails)Type

    Limit

    • Strategy

    Action upon Failure

    Class A (mustCritical)

    Missing neverComponent, escape):Polarity Reversal, BGA Bridge.

    Tight Limits. polarity,Immediate missinghard parts,stop BGAor bridges,rejection criticalto netthe shortsrework →station.

    Must tightbe limits,addressed reviewbefore ifthe unsure.next panel begins.

    • Class B (quality/cosmetics):Major Quality)

    Tombstoning, Insufficient Fillet, Minor Skew.

    Moderately Wide Limits. smallRequires wettingverification defects,and filletrework; cosmeticsdata →is widerlogged band,for biasSPC toward(Chapter pass4.5).

    Used withto trendmonitor tracking.process drift.

    • Class C (informational):Informational)

    Smudges, Silk Screen Nicks, Flux Residue.

    Chart Only. smudges,Never silks nicks → chart only; never blockblocks WIP. Used for periodic housekeeping checks.

    PutData is captured but does not trigger an alarm.

    4.4.4 Tuning and the classFalse onCall everyReduction rule.Loop

    When

    Tuning inis doubt,the promotecontinuous process of adjusting parameters to areduce stricterFalse classCalls for(False NPI,Positives) thenwithout relaxintroducing afterEscapes (False Negatives). This must be driven by data.



    4.4.5 The fast iteration loop (Pareto → change → recheck)

    Run 50–200 boards, then:

    1. Pareto false callsAnalysis: byRun refdesa &sample reason,lot (50–200 boards) and generate a separatePareto chart fromof trueall defects.false calls, classified by component reference designator (RefDes) and specific failure reason (e.g., "U10 fillet too reflective").
    2. Iterative Fixes:Fix Address the top 3 false-call sources with the smallest possible change (e.g., adjust a single lighting gain,angle, ROIslightly size,widen one thresholdcomponent's notch)acceptable X/Y window). Avoid global relax.threshold relaxation.
    3. Verification: Re-run a small mini-lot (10 boards) →to confirm the false callscall droprate drops and escapesthat staythe flat.detection of true defects remains intact.
    4. Process vs. Program:Version-bump theIf program;tuning addcannot resolve a 1-line “why” in the changelog.

    Repeat weekly untilconsistent false calls settle under your targetcall (e.g., ≤0.5–1.0/board)recurring Insufficient Solder calls on a specific component), thenthe switchissue is likely Process Drift (low SPI volume, poor paste release) and must be referred to monthlyManufacturing trims.

    Engineering



    4.4.6 When to touch process, not code

    Use inspection asfor a thermometerstencil or printer recipe fix, not(Chapter a1.5).

    • hammer:

    4.4.5 Change Control and Program Maintenance

    Inspection programs must be treated with the same change control rigor as PnP programs (Chapter 2.2).

    • BridgingVersion callsControl: clusterThe →AOI/AXI checkprogram revision must increment with any change to the libraries, limits, or algorithm settings. The program version is tied directly to the product's SPIGolden area/cleaning and separationRecipe (7.Chapter 2.5) before widening AOI thresholds..
    • BGA voidsDocumentation: highEvery acrossprogram change must include a lotone-line →explanation revisitin QFN/BGAa windowingchangelog (7.4) or soak/TAL (9.2)e.g., not"v2.1: AXIReduced limits.false bridge calls on C47-C49 by softening lighting angle per QE approval").
    • TombstonesMachine Learning (ML): → re-balance chip apertures (7.4) and confirm ramp rate (9.1).

    If the Paretosystem smellsuses likeML/AI afor processdefect drift, fixclassification, the linetraining set must be diverse, and keep inspection tight.



    4.4.7 ML/auto-learn without regret

    • Train on diverse lots (mask colors, finishes, vendors).
    • Keep a holdout set (neverof trainedimages on)must be reserved for spotregular, checksindependent everyverification rev.of the model's accuracy.

    Final Checklist: Inspection Program Health


    Action

    Control Point

    Responsibility

    Library Management

    SaveComponent library items are reused, standardized, and version-controlled.

    AOI/AXI Programmer

    False Call Control

    False call rate is measured weekly and falls below the exampletarget imagesthreshold with pass/fail scores; if a model decision isn’t explainable, don’t use it on Class A items.

  • Re-train only on a schedule (e.g., monthly),≤ not1% mid-shift.per board).

    • Quality



    4.4.8 Change controlEngineer (tieQE)

    Process toFeedback

    False ECNscall soPareto docs match the floor)

    • AOI/AXI programflags revprocess drift increments(e.g., withrecurring anyshorts) ECNinstead thatof changesa landtuning patterns,flaw.

    Process silks,Engineer or(PE)

    Recipe components.

  • Integrity

    • Bundle:Program program file,revision, lighting/slice settings, and limits sheet,sheet goldenare images.

  • Storelocked underwithin the product’product's Golden Recipe so stations load by ID, not memory.bundle.



    • 4.4.9 Roles & cadence (who does what)

    • AOI/AXI programmer (R): builds/edits libraries, runs Pareto loops.
    • QE (A): sets limits, classes, and approves changes.
    • PE/ME (C): decides when issues are process not program and triggers stencil/profile fixes.
    • Operators (I): log nuisance calls accurately; don’t “teach around” defects.

    Daily:Manufacturing clear review queue.
     Weekly: Pareto + top-3 trims.
     Monthly: program health (false calls/escapes trend) + ML retrain (if used).Engineering



    4.4.10 Pocket checklists

    Before first lot

    • Golden/near-limit/defect packs ready
    • Libraries reused (no one-offs); risk classes set
    • Limits sheet posted; lighting/slices saved in recipe

    After 50–200 boards

    • False-call Pareto built (by refdes/reason)
    • Smallest fixes applied; mini-lot verified
    • Program rev’d with a clear “why”

    Ongoing

    • Escapes cross-checked at ICT/FCT (with image backlinks)
    • Process drift flagged upstream (SPI/oven) before loosening rules
    • ECN ties intact; Golden Recipe bundle current


    When inspection programs are built from trusted golden sets, refined with disciplined iteration, and connected to process feedback, they deliver stable, low-noise performance. The result is inspection that works as an efficient safeguard—quietly filtering risk, accelerating debug, and keeping production flow steady.

    Back to top