Sub-system
Inspection is where machine vision earns its keep, and it splits cleanly in two. Measuring a dimension against a tolerance is deterministic and solved. Deciding whether a surface looks wrong is a judgement, and the hard part is that the factory produces very few examples of wrong.
In one sentence
Inspection software analyses industrial images to measure features against tolerances and to classify defects, running within the cycle time of the production line.
Rule-based measurement finds edges, fits geometry and compares against a specification. It is fast, deterministic, explainable to a quality auditor and unchanged for decades — and it is the right tool whenever the property being checked is a dimension.
Learned inspection handles the cases rules cannot express: a scratch, a smear, a subtle texture difference. Its practical obstacle is data. A line running at high yield produces almost no defective examples, so the interesting class is rare by construction, and collecting a balanced dataset can take months of production.
Because defects are rare and varied, the more workable framing is often to model what normal looks like and flag departures from it. That needs only good parts to train on, which the factory has in abundance, and it catches defect types nobody anticipated — at the cost of flagging harmless variation until the threshold is tuned.
A false reject scraps a good part; a false accept ships a bad one. Those costs are rarely equal, and the operating threshold should reflect that rather than maximising accuracy. Systems configured to a balanced metric are frequently mistuned for the economics of the line they run on.
In regulated production, a rejection has to be justifiable and the inspection method has to be validated. A model that outputs a score with no visible reason is difficult to qualify, which is why heat maps showing what triggered a rejection are a practical requirement rather than a nicety.
1 of these is marked as a chokepoint: a handful of qualified suppliers, a multi-year lead time, or a single geography.
Any inspection model assumes the scene looks the same every cycle; lighting drift breaks it silently.
Industrial imagingThe scarce input: a high-yield line produces very few defects, which is exactly what a classifier needs.
Human data operationsDecisions must be made inside the cycle time, on hardware in the cell rather than in a data centre.
Inference siliconHeight, flatness and volume cannot be read from a projection, so the checks that need them depend on a laser profile or a point cloud rather than an image.
3D sensingWhat each company supplies at this step, and — where a public figure exists — its share of this specific market — with what that share measures, the period it covers and who published it. Some rows also show the company’s own reported revenue for the segment covering this step, which is a different thing: it says how much this business matters to that company, not how much of the market it holds. Not a ranking and not a recommendation.
Supplies vision software spanning rule-based tools and learned inspection.
Supplies integrated inspection systems and vision software.
Supplies inspection and process control for semiconductor and packaging production.
Supplies inspection and metrology systems for electronics manufacturing.
Supplies the vision libraries a large share of integrators build their own inspection on rather than buying a system.
Supplies vision software alongside the cameras, which is how most of this market is actually sold.
Supplies the metrology software that decides whether a measured deviation is out of tolerance.
Sells learned defect detection for the cases where the defect cannot be written down as a rule.
Whether anomaly-detection approaches displace supervised defect classification in production.
Whether validated learned inspection is accepted in regulated manufacturing.
Whether inspection models transfer between lines without retraining on each.
Each page explains one technology in plain language, states what it depends on, and names companies by what they supply at that step. Company roles are described qualitatively and deliberately carry no market shares, revenue figures or rankings — those change faster than an explainer can, and a stale number is worse than none. Ticker links point at company pages on this site and are provided for reference only.
Nothing here is investment advice, a recommendation, or a forecast. A company named on a page about a technology is not thereby a good investment, and the chokepoints described are structural facts about supply chains rather than predictions about prices. Technology moves; where a page describes something as unresolved or in development, that was true when it was written.
Plutux is not an investment adviser. Market data and AI-generated analysis are for information and education only, not investment advice. Disclaimer