Machine vision / Article

How to choose an AOI vendor in Malaysia: 8 questions before you sign

Choose an AOI vendor on proof, not on the specification sheet. Before you release capital, make them run a feasibility study on your own parts, show you a measured false-call rate alongside the detection rate, name what the system cannot catch, and confirm that an engineer who can change the recipe is close enough to your plant to be on the floor the same week. The eight questions below are the ones that separate an inspection partner from an equipment reseller, and any honest vendor will welcome all of them.

The context you are buying into

RM711b
Malaysian E&E exports in 2025, up from RM601 billion in 2024
40%
Share of national export output the E&E sector drives
13%
Of global semiconductor assembly, test and packaging happens in Malaysia
8
Questions to ask before a purchase order, all answerable in one meeting

01 / What you are actually buying

The camera is not the risk

Malaysian manufacturers are inspecting under real pressure. E&E exports rose from RM601 billion in 2024 to RM711 billion in 2025, with semiconductors accounting for roughly 65% of that total, and the sector accounts for over 40% of the nation's exports, with Malaysia contributing 13% of global assembly, testing and packaging volume. Export customers audit quality escapes. That is the environment your AOI purchase has to survive.

In that environment the sensor is rarely what fails. What fails is the engineering around it: a recipe programmed on a golden sample that no longer represents production, a false-call rate that trains operators to wave alarms through, a defect class nobody validated, and a support engineer three time zones away. AOI systems are known to require initial programming with golden samples, with unprogrammed defects going undetected particularly in highly variable production, and setup time and cost that may not suit low-volume lines. Every one of the eight questions below targets one of those failure modes.

02 / The eight questions

Ask these before the purchase order

01Will you test on my parts first  Not a demonstration board, not a similar part from another customer. Yours, including the defective ones you have kept. If the answer is no, the rest of the conversation is theoretical.
02What is the false-call rate  A detection rate quoted without a false-call rate is half a number. Ask for both, measured on the same sample set, and ask what happens to both when the recipe is tightened.
03What can it not catch  Every optical system has a blind spot. A vendor who names theirs, hidden joints under a BGA for example, is telling you where your second gate has to go.
04Who owns the recipe  Can your engineers retrain and retune, or is every change a service call. Recipes drift as materials and variants change, so ownership is an operating cost, not a legal detail.
05How long is a new part number  Ask for the hours to bring up a variant, and who does it. This is the number that decides whether the machine keeps up with your product mix.
06Where is support based  Ask for the physical location of the engineer who will retune your recipe, and the response time in the contract. Distance is a specification.
07Does it fit my takt and my line  Cycle time at the required resolution, footprint, handling, and how rejects leave the line. Resolution and throughput trade against each other, so ask for both at once.
08What data comes out  Defect images, classification, traceability by unit, and whether it reaches your MES. Inspection data you cannot query is inspection you cannot improve.

03 / Answers that should worry you

Four replies that mean keep looking

"Our accuracy is 99.9%." On what parts, against which defect classes, at what false-call rate. A single headline figure with no sample set behind it is marketing. Ask for the number on your parts or treat it as unquoted. "We will tune it during installation." Tuning during installation means the feasibility work never happened, and you are funding it with production time. "It catches all standard defects." Your escapes are not standard, or you would have caught them.

And the most expensive one: "You do not need to change the lighting." Illumination is the first decision in any optical inspection, and a vendor who skips it is selling you a camera and hoping. Resolution has a cost too, since AOI equipment comes in resolutions from 10 to 20 micrometres, with higher resolution providing finer defect detection at the cost of throughput. Any vendor promising finer detection and higher speed with no trade-off is describing a machine that does not exist. The mechanics of the false-call trap are in our piece on reducing AOI false calls without letting defects escape.

04 / What proof looks like

A feasibility study, not a demonstration

A demonstration shows you a machine finding a defect it was set up to find. A feasibility study tells you what will happen on your line. It should come back with images of your parts under the lighting and optics the production system will use, a defect-by-defect detection result, a measured false-call rate on the same samples, the cycle time achieved at that resolution, and a written statement of what could not be detected.

Read the last item first. If a vendor tells you a class of defect is out of reach optically and recommends X-ray or a manual gate for it, that is an engineering answer and it is worth more than a clean pass on everything. It also gives you a defensible inspection plan to show an export customer during an audit. Where the honest limits of optical inspection sit is covered in our comparison of 2D and 3D AOI, and the baseline you are improving on is in what manual inspection actually catches.

05 / Support is geography

Who is on your floor on Thursday

An inspection system is not finished at handover, and this is where most AOI disappointment actually originates. A supplier changes a material finish, a new variant enters the line, ambient light shifts near a shutter door, and the false-call rate climbs. If the person who can retune the recipe is in another country, the machine gets bypassed long before the service visit is scheduled, and you are back to manual inspection with a capital asset standing idle.

Malaysia is a good place to solve this, because the local machinery and equipment ecosystem is real: homegrown firms are delivering automated optical inspection systems, smart automation platforms and precision engineering tools into front-end fabrication and advanced packaging lines. For a plant in Selangor or the Klang Valley, the practical question is simply which of those engineering teams can be at your machine this week. CODETRACE keeps engineering and support in Shah Alam, Selangor and Batu Kawan, Penang for exactly that reason.

06 / Our own answers

How CODETRACE answers its own eight questions

We do not sell hardware on day one. The first step is a feasibility study on your parts, and the deliverable includes what we could not catch. CODETRACE machine vision and AOI runs on AONIA, our optical inspection machine, which is designed, built and supported in-house in Malaysia, so optics, lighting, handling and the defect model can all be changed for your part rather than configured around a catalogue specification. Defect models are trained on your real defect images, and where those images are too few, NVIDIA defect image generation for AOI expands the training set.

CODETRACE is a member of the NVIDIA Inception program, and was at COMPUTEX 2026. Building the machine ourselves is what makes the support answer credible: the team that designed your inspection cell is the team that retunes it, from Shah Alam and Batu Kawan. Inspection also sits inside factory automation, so handling, rejection and traceability get designed with the vision station rather than after it. Bring us a defect you keep missing. We will tell you if vision can catch it.

Buy the proof on your parts, not the spec sheet.

FAQ / Choosing an AOI vendor

Questions, answered.

01

How do I choose an AOI inspection vendor in Malaysia?

Judge the vendor on proof, not on the specification sheet. Ask them to run a feasibility study on your own parts, including the defects you keep missing, and to show you measured detection and false-call rates from that study before you release capital. Then check that support is physically close to your plant, that the inspection recipe belongs to you, and that someone who can change the recipe can be on your floor the same week. A vendor who will not test on your parts is selling hardware, not inspection.

02

What should an AOI feasibility study include?

Images of your real parts under the lighting and optics the production system will use, a list of defect classes with the detection result for each, a measured false-call rate, the cycle time achieved, and a clear statement of what the system could not detect. That last item is the most valuable part. It tells you where you still need a second gate such as X-ray or a manual station, and it separates an engineering assessment from a sales demonstration.

03

Why does local support matter for AOI in Malaysia?

Because an inspection system is not finished at handover. Recipes drift as materials, suppliers and product variants change, and a machine that over-rejects gets bypassed within a week. When the engineer who can retune the recipe is in Shah Alam or Batu Kawan rather than another time zone, a drifting false-call rate is a same-week fix instead of a quarterly service visit. Response distance is a specification, and it belongs in the purchase decision.

04

Should I buy AOI from a local Malaysian vendor or an overseas brand?

Both can work, and the question to settle is who owns the engineering that makes it work on your part. Malaysia has a deep machinery and equipment ecosystem, with homegrown firms delivering automated optical inspection systems and smart automation platforms into semiconductor and advanced packaging lines. What matters is whether the party you sign with can modify optics, lighting, handling and the defect model for your part, or can only resell and configure someone else standard machine.

05

How does CODETRACE work with a new AOI customer?

Feasibility first. We do not sell hardware on day one. We take your parts and the defect you keep missing, prove on real samples whether vision can catch it, and tell you honestly when it cannot. AONIA, our optical inspection machine, is designed, built and supported in-house in Malaysia, and models are trained on your real defect images. CODETRACE is a member of the NVIDIA Inception program, with engineering and support teams in Shah Alam, Selangor and Batu Kawan, Penang.

Ask us all eight. Then ask us to prove it on your parts.

Talk to our engineers
Scroll to Top