AOI False Calls And Program Tuning

Automated optical inspection, usually shortened to AOI, is a measurement system that uses cameras and algorithms to compare what is on the board with what the program expects. Its usefulness depends less on the hardware than on the program, because a program that flags good joints as defects sends product to unnecessary rework and teaches the operators to disregard the results. The two numbers that describe a program are the false call rate and the escape rate, and they pull in opposite directions.

This article explains where the errors come from, how three dimensional measurement changes the picture, how a program is tuned without hiding real defects, and which layout decisions make inspection easier.

Where The Errors Come From

A false call is a good joint that is reported as defective, and its origin is usually a library threshold that does not match the product. Each component and each footprint has parameters such as the expected fillet length, the acceptable colour, and the permitted offset, and those values are set from a template rather than from the board in front of the machine. Where the template was taught on a different laminate, a different finish, or a different paste, the first production run produces calls that are not defects.

The second source is variation in the process itself. A paste deposit that varies across the panel changes the fillet on every joint, a board that warps changes the height that the camera sees, and lighting that drifts with the age of the lamps changes the contrast. The inspection system detects all of that faithfully, which is why reducing false calls is often a matter of reducing process variation rather than of adjusting the algorithm.

Two Dimensional And Three Dimensional Systems

A two dimensional system judges a joint from its colour and its outline in a single image. That approach works when the solder is uniform in appearance and the lighting is controlled, and it fails when a finish is bright, when a component casts a shadow, or when the joint colour varies with the profile. It is also the reason a two dimensional program has to be taught with a narrow tolerance, which raises the false call rate.

A three dimensional system measures the height of the joint and reconstructs the fillet from the geometry, so the decision is based on shape rather than on colour. That removes most of the dependence on the finish and on the lighting, allows the limits to be expressed as a height and a volume, and makes the program far more portable between products. Where a board mixes reflective and matt surfaces, the three dimensional measurement is usually the difference between a usable program and a permanent argument about acceptable joints.

Tuning Without Hiding Defects

The first step is to correct the library rather than the threshold. A call rate that is concentrated on one footprint usually means the footprint parameters are wrong, and adjusting them fixes many calls at once without affecting anything else. The second step is to teach the program from the actual product, using boards from the beginning, the middle, and the end of a production run rather than a single golden board, so that the natural variation is inside the limits from the start.

The third step is to separate the two failure directions. Calls that are caused by genuine process variation should be reduced by fixing the process, not by widening the limit, because widening the limit to silence a call also widens the window through which a real defect can escape. Where a call is unavoidable and harmless, it is better to reclassify it as a review item than to remove the check, so that the operator knows the machine is working as designed.

Escapes And Why They Happen

An escape is a defective joint that the system passes, and it is the more dangerous error because it is invisible until the customer finds it. Escapes come from joints that are hidden under a component, from shadowed areas where the camera cannot see the fillet, from low contrast between solder and the surrounding surface, and from defects whose signature is outside the algorithm entirely, such as a cold joint with a normal outline or a crack inside a ball.

The escape rate can only be measured by an independent check. A sample of passed boards is inspected under a microscope, or with X-ray where the joints are hidden, and the results are compared with the machine output. That correlation study is what gives a program its credibility, and it should be repeated when the product, the paste, the stencil, or the profile changes, because a program that was verified against one process is not verified against another.

Program Management

A program is a document and should be treated as one. It is versioned, the change history is recorded, and any modification is verified against a known sample before it is released to production. The call rate is tracked as a process indicator, and the positions that generate the most calls are reviewed at a regular interval, because a position that calls repeatedly is either a library error or a process trend rather than an occasional false alarm.

Retraining is required whenever the inputs change. A new paste, a new stencil thickness, a new laminate finish, or a new reflow profile changes the appearance of every joint, and a program that is not retrained after those changes will drift out of its limits. The review meeting that follows a shift in the call rate is the cheapest tool available, because the cause is usually visible in the data before it is visible in the yield.

Layout And Process Measures That Help

Inspection is easier when the board is designed for it. Components that are spaced apart do not shadow each other, a silkscreen that is clear and does not overlap a pad gives the camera a reference, and a solder mask with good contrast to the copper makes the joint boundary visible. Polarity markings that are covered by the component body, or by a connector that is placed over them, remove the only evidence the system has about orientation.

The layout choices that make a board easy to inspect are the same ones that make it easy to assemble, and they are collected under manufacturable design guidelines and layout decisions that affect production. The quality framework that surrounds the acceptance criteria is described under PCB design quality characteristics.

FAQ

Is a lower false call rate always better? No. A program can be made quiet by widening its limits, and the same change lets real defects pass. The two rates have to be managed together rather than traded silently.

Why does a three dimensional system call fewer false defects? Because it measures the height and the shape of the joint instead of its colour, so the decision does not depend on the finish, on the lighting, or on the reflective properties of the solder.

How often should an inspection program be retrained? Whenever the inputs change: a new paste, stencil, laminate finish, profile, or component. A program that is not retrained after those changes drifts out of its limits without any change to the machine.