Round the clock: a coupon's actual misregistration is always worse than can be seen in any single mount's cross-section.

Circuit boards are mature. Specifications and industry classifications have been established; supply-chain partnerships developed; processes and materials qualified; data transfer worked out; workmanship and process control finessed into non-issues. Nothing else needs doing. Or does it?

We still struggle with spec details, drawing notes, troubleshooting and DfM interactions between fabricator and assembler. Better materials are developed, and external costs and "green" pressures must be dealt with. The pressure worsens as the move is made into high-reliability products, harsh environments, microvia/HDI technologies, shrinking geometries, smarter devices and extended supply chains.

Fabricators control their processes and shipments based on internal test results. Similarly, customers must evaluate workmanship quality. Lots are accepted or rejected, equipment and processes are evaluated, business relationships defined, even careers hang in the balance. One set of attributes requires viewing a pot-and-polished "mount," prepared from a coupon, that is positioned within and represents the quality of a panel of boards. The mount reveals important quality attributes in cross-section: delamination, voids, separation, smear and etchback, misregistration/eccentricity, (annular ring, breakout, etc). Important decisions are made based on careful inspection of the mount. Therein lies the problem.

Eccentricity is the issue: the misregistration of one circular PCB feature to another circle that should share a common axis. Judging the coupon by looking at cross-sections of mounts should be easy. In fact, it's too easy. The inspector or process engineer simply looks at the mount and decides "yes" or "no": Does breakout exist? Does the annular ring meet spec? Is something excessively misregistered? Simple. These things can be observed and measured directly from the mount.

So what's the problem? The problem arises because the mount is profoundly and favorably biased. A coupon's actual misregistration is always worse than tan be seen in any single mount's cross-section. Well, maybe not always. Only rarely (trigonometrically speaking: 1[degrees] out of 360[degrees]), the mount cross-section reveals the precise and actual misregistration of the coupon. In all other cases, the coupon is worse. Conversely, and importantly, the coupon can never be better than what the mount reveals. Therefore, the decision-maker is urged to realize that the coupon's misregistration is almost always worse--how much is unknown--than the mount's cross-section.

The discussion describes the geometry of the mount and the coupon from which it is cut. It does not cover the other serious and lurking possibility, that the coupon itself is not representative of the panel.

The accident of "clocking," or rotational orientation, of the cut made to prepare the mount influences all accept/reject decisions on attributes related to eccentricity. By pure geometry, the actual eccentricity of the coupon will, in most cases, be worse than the view revealed in the mount. The reason lies in the rotational location of the cut that is made as the mount is prepared. For example, FIGURE 1 shows a misregistration as would be seen in an x-ray or some true view of an eccentric feature in a coupon. The arrow indicates a tangency feature, called zero annular ring or breakout. We can see that the large circle goes tangent to the inner circle. The actual misregistration could very well occur in any position, as shown. For the purposes of our example, call this a rejectable condition. Will that rejectable condition be caught by a conventional coupon-mount inspection? Probably not. Reason: The usual mount is cut orthogonally; i.e., XX or YY depending on how the coupon is oriented in the panel and which hole pattern in the coupon is cut through for the mount. And any given cut is likely to miss the worst orientation.

[FIGURE 1 OMITTED]

FIGURE 2A shows the mount view if the cut were at AA. The registration is OK. There is no breakout or apparent insufficient annular ring. Ship it! FIGURE 2B shows what the mount would reveal if it were cut at BB. Still OK; not even close to tangency or breakout. If the cut had been at CC, FIGURE 2C, the registration would appear perfect! For this exercise, we can say we've just accepted the quality of these attributes, based on inspection of the mount(s). So, far, the cuts have revealed no problematic tangency or breakout.

[FIGURE 2 OMITTED]

But what if the mount had been accidentally and exactly clocked at DD? Equivalently, what if the actual misregistration had been oriented exactly such that it lined up with a typical orthogonal mount cut (FIGURE 3)? In this case, the inspector sees the mount at its worst; i.e., at exactly the orientation that shows the full extent of the misregistration. Reject! The same amount of actual misregistration, but a completely different accept/reject decision. Simple geometrical inevitability: Most decisions on misregistration, based on a single vertical mount cut, are wrong. This error is biased: the situation is usually worse, never better. The coupon can never be better than the evidence presented by a single vertically cut mount. Importantly, this is not a typical statistical situation, which can be managed by increasing the sample size. Misregistration is typically not random: whole panels are laid up a couple of mils off; drills "walk" off-center, usually in a certain direction; artwork on an entire panel is shifted a bit. One panel is off in the northeast direction, the next is off due south. The next is dead on. Simply cutting more mounts to be "sure" will likely result in more "certainty" about a wrong decision.

[FIGURE 3 OMITTED]

As a quiz, assume zero annular ring is rejectable. Look at FIGURE 4. Is the coupon from which this mount was cut acceptable? It is? Why?

[FIGURE 4 OMITTED]

(Although not detailed here, it is crucial to cut the mount through the centerline of the hole. All sorts of errors will always arise from a mount that is not cut exactly through the centerline. The unconvinced reader is urged to play with some simple cartoons to get a feel for the potential misinformation spawned by off-center cuts. These effects will impact decisions on wall thicknesses, registration, diameters and more, certainly compounding the clocking issues described.)

The Solutions

We've defined the problem. What are the solutions? Take more similar cuts/mounts? No, per the above reasoning. Tighten the spec? That's a clumsy, but occasionally effective, way to weed out bad product, and weed out good product. Take a second cut, orthogonal to the first? That's geometrically much better. Take many vertical cuts, clocked around the dial? Sure, but impossibly expensive. Note that these must not be averaged. It is appropriate to look for, and to make decisions on the worst mount attribute condition. Use x-ray? Also expensive, but perhaps a combination of an x-ray and good pattern-recognition software and protocols could be cost-effective. Horizontal cuts? Also expensive and requiring practiced finesse. Some combination is probably a good bet.

This article dramatizes a fact that is recognized but possibly underappreciated. Decisions on registration quality made by looking at mounts are positively biased. The actual attribute is almost always worse than it appears. The reason stems from simple geometry. Any vertical cut will likely miss the true extent of any eccentricity. To properly evaluate a coupon based on mount evidence, several additional cuts should be made: at 90[degrees], preferably also 45[degrees] and 135[degrees]--and these are not to be averaged. Referee actions must acknowledge the consequence of the bias: at what point is special action to be taken? (Perhaps more vertical cuts when and where the worst misregistered cut appears.)

ACKNOWLEDGMENTS

The author would like to acknowledge the assistance of Mike Green and the Sunnyvale Lockheed Martin PWB Commodity Team in providing the impetus, the challenge and the samples, and the interactions with John Perry and Jack Crawford and the other good folks in IPC, who also struggle with this situation.

TOM CLIFFORD is a process engineer at Lockheed Martin (Imco.com). He can be reached at [e-mail].