May 1998 - ChipScale Review

May 1998


eMail the Editor

Machine Visio Inspection (MVI) Enhances CSP Yield and Reliability

While chip-scale package mvi is similar to the inspection of traditional bga packages, there are some critical differences.

By Ed Caracappa, Acuity Imaging Inc., Nashua, N.H.

Although relatively few chip-scale packages (CSPs) are now in production, their widespread development is moving steadily forward, and within a few years their benefits will be taken for granted by consumers worldwide.

There are many steps between the conceptual promise of CSPs and the technologically and economically successful manufacture of these packages.


Figure 1. The typical CSP process offers several opportunities for machine vision inspection.

Like BGAs, but More Fragile
Since the inception of the earliest pilot lines, machine vision inspection (MVI) has been widely used to monitor CSP production processes. In some ways, machine vision inspection of CSPs is similar to the inspection of traditional BGA packages. At some critical process points, however, machine vision inspection is more important for CSPs.

The differences stem, in part, from the more fragile design of CSPs, and, in part, from the transport method. In pilot lines at companies preparing for CSP production, the packages are variously carried on segmented tape in metal stiffeners, unsingulated arrays in strip form or custom boats or fully singulated in JEDEC trays.

Machine vision inspection of CSPs becomes significant as soon as the eutectic solder balls are placed on their pad locations, and continues to be significant until final packing of the device for delivery. Throughout these processes, MVI has three purposes:

  • It identifies individual anomalies at the earliest point to permit rework and prevent adding value to a flawed device
  • It permits identification of repeating defects or defect trends
  • Finally, it gives manufacturers the data needed to control processes

Inspection After Ball Placement
MVI dynamics can be examined at the point in the process just after the placement of solder balls on the CSP. This is usually the first of as many as five inspection points. Flux has previously been placed on the pad locations by screening or pin transfer.

Ball diameters for CSPs are currently .008" to .012". The pins used in the pin transfer method require diameters of .005" to .008", and pins this small are easily damaged.

These constraints make flux deposition by screening more attractive.

Placing the right amount of flux is also critical, because excess flux is known to cause bridging of the balls during reflow. Additionally, the method of flux deposition can impact reliability.

The solder balls themselves can be placed by vacuum transfer or gravity feed. Vacuum transfer uses a tooling head which has been machined to match the pad array matrix of the device. Moving to the solder ball reservoir, the tooling head engages vacuum to pick up the balls. The tooling head then moves precisely over the pad locations, releases the vacuum and deposits the balls.

Gravity feed systems use a tooling plate into which solder balls are fed from a constant ball supply. Like the vacuum tooling head, the plate is machined to match the pad array matrix onto which it deposits the balls.

Image Collection
After they are placed, the balls are inspected by machine vision. During inspection, the machine vision illumination system strobes the balls and collects the image.

Images are typically collected at 1,200 solder balls per second. This speed is essentially unrelated to the number of balls per device, and means that MVI is inherently faster than the throughput of the line itself. This speed is important for CSP production; at 1,200 balls per second, the inspection throughput of CSPs with 46 balls each is over 30,000 devices per hour, including time for motion.

At this speed, machine vision inspects the placed balls for several parameters. Proprietary algorithms are used which tell the system what it should expect to see. One algorithm, for example, spots both missing and extra balls. Both of these anomalies, if not found by inspection, can result in device failure during electrical testing.

Both anomalies can have many causes. A missing ball, for example, may result when the tool plate becomes contaminated, and a ball fails to fill its location in the plate pattern.

Vanguard Automation, a customer of Acuity Imaging,* has reported to us that they have occasionally observed this phenomenon. MVI can determine that a ball is missing and where in the placement head the error is being generated.

In addition to spotting missing balls and extra balls, machine vision measures the diameter of each ball to identify undersized and oversized balls, both of which are obvious potential sources of problems.

While measuring diameter, algorithms also measure the circularity of each ball. This is important because, while a ball may have the correct overall diameter, out-of-round balls (which are usually oval) may also result in rejects.

The evenness of illumination for the system's CCD camera is particularly important in acquiring the image data for these measurements. Field of view and spatial resolution are critical: The more pixels per ball, the more accurate the image data.


Figure 3. The CSP can be examined after the placement for solder balls.

Checking for Surface Oxidation
Algorithms also measure the intensity of the light reflected from each ball. This is primarily a check for possible oxidation of the ball surface. Oxidation depends largely on handling; balls which move in contact with each other can become oxidized very quickly.

Marygrace Stevens, Vanguard Automation, notes that oxidation is not usually an isolated event. If one ball is oxidized, they all are, and the source of oxidation is likely to rest with the vendor. Oxidation darkens the ball surface. The machine vision system's CCD camera uses 256 gray levels (needed, among other things, for finding edges of structures), so distinguishing darkened balls is a straightforward matter.

The precise algorithms used in these measurements- and in measurements following later process steps- must be robust. This, in turn, requires extensive processing of the data, since algorithms which follow abbreviated paths to generate output data are generally inaccurate or misleading.

A fast processor speed for the machine vision system is therefore important, but the results gained from truly robust algorithms are so much more credible that users will even accept minor time delays to acquire the right data.

System Reporting
MVI systems typically report at one of two levels: the device level or the ball level. Reporting at the device level simply gives the overall status of the device-accept or reject.

Machine vision systems, the type of system which Vanguard incorporates into its workstations, is an example- are nearly always operated at the device level. When a defective device is reported, the operator can then access data at the ball level to characterize the defect.

In the future, it may become feasible to run systems more extensively with ball-level reporting. The reason: The data gained has the potential to be used as feedback to control process equipment. Quick diagnosis and quick correction of a process anomaly, such as a contaminated tool plate, would be an effective way to keep a CSP line running and to keep costs down.


Figure 4. MVI setup checks for singulation-caused defects.

MVI After Reflow
For both CSPs and traditional BGAs, reflow is the process step which fixes in place any existing anomalies. Rework after reflow may be unfeasible, depending on the value of the device. Alternatively, it may be physically very difficult. At least in early production runs of CSPs it is advisable to perform MVI post-reflow.

Post-Singulation Inspection
In the singulation of CSPs, a critical concern is alignment- that is, cutting between die without causing damage to any part of the device. Various methods of alignment have been tried for CSPs: One positions the devices with the balls down and looks at the backside of the die for alignment.

Another method aligns the devices for singulation using the solder spheres as the alignment feature. Both mechanical and laser cutting methods have been used in the development of singulation systems.

After singulation, the silicone seal around the die as well as the die itself and the solder spheres should be inspected.


Figure 5.Edge chips and sealant delamination are the major flaws that may result from the singulation process.
Process Parameters Subject to subtle variations
According to industry consultant Tony Perrotta (who is also a past president of Amkor), process parameters are very stringent, especially for high I/O-count CSPs.

At the same time, variations within those parameters may be caused by phenomena which are extremely subtle. "It's not unusual," Perrotta notes, "to have two more or less identical production lines, one of which experiences few anomalies, while the other has numerous anomalies."

Perrotta points to temperature variation as one of the many parameters that can impact the behavior of solder balls. "Temperature variation can have many causes," he observes. "It could be a simply an incorrect temperature, or it could be a control failure.

Solder balls are very sensitive to temperature, and one of the results of temperature variation is joined balls. Identifying the cause of joined balls is fairly simple- a statistical analysis can pinpoint the place and time of the event within the process flow fairly quickly. But in order to characterize this defect you must first be able to see it.

"There are many other potential anomalies," he adds. "The dangers of non-similarity in solder balls are legion. Perotta's conclusion: "I wouldn't build a BGA or CSP without MVI."

Post-Test Inspection
For CSPs the next key MVI point after reflow is typically after test. Procedures for handling CSPs during test are still under development.

CSPs designed leaving the backside of the die exposed makes handling at this point vastly different from that of traditional encapsulated BGAs.

Several problems are inherent in this procedure, the most important of which is that the device, with the backside of the die exposed, must be lifted and moved with a degree of mechanical force into the test socket.

The potential for damage is great. The silicone seal around the die can be broken, and, in some CSP designs, the seal is the only moisture barrier present. In addition, the die itself can be cracked, chipped or scratched.

MVI after test can find both cracked die and seal damage, although there are probably some limitations as to the degree of damage which can be detected. Die cracks in CSPs being handled for test are most likely to occur on the back of the die. Die cracks in traditional devices have numerous causes and can range in size from very small to catastrophic. If the release of stresses caused by handling creates a significant difference in elevation on the opposite sides of a die crack-that is, an edge-then the crack will be visible to machine vision. A very small crack which has essentially no difference in elevation may not be visible.

In measuring the integrity of the silicone seal around the die, machine vision makes a relative measurement of the width of the seal from the edge of the die. Above a known minimum, the actual quantity of silicone laid down to form the seal may not be important. In addition, machine vision identifies significant voids and bubbles within the seal and in areas where the sealing material is missing.

Final Inspection
The final MVI for traditional BGAs occurs typically just before packing for shipment. Inspection at this point in the process is the last chance to capture any defects prior to shipment.

The critical inspection features are damage to the backside of the die, such as edge chips, scratches, and cracks. Die cracks are particularly dangerous because their dimensions can initially be very small and because they are good candidates for growth from thermal cycling.

When a crack affects the active area of the die, it typically results in outright failure. Additionally, the solder spheres, mark and orientation of the device should also be inspected.

Many CSPs will be built on a more rigid substrate than flexible tape. This substrate will nonetheless be very thin to minimize overall device thickness. These CSPs require additional inspection for overall substrate flatness. The basic risk is that the very thin substrate may become warped during processing, adversely affecting ball-top co-planarity.

The quality of solder spheres also affects CSPs. Difficulties in manufacturing solder spheres under .020 inches have generated a number of anomalies in sphere shape, diameter and volume.

The greatest chance for successfully capturing these defects is to measure the individual singulated devices for co-planarity prior to shipment. This inspection is performed using a 3D laser scanner which allows for substrate flatness measurements to be made at the same time.

By recording these two critical measurements, co-planarity and substrate flatness, the device manufacturer can decouple the contribution of substrate defects from those of traditional ball-top co-planarity. This allows appropriate repairs to be made to the process or materials, thereby improving overall yield.


Figure 6.Inspection during test is still under development, but will likely include inspecting the part for sphere diameter, quality and the presence of spheres.

Two Key Defect Categories
The overall payback from MVI of CSPs can be examined more closely by looking at events subsequent to ball placement and the consequences of the various anomalies which may occur.

Ball placement necessarily occurs near the end of production, with the result that each device has-at this point- a relatively high intrinsic value.

Suppose an anomaly, such as a missing ball or an extra ball, occurs during ball placement. If the defect is detected before reflow, rework is simple and inexpensive, and may be feasible even if the intrinsic value of the whole device is relatively low. But if the anomaly is not detected until after reflow, rework is much more difficult, and may be so costly that it is not feasible, even for devices with relatively high intrinsic value.

Our experience with many manufacturers has resulted in a useful evaluation of specific anomalies. While the number of machine vision inspectable anomalies is fairly large (missing balls, oversized and undersized balls, out-of-round balls, misplaced balls, edge chips and cracks on die, etc.), all anomalies can be placed into one of two categories:

1. Anomalies which will cause an immediate electrical failure in the packaged device
2. Anomalies which cause no immediate electrical failure in the packaged device

Anomalies in the first category include missing balls and bridged balls, severely misplaced balls, and, in some instances, other anomalies such as ball oxidation.

These anomalies-or, more precisely, the performance consequences of these anomalies- will be found by routine electrical tests. While these devices will cause economic loss by being rejected, they will not be shipped or installed, and thus will not cause field failures later.

Anomalies in the second category include undersized balls, and, in most instances, out-of-round balls, oversized balls, the presence of foreign particles such as dust, some degrees of ball oxidation and cracks in the die.

These anomalies will frequently pass initial electrical tests, and the devices carrying these anomalies will therefore be shipped or installed. But since these anomalies represent stresses which will be subject to both thermal cycling and normal environmental hazards, they are very likely to cause field failures relatively early in the expected lifetime of the device. Manufacturers can, therefore, employ MVI simultaneously to detect anomalies within both classes.

Anomalies in both categories may or may not represent isolated events. If a given process is drifting badly, the result may be a defect trend, which is most usefully identified in its early stages.

The high speed of MVI, along with the high correlation of its measurements with both types of anomalies, gives manufacturers the flexibility to make the best economic decisions to achieve the highest yield and long-term reliability.

Ed Caracappa is Semiconductor Pro-gram Manager at Acuity Imaging Inc., an RVSI company. He can be reached at 603.598.8400, fax 603.598.4684.

*Vanguard Automation Inc. is a subsidiary of RVSI.



Chip Scale Review o 7291 Coronado Drive, Suite 8 o San Jose, CA 95129 o Email: editor@chipscalereview.com



Feature, 05/13/99, 05/13/99, ID=9805/caracappa1
Keywords=fc00 bc00

© 1998 ChipScale REVIEW