Sounding out flip chip analysis
03 October 2008
The long-term reliability of a flip chip assembly strongly depends on the condition of the solder bumps, the solder bump bonds, and the underfill material.
All of these features are sandwiched between the silicon chip and the substrate. They would be hard to inspect without cutting the flip chip assembly open if not for the fact that the silicon chip is nearly transparent to VHF and UHF ultrasound. Transparency means that a suitably equipped acoustic microscope can image and evaluate these features, and can identify threats to reliability such as delaminations and voids.
Flip chips can be imaged rapidly in a production environment and less rapidly in an analytical mode. In the analytical mode, evaluation is made by a human operator; in a production environment, evaluation is carried out by software.
Where large numbers of flip chip assemblies must be inspected and evaluated in a production environment, a different system is needed. Sonoscan has just completed development of software that works with its automated acoustic microscope systems to image and analyse large numbers of flip chip devices. Working with definitions made by the user of the system, the software identifies each flip chip as acceptable or rejectable.
The new software is integrated into the automated acoustic microscope and makes it possible to enhance the long-term reliability of very large numbers of flip chip assemblies in a short time. Automated handling of JEDEC-style trays of flip chip assemblies is combined with the same precise, high-resolution imaging used in manual systems.
When scanning, the ultrasonic transducer performs two functions: it pulses ultrasound into the flip chip assembly and receives the return echoes. Echoes are returned only from material discontinuities and interfaces within the assembly. The pulsed ultrasound passes through the silicon chip with very little absorption, and with no internal reflections unless physical defects such as cracks are present within the silicon structure. As it exits the silicon, the pulse is reflected by material interfaces and discontinuities –die-to-bond pad interface, chip-to-underfill, chip-to-solder bump, solder-bump-to-substrate and underfill-to-substrate are the major interfaces. Critically for inspection purposes, reflections are also returned from voids, delaminations or cracks within the underfill.
Voids, delaminations and cracks are all gaps. They reflect ultrasound much more efficiently than other types of interfaces. The silicon-to-underfill interface, for example, reflects a portion of the ultrasound, while the majority of ultrasound crosses the interface and travels deeper, where some of it will be reflected by the next interface. But a gap at any depth within the underfill layer reflects virtually all of the ultrasonic pulse. Since the amplitude of the reflected ultrasound determines the brightness of the pixel, gaps such as voids, delaminations and cracks are typically the brightest features in the acoustic image of a flip chip assembly.
Development of an automated flip chip inspection system involved examining the true inspection needs. It would be fairly easy, for example, for software to simply calculate the percentage of a given area occupied by voids or other defects. The given area might be the whole area of the flip chip, or one quadrant, as is used in the much simpler process of evaluating die attach. But such an approach would ignore one of the most significant threats to flip chip reliability: voids adjacent to chips. Repeated thermal cycling or the natural creep of solder may cause solder to slump into the void, with the result that the bump eventually collapses and breaks it electrical connection. Simply measuring the percentage of high-amplitude pixels over a large area would not inspect for this problem.
The new software inspects the flip chip device by inspecting cells, a cell being the area enclosed by the centers of four solder bumps (see figure). A cell may be a square, a rectangle, a parallelogram, or a trapezoid, depending on the layout of the solder bumps. The software measures the percentage of the area within the cell that contains the high-amplitude pixels that are associated with defects such as voids, delaminations and cracks. When a defined cell is considered a failure by the specified user’s criteria, the percent of high-amplitude defect area within one cell might be as high as 50% without falling into the 'reject' classification, although limits around 25% are more usual.
Two more steps are involved in determining the status of the whole flip chip. First, the total number of reject cells is counted. Second, the number of solder bumps that are unsupported by underfill on one or more sides is counted. The less support it has, the more likely it is that a given solder bump will eventually collapse and fail electrically. The software accordingly counts the number of bumps that are unsupported on one, two, three, or four sides. The final accept /reject decision for the flip chip as a whole depends on the user’s definitions for that particular type of flip chip. Those definitions in turn will be based on the user’s reliability requirements for the flip chip, and on its previous reliability record.
Because flip chips are often used for complex, high-performance devices, the arrangement of the solder bump interconnects is also likely to be complex. Solder bumps in early flip chips were typically arranged in a single, simple pattern, but in today’s designs different regions of the flip chip may use different patterns for arranging the solder bumps. In addition, there may be regions that have no solder bumps. For these reasons, the acoustic inspection software has been designed to identify the various regions, and to apply separate algorithms to each region. The recipe for the acoustic imaging and analysis of a given flip chip can be stored, to be called up when needed. It an also be locked to prevent unauthorized modification.
Anyone who has used an acoustic microscope to image flip chips knows that there are variations from chip to chip or from lot to lot. For example, the amplitude of return echoes may change from one lot to the next because the density of the underfill material has changed. More locally, the relative brightness of various features may change because of a change in the distribution of the filler particles within the underfill material. To accommodate these variations, the software contains a dynamic threshold function that adjusts the threshold range in response to variations in brightness.
Steve Martell is Manager of Technical Support Services at Sonoscan, and Tom Adams is a Consultant.
Sonoscan, Inc.
www.sonoscan.com
Contact Details and Archive...
Related Articles...
Most Viewed Articles...