The paper, presented by Ralph Binner, R&D director at ESEC Asia Pacific Pte. Ltd. (Singapore), reported test wafers that showed a narrow process window and a large proportion of completely unbondable sites. The tests were conducted on test chips provided by "a major industrial client," Binner said.

Such problems have been reported before although changes to both materials structures and wirebonding processes have been made to try and prevent damage to die. The latest test results reference chips made used Black Diamond 2 low-k material, which is supplied by Applied Materials Inc., but the results would appear to be applicable to dice using any porous low-k material with similar mechanical properties.

The test chips provided a large number of bonding sites with various types of metal and inter-layer dielectric structures underneath the pad, including solid metal, vertical or horizontal strips, lattice patterns and areas with no metal at all.

The experiments were performed with precision bonding equipment using a 40-micron gold-wire process with approximately a 20-micron solder ball diameter. Three independent variables were studied: the search velocity; the speed with which the head made its final approach to the pad, and hence, the impact force when the head touched the pad; the down-force applied to the bond; and the ultrasonic power used during the bonding.

The researchers looked for three specific failure modes on the resulting bonds: metal peeling, cracks in the inter-metal dielectric and reduced ability to withstand sheer force. Each of these mechanisms has previously been identified as a potential problem for highly-porous low-k dielectric materials.

In describing the results of the experiments, Binner presented a photomicrographic chamber of horrors. Images showed pad metal separated from the underlying stack, wires ripped from pads and damage extending deep into the interconnect stack under a damaged pad.

The statistical data were no more encouraging. Binner separated the sites on the test dice into three categories: those that were bondable under reasonable conditions, those that were bondable with very narrow process windows, and those that were essentially unbondable.

Binner said that some sites, with some particular patterns of metal and dielectric beneath them, responded well. However many sites-representing a variety of patterns-could only be bonded successfully if the three process parameters were held within about plus or minus two per cent of the most successful values.

"The most critical parameter appears to be search velocity," Binner said, "because it is directly related to the force with which the head impacts the pad, and this in turn appears to be directly related to metal peeling."

The worst news from the study was that some sites were simply unbondable. The down-force and ultrasonic power necessary to achieve an acceptable solder-ball compression ratio -- about 3.5 to 4 -- were already too much for the interconnect stack to handle, and repeatedly caused mechanical failures.

Binner suggested changes on both the manufacturing and design sides of the equation. There would have to be "a revolutionary change in the design of wirebonder heads," Binner said, and speaking after the presentation, Binner said that he believed the experimental results would have to be systematized and captured in additional design rules to prevent unbondable designs from being created.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved