The first part of this article series identified the primary drivers for the on-going development of camera modules. For the consumer, the driver is a better quality picture from single click 'point-and-shoot' operation. The handset manufacturer is more interested in reduced cost and height, as thin, portable electronics products is currently the fashion. However, the existing evolutionary roadmap of smaller and cheaper camera modules, without degraded picture quality, has come to an end. Achieving a step-wise reduction in cost and height, while simultaneously boosting optical performance, requires a radical new approach to camera module architecture, materials choices and assembly techniques.

Obsolescence of Discrete Assembly
Manufacture of a conventional camera module involves serial assembly from discrete parts. This has several fundamental limitations. First, the costs of assembly increase with every part manufactured. Once a high-volume line has been established, further gains in productivity can only be realized by diluting the administrative overhead for the facility by installing a parallel line with all the attendant capital costs involved.

Second, the precision of assembly increases with the quality of the image required. Image quality and imager resolution are linked, so manufacturing a megapixel camera module generally requires far greater mechanical precision than one of VGA resolution. The specifications of camera module assembly are already far more exacting than for conventional electronic components, necessitating more expensive machines, and generally compounded by slower throughput.

Third, each component must be manufacturable at the lowest possible cost. This means, for example, lenses must be radially symmetrical about the optical axis because this is the least expensive shape to produce and is compatible with using a screw thread mechanism to set camera focus. Imager die, however, always have a rectangular format, so there is clearly a mismatch between the aspect ratio of the lens train and the imager, which usually manifests as reduced image quality in the picture corners.

Finally, there are the consequences arising from the inability to make the product right the first time. While it is possible to set the optical axis of the lens train sufficiently close to that of the imager so as to not require adjustment, the same is not true of its height. The multiple glue line thicknesses and other variables can not be controlled with sufficient accuracy to fix the lens turret in position during manufacture. This means that every camera module must be placed in a test fixture, powered up and a series of images acquired while the focus is adjusted and then set. Clearly, this is a slow undertaking requiring a known good die tester equipped with an optical head, making final set-up of conventional camera modules a significant and unavoidable contributor to the manufacturing cost.

Wafer-Level Packaging
Wafer-level packaging of imagers is not a new architecture for camera modules, but, as it will be explained, it continues to enable design advancements.

The original basis of the move to wafer-level packaging of imagers was the need to decrease cost. While the adoption of wafer-level packaging actually increases assembly costs, it helps eliminate the major cause of yield loss from camera module manufacturing--particle contamination of, or damage to, the optically sensitive area of the imager. The benefit comes from attaching a cover glass to the face of the image sensor wafer as the very first process step (Figure 4). From that point on, the imager is totally protected from environmental and mechanical damage so the vast majority of good optical die on the wafer can be converted into yielded camera modules. (An imager die housed in a modern wafer-level package is shown in Figure 3 of Part 1 of this article series.)

A wafer-level package can be provided with a ball grid array interface on its rear surface. This allows the imager to be placed, with other components, on a common printed circuit board and attached by a single reflow soldering cycle, further minimizing assembly cost. The reliability of a ball grid array interface is also superior to the flexible circuit and connector of a conventional camera module built using chip-on-board assembly.

The wafer-level package must provide connectivity between the bond pads on the front face of the imager die and the ball grid array interface on the rear face of the package. There are a number of means by which this may be accomplished, and one has gained commercial acceptance in the form o a family of imager package solutions. More than 1 billion imagers have been included in these wafer-level packages since the technology was introduced in 2001. Not only will the packages pass the specification for cell phone parts with ease, they also exhibit a margin of safety of more than two over the more arduous JEDEC Level-1 automotive reliability standard.

Wafer-level packaging fails to be a low-cost solution if the materials and equipment set are derived from standard semiconductor practice. To achieve substantial cost reduction requires innovation in both of these areas. One approach that has proved highly successful is the adoption of mature materials produced in extremely high tonnage for an entirely different industry and purpose, and a tool set developed for the PCB industry [Humpston, G, et al., IEEE Trans. Advanced Packaging, 2008, 31(1), 33-38]. By this means, the packaging cost per die, including depreciation of the equipment, works out to a few cents per die, which is two orders of magnitude cheaper than discrete packaging.