Image sensors adopt wafer-level packaging for mobile phone cameras
07/01/2008
This month, SST presents a preview of the July edition of Chip Forensics, an online column by Dick James, senior technology adviser at Chipworks, a specialty reverse-engineering company that takes apart ICs and electronics systems in order to provide engineering information for its customers. In this column, James examines how wafer-level packaging is finding a niche in image sensors for mobile phone cameras.
One of the challenges to adoption of cameras into cell/mobile phones has been the development of suitable packaging that protects the image sensor, lets the light in, and can be integrated cost-effectively into a camera module. Most of the sensors we have seen use some sort of chip-on-board technique in which the chip is wire-bonded to a PCB, which is then assembled into the camera module. Even though this is now highly automated and performed under rigorous clean-room conditions, the die is exposed until the module is sealed, so some form of wafer level packaging (WLP) would seem to have an inherent advantage.
In 1999, Israeli company Shellcase announced its ShellOP WLP technology [1], targeted on image sensors. Shellcase then licensed it to Xintec in Taiwan, and set up its own packaging facility in Jerusalem. The package was slowly picked up by companies such as Sanyo and Wizcom, and by 2004, they announced that 80 million image sensors (in all sectors) had been packaged using their technology.
Perhaps by this point the guys at Shellcase were a little disappointed that their system had not achieved the commercial take-up that they hoped for, since in 2005, they changed from a packaging company into an IP licensing and consulting one. This provided the classic exit strategy, and later in the year, Shellcase was bought by Tessera, becoming a WLP “Center of Excellence” within Tessera.
 Figure 1. Sony Ericsson V630i and W710i mobile phones, with cases removed. |
We found a good example of the ShellOP package used with the 2-megapixel OmniVision OV2640 CIS that we extracted from two Sony Ericsson mobile phones-a V630i and a W710i. OmniVision seems to have adopted the ShellOP technology in this sector, because we have now seen in it most of the company’s recent phone CIS chips. Figure 1 shows the different form factors of the two phones. The form factor affects the way that the camera module is mounted–in the “candy bar” V630i, it can be set directly on the main board, whereas in the W710i flip-phone it has to be remote from the PCB on a flex-circuit that also connects the display.
 |
null
 Figure 2. De-mounted CIS, X-ray and optical cross sections. |
In Fig. 2 we see the de-mounted module, and an X-ray and optical cross section of the assembly (the metal shield around the module is still present in the optical section). The module is ~8.2mm square ?? 7mm high; the three lenses and the infrared filter are visible above the CIS die.
 Figure 3. Exposed CIS die and die edge. |
After removal of the super-structure, the die is exposed on the 38-lead CSP2 base package in Fig. 3. Around the edges of the die we can see bright spots–actually metallized extensions to the bond pads. We can also see the glue layer to attach the cover glass to the die. Looking at the die layout reveals that a relatively small area (~28%) is taken up by the active pixel array, and the rest of the die is taken up by an impressive amount of integrated image processing circuitry. According to OmniVision’s product brief [2], the chip includes a 10-bit ADC, a PLL, balance controls, a DSP and compression engine, the necessary blocks of SRAM, and a microcontroller to help run all of that stuff. The bottom line is that the OV2640 is capable of a 15 frame/sec output at UXGA (1600 ??1200) resolution, and can output anything from the raw RGB data to a number of compressed formats.
The sensor is actually fabbed by TSMC in its long-standing relationship with OmniVision, likely in its 150nm or 180nm process, and uses a 2.2 ?? 2.2µm pixel size in a novel design, which uses six transistors for every four pixels, or effectively a 1.5T pixel.
 Figure 4. Details of ShellOP package in OmniVision CIS. |
However, the topic here is the WLP used for the part; if we zoom in from Fig. 2 and look at the cross section of the image sensor die (Fig. 4), we can get an idea of how things have been put together. In Figs. 4b-d the solder bump connects to a metal lead that goes up the beveled side of the glass substrate, and then up the side of the glue that attaches the glass to the die. At the top this lead contacts the bond pad extensions (Fig. 4c) that we see in Fig. 3, providing inter-connection from the top surface of the die to the FR4 substrate.
But, the question is: how is it done? Xintec has put the basic steps of the process up on its Website [3].
What we are seeing is a cover glass actually used as a handle layer. Adhesive is applied to the scribe lanes, overlapping the die edges, and the 0.38mm-thick glass plate is glued on, sealing the image sensor surface. The device wafer is then ground to ~125µm thick, and the scribe lane masked and etched from the backside, likely using potassium hydroxide (KOH) since the angle of the etched face is ~55o. The sub-assembly is then glued to the second glass plate/substrate (probably pressed into the glue, since we can see filler particles adjacent to the lower glass surface), inverted, and a barrier layer is applied and patterned into pillars that support the lead-free solder bumps. Next, the sandwich is notched along the scribe lanes, into the glue layer above the die surface, exposing the cross section of the bond pad extensions.
The metal lead material is applied and patterned to link the solder bump sites to the extensions; a solder mask is spun on and patterned, the bumps are formed, and finally the glass/silicon/glass sandwich is diced into the individual WLP image sensors.
This sequence has the advantage of sealing the wafer surface right at the start of the process, but it poses the question as to what is used as the pad extension layer, since it has to be part of the wafer processing. Xintec implies that it is applied as an extra layer at the end of normal wafer processing, but this is not the case.
 Figure 5. Top edge of die and metal lead. |
Figure 5 shows a cross section of top edge of the die (right) and the lead from the solder bump below. The metal lead to the backside is a bi-layer of aluminium and nickel-iron. Unfortunately we had no cross section showing the link from the extension to an internal pad; this image is offset from the link. However, we can see that the bond pad extensions are actually formed from the same three metal layers as are used in the image sensor, using the conventional Ti-TiN-Al-Ti-TiN sandwich structure, and also that the KOH etch has used the STI as an etch stop. When we compare Fig. 5 with Fig. 3b it appears that the die seal is continuous across the pad extensions, so the top metal cannot be used for the link. Close examination of a plan-view sample beveled down to the substrate shows that the link layer is Metal 2 (Fig. 6), with a break in the die seal to allow the connection. The Metal 1 die seal is also continuous.
 Figure 6. Metal 2 bond pad extention and die seal. |
Since the link is aluminium, it makes sense that the base layer of the track to the solder bump is also aluminium. A look at the base of a solder bump in Fig. 7 reveals that the aluminium step coverage is characteristic of an evaporated or sputtered layer, and the patterning of the leads was performed after the aluminium deposition. The NiFe was then plated onto the Al tracks. It also appears that the support pillars are made from the same material as the solder mask.
 Figure 7. Base of solder bump |
There appears to be no discrete under-bump metallization, so after patterning the solder mask, the bumps are formed from a predominantly tin-based solder, likely a tin-silver-copper combination. The final step is the die separation.
Most of these steps are discussed in the Shellcase patent [4], filed in 1995, but at least for us reverse engineers, seeing the real thing is far preferable to reading about it in the dry language of a patent lawyer! We could also say this is a good example of an interesting invention waiting for an application–it wasn’t until mobile phone cameras took off that the ShellOP package really found a big-time commercial application.
Addendum: OmniVision recently announced [5] a sensor that uses backside illumination. Apart from the technical challenges of reversing the sensor structure to make the backside light-sensitive, it obviously cannot use the ShellOP package, so one would think the Tessera/Xintec alliance would be out of the loop for these new parts. On the other hand, the new structure will obviously require massive wafer thinning and an innovative package style, so maybe we will see more creative wafer level packaging from the OmniVision/Tessera/Xintec grouping; they have already proved they are resourceful in this field.
References
- Chip Scale Review, Jan.-Feb. 1999, www.chipscale review.com/9901/forumb1.htm.
- www.ovt.com/data/parts/pdf/Brief2640V2.4.pdf.
- www.xintec.com.tw/shellOPEng/ShellOP.
- US Patent 6,040,235, “Methods and Apparatus for Producing Integrated Circuit Devices,” P. Badehi.
- www.ovt.com/data/newsreleases/english/BSI%20Technology%20launch%20release_FINAL.pdf.
DICK JAMES is a 30-year veteran of the semiconductor industry and the senior tech analyst for Chipworks, an Ottawa, Canada-based specialty reverse engineering company. Contact him at 3685 Richmond Road, Suite 500, Ottawa, ON, K2H 5B7, Canada; ph 613/829-0414, [email protected].