New process for flip chip assemblies
BY ARONNE J. CAMILLERI, YUNLONG SUN AND EDWARD SWENSON
Functional laser tuning of traditional ceramic-based hybrid circuits has been accomplished with the use of a standard 1.064-µm wavelength Nd:YAG laser for more than two decades. Bare silicon chips integrated on a hybrid circuit have to be covered with a protective shield during laser tuning to reduce drift or performance shift because of the photoelectric response of the devices to the laser light.
As the industry moves from leaded devices to flip chip technology, the challenge of reducing this drift or per-formance shift during functional tun-ing becomes an even greater issue. With flip chip devices, the silicon is face down on the ceramic, and shield-ing becomes extremely difficult. It is now established that the process of laser functional tuning monolithic integrated circuits (ICs) can be signifi-cantly enhanced with the use of a Nd:YAG laser with a longer wavelength, 1.318 µm, because of the elimination of the photoelectric response of the sili-con device at the new laser wavelength. This article describes the basic principle and reports process results on a production thick-film module. The results show an ability to functionally tune the module with the laser when using bare flip chip technology.
Flip Chip Complicates Current Approach
Traditional 1.064-µm Nd:YAG lasers have been employed by the industry for more than 20 years to functionally tune packaged hybrids to an explicit performance specification. Lower tolerance components – and therefore less expensive ones – can be used, because tolerance variations are subsequently tuned out by laser trimming thick-film resistors to achieve the desired final device specifications.
Since the early days, engineers have known that silicon circuits are photon sensitive. Even system-viewing illumination must be shut off during testing and tuning. A more difficult problem is the photo response induced by the 1.064-µm Nd:YAG laser used to do the tuning, which requires special covers or shields for the ICs, as well as complex tuning sequences to work around the laser-induced photo response. This results in a much lower process throughput.
Figure 1. Responsivity of silicon as a function of the wavelength of light. |
To increase package density and functionality and lower assembly costs, packages are migrating from wire bonded ICs to flip chip technologies, which greatly increase the chip's expo-sure to the 1.064-µm laser light. The bumped flip chips are mounted face down on the ceramic substrate, making them much more difficult to shield. Additionally, the ceramic substrate tends to scatter the output beam from the Nd:YAG laser across the substrate. Because the IC is facing the ceramic, it is exposed to or “sees” the laser output and reacts to it, and moving the resistors to be trimmed farther away from the IC is of little help. Making a change in the Nd:YAG laser wavelength solves this problem.
Laser Wavelength Solution
Figure 1 shows the light responsivity of silicon versus the light's wavelength. The plot illustrates the high responsivity to the 1.064-µm wavelength of the standard Nd:YAG laser used for hybrid tuning since the 1970s.1 This responsivity is the source of the problem. In most cases, this upsets the circuit's performance, creating serious offsets or preventing the normal tuning process altogether.
Figure 2. Automotive module with flip chip IC used for laser trimming evaluations. |
Note that in Figure 1, by shifting the laser wave-length to 1.318 µm, which is a secondary line of an Nd:YAG laser,2 silicon becomes “blind” to the laser output and thereby is not affected by the functional tuning process. This innovation was first introduced in 19963,4 for trimming thin films on silicon to functionally tune devices directly on silicon sub-strates. This almost immediately became a volume produc-tion process, because the throughput and yield improve-ments were dramatic.
Figure 3. Output of flip chip silicon IC with a 1.064 µm laser pulse for (above) thin film on silicon and (below) thin film on ceramic assemblies. |
There have been suggestions (or suspicions) that shifting the wavelength of an Nd:YAG from IR to green at 0.532 µm, or from a UV at either 0.355 µm or 0.266 µm, would solve the photo response problem. Once again referring to Figure 1, the green light exacerbates the respon-siveness of the silicon IC, and the UV light is no better than the stan-dard 1.064-µm Nd:YAG laser.
|
Why has this elementary solu-tion been so long in coming? The problem has been the very low gain available at 1.318 µm from the lasing material of Nd:YAG. The power needed to trim thin-film materials on silicon is two orders of magnitude lower than that required to trim standard thick-film resistors. The original diode- pumped lasers and the older lamp-pumped lasers just could not deliv-er adequate power at 1.318 µm for thick-film trimming. In recent years, the development of high-power diode arrays has allowed the production of a wide range of high-power, diode-pumped lasers, operating from UV to 1.318 µm. The new classes of diode-pumped Nd:YAG lasers require no external water-cooling, are wall-plug compatible (110 volt, single phase), and easily exceed one year of service-free continuous operation.
Thick-film Trimming at 1.318 µm
The new high-power 1.318-µm diode pumped Nd:YAG lasers have an output specification of 3 watts at 10 kHz. Although this is less power than is available at the conventional 1.064-µm wavelength, it is more than adequate to trim today's “thinner” thick-film resistors, particularly for functional tuning where trimming speeds are modest. Shifting the laser wavelength from 1.064 µm to 1.318 µm does not raise the laser power or the energy needed for the resistor trimming process.
When comparing the optical energy absorption versus wavelength for different metals and glasses commonly used in thick-film resistors, the absorption remains relatively flat between 1.064 µm and 1.318 µm for all of the materials. Also, other materials used for thick-film resistors, such as ruthenium oxide (RuO), are not semiconductor materials, and their absorption will not change significantly between 1 µm and 1.3 µm. Additionally, there is no reason to suspect that laser trim quality, such as the trim kerf quality and characteristic post-trim drift, will change from moving the laser wavelength from 1.064 µm to 1.318 µm. Short-term drift experiments have confirmed that no difference in the trim kerf or drift characteristics was observed. This is also consistent with an earlier, more detailed study done on trimming thin films on silicon with the 1.318-µm laser.3
An Example
Figure 2 is an automotive electronics module with a flip chip silicon IC. The circuit was powered up, and the output monitored during tuning with 1.064 and 1.318-µm laser wavelengths.
Figure 4. Output of flip chip silicon IC with a 1.318 µm laser pulse for (above) laser scribing and (below) resistor trimming. |
To illustrate the problem, the output from 1.064-µm and 1.318-µm lasers were scanned across the bare ceramic sub-strate area in front of R1, as shown in Figure 2. Figures 3a and 3b show the oscillations in the output created by the 1.064-µm laser when scanned at power levels required for trim- ming thin-film on silicon and thin-film on ceramic resistors, typically tens of milliwatts to a few hundreds of milliwatts respectively. (This is well below the power needed to trim thick-film resistors.)
|
Figure 4a shows the output of the circuit when the laser output is increased sufficiently to scribe two lines in the ceramic at scanned area 1. Note that in Figure 4a, exposure to the 1.318-µm laser wavelength at this much higher power has no effect on the output of the circuit. When trimming the thick-film resistors at normal trim power, a “plume” is gener-ated during the trimming process and is seen as a bright spot on the CCTV monitor. Figure 4b indicates that there is no photo response when trimming the thick resistor films, con-firming that the intensity of the plume is insufficient to cause a problem. Similar photo induced responses to those shown in Figure 3a and 3b were observed when exposing all other resis-tor locations to the 1.064-µm laser output. Trimming any of the resistors with the 1.318-µm laser wavelength displayed no photo induced response at all.
Conclusion
A high-power 1.318-µm YAG laser has been tested and documented for tuning thick-film modules. The same method can also be used when trimming SMT modules on organic substrates, where screened thick-film resistors are replaced with surface mounted discrete chip resistors. Flip chip trimming and functional tuning of thick-film resistors with 1.318-µm wavelength of light instead of the more traditional 1.064-µm wavelength virtually eliminates the photo response in the silicon flip chip devices. This solves the problem created when unshielded flip chips are trimmed and enables direct tuning of thick-film modules. AP
References
- Arthur G. Albin, Edward Swenson, et al., “Laser Resistance Trimming from the Measurement Point of View,” IEEE Transactions on Parts, Hybrids, and Packaging, June 1972, Volume PHP-8, Number 2.
- Walter Koechner, Solid State Laser Engineering, Second Edition, Springer-Verlag, 1985.
- Russ Barcey, Edward J. Swensen and Yunlong Sun, “Reducing Optoelectric Response on Silicon Integrated Circuits,” Proceedings of the IEEE Electronic Components and Technology Conference, May 28, 1996.
- Edward J. Swenson and Yunlong Sun, “Method for Laser Functional Trimming of Films and Devices,” US Patent Number 5,685,995., Nov. 11, 1997.
Aronne J. Camilleri, senior product/test engineer, can be contacted at Standard Motor Products Electronics, 170 Sunport Lane, Orlando, FL 32809; 407-541-2255; E-mail: [email protected]. Yunlong Sun, corporate fellow and director of photonics R&D, and Edward Swenson, senior VP of research and development, can be contacted at Electro Scientific Industries, 13900 NW Science Park Drive, Portland, OR 97229; 503-641-4141; 503-643-4873; www.esi.com.