Nanobump Flip Chips: Realizing the Advantages
04/01/2007
BY GEORGE A. RILEY, Ph.D., Contributing Editor
While research in nano-transistors and nanotube wiring has been widely publicized, few seem to realize that a less glamorous application of nanotechnology in microelectronics device packaging appears closer to commercial reality. In December 2005, Fujitsu Limited and Fujitsu Laboratories LTD announced successful laboratory operation of a flip chip high-power amplifier (HPA) assembled with carbon nanotube (CNT) bumps.1 This is the first reported micro-packaging application to take advantage of the high thermal conductivity of CNTs for improving device performance. It has the potential of doubling the power handling capability of high-frequency amplifiers in next-generation base stations for mobile and satellite communications.
The prototype semiconductor is a stock gallium nitride high-electron-mobility transistor. Because these high-voltage devices provide an output power of over 100 watts from a few square millimeters of chip, thermal resistance is a major concern. To minimize thermal resistance, these power devices are usually mounted face-up on a grounded metal package that is directly attached to a heatsink. However, the long bond wires required in this configuration create a high inductance to ground, degrading performance as application frequencies move towards 5 GHz and beyond. Flip chip assembly eliminates high-inductance bond wires, but the thermal conductivity of conventional gold flip chip bumps is inadequate for highest power operation. Replacing gold bumps with CNT bumps both reduces inductance and provides adequate heat transfer with the much higher thermal conductivity of CNT.
 Figure 1.Conceptual view of CNT layout on substrate. |
The small feature size and high aspect ratio attainable by growing CNT bumps in place provide a further benefit. CNT bumps may be bonded directly over the 10-μm-wide source electrodes, where most of the heat is generated. Gold bumps are too big to be located there, and must be placed on the bond pads, extending the heat path within the die by several hundred microns before reaching a heatsink.
Figure 1 shows how multiple CNT bumps are arranged on a high-thermal-conductivity aluminum nitride (AlN) substrate. The multiple parallel source bumps make only a small contribution to ground inductance, while removing heat directly from the source to the substrate.
 Figure 2.SEM photo of nanotube bumps on substrate |
To create the bumps, the substrate was first patterned using a lift-off process. A thin film of iron catalyst was deposited on the patterned metal film. Iron is preferred for this application, because iron catalysts tend to grow taller, higher-aspect-ratio CNTs. The multi-walled CNTs were grown on the deposited catalyst by hot-filament chemical vapor deposition (CVD), using a mixture of acetylene and argon gases as the carbon source. The resulting connection density is about 1011 CNTs per square centimeter, with a typical tube length of 15 μm.
 Figure 3.SEM close-up photo of the end of one source bump. |
Figure 2 is a SEM photo of the CNT bumps on a portion of the substrate. The long, thin source bumps are less than 10 μm wide and about 15 μm high. The drain bumps across the top are about 185 μm long. Figure 3 is a close-up of the end of one source bump, showing the structure of individual CNTs.
The substrate nanobumps were plated with 1 μm of gold before a standard HPA chip was thermocompression bonded to the bumps at a bonding temperature of 345°C. The resulting chip-to-substrate gap was 12 μm.
Tested and Proven
Extensive comparative testing proved that the electrical characteristics of the HPA die were essentially the same before and after flip chip bonding. The unchanged turn-on resistance of the flipped device shows that the resistance contribution of the CNT bumps is negligible. The bumps are effective in removing heat from the HPA, with a calculated thermal conductivity of 1,400 W/mK - almost five times higher than gold.
 Figure 4.Diameter distribution of catalyst particles and CNT after particle sorting. |
The inductance of the flip chip device to ground was less than half that of the wire-bonded device, giving the bumped HPA a high-frequency gain advantage over wire-bonded. Above 5 GHz, the flip chip HPA gain is 2 dB higher than wire-bonded.
It Doesn’t Stop There
While this prototype performance is encouraging, researchers are continuing their efforts to obtain even higher thermal conductivity. A key to increasing thermal conductivity is to grow a higher density of carbon nanotubes. Although a density of 1011 per square centimeter sounds high, it is more than an order of magnitude below the theoretical maximum density.
A direct method to increase the number of tubes per unit area is to reduce the average diameter of the nanotubes, allowing closer packing. Each nanotube grows from a single catalyst particle, like a plant from a seed. Sowing a higher density of smaller “seeds” thus results in more tubes per unit area.
A classical molecular dynamics study of nanotube growth, verified by experimental results, concluded that a cap-shaped nanotube nucleus could be formed with almost the same diameter as the catalyst nanoparticle. This study established that the key to smaller, higher-density tubes is smaller, more-uniform catalyst particles.
The iron catalyst for the described prototype device was deposited as a thin-metal film sputtered onto the substrate. There was no control over the deposited particle size in the film, and smaller iron particles tended to agglomerate into larger ones on the hot substrate. The new catalyst deposition method creates a stream of distinct particles before depositing them, and selects only a narrow range of small particles from the particle stream for deposition on a cooler substrate.
A differential mobility analyzer was developed to allow sorting and selecting of nanoparticles by size.2 This machine takes advantage of the difference in momentum of small and large particles moving in a curving gas stream to isolate the smaller-diameter particles and direct them to the substrate, diverting and discarding the larger particles. Figure 4 shows a typical distribution of catalyst particle sizes deposited by this method, and the related distribution of tube diameters. Selecting a narrow distribution of particle sizes, and depositing them as discrete particles rather than as a continuous thin film, increased CNT densities in localized areas to 9 × 1011 cm-2. The density of the deposited catalyst particles was about 5 × 1012 cm-2, so there is a possibility of even further increases in tube densities.
Conclusion
If these and related laboratory advances can be realized in commercial production, today’s HPAs could double in high-frequency power. Substrates bumped with CNTs may set the new performance standard for the next-generation of communication base stations.
References
1. T. Iwai, H. Shioya, D. Kondo, S. Hirose, A. Kawabata, S. Sato, M. Nihei, T. Kikkawa, K. Joshin, Y. Awano, and N. Yokoyama, “Thermal and Source Bumps Utilizing Carbon Nanotubes for Flip-chip High-power Amplifier,” IEEE International Electronic Devices Meeting 2005.
2. S. Sato, M. Nihei, A. Mimura, A. Kawabata, D. Kondo, H. Shioya, T. Iwai, M. Mishima, M. Ohfuti, Y. Awano, “Novel approach to fabricating carbon nanotube vie interconnects using size-controlled catalyst nanoparticles,” IEEE International Interconnect Technology Conference 2006.
GEORGE A. RILEY, Ph.D., contributing editor, may be contacted at Flip Chips Dot Com, 210 Park Ave. #300, Worcester, MA 01609; 508/753-3572; E-mail: [email protected].
AP Exclusive: Interview with Yuji Awano, Ph.D.
Advanced Packaging contributing editor, George Riley, Ph.D., had the opportunity to talk with Yuji Awano, Ph.D., research fellow in the Nanotechnology Research Center of Fujitsu Laboratories Ltd, and program manager of the Carbon Nanotube Interconnect Program, about his work with CNT applications.
Doctor Awano, what is your company’s position in carbon nanotubes?
We did not pioneer CNT materials, but we want to pioneer CNT applications for electronics, so our team began studying this technology in 2000. In 2005, we announced the first successful CNT application, in a high-power amplifier for cellular-phone base stations.
Why was a high-power amplifier chosen as the first application?
We were able to assemble a research team combining the strengths of both our high-power amplifier group and our CNT group. Power devices benefit most from the high thermal conductivity of CNT.
What progress has the company made with the HPA since 2005?
Work on increasing the density of CNT led to our deposition process that gives close control over the size of catalyst particles. This should allow for at least doubling the CNT density, and related thermal conductivity.
What is your company’s goal in this program?
We expect that the next generation of base stations will require a higher frequency band than is presently needed. We can’t wait for the new frequency announcement to begin development, so our goal is to have a higher power, higher frequency amplifier ready before the announcement.