Green wavelength laser processing of memory metal fuses
09/01/2006
Metal fuses in redundancy circuitry can be laser repaired to enhance yield. However, shrinking IC design rules have necessitated fuse pitches smaller than the 2µm pitch lines that can be processed by traditional infrared lasers. Now metal fuse structures down to 1.0µm pitch can be cut using a newly developed, 532nm green-wavelength laser source on materials including poly-Si, W, and Al alloy. FIB/SEM observations and electrical measurements confirm that this is a viable production solution for the 65nm node and below.
Joohan Lee, Joseph J. Griffiths, James Cordingley, GSI Group, Wilmington, Delaware Kyoung-Suk Lyu, Kyeongseon Shin, Samsung Electronics, Hwasung-City, Korea
One of the more established final steps in fabricating ICs with circuit redundancy, such as memory, involves laser processing of fuses (or links) to disable defective circuit elements and replace them with good redundant ones [1]. Polysilicon has been widely used for link material due to its superior cut quality from unique material properties. IR wavelength lasers (1.0~1.3µm) have been successful in processing these polysilicon links.
However, due to the high resistance and complex fabrication process limitations of deeply buried polysilicon fuses, the industry has driven the migration of fuse material to aluminum. For high-performance logic devices and high-speed SRAM, copper has been also investigated as link material due to its enormous benefits when compared to aluminum, such as low resistance, power dissipation, manufacturing cost, and superior resistance to electromigration.
In previous works, we explored the failure mechanisms and status of current laser cut processes of aluminum and copper [1-4]. Lower corner cracking as well as substrate and neighbor damages were seen, and we defined the laser-energy process window based on finite-element simulation results and various experimental observations.
In the laser processing of Al or Cu links, IR wavelengths such as 1.064µm (Nd:YAG), 1.047µm (Nd:YLF), and 1.342µm (Nd:YVO4) lasers have been widely used due to their relative stability, favorable IR-absorption characteristics of Si, and acceptable process windows.
However, the continuous shrinking of device dimensions of Si MOS technology has resulted in the scaling of repair technology. Because the density of memory cells has been of primary importance in reducing their cost, the reduction in cell size has been achieved by the use of smaller interconnect linewidths, as well as by cell structure complexity [5]. With this trend, decreased fuse pitches have been required in the redundancy circuitry, and the demand for even smaller pitch has increased over the generations.
Decreased fuse pitch and size have necessitated a laser process with smaller minimum spot sizes and better focus margin in order to avoid damages to adjacent fuse structures. In addition, clean cutting with low-laser energies has become more desirable for shrunk-pitch fuse structures that tend to be more susceptible to laser damage than the Si substrate.
Today’s 1.0~1.3µm wavelength lasers have shown limited capability as acceptable and reliable production tools down to about 2.0µm pitch structures, due to their diffraction-limited spot size and depth of focus (DOF) capabilities. As fuse pitches continue to decrease, the limitation from neighbor-fuse damage as a major failure mode for fine-pitch fuse structures has eliminated the advantage of less damage to the Si substrate from a 1.3µm laser. As a result, new production-proven solutions needed to be established to support advanced generations of device.
Extensive manufacturing and qualification studies have been conducted to achieve a robust laser-fuse processing step with small spot sizes and large focus margins, for both current and future fine pitch processes.
Laser-energy process window
Figure 1 displays experimental results showing how to understand the laser-energy process window of a metal cut process with a variation of fuse pitches. There were seven different fuse pitches (0.8, 1.0, 1.2, 1.5, 1.8, 2.0, and 2.2µm) and five different fuse widths (0.2, 0.24, 0.3, 0.4, 0.5, and 0.6µm) for each pitch resulting in a total of 35 fuse structures. Each data point in Fig. 1 indicates an average value from 5 different structures with different widths at each specified pitch. A 1065nm wavelength IR laser beam with a 1.5µm 1/e2 spot size and 21nsec pulsewidth was used to perform this experiment.
The Elow curve shows the minimum energy levels required to cut successfully without material remaining at the bottom of cut site. The Sub-damage and Neigh-damage curves indicate the energy levels that will damage the Si substrate and adjacent fuses, respectively. These two curves show that the energy level for damage to adjacent fuse structures decreases with shrinking fuse pitch, whereas the energy level for substrate damage stays about the same.
The Ehigh curve thus indicates the maximum energy level that can be used to process each structure without any damage, and the results were determined based on two failure modes. When the pitches are larger than ~1.5µm, Ehigh was limited by Si substrate damage (Sub-damage curve), while neighbor-fuse damage (Neigh-damage curve) occurred at higher levels. However, neighbor-fuse damage occurred at lower energy levels than Si substrate damages with a decrease of pitch to <1.5µm. Therefore, a smaller spot size and lower energy levels are required to process tight pitch structures of ≤1.5µm.
These data were based on controlled, accurate alignments, and the actual cross-over pitch of neighbor fuses will be likely even larger assuming a real production process. Lower corner cracking of aluminum links was not evaluated because the data were based on top view observations. However, the link structures were very thin relative to the link width, with the aspect ratio <1. Therefore, cracking at lower corners is unlikely and the data are considered to be valid [2].
For example, the second harmonic of the 1.064µm source yields a wavelength in the green portion (532nm) of the visible spectrum, which can generate a minimum 0.7µm 1/e2 diameter spot with double DOF compared to IR at equivalent spot size.
The third harmonic of the 1.064µm source generates a 355nm ultraviolet wavelength, which can decrease the diffraction limited minimum spot size even more with better DOF at the equivalent spot size. However, the absorption into SiO2 causes damages to optics and significant power-drop over time. Furthermore, 355nm typically generates severe damage to the Si substrate due to its high absorption.
Laser-energy experiments
The test wafers with aluminum lines were fabricated using a standard two-level metal CMOS process. The metallization used for this study was 0.6µm thick sputtered Al (1% Si, 0.5% Cu) etched to form the fuses. The Al lines were originally undercoated and overcoated with 0.05µm thick TiN. However, an ARC over-coating TiN layer was etched away to optimize the fuse thickness to form 0.35µm thick metal lines. During this etching process, surrounding SiO2 was recessed due to etch selectivity compared to aluminum. A passivation layer, consisting of 0.2µm of Si3N4, covered the metallization for the purpose of reliability after the laser process.
There are four pitches with a 0.1µm step in the 1.0-1.3µm range. Each pitch has six different fuse widths (0.1~0.6µm with a 0.1µm step). Therefore, there are a total of 24 different linear aluminum fuse structures. Each structure was designed to have two different formats: to check the cut quality (parallel) and to check for any damage to adjacent structures to ensure the acceptability of the cut (serial). Electrical measurements were conducted after microscopic observations of the processed fuse structures.
The laser system used to perform these experiments was a GSI Group M455, employing a diode-pumped, Q-switched, frequency-doubled Nd:YVO4 laser (532nm) operated in the saturated single-pulse mode. Pulses, with lengths of approximately 19nsec in FWHM scale, were directed through focusing optics to produce a beam of 1/ee diameter of approximately 0.7µm spot at focus. The positioning accuracy of the laser system was approximately <0.15µm. Three optimum energies irradiated each structure (nominal, and two slightly higher), based on laser-energy studies.
Figure 2 shows a series of links that were a 0.2µm wide fuse structure with 1.0µm pitch. They were processed with various laser-energy levels, from 0.005-0.090µJ with 0.005µJ steps, to decide the nominal energy at a spot size of 0.7µm 1/e2 in diameter. One link out of every four was blasted to see damage to adjacent links. From visual inspections, we noticed that links started to open at 0.015µJ, and damage to the adjacent links due to excessive laser energy occurred at 0.055µJ and above. Therefore, the nominal energy is (0.015+0.050)/2 = 0.0325µJ. We rounded off the value and 0.030µJ was selected as a nominal energy for this laser setting. Two slightly higher energies (0.040µJ and 0.050µJ) were also tried in order to see the susceptibility of adjacent links.
Each laser energy was used to blast two sets of 600 links (a parallel set of 300 links and a serial set of 300 links) to ensure cut quality and no damage to adjacent links, as mentioned earlier.
 Figure 3. FIB images of laser-cut sites processed with 0.04µJ at 0.7µm 1/e2 spot in diameter, with a) top views of the processed fuses, and b) cross-sectional views. |
Figure 3 shows SEM and FIB cross-sectional images of the 1.0µm pitched 0.3µm wide aluminum fuses that were processed with a laser energy of 0.04µJ and 0.7µm 1/e2 spot in diameter. Figure 3a is a top-view image showing that every other link was processed to check for any adjacent damage. The top-view image reveals that fuses look wider than actual size because of the Si3N4 layer deposited after aluminum etching. The Si3N4 passivation layer can be seen in Fig. 3b as the bright layer on the top. Figure 3b shows that the fuse in the middle was not blown, whereas the two fuses on the sides were blown and aluminum was removed. The image also reveals aluminum debris around the cut sites, which was generated during the rupture of the aluminum links by the laser energy. The debris was one of the reasons for using slightly higher energy than nominal for actual processing of the metal-link structures. All of the aluminum, as well as the TiN undercoating, was removed by the laser cutting process.
Electrical test results
Various metal structure designs and laser parameters, as previously described, were tried and the results were measured electrically. Figure 4 shows electrical measurement data from 1.0µm pitched metal fuse structures with various fuse widths, using three energy levels (0.03, 0.04, and 0.05µJ). Figure 4 displays the resistance measurement data of 300 laser-processed paralleled links, indicating that all of the processed links were cut successfully throughout the various widths. All the data show around 100GΩ, which is well beyond the value for acceptable laser link processing.
 Figure 4. Electrical measurement results of 300 parallel link structures at 1.0µm pitch (each data point) processed with 0.7µm 1/e2 diameter spot showing cut qualities. |
The resistances of a series of fuses (the serial structures), which are next to processed fuses, were also measured. The results from electrical measurements of the unprocessed links reveal that none of the three sets (one set of 300 adjacent links next to 300 process links) were damaged and kept their original resistance (below 60Ω per 300 links) after laser processing.
Electrical test results for all the link structures, including the data shown in Fig. 4, were obtained and are summarized in Figure 5. This time, each curve indicates an average value of three sets (900 links) that were processed with three different laser-energy levels. Energies of 0.03, 0.04, and 0.05µJ were used for 1.0µm and 1.1µm pitched structures, while energies of 0.04, 0.05, and 0.06µJ were used for 1.2µm and 1.3µm pitched structures. Metal-link structures pitched 1.4µm and larger were also processed and showed successful results, but the data were not included in this article in order to focus on fine-pitch structures.
All metal structures pitched from 1.0µm to 1.3µm were successfully processed without any damage to adjacent link structures. There are small fluctuations in resistance seen in Fig. 5 that are considered to be from many factors such as laser processing system accuracy and imperfect fabrication processes. However, this variation does not have statistical significance, and it is considered to be negligible for the implementation of this technology.
Conclusion
It has been shown through microscopic observations and electrical measurements that a 532nm wavelength laser is fully production-capable of processing very fine-pitch metal-link structures down to 1.0µm without any changes in current IC fabrication processes. The advantages of the 532nm laser include larger DOF with smaller spot size compared with the current IR lasers and moderate susceptibility to Si substrate and dielectrics.
This technology has been implemented into mass production since 2005, and has successfully processed various memory devices at the 90nm technology node and above. So far, the processed production devices include fuse pitches down to 2.0µm and fuse materials of poly-Si, W, and AlCu. This process should be capable of seamless incorporation with the shrunken linewidths and spacings needed for the 65nm technology node and below.
References
- J. Lee, J. Ehrmann, D. Smart, J. Griffiths, J. Bernstein, “Analyzing the Process Window for Laser Copper-link Processing,” Solid State Technology, pp. 63-66, December 2002.
- J. B. Bernstein, J. Lee, G. Yang, T. Dahmas, “Analysis of Laser Metal-cut Energy Process Window,” IEEE Semicondut. Manufact., Vol. 13, No. 2, pp. 228-234, 2000.
- J. Lee, J.B. Bernstein, “Analysis of Energy Process Window of Laser Metal Pad-cut Link Structure,” IEEE Semicondut. Manufact., Vol. 16, No. 2, pp. 299-306, May 2003.
- J. Lee, J. Griffiths, “Analysis of Laser Metal-cut Energy Process Window and Improvement of Cu Link Process by Unique Fast Rise Time Laser Pulse,” Proceedings of Semiconductor Manufacturing Technology Workshop, pp. 171-174, Hsinchu, Taiwan, December 2002.
- T. Kikkawa, “Quarter-micron Interconnection Technologies for 256M DRAMs,” Extended Abstracts, Int. Conf. Solid Devices and Materials, pp. 90-92, 1992.
Dr. Joohan Lee received his BS degree in chemical engineering from Yonsei University, Seoul, Korea, in 1996 and his MS and PhD degrees in chemical engineering from U. of Maryland in 1999 and 2001, respectively. He has been in the Laser Systems Division of GSI Group as a principal applications engineer since 2001, developing newlaser technologies for memory repair. GSI Group, 60 Fordham Rd. Wilmington, MA 01887; ph 978/661-4567, e-mail [email protected].
Joseph J. Griffiths received his BS degree in electrical engineering from Northeastern U. He is a technical product market manager in the Laser Systems Division of GSI Group. His technical expertise is in the area of laser system design and laser/material interaction in semiconductor devices.
James Cordingley received his BS degree in electrical engineering from the U. of Rhode Island. He is a senior principal engineer in the memory repair engineering group at GSI Group. He has been responsible for optics, laser design, and integration, as well as memory repair systems engineering and design.
Kyoung-Suk Lyu is a senior engineer in the memory test solution group, frontend test engineering team, Interconnect Product & Technology Division, at Samsung Electronics, Hwasung-City, Korea. He has worked on the development of new laser technology and the memory repair algorithm since 1995.
Kyeongseon Shin is a senior manager in the frontend test engineering team, Interconnect Product & Technology Division at Samsung Electronics, Hwasung-City, Korea. He has worked in the R&D of memory laser repair and test for more than 20 years at Samsung and is in charge of the memory test solutions group.