alysis and esign Consideratio. Yu-Lin Shen, Subra Suresh, and Joseph B. Bernstein. Abstract-Lateral connections between adjacent lines of met- allization have ...
402
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 43, NO. 3, MARCH 1996
inklng of eta1 Interconnects: alysis and esign Consideratio Yu-Lin Shen, Subra Suresh, and Joseph B. Bernstein
Abstract-Lateral connections between adjacent lines of met- pability due to the isotropic nature of laser heating and allization have been developed in order to achieve high density ablation. linking for customization in programmable gate arrays and for In order for laser programmable gate arrays to become additive redundancy in restructurable integrated circuits. Links a widely applicable technology, the ability to scale down were formed by focusing a pulsed laser between two same-level aluminum lines. The mechanism of link formation appears to the size of each link with finer lithography while using be the nucleation of a fissure, induced by the thermal expansion standard processing for the metal lines and the insulation, such mismatch between the metallization and the surrounding dielec- as the CMOS (complementary metal oxide semiconductor) tric (Si&) and passivation layer (Si3N4); molten aluminum fills processing [2], [3], is crucial. Diffused links, which involve the crack. Numerical simulation by the finite element method was carried out using a plane strain model. The probable path conduction through the silicon, have been used to achieve the for the link-forming fissure, as predicted by the model on the compatibility goal [4], [5]. However, there are disadvantages premise that the local maximum tensile stress determines crack- associated with diffused links. Firstly, a severe limit to the ing, is shown to be consistent with experimental observations. scalability exists because of the need to connect two levels Parametric analyses were performed to gain insights into the of metal through a buried level of silicon. Secondly, diffused linking processes. It is found that damage in the passivation can be avoided by increasing the thickness of the dielectric between links have inherently higher resistances. Earlier experimental investigations have documented laser the aluminum and the passivation. Reducing the spacing between the metal lines increases the chance of successfully forming the formed connections between two metallization layers through link. Under certain conditions, the linking propensity can also be the interlevel dielectric [6], but the success rate did not exceed increased by reducing the metal width. In addition, the link is 99%. More recently, laser programmed lateral connections much easier to form when symmetric laser heating between the two metal lines can be achieved. These findings can be directly between adjacent lines on the same level of interconnect have applied to improving the design of the laser linking processes been developed, which showed extremely high yields [7]. This technique has potential for practical implementation of and devices. “make-links”. In this paper, the experimental aspects of forming lateral I. INTRODUCTION connections between adjacent lines of aluminum-based metalASER line cutting has been used to build in redundancy lization are briefly reviewed. Then a numerical analysis of the for large area repetitive circuitry. Until now, only laser linking process is presented. The analytical technique we used deletive “break-links’’ have been found to be commercially is the finite element method. The simulation was carried out viable. However, there are problems associated with laser within the context of plane strain formulation, for appropriate metal ablation, such as metallic debris left on the surface of thermal loading conditions. Particular attention is devoted the chips, unpassivated circuitry, and extra processing steps to providing guidance for improving the process design. It necessary for wafer preparation and cleaning. should be emphasized here that the present analyses have been The laser break-link process is, nevertheless, extremely predicated upon continuum level modeling, with the role of reliable and has been applied to fully programmable cir- microstructural changes in influencing the onset of cracking cuits. A commercial cut-only laser programmable gate ar- from A1 ignored in the analyses. While such simple continuum ray (LPGA) technology has been in operation by Chip Ex- level formulations appear to capture the salient features of link press, Santa Clara, CA [l]. However, as metal lines become formation and crack evolution, a more complete solution to the narrower with deep sub-micron design rules, the break-link problem must inevitably address the microscopic phenomena technology has essentially reached its maximum density ca- not considered in the present work. This paper is arranged in the following sequence. Section I1 describes the experimental observations, following the work reported in [7]. The details Manuscript received April 6, 1995; revised October 9, 1995. The review of this paper was arranged by Editor A. H. Marshak. This work was supported of the finite element formulation, together with the underlying by a subcontract from the Lincoln Laboratory to MIT and by the National assumptions of the simulation, are described in Section 111. Security Agency. Comparisons of numerical predictions with experiments are Y.-L. Shen and S. Suresh are with the Department of Materials Science and Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139 the subject of Section IV. Section V presents parametric USA. analyses for optimizing the linking process through geometric J. B. Bernstein is with the Department of Materials and Nuclear Engineerconsiderations. This paper concludes in Section VI where the ing, University of Maryland, College Park, MD 20742 USA. key findings are summarized. Publisher Item Identifier S OOIX-93X3(96)01725-X. 0018-9383/96$05.00 0 1996 IEEE
SHEN
et
403
al.: LASER LINKING OF METAL INTERCONNECTS
Via.
Metal 1,
Fig. 1. Schematic of a continuous chain of metal links to allow 4-point measurements of one link in series with a short segment of metal 1 and two vias to metal 2.
11. EXPERIMENTAL OBSERVATIONS
Fig. 1 shows the test structure manufactured using a standard two-level metal CMOS process [7]. An array of parallel metal 1 (tungsten) lines was connected by a common metal 2 (aluminum) line through vias, as shown in the top areas of Fig. 1. A short segment of 2 pm-wide metal 2 was also connected to each metal 1 line. These segments could be linked to a continuous 2 pm-wide metal 2 track, which was not directly connected to metal 1 lines initially, by way of the laser heating operation. The link region consisted of a 1 pm-wide gap. A diode pumped Q-switched Nd:YLF solid state laser system was used. The laser was focused to a l / e 2 diameter of approximately 4 pm and could be expanded to 8 pm, with a pulse length of approximately 8 ns. The positioning accuracy was better than 0.1 pm across the test chip. The only variable parameters were focus and diode pump power. The horizontal metal 2 tracks in Fig. 1 were bonded out to pads on both sides so that four-point resistance measurements could be made after pulsing the laser at each link site. If the link was successfully formed, the laser then cuts away the link. With a laser beam diameter of approximately 8 pm and pump power of 520 mW, it has been found that over 30 000 successful connections have been achieved without failure. After testing, the links demonstrated an average current handling capacity of 78 mA (individual links) and showed a low average resistance (, stresses in these areas become lower. This is because the constraint imposed by the metals on the Si02 region between them is higher
407
SHEN et al.: LASER LINKING OF METAL INTERCONNECTS
Fig. 9. Contours of the constant maximum principal stress for the case of a 1.5-pm-wide AI. See text for details.
Si3N4
1: -I L.l4 0.5pm
Si02
sio2
1.0pm
I
I
@)
Fig. 8. (a) Contours of the constant maximum principal stress, and (h) the predicted crack path, for the case of a 1.0-pm wide A1 for a temperature increase of 20 to 3OOOC in Al.
when the two A1 domains are closer together. Thus a higher tendency of cracking can be expected in the case of Fig. 11. This, along with the fact that the molten A1 has to travel a longer distance to fill the crack if the metals are farther apart, suggests that the link is easier to form when the interspacing is decreased. C. Effects of Asymmetric Laser Heating
In previous calculations, a symmetric heating condition was assumed, i.e., equal amounts of laser energy were absorbed by each of the two A1 lines. During the actual processing this may not occur due to the inaccuracy in laser positioning. In the following, we investigate the effects of uneven heating of the metals on the cracking feature. Because of the asymmetry, the full structure, instead of only a half of it, has to be used for the calculation. Here the geometry for the case of Fig. 6 is considered. The result for a symmetric heating condition is
Fig. 10. Contours of the constant maximum principal stress for the case of a 1.0 pm-wide AI. See text for details.
shown in Fig. 13(a). The indicated percentages correspond to the portions of laser energy absorbed by each metal segment. This figure shows exactly the same situation as in Fig. 6(b). A crack connecting the neighboring upper corners of the two metals can form. A link can thus be developed by metal infiltration from both sides. Fig. 13(b) shows the case when 60% of the laser energy goes to the right A1 domain while 40% goes to the left one. The crack originates in the upperleft corner of the right Al. It curves downward, following an asymmetric path, and terminates at the top surface of the left Al. This means that the chance of forming a link is reduced, because a relatively sharp bending of the crack near the left A1 is required. Fig. 13(c) shows the case when the extent of asymmetric heating is further increased so that 75% and 25% of the laser energy are absorbed by the right and left metals, respectively. The predicted crack can be seen to initiate from the right A1 and terminate in the middle of Si02 without connecting to the left Al. This appears to be a consequence of the fact that the temperature increase, and thus the thermal
~
[EEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 43, NO. 3, MARCH 1996
408
I Si3N4
I
Si02
Si02 1.0 pm
(b)
Fig. 11. (a) Contours of the constant maximum principal stress, and (b) the predicted crack path, for the case of a metal interspacing of 0.8 p m for a temperature increase of 20 to 3OOOC in Al.
expansion, in the left A1 are too small to generate enough tensile stresses in the dielectric around the metal. In fact, the “streamline” (crack) ends because it encounters the area where the maximum principal stresses are compressive. Hence we can deduce that it will be very difficult to form a link between the two metals under this circumstance. In the extreme case that all the laser energy goes to one metal, as in Fig. 13(d), the predicted crack is seen being oriented at approximately the 45’ direction and extending into the nitride layer. Therefore, it is certain that no link can possibly form in this case. From the above analysis, we can conclude that deviating from the symmetric heating condition significantly reduces the chance of successfully forming a link. Hence, increasing the laser positioning accuracy is essential for improving the linking yield. This can also be achieved by expanding the laser beam diameter, which leads to a more uniform heating of the two A1 lines. This finding has been applied to actual laser linking tests, and significantly improved results were obtained [7].
Fig. 12. (a) Contours of the constant maximum principal stress and (b) the predicted crack path, for the case of a metal interspacing of 2.0 pm for a temperature increase of 20 to 3OOOC in AI
VI. CONCLUDING REMARKS In this paper we report successful metal-to-metal links from a standard two-level metallization process. The links are between adjacent lines on the same level of aluminum metallization. The link mechanism is understood to be a thermo-mechanical fracture of the dielectric and passivation allowing molten aluminum to create a connection. A mechanistic justification of the laser linking process is presented with the aid of detailed finite element analyses. To avoid the passivation damage, it is proposed that the thickness of the oxide layer between the aluminum and the passivation should be significantly increased. Reducing the interspacing of the metal lines increases the linking propensity. Under appropriate conditions, the chance of forming the link can also be increased by reducing the width of the metal lines. It is also found that increasing the extent of symmetric heating by increasing the laser positioning accuracy andlor expanding the laser beam diameter can significantly improve the rate of successfully
409
SHEN et al.: LASER LINKING OF METAL INTERCONNECTS
REFERENCES
p-l--TI 50%
50%
Si02
SiSN4
60%
40%
Si02 (b)
Si3N4
I
25%
I
r-Xi-1
Si02 (C)
Si3N4
(All
0%
[ l ] D. Bursky, “Laser programming turns 18-K gate arrays around fast,” Electron. Design, vol. 40, pp. 30-31, May 28, 1992. [2] H. D. Hartmann and Th. Hillmann-Ruge, “Yield and reliability of laser formed vertical links,” in Proc. of Multilevel Interconnections: Issues that Impact Competitiveness, 1993, SPIE, vol. 2090, pp. 146160. [3] M. Rouillon-Martin, M. Chambon and A. Boudou, “Laser programmable vias for reconfiguration of integrated circuits,” in Proc. of Optical Microlithography and Metrology for Microcircuit Fabrication, 1989, SPIE, vol. 1138, pp. 190-197. [4] J. Raffel, A. H. Anderson and G . H. Chapman, “Laser restrncturable technology and design,” Wafer Scale Integration. Norwell, MA: Kluwer Academic, 1989, Chap. 7, p. 363. [5] S. S. Cohen, P. W. Wyatt, G. H. Chapman and J. M. Canter, “Laserinduced diode linking for wafer-scale integration,” ZEEE Trans. Electron Devices, vol. 35, pp. 1533-1550, 1988. [6] J. B. Bernstein, T. M. Ventura and A. T. Radomski, “High density laser linking of metal interconnect,” IEEE Trans. Comp., Packug., Munufuct. Techno1.-Part A, vol. 17, pp. 590-593, 1994. [7] J. B..Bemstein and B. D. Colella, “Laser formed metallic connections employing a lateral link strncture,” IEEE Trans. Comp., Packag., Manufact. TechnoL-Part A, vol. 18, pp. 690-692, 1995. [8] Y.-L. Shen, A. Needleman, and S. Suresh, “Coefficient of thermal expansion of metal-matrix composites for electronic packaging,” Metall. Muter. Trans. A , vol. 25A, pp. 839-850, 1994. [9] Y.-L. Shen and S. Suresh, “Thermal cycling and stress relaxation response of Si-AI and Si-AI-Si02 layered thin films,” Acta Metall. Mater., vol. 43, pp. 3915-3926, 1995. [lo] A. Bartlett, A. G. Evans and M. Ruhle, “Residual stress cracking of metal/ceramic bonds,” Acta Metall. Mater., vol. 39, pp. 1579-1585, 1991. [ l l ] J. H. Selverian and S. Kang, “Ceramic-to-metal joints: Part 11-Performance and strength prediction,” Amer. Cerum. Soc. Bull, vol. 71, pp. 1511-1520, 1992. [12] M. Finot, Y.-L. Shen, A. Needleman and S. Suresh, “Micromechanical modeling of reinforcement fracture in particle-reinforced metalmatrix composites,” Metall. Mater. Trans. A , vol. 25A, pp. 2403-2420, 1994. [13] S. Y. Kweon and S. K. Choi, “Prediction of residual stress-induced cracking by finite element analysis,” Scripta Metall. Mater., vol. 32, pp. 359-364, 1995. [14] Y. Sumi, S. Nemat-Nasser, and L. M. Keer, “On crack path stability in a finite body,” Engng. Fract. Mech., vol. 22, pp. 759-771, 1985. [ 151 A. R. Ingraffea, “Interactive computer simulation of fracture processes,” in Proc. Fourth Znt. Con$ Numerical Methods in Fracture Mechanics, Pineridge Press, Swansea, U.K., 1987, pp. 677499. [16] D. V. Swenson and A. R. Ingraffea, “Modeling mixed-mode dynamic crack propagation using finite elements: Theory and applications,” Computat. Mech., vol. 3, pp. 381-397, 1988. [17] X.-P. Xu and A. Needleman, “Numerical simulations of fast crack growth in brittle solids,” J. Mech. Phys. Solids, vol. 42, pp. 1397-1434, 1994.
100%
Si02 Fig. 13. (a)-(d) The predicted crack paths under asymmetric heating conditions. The indicated percentages correspond to the portions of laser energy absorbed by the metal segments.
forming the link. These findings can be directly applied to the design of the laser linking processes and devices. ACKNOWLEDGMENT
The authors wish to gratefully acknowledge the assistance of John Rimshaw and Aaron T. Radomski of the Department of Defense.
Yu-Lin Shen received the B.S. and M.S. degrees in materials science and engineering from National Tsing Hua University, Taiwan, in 1986 and 1988, respectively, the M.S. degree in applied mathematics from Brown University, Providence, RI, in 1993, and the Ph.D. degree in engineering (materials science) from Brown University in 1994 He is currently a Post-doctoral Research Associate, Department of Materials Science and Engineering, Massachusetts Institute of Technology, Cambridge. He has an extensive research experience in mechanical behavior of materials and numerical simulation. His current interests include modeling of thermal processing in microelectronic devices, thermo-mechanical properties of multl-layered materials and thin films, and micromechanical simulation of deformation in advanced composite materials.
410
Subra Suresh received the Sc.D. degree from the Massachusetts Institute of Technology, Cambridge, in 1981. He is the R. P. Simmons Professor, Department of Materials Science and Engineering, and Professor of Mechanical Engineering, MIT. He is also the CoDirector of the MIT-Harvard Program on Modeling of Materials. Prior to joining MIT in 1993, he was a Professor of Engineering at Brown University, Providence, RI. His current research interests focus on quantitative investigations of the microscopic and macroscopic aspects of mechanical behavior of metals, ceramics, thin films and composites. Dr. Suresh is currently a member of the Executive Committee, Materials Division, ASME, a Principal Editor of Acta Metullurgica et Maferialia and Scripta Metallurgica at Materialia, an Associate Editor of Materials Science and Engineering A, and a Series Editor for the Cambridge University Press Solid State Science Series.
[EEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 43, NO 3, MARCH 1996
Joseph B. Bernstein received the M S , E E , and P h D degrees in electrical engineering from the Massachusetts Institute of Technology, Cambridge, in 1986, 1987, and 1990, respectively He received the B S degree in electrical engineering from Union College, Schenectady, NY He is currently an assistant Professor in the reliability engineering program at the University of Maryland, College Park He is building a laboratory for laser processing of microelectronic devices in the Department of Materials and Nuclear Engineering, sponsored by the Department of Defense His research areas include studies of thermal, mechanical, and electrical interactions and failure mechanisms of dielectnc and metallic matenals used in microelectronics and laser processing of completed circuitry for defect avoidance and repair
