Redundancy by laser cutting of polysilicon fuses has been used by the memory industry for many years. As the levels of metalization layers increases, it becomes more difficult and expensive to delete buried polysilicon lines. Ideally, metal fuses will be cut exclusively. However, to achieve reliable metal line cutting, a wide process window has to be found that can cut metal lines buried beneath the passivation layers. The upper energy limit has previously been thought to result from excess laser energy absorbed by the substrate. We show that another failure mode exists at energies far below the threshold to cause substrate damage directly. The same laser pulse which ejects the passivation and removes the metal is also likely to crack the dielectric material below the metal. Molten metal then fills the crack and maintains an electrical short circuit, preventing the line from being disconnected.

Semiconductor memories have been designed with redundancy since early in 1964. Although, they were not commercially used until 1979 when the Bell group reported the first fault tolerant 64K DRAM[1]. In that process, a laser was employed to remove defective elements and replace them with redundant ones. Since then, research on laser memory repair has attracted extensive interest. Polysilicon has deep light absorption in the 1um wavelength range, so the vertical temperature distribution is relatively uniform. This eases the clean removal of the cut material by laser explosion[2]. Some intrinsic deficiencies of polysilicon, such as high resistance and complex process, ultimately limit its deep sub-micron application.

A new make-link technology [3][4][5][6] is competitive and becoming a strong candidate to replace non-metal cut technologies for several practical advantages; standard CMOS compatible process, very low electrical resistance and small link area. To this day, however, polysilicon fuses overwhelmingly dominate laser repair. Metal cuts would be preferred by industry if not for concerns about yield and reliability.

Laser processing involves complex thermal and mechanical coupled mechanisms and it is essential to understand the kinetics in order to improve the reliability of laser processing. Here, we present an analysis of the laser cutting of metal lines. A series of laser energies with fine steps were applied to an array of metal lines to investigate the energy dependence of the process. Focussed Ion Beam (FIB) analysis was used to determine the effects of laser pulse energy and to show the existence of a newly discovered failure mode.

Insights gained through development of the laser make-link have lead to postulation of a potential failure by cracks emanating from the lower corners of a metal cut. Although the predominant crack is in the upper corners, there is a likelihood that the lower corners of the metal will also fracture due to the laser heating. A lower corner crack would fill with metal, as occurs with the make-link process[3][4] and prevent the disconnection from being realized.

Sample wafers were fabricated by a standard commercial two-level metal CMOS process. Only upper-level metallization was used in this study, which was deposited by sputtered Al (1%Si-0.5%Cu) and etched to form 2 um wide and 0.9um thick lines. The Al lines were under coated by a 0.09 um layer of TiW and over coated by 0.05 um of Ti. The experiment was conducted on an XRL 525 laser process system which employs a Spectra physics diode-pumped Q- switched Nd:YLF laser (1047 nano-meter) operated in the saturated single pulse mode. The laser pulses of 15 ns were directed through focusing optics and brought to impinge upon the chip surface. Position accuracy of the laser system is approximately 0.5 um.

Fig.1 Schematic of the FIB cross-sections taken across the metal lines that have been hit with an elliptical laser spot.

Two groups of energy levels were used to investigate the precise dependence of the laser cutting process. In the first group, energy was varied from 0.1 uJ to 0.69 uJ, in steps of 0.01uJ and the spot was focused to 2.5 um by 5.0 um ellipse. In the second group, higher laser energy levels were chosen, ranging from 0.2 uJ to 1.675 uJ in steps of 0.025 uJ. Dual-beam (FIB and SEM) observations were conducted to study the laser process from surface and cross sectional images. The FIB cross-sections were made perpendicular to the line in the center of the laser spot as shown schematically in Fig. 1, and viewed at 45 degree from normal along the line.

At energies below 0.14 uJ, no detectable effect of the laser was observed. In the range of 0.14-0.15uJ, cracks were observed to have penetrated the dielectric layers and extend to the free surface. At no point were cracks formed without metal flow. Molten metal was ejected though the cracks and re-solidified. Fig. 2 is a cross sectional image resulting from a laser energy of 0.33uJ. From the image, one can see that the crack trajectories extended from the upper metal corners to the passivation step corners, which was the route of maximum stress concentration. A void remains in the line due to the reduced volume of remaining material.

Fig. 2 FIB cross-section of a 2um wide metal line after 0.33uJ of laser energy showing ejected metal at the surface and a void remaining in the line.	Fig. 3FIB cross-section of a 2um wide metal line after 0.875uJ of laser energy representing a threshold of clean cutting.	Fig. 4FIB cross-section of a line after 1.6uJ of laser energy where the metal-filled lower corner cracks are evident

When the laser energy reached 0.475uJ, the passivation was blown away. Energy higher than this value is necessary to make a metal cut possible. Since the TiW underlayer is refractory and does not melt, the lower limit of the laser process window is determined by the energy to remove TiW as well as aluminum. Fig. 3 is the FIB image of 0.875 uJ which is found to be the cut energy threshold. Ideally, the optimal laser energy should be higher to assure that all the cuts are made successfully. However, an increase in the laser energy to 1.6uJ resulted in cracks forming at the lower corners of the line as seen in cross-section in Fig. 4. This lower crack is filled with aluminum and represents a short circuit across the cut line, and therefore, a failed cut.

Due to the differences of thermal conductivity between aluminum and surrounding materials (silicon dioxide and silicon nitride), the aluminum metalization experiences a substantial temperature increase when the laser impinges upon the surface. Silicon oxide and nitride can be treated approximately as remaining at room temperature throughout the whole process which lasts for several nano-seconds[3]. Because of the temperature difference and aluminum's high thermal expansion coefficient, there are tensile stress concentrations inside dielectric in the vicinity of the corners. When the laser heats up the metal, cracks are initiated from the corners and molten aluminum fills the cracks.

We have learned from finite element analysis[6][7] that the stress in the upper corners is higher than in the lower corners. Hence, the crack is favored to initiate from the upper corners and terminate at the free surface. This is consistent with our observations in a vertical linking processes[4]. The low end of a vertical linking processing window would be determined by this threshold to crack, and the high end would be determined by the ejection of passivation. In our experiment, the energy range for linking should be from 0.15uJ to 0.45uJ; a much wider process window than for cutting.

Previous work[7] assumed damage to the substrate is due to the laser absorption in the silicon. They have claimed that since light absorption in the silicon decreases when the laser wavelength increases from 1.0 um to 1.3 um, then the upper limit of the energy to cut should be widened by a factor of three. Such an improvement has not been realized. The major problem associated with increasing the laser wavelength is an inherent loss of positioning accuracy and an automatic expansion of the smallest possible laser spot size. Whereas the absorption of laser energy by the substrate may be a concern for some processes, our experiment proves that damage to the substrate may result as a consequence of the laser absorption in the metal alone.

Stress concentrations are known to exist in all four corners of an aluminum cross-section. The crack is favored to initiate from the upper corners where the stress concentration is highest, but with enough energy, cracking will also take place at the lower corners and molten aluminum will fill these cracks. Ultimately, as the energy is further increased, the cracks can propagate down to the silicon substrate and cause significant damage to the underlying circuitry. This newly determined failure mode is likely to be the major causes of cutting failure with a threshold much lower than what is required to damage the substrate directly. This energy limit will not be affected by laser wavelength. In our experiment, the cut energy threshold is 0.875uJ and the upper energy limit is 1.6uJ, beyond which this new failure mode becomes likely.

The energy processing window of laser induced cutting is discussed for the first time by consideration of a new failure mode. The high end of a cut processing window based on damage to the substrate is not determined by the laser wavelength, but rather by the fracture dynamics of the metal-dielectric system. We have shown that the upper energy limit is determined by the kinetics of the laser-metal interaction, independent of the wavelength. Vertical linking to metal positioned above the crack trajectory should result in a more robust process, while the energy window for cutting may be less favorable.

[1] R.T.Smith, J.D.Chlipala, ``Laser Programmable Redundancy and Yield Improvement in a 64K DRAM,'' IEEE J. of Solid-State Circuits, vol. SC-16, pp.~506--514, Oct.1981

[2] J.D.Chlipala, L.M.Scarfone and C.Y.Lu, ``Computer-simulated explosion of Poly-Silicide Links in Laser-Programmable Redundancy for VLSI Memory Repair,'' IEEE Trans. on Elect. Dev., Vol.36, pp.~2433--2439, Nov.1989

[3] J.B. Bernstein, B.D.Colella, ``Laser-Formed Metallic Connections Employing a Lateral Link Structure'', IEEE Trans. on Comp., Pack. and Manuf. Tech., Part A, Vol.18, pp.~690-692, Sep.1995

[3] R.L. Rasera and J.B. Bernstein, ``Laser Linking of Metal Interconnect: Linking Dynamics and Failure Analysis,'' IEEE Trans. on Comp., Pack. and Manuf. Tech., Part A, Vol.19, pp.~554--561, Dec. 1996

[4] J. B. Bernstein, T.M.Ventura, and A.T.Radomski, ``High Density Laser Linking of Metal Interconnect,'' IEEE Trans. on Comp., Pack., and Manuf. Tech., vol.14, pp.~590--593 Dec.1994

[5] J.B.Bernstein, T.M.Ventura, and A.T.Radomski, ``High Density Laser Linking of Metal Interconnect,'' IEEE Trans. on Comp., Pack., and Manuf. Tech., vol.14, pp.~590--593 Dec.1994

[6] Y.L. Shen, S. Suresh, and J.B. Bernstein, ``Laser Linking of Metal Interconnect: Analysis and Design Considerations,'' IEEE Trans. on Elect. Dev., Vol.43, pp.~402--410 Mar. 1996

[7] Y.Sun, R.Harris, E.Swenson, C.Hutchens, ``Optimization of Memory Redundancy Laser Link Processing,'' SPIE vol.2636, pp.~152-164, 1995