Analyzing the process window for laser copper-link processing
12/01/2002
overview
Control of laser pulse shape and polarization improves the reliability of laser processing of copper links used for circuit redundancy. Using finite element modeling, simulation shows that fast-rise-time laser pulses give a unique pulse shape that helps avoid lower corner cracking by initiating upper corner cracking and stress relief. Polarization variation enables control of hole size by changing energy coupling into the link, thereby increasing the process window significantly.
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. Deep absorption of laser energy in the 1µm wavelength range provides relatively uniform temperature distribution that promotes clean removal of link material. The high resistance and complex processing of polysilicon, however, limits its use for deep submicron application [2].
Successful laser cutting of an aluminum link involves addressing material left over at the bottom of cut sites and lower corner cracks, which can be major reliability issues. These two failure modes have been defined as the low and high end of the laser process window in previous work [3]. Failures can also occur from damage to neighboring fuses or the substrate due to excessive spot size and high laser energy.
Figure 1 shows a long, undesirable lower-left corner crack that poses a reliability concern since it may form a short-circuit or damage to surrounding structures. Asymmetric cracking is caused by laser-spot positioning error. Damages to neighboring fuse structures or the substrate due to excessive spot size and energy is also another failure mode at high laser energy.
Today trends clearly favor using copper along with low-k dielectrics — particularly for high performance logic devices and high speed SRAM — as link material due to its enormous benefits when compared to aluminum, such as its low resistance, power dissipation, manufacturing cost, and superior resistance to electromigration. However, there are some difficulties with laser processing of copper fuses because of different material properties and the fabrication of copper metallization, such as lower coefficient of thermal expansion, higher melting point, and copper's thick structure.
Accordingly, we set out to analyze copper-fuse laser processing with the objective of improving process results when using a unique fast-rise-and-fall-time laser pulse shape and various polarization modes.
null
Important copper considerations
Compared to aluminum, there are a few important considerations regarding the mechanical phenomena of copper under laser heat. Copper is highly reflective in 1µm wavelength range and, therefore, higher laser energy is required to blow a link compared to aluminum. In addition, copper has a low coefficient of thermal expansion (CTE) and requires a higher temperature to initiate cracking at upper corners.
We must also consider that the commonly used damascene copper metallization process includes silicon nitride etch-stop and hard-mask layers: each copper metallization layer is sandwiched between silicon nitride layers. At certain wavelengths, these silicon nitride layers reduce damage to the substrate even for high power laser processing. This is due in part to the high reflectance of the laser beam by the multiple silicon nitride layers.
Finally, the fuse-pitch for copper links is relatively large and neighbor structure damage is not likely to be the limiting factor. Thus, the high end of the energy process window is limited by lower corner cracking and the low end is defined by clean material removal.
 Figure 2. a) Gaussian and b) fast-rise-and-fall laser-pulse power profiles used for finite modeling. |
null
Finite element simulation
In some cases, the high end of a laser-cut processing window based on damage to the substrate is not determined by laser wavelength, but by the fracture dynamics of the metal-dielectric system, even in aluminum based processing [2]. Moreover, with copper-link processing, substrate damage is not likely to be the limit for the high end of the process window due to the multiple highly reflective silicon nitride layers. Therefore, the high end for the process window is more likely to be lower corner cracking in copper link processing.
The dynamics of crack formation, including lower corner cracking and stress relief effects are detailed in previous work [3]. With the understanding of this stress relief effect, we used the MARC Mentat finite element model to simulate two unique pulse shapes with fast rise and fall time to see if it would help avoid lower corner cracking.
Since most of our concerns centered at the upper and lower corners of copper metal structures, we used a triangle mesh for our simulations. This provided denser elements at the corners versus other regions of the model. In addition, we only simulated half of the structure to save computation time.
Figure 2 illustrates the power profiles of a Gaussian laser pulse used in our first simulation and a fast-rise-and-fall-time laser pulse used in our second simulation. The pulse width is 7 nsec in FWHM for both pulses. With a Gaussian pulse, the peak is on about 1/3 of the whole pulse time. With a fast-rise-and-fall pulse, the rise and fall times are ~1.5 and 2 nsec.
 Figure 3. Stress profiles under laser heating with a) a Gaussian laser pulse and b) a fast-rise-and-fall-time laser pulse. |
null
A fast-rise-time laser pulse will heat the upper part of a link rapidly and promote fast top oxide cracking that relieves lower corner stress faster than a regular Gaussian pulse. In addition, with this pulse the reflectivity of metal decreases due to fast heating of the line; this promotes efficient absorption of laser energy. Overall, with a fast-rise pulse, metal links are cut more efficiently at lower nominal energies and are less likely to have lower corner cracking beneath the link.
Figure 3 shows stress profile results from finite element analyses. Compared to the Gaussian pulse result, the fast-rise-time laser pulse shows faster heating of the upper part of a link within a shorter time. This fast heating develops upper corner stress and cracking faster as can be seen in the steeper angle of upper corner stress curve (see Fig. 3b). This faster cracking and subsequent stress release delays lower corner cracking ~1 nsec. So, the time interval between upper and lower corner cracking for fast-rise-time laser pulses is longer than that for Gaussian pulses.
We note that our simulations did not include the flow of molten metal after melting and material removal with passivation explosion. We believe that the delay in lower corner cracking should be even larger when we account for the stress relief effect from material removal.
To validate our simulation result, we processed actual wafers with both Gaussian and fast-rise-and-fall-time laser pulses and found the results to be consistent with simulation results.
In copper processing, lower corner cracking was found to be a critical failure mode for low-k material underneath, hence the use of a fast-rise-and-fall-time laser pulse is important. Clearly, the timeline for low-k implementation is accelerating. Most low-k dielectrics are porous and fragile and exhibit low cracking resistance and yield strength. Therefore, having a low-k material under a link significantly increases the chance for lower corner cracks. Preventing them is a critical requirement.
null
Laser beam polarization
Because copper processing requires high laser energy to perform a reliable cut, typically big holes are left at a cut site. When tensile stress reaches a critical level for the dielectric used, cracks initiate and propagate within the dielectric perpendicular to the local maximum principle tensile stress. However, a dielectric layer has a weak point around the corners of a fuse due to chemical mechanical polishing and the interface between SiN and SiO2. This results in the critical stress for cracking being lower than for other types of metal fuses. For this reason, cracks tend to follow a path along the weak interface of SiN and SiO2 and upper cracks propagate laterally. These laterally propagating upper cracks contribute to the large cut. In addition, thermal diffusion from the link into dielectric layers on top of the link from the relatively high laser energy might account for severe delamination of SiN and SiO2 layers, also contributing to the large cut site.
In the end, hole size is also a limiting factor for the high end of an energy process window, along with a lower corner cracking. For this reason, copper fuses must have at least a few times larger pitch than aluminum fuses.
To control and minimize hole size after laser processing, we investigated the effect of polarization. We applied laser pulses with three different polarizations — along the link length, circular, and across the link length; we applied these to adjacent similar links and examined the resulting process windows. In Fig. 4a, the vertical lines indicate absolute laser energy ranges where links were processed successfully. The energy range was determined by optical microscopy (i.e., the lower limit was marked when we observed link material remaining and the upper limit when we observed damage larger than the fuse pitch caused by a large hole size). We found we could greatly improve our process window by choosing the correct polarization.
In Fig. 4a, the dotted line indicates the changes of relative process windows depending on polarization mode. Relative process window is the ratio of the difference between the high and low ends of the process window (Eh - El) to the average energy (Ea = (Eh + El)/2 ) or (Eh - El)/Ea. This normalized, nondimensional term considers the performance of laser systems clearly and eliminates the dependence of the absolute energy window on the characteristics of different laser systems [4].
The data in Fig. 4a demonstrate that the energy process window is strongly dependent on polarization. Specifically, it indicates that across-link polarization is optimum for this particular structure. The relative process window with a variation of polarization changes significantly depending on the polarization mode. We believe that changing polarization results in a unique heat distribution in the link, and we are studying this polarization coupling effect on link structure.
We also plotted the trends of relative process windows with a variation of spot size and the impact of changing spot size with polarization (Fig. 4b). These data show the same polarization trend and that the minimum process window (polarization along the link) of a 2.3µm-dia. spot is smaller than that of 2.8µm spot. With across-the-link polarization, however, the maximum relative process windows for two spots have almost the same value. This indicates that polarization is a critical factor when determining a laser energy window for copper processing, especially when a smaller spot size is required.
Conclusion
We have simulated fast-rise-and-fall-time laser pulses to study the ability to cut copper fuses. Our results show that lower corner cracking can be delayed ~1 nsec due to faster upper cracking and subsequent stress release. We have also found that the polarization of the laser beam used has a critical effect on the cut hole size and material removal; changing energy coupling into the link increases the process window significantly.
References
- 1. R.T. Smith, J.D. Chlipala, IEEE J. Solid State Circ., SC-16, p. 506, Oct. 1981.
- 2. J.B. Bernstein, et al., IEEE Electron Device Lett., Vol. 19, No. 1, p. 4, 1998.
- 3. J.B. Bernstein, et al., IEEE Semiconduct. Mfg., Vol. 13, No. 2, p. 228, 2000.
- 4. J. Lee, et al., IEEE Semiconduct. Mfg., submitted for publication.
Joohan Lee is a senior application engineer at GSI Lumonics, 60 Fordham Rd., Wilmington, MA 01887; ph 978/661-4567, fax 978/988-9353, e-mail [email protected].
Jonathan Ehrmann is a senior principal engineer at GSI Lumonics.
Donald Smart is a senior principal scientist at GSI Lumonics.
Joseph Griffiths is technical product market manager at GSI Lumonics.
Joseph Bernstein is an associate professor at the University of Maryland, College Park.
Joohan Lee, Jonathan S. Ehrmann, Donald V. Smart, Joseph J. Griffiths, GSI Lumonics, Wilmington, MassachusettsJoseph B. Bernstein, University of Maryland, College Park, Maryland