Laser Formed Metallic Connections

Solid metallic connections have been successfully formed between two standard levels of metalization using a focussed IR laser. This new process of laser formed connections has been used to link continuous chains and with resistances of less than 0.8ohm per connection. A commercial laser repair system used extensively by the memory industry was employed to perform approximately 50,000 individual links without failure. The electromigration resistance is comparable to standard metal interconnect. This technology has the potential to replace laser fuse cutting techniques to implement repair schemes and it can be used to program wiring in MCM-D or wafer scale integration applications implemented on silicon substrates. Furthermore, because it is an additive process, it lends itself to redundancy for higher yield and reliability.

Lasers have been used to form electrical connections in electronic circuits for a number of years. Previous make-link technologies are explained in an article by S.S Cohen and G.H. Chapman[1]. The most reliable type of link that has been formed works on the principle that two adjacent lines of metal can be exposed to a laser pulse and thermal expansion of the metal fractures the dielectric separating them. At the same time, rapidly expanding metal flows through the crack to connect the lines[2-4]. The process has been shown to work between adjacent lines on the same level of metalization, although, previous attempts to link between different levels of metal had not produced sufficient yield to be practical[5]. We present here the vertical link technology and show that it is robust, easy to design and implementable on most multi-level metal processes. Potential applications of a laser make-link are at least as extensive as the laser break-link. The ability to both make and break links allows for a much broader range of repair algorithms. If repair is the goal, many more combinations of replacing non-functioning circuitry with functional circuitry can be achieved. Furthermore, a design which incorporates additive as well removable circuit elements can be made very large, which is the goal of wafer scale integration (WSI) and is well described in[6]. This same technology could be applied to programmable interconnect for MCM-D or other similar applications which do not use organic dielectric materials.

Many ideas have been proposed in the past to make connections between two otherwise insulated interconnect lines. The first method for making a link was disclosed in a patent from Hitachi in June, 1984[7]. They proposed creating two back-to-back bipolar junctions on a polysilicon (poly) island. The poly could then be heated with a laser, causing the thermally isolated block of material to heat up. The n+ diffusions would then extend across the bar, resulting in an n-doped polysilicon link. The resistance of this link will be higher than an in-situ doped poly line, but, the process requires a modified CMOS process. Another problem with this process is the narrow energy range over which efficient heating and dopant diffusion can occur without damaging or blowing out the link. This `blow-out' phenomenon was recognized as a problem with the polysilicon make-link, and in the same patent, the authors suggested that the diffusions could be placed in the silicon itself. This has become known as a diffused link and it is compatible with standard silicon processing, but it requires that the circuit be restructured on a silicon devices. It also requires a several microsecond long laser pulse that is highly absorbed in the silicon surface, such as a CW argon-ion laser[1]. Other means of making connections included the use of alternate materials, such as amorphous silicon between two levels of metalization, which is not compatible with standard process methods. Large pads of metal have been connected with an excimer laser, but the yield and reliability has never justified the expense of processing[8]. In the end, laser formed connections have not been found in commercial applications, until the advent of the laser induced cracked-passivation make link, which is described in the following section.

The formation of a metallic connection is based on a set of coincidental properties relating to the materials used in microelectronics manufacture. Metalization in silicon circuitry almost universally consists of aluminum based alloys. More advanced metalization schemes use copper, but the fundamental properties required to form the links are the same, provided the dielectric is amorphous, inorganic and brittle. The following properties of metal are salient to the creation of metallic links:

1. Highly thermally conductive and thermally expansive.
2. Plastic and easily conform to the shape of their surroundings.
3. Low melting temperatures relative to inorganic dielectrics (such as SiO2 or Si3N4, this is not necessarily true for organic dielectrics such as PI).

These properties contrast to the very low thermal expansion coefficient and thermal conductivity silicon based dielectrics. The metal, which is encased by dielectric, can absorb focussed laser energy that is transmitted through the oxide and nitride and expand within the nanosecond time scale of a Q-switched laser pulse. However, the brittle dielectric will not be able to support the expansion without fracturing. Once a fracture is initiated, the molten metal is able to expand and flow through the resulting crack. One characteristic of dry-etched metalization manufacturing is a sharp corner along the side-wall of the metal line. These corners represent mechanical stress concentration points for thermally expanding metal, heated by the laser, contained within the dielectric. The amorphous nature of the dielectric removes any consideration of grain boundaries associated with a crack trajectory. Thus, a crack will form generally between points of greatest stress concentration and follow an energetically favorable path.

Fig. 1 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.

The phenomenon of dielectric fracture and metal flow is seen in Fig. 1. This is a cross-sectional image, taken with a focussed ion-beam (FIB), of a 2um wide metal line that was exposed to a 15ns Nd:YLF laser pulse. The energy was 0.45uJ with a spot size of 3um 1/e2 diameter. The metal was passivated by a 1.2um thick layer of plasma deposited Si3N4 which cracked from the upper corners of the metal line to the upper surface. Metal is seen to have flown through the cracks and re-solidified at the surface. This phenomenon is also responsible for failures to cut metal lines, as described in[9], when cracks form in the lower corners of the lines which then fill with metal.

Fig. 2 Schematic of the laser spot and cross-sectional image of a segment from two link points in the the vertical link chain, showing the laser spot and cross-section plane.

The most straight forward type of direct metallic connection that can be made to take advantage of the dielectric fracture and metal fill phenomenon employs a vertical link configuration. An upper layer of metal crosses over the lower level with an opening for the laser to heat up the lower line. This is the first metallic link that was proposed[5]. The structure used in this study is shown schematically in Fig. 2. The upper level of metal must overlap the region where the crack would terminate. The laser spot must also be small enough to impinge mostly inside the annular opening so the crack will be sure to terminate on metal 1.

Samples were fabricated by a standard commercial two-level metal CMOS process. First level metal (metal 1) was deposited by sputtered Al (1%Si--0.5%Cu) to a thickness of 550nm. The Al lines were under coated by a 50nm layer and over coated by 25nm layer of TiN. The second level of metal (metal 2) had the same constituents except the aluminum alloy layer was 750nm thick. The inter-level dielectric was TEOS based SiO2 of approximately 0.8um thickness and metal 2 was passivated with 0.9um of plasma deposited Si3N4. The experiments were performed on an XRL 525 laser process system which employs a Spectra physics diode-pumped Q- switched Nd:YLF laser (1047nm wavelength) operated in the saturated single pulse mode with a 0.5W diode pump. The laser pulses of 15ns duration were directed through focusing optics and brought to impinge upon the chip surface. Position accuracy of the laser system is approximately 0.8um along the metal 1 axis (Y) and 0.3um along the orthogonal axis (X). The spot size was adjusted from 2.5um to 4um full width half maximum (FWHM) diameter of the circular Gaussian beam. The spot was focussed to the center of the 6x6um square as shown in Fig. 2. For reference, the 1/e2 diameter is found by multiplying the FWHM value by 1.7. The link structure tested in this study consisted of 4.0um wide metal1 lines centered beneath the hole in metal 2. Links were connected in series chains of 1320 connections. Every link had to be made for there to be continuity. There were taps off the series chain after the first link and at links numbered 24 and 216 in order to converge quickly on optimal laser parameters for the most robust connections. Several die were processed at a range of spot sizes and energies before being probed for continuity.

The data consisted of two parts; link yield and resistance per long chain. If full chains were not connected, the resistance of the chain was not measured. However, if chains were consistently linked across many chains, the laser parameters could be optimized to minimize the resistance per link. If the chains were not fully linked, then we approximated the link yield by measuring the tap points. If 50% of the long chains were linked, then we describe the yield as approximately 1000 links per failure. Similarly, if none of the long chains were linked, but 50% of the 216 link chains were linked, then the assumption can be made that the yield is approximately 200 links per failure and so forth.

The relative yield is plotted in Fig. 3 A) as a function of laser energy and spot size. No error bars are included with this data since these are approximations to lead to the 100% yield regions. From the data, it is clear that the 2.5um spot is too narrow due to the size of the line and the positioning inaccuracy. But, from 3.0 to 4.0um spot size, there is a clear region of very high yields for a large range of laser energies. Since the system is rated at 1% laser energy, there is a broad range of acceptable laser energy to produce links. The region where no failures occurred represents the range of energy parameters where perfect linking was achieved on 3 to 8 chains without a single missed connection. In total, within the range of perfect link yield, more than 50,000 connections have been made thus far. The resistance of chains in the region of no failures were measured and divided by 1320 (the number of links per chain) and plotted in Fig. 3 B) as a function of laser energy. These are average resistances measured across the long chains and divided by the number of links; thus, it over-estimates the resistance per link. These were not 4-point measurements but simply series resistance of both the metal 1 and metal 2 wires as well as lines connecting to the pads, so the links themselves are a fraction of this value. Nonetheless, this resistance value is approximately 0.8ohm per link across the entire range of laser energies from 0.60 to 0.85uJ with a 3.5um diameter round spot.

Fig. 4 FIB cross-section of a vertical link between the 4um wide metal 1 and two sides of the hole in metal 2.

A solid metallic connection forms as a result of the cracked dielectric between the two levels and the molten metal filling of the crack. The crack forms on both sides of the line as shown in cross-section in Fig. 4. The major advantages of the vertical link is that the cracks are completely terminated on metal lines and there is no damage to the passivation. Since metal is highly plastic, any stresses induced in the structure will be absorbed in the metal without causing additional pressure on the upper dielectric. This results in a broad process window and a robust link. Since most of the material that flows through the cracks originated from the lower layer, there will be a deficit of metal remaining in the line. The dark region in the middle of the lower line is a result of the displaced metal. The metal flows due to the thermal expansion while it is hot. Then, as it cools, there will be remaining voids. The metallic sheet which forms the link is approximately 200nm thick and is continuous around the annulus, with distributed voids in the volume of metal one, in the center of the link region. The voids are due to the metal that was displaced to form the connection. The volume reduction of metal from the first metal level accounts for the increased resistance due to too much laser energy. Similarly, if not enough energy is delivered to metal 1, there is insufficient metal in the link itself accounting for the higher resistance on the low energy side. Nonetheless, there is approximately a 2 to 1 energy window for the optimized spot size where the resistance varies very little across the wafer.

Fig. 5 Result of an open circuit from an accelerated electromigration test.

The reliability of these link chains were tested under high current and temperature acceleration. Preliminary results suggest that the effective current density of the link chains is increased by less than a factor of 2 over the original area of metal 1. Thus, the lifetime is limited primarily by the void growth and eventual electromigration in the whole link region. Fig. 5 shows an FIB cross-section of a link that opened up after constant current stress at 80mA for 7 hours at 300C. We have found that the lifetime at this current density extrapolates to greater than 100 years at room temperature. The presence of three voided regions indicates that the link material is not the primary reliability limitation. The void that formed in metal 2 is entirely on top of the TiN barrier metal, suggesting that normal electromigration is occurring in the upper layer at approximately the same rate as in the metal 1 and the link region. Whereas the failure in this example is clearly due to the voided metal from the link itself, the robustness of the link is comparable to that of the metalization. Thus, if a metal with higher electromigration resistance is used, such as copper, the link will be similarly improved.

Fig. 6 Schematic of another vertical link implementation with increased metal 1 under the metal 2 link annulus.

One way to increase the reliability of the link is to include extra material in the link region by increasing the width of the line beneath the opening, as shown in Fig. 6, for example. The wider area of metal under the link would serve several purposes; to provide more edge area to initiate linking, contain the laser energy to the inner region, and provide more material to flow through the cracks to form a more reliable link. The dimension of the pad should be no larger than the size of the hole. The preferred overlap of the metal 2 opening around the metal 1 pad should be approximately 1/2 the thickness of the inter-level dielectric. For metal and dielectric layers on the order of 1um thick, the overlap can be as small as 0.5 um within the walls of the metal 2 opening.

Fig. 7 Schematic of a 20um metal pitch interconnect layout with two vertical links per crossing.

As for scalability of the link structure, this phenomenon occurs on the scale of smallest hole in metal 2 through which one can accurately position a laser. Most of the laser energy must impinge on the lower level metal so that the fracture occurs beneath the upper level, and no cracks originate from the top of metal 2 toward the outer surface. The smallest structures that have been successfully linked have a 2.3um opening to a 1.2um wide lower level of metal. This was performed with the same laser using a 2.5u FWHM laser spot. A smaller spot will allow higher density scaling. The largest practical links also depend on how large a spot can be made by the laser system. The largest metal one lines that have been connected were 6um situated beneath an 8um opening in metal 2. No other metal thicknesses were tried, yet it is clear that for packaging applications, thicker metal and dielectric layers will be required. A sample layout of an interconnect array using vertical links is shown in Fig. 7. There are two links per crossing to maintain parity of metal pitch in both vertical and horizontal directions. Of course, the lines can be broken up to have single metal 1 tracks running perpendicular to the metal 2 tracks. The actual size of the lines depends on the density requirement of the interconnect and the design rules of the fabrication process. Lines of this width and pitch would be appropriate for chip-to-chip routing. The metal thickness in this example could be 1 to 1.5um thick. If thicker metal is to be used, the dimensions should be scaled up proportionally

The mechanism of laser induced fracture and metal flow has been used to form metallic connections between standard lines of aluminum based metalization. This process can be used in laser repair or to program cells in a circuit. The connections do not breach the passivation and can be applied to any standard multi-level metalization process. Connections can be made between any two levels of metal processed by standard silicon fabrication techniques. It is also scalable to the minimum dimensions of metal line lithography and the smallest laser spot that can be focussed through an opening of the upper level metal.

[1] S.S Cohen and G.H. Chapman, ``Laser Beam Processing and Wafer-Scale Integration,'' from Beam Processing Technologies, edited by E.G. Einspruch, S.S. Cohen, and R.N. Singh, Academic Press (1989)

[2] 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

[3] 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

[4] 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

[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] J. Raffel, A.H. Anderson, and G.H. Chapman, Wafer Scale Integration, chapter 7, 363, (Kluwer Academic 1989)

[7] US Patent # 4,455,495 (filed 4 years earlier)

[8] H.D. Hartmann and Th. Hillmann-Ruge, ``Yield and Reliability of Laser-Formed Vertical Links,'' from Proc. of Multilevel Interconnections: Issues that Impact Competitiveness, SPIE Vol. 2090, pp. 146-160 (1993)

[9] J. B. Bernstein, Y. Hua, and W. Zhang, ``Laser Energy Limitation for Buried Metal Cuts,'' IEEE Elect. Dev. Let., Vol. 19, pp. 4-6, Jan. 1998