Logo
The following article appears in the journal JOM,
51 (3) (1999), pp. 22-27.

Microelectronics Overview

Materials Issues in Area-Array Microelectronic Packaging

Darrel R. Frear
TABLE OF CONTENTS
The important issues in advanced area-array electronic packaging for semiconductor devices are materials driven. Some of the processing-driven materials issues include the effect of introducing a silicon device interface with copper pads and a low- dielectric, the effect of decreasing pitch and feature size on the package interconnects, the development and implementation of organic substrates, and advanced underfills for fine-pitch flip-chip applications. From a materials reliability aspect, important materials issues include enhanced solder interconnect reliability, -particle-induced soft errors, and the introduction of lead-free solder alloys.

INTRODUCTION

Electronic packages, the housings for semiconductor devices, can be described as all of the materials around the semiconductor up to, and often including, a printed wiring board. A package typically consists of the semiconductor mounted and wired (with solder or gold wires) to a dielectric substrate (with a leadframe or metal traces) that is encapsulated to seal the device from the environment. The electronic package must serve four functions: provide electrical (or photonic) contact to and from the chip; act as an electrical space transformer to take electronic functionality from the dense surface of the chip to the coarser pitched "outside world;" provide environmental isolation to the semiconductor; and provide an avenue for heat conduction away from the device.

In the past, advances in silicon semiconductors had little or no effect on the electronic-package design or process. New advances in silicon devices have driven increases in chip speed, device density (e.g., the number of transistors/mm2), and power dissipation. These changes, and those planned in the future, require larger numbers of interconnects to and from the devices, improved electrical performance, more effective thermal management, software systems to design and model the assembly, and different sets of materials to make these changes happen. Furthermore, the package is shrinking to a minimalist version of the products that traditionally have been called packaging. The chip and board have become a conjugate packaging system that must provide the traditional levels of electrical, mechanical, thermal interfaces, and environmental protection, with the additional challenge of increased performance.1 In addition, the boundary between where the device ends and where the package begins has become blurred with passive devices (e.g., resistors, capacitors, etc.) in the package and redistribution layers that act as space transformers on the chip. Electronic packaging is becoming one of the greatest challenges in manufac-turability, performance, and reliability in advanced electronics applications.

Some of the needs for improved performance can be met with some version of the packaging solutions that have been developed over the last 5-10 years. These solutions have typically focused on peripheral packages with wirebond chip-to-package interconnects. A schematic illustration of one such package is shown in Figure 1. The devices on the silicon die are routed from the center location on the chip to the wirebond pads on the periphery. The chip is mechanically bonded to a substrate and is gold-wirebonded to electrically connect the chip to the leadframe. The leadframe fans interconnects through a dielectric (ceramic or organic) into a peripheral pattern, where it can be soldered to a printed wiring board. As the number of leads increases, this type of package becomes problematical. The minimum wirebond pitch (the distance from the center of one wirebond lead to the center of another) is currently 70 mm, with a decrease to 50 mm expected in 1999.2

Figure 1. An isometric-section schematic that illustrates a peripheral package showing a dual-in-line-package (DiP) that is wirebonded.
An alternative to peripheral interconnect packaging is to access the unused area under the chip and package for interconnects in an area-array solution. A schematic example of an area-array package is shown in Figure 2. In area-array packaging, the surface of the chip has an array of solder interconnects (solder bumps) that are joined to a substrate when the chip is flipped over (flip-chip packaging).3 The interconnects can then be fanned-out through the package to another area-array geometry on the bottom of the package called a ball-grid array (BGA).4 An organic underfill fills the gap between the silicon die and substrate to enhance mechanical adhesion. The pitch of the solder bumps is much coarser than for wirebonding-200 µm to 250 µm.

One of the great advantages of the area-array package is an increase in the number of available input/output (I/O) leads over peripheral designs. For example, at a 70 µm pitch, the maximum number of wirebonds on a 10 mm chip is less than 600. For an area-array package on the same 10 mm chip, even at the coarser bump pitch of 250 µm, the number of available I/O interconnects increases to 1,600.

Furthermore, it is difficult to design the sweep of wires from the chip to the leadframe in peripheral packaging to eliminate shorts and minimize inductance between adjacent wires. These issues are not present in the much coarser pitched area-array package. In a flip-chip package, the solder interconnects can also act as heat sinks because they are attached to the active surface of the silicon and are massive enough to draw heat away from the silicon.

Figure 2. A schematic of an area-array package with a flip-chip device interconnect and a ball-grid array package.
Finally, the cost of current peripheral array solutions have generally exceeded customer targets while falling short of advanced requirements in performance and reliability. Additional development is needed to meet the target of reduced cost with increasing functionality and performance. This cost/performance tradeoff is summarized in Figure 3. The number of pins, or I/O interconnects, must increase dramatically with functionality while the cost per pin must decrease at a rate comparable to that of the silicon device. Figure 3 also shows that future package feature size scales with silicon feature size (e.g., a 150 µm package feature scales with a silicon feature of 150 nm).

The requirements for improvements in the functionality of electronic devices will also drive an area-array package solution. High-performance applications have an I/O range of 1,000-2,000 with a voltage drop power requirement and ground availability in the middle of the chip, rather than at the edges, which results in a drive toward a flip-chip area-array solution. Hand-held (e.g., cellular telephones) and cost-performance (e.g., desktop and portable computers) applications have a lower I/O requirement, with a form factor that requires a tighter density of I/Os provided by an area-array package.

Even with the advantages created by area-array packaging, performance, reliability, and cost requirements are challenging currently available area-array electronic package design. The majority of these challenges are materials driven.

THE MATERIALS CHALLENGE OF ELECTRONIC PACKAGING

Electronic packaging, and to a greater degree area-array packaging, is arguably one of the most materials-intensive applications today. The families of materials included in a package include semiconductors, ceramics, glasses, polymers, and metals. A list of the types of materials used in an electronic package are shown in Table I. The processes that are required to assemble a package are equally variedwelding, soldering, curing, cold and hot working, sintering, adhesive bonding, laser drilling, and etching.

Figure 3. The assembly and packaging challenge. (Derived from the The National Technology Roadmap for Semiconductors Technology, 1997 edition, published by SIA).2
Consider a generic example of the processes and materials required to manufacture an area-array package. Deposition of metal on the aluminum or copper pads on the silicon device is used to provide a solder wettable surface for the flip-chip solder interconnects and a diffusion barrier to the silicon (Figure 4). The metals are deposited in an area array at the wafer level and can be plated or deposited by physical means (evaporating or sputtering). The metal stack, or under bump metallurgy (UBM), typically consists of an adhesion layer to the pad (chromium, titanium, or zincated aluminum), a diffusion barrier (chromium or nickel), and a solder wettable layer (nickel, copper, or gold). These layers are typically a few thousand angstroms thick and are patterned on the wafer by photolithography. Solder is deposited on the UBM by evaporation (high-lead-content solders), plating, or solder paste.

The flip-chip solder interconnects are typically a high-lead-content solder alloy (e.g., Pb-5Sn or Sn-3.5Pb, in wt.%) that has a melting point in excess of 300°C. This high-melting-temperature solder alloy does not reflow during subsequent soldering processes (e.g., substrate to board reflow). The solder is reflowed to form solder balls, and the wafer is then diced. For assembly into a package, the silicon die is soldered to a ceramic, or organic, substrate. A silica-filled anhydride resin underfill is then flowed and cured under the die to enhance mechanical adhesion. The substrate is patterned to mirror the configuration of the solder balls on the die. The substrate pads are metallized to be wettable by the solder. The substrate acts to redistribute the leads to a coarser pitch on the BGA side.

To form the BGA interconnects, solder balls are mechanically placed on the substrate lands using mechanical fixtures. The solder alloy for the BGA interconnects is typically a near eutectic Sn-Pb alloy that melts at 183°C. The balls are fluxed and reflowed to join to the substrate. Welding, or brazing, or glass sealing a lid on the die or by flowing and curing an organic encapsulent (glob top) over the die completes the package. The package is then ready for assembly to a printed wiring board.

Table I. Materials Used in Electronic Packaging
Material Application
Semiconductors Si, GaAs
Metals Solders for interconnects (Sn-Pb, Sn-Ag; gold wirebonds; copper leadframes (Kovar, CuBe, Alloy 42); copper traces in substrates; tungsten, molybdenum traces in co-fired ceramics; Ag, Au, Pd for thin/thick films on ceramics; and nickel diffusion barrier metallizations.
Ceramics Al2O3 substrates modified with BaO, SiO2, CuO,etc.; SiN dielectrics; diamond heat sinks.
Polymers Epoxies (overmold); filled epoxies (overmold); silica-filled anhydride resin (underfills); conductive adhesives (die bonding, interconnects); laminated epoxy/glass substrates; polyimide dielectric; benzoyclobutene; silicones; and photosensitive polymers for photomasks.
Glasses SiO2 fibers for optoelectronics; silicate glasses for sealing; borosilicate glass substrates; and glass fibers for epoxy/glass substrates (F4-4).

The use environment for electronic packages can be quite severe, depending upon the application, and the package must perform reliably in the required environment. The package can be exposed to ambient thermal cycles that can range from a few degrees for a desk-top computer to almost 200°C for automobiles or avionics applications. The device itself can create thermal cycles through Joule heating (turning the device on and off). These temperature fluctuations result in thermomechanical cycles because the materials joined in an electronic package have a wide range of linear coefficients of thermal expansion. The package can be exposed to long periods of time at elevated temperatures due to a combination of high-temperature ambient conditions (e.g., under the hood of an automobile) and Joule heating. These temperatures (~125°C to 150°C) can be well in excess of 90 percent of the melting point of the solders used in the package. The package can also be exposed to mechanical strains due to vibration or shock (e.g., dropping the device). These environmental requirements are complicated in an area-array package where the solder bonds are both electrical and mechanical interconnects.

CRITICAL MATERIALS AND PROCESSING ISSUES

Figure 4. A cross-section schematic of a UBM and solder bump for a flip-chip interconnect.
The processes discussed above for currently used area-array applications have been optimized and, although challenging, result in manufacturable and reliable assemblies. However, there are a number of requirements on the horizon that relate to device performance, size, number of I/Os, cost, and reliability that will challenge and modify these processes.

Packaging Devices with Copper Metallization with a Low- Dielectric

New metallization materials are being incorporated on silicon devices to improve performance, while simultaneously increasing electromigration resistance. Plated-copper metallization offers the lowest cost with greater electrical conductivity than sputter-deposited aluminum metallization. Additionally, low-permitivity insulators (low- dielectrics with < 3) provide a reduced capacitance and cross-talk between conductors. The low- material is used as a dielectric insulator in the interlevel dielectric (wiring the silicon out to the first package interconnect) with a matrix of copper or aluminum interlaced, typically using a dual damascene process.5-7 An example of a low- material being considered for semiconductor devices is the photosensitive benzocyclobutenes (photo-BCB) polymer8 that has a = 2.65. The damascene process and resultant one-layer metal structure is shown in Figure 5.

The devices for which copper and low- dielectric are the most suitable also require area-array packaging, involving flip-chip interconnects, in order to accommodate the large number of I/Os expected with these devices. The flip-chip array also reduces cross talk that could occur in wirebonded packages, because the long wires can act as antennas and create an inductive current in nearby conductors.

The move toward the use of copper and low on the die could potentially affect the manufacturability and reliability of the packaged devices. The significant issues involved in packaging copper and low k involve bonding with the copper and structural stability of the dielectric during processing and in service. Many of the currently used low- materials have properties that degrade at processing temperatures in excess of 250°C due to extensive cross-linking in the polymer. Therefore, the solders that must be used to interconnect with the copper/low- silicon device must have a suitable process temperature. The solder alloy proposed for this application is the near eutectic Sn-Pb alloy that melts at 183°C.

Figure 5. Process steps to fabricate a dual damascene structure with copper and a low- dielectric. (Adapted from Hu and Harper.)7 SiN and low- dielectric deposition; via definition by etch; pad definition by etch; barrier (e.g., tantalum) seed (Cu) physical vapor deposition; and copper plating and chemical-mechanical polishing.
There are a number of challenges associated with this proposed change. The UBM must be redeveloped in changing from a high-lead solder to the eutectic sol-der alloy. The greater amount of tin (~63 wt.%) in the eutectic solder alloy results in a much faster reaction with the wettable UBM metallization and could result in dewetting if the appropriate metal types and thicknesses are not present. The UBM metals are deposited in what could be highly stressed conditions that could affect the performance and reliability of the bumps.

These UBM structures were optimized for aluminum pad metallizations and may require changes for adhesion to copper pads. One metallization that has seen a great deal of development over the last few years and has shown a great deal of promise is electroless Ni-P. Electroless Ni-P has the advantage that it only plates over metal and, so, is a selective-metal-deposition technique as opposed to evaporation or sputtering, which require patterning and etching process steps. Electroless nickel has potential as a low-cost UBM and has a very slow reaction rate with tin. However, implementation of the electroless nickel as a UBM is not widespread because processing and reliability require further characterization efforts, but there are significant R&D efforts underway. Furthermore, electroless nickel is typically preceded by zincation on aluminum pads, which can not be performed on copper. A new process will need to be developed.

Fine-Pitch Packaging

The trend in flip-chip interconnect pitch, as represented in Figure 3, indicates that the current 250 µm pitch will shrink to 150 µm by 2000 and to 50 µm by 2010. The driving force for this pitch shrink is to satisfy the requirements for high-performance silicon devices. These requirements include a dramatic increase in the number of I/Os due to increases in the number of signal lines and power requirements. Higher power devices require more signal and ground lines and, to limit point sources of heat, the power and ground interconnects should be spread evenly across the area array.

There are a number of materials and processing challenges associated with finer pitches. As the pitch shrinks, the methods to deposit the solder become more limited. Solder paste is very difficult to deposit using a silk-screen method at pitches below 150 µm due to rheological limitations of forcing a semi-solid (the paste) into small holes (the silk screen). Evaporation is difficult because developing a metal screen mask with the required tolerances is prohibitively expensive. Solder plating is still a good option, but the solder must be very uniform across each die. Solder ball uniformity is critical because large variations between die could result in electrical opens for small balls and shorts for large balls (Figure 6).

Figure 6. The uniformity of ball size on the flip chip is important. (a) If the ball is undersized, electrical opens occur. (b) If the ball is oversized, electrical shorting is possible.
An additional issue with a decrease in ball size is the joint gap between the substrate and the die decreases to the point that it may become very difficult to flow underfill completely under the die. At 100 µm pitch, the gap between the die and substrate could be significantly less than 25 µm, the limit of underfill flow. For these very fine-pitch applications, an alternative underfill technique will need to be developed or flow under the chip will not be possible. One alternative would be to deposit the underfill material on the wafer immediately after flip-chip solder bumping. New underfill materials and processes would have to be developed to implement this process.

Organic Substrates

Additional cost savings are expected by replacing ceramic substrates with organic substrates. One stated reason for moving to an organic substrate is based on the belief that, in volume production, organic materials are less expensive than ceramics. The organic substrate also offers a lower dielectric constant (therefore, less capacitance and better speed) of 3.5 for organic vs. 5 for ceramics.

The move to an organic substrate is the focus of an extensive development effort worldwide. The board interconnect density for flip-chip substrates must accommodate the increasing density of off-chip interconnect, and cost-effective substrate capability that combines the necessary fine line and micro-via features must be developed. Micro-vias are the metal-filled holes that provide a conduction path between copper lines in the multiple layers of an organic substrate. Micro-via capability must also scale with line width/spacing in order to provide the "via in-line" structures necessary to support bump-pitch densification. At current 250 µm pitches, micro-vias can be created using a photolithographic process. At finer pitches, laser-drilling techniques must be developed to create the required small via size.

Advanced substrates for flip chips must also provide increased wireability while delivering improved electrical performance with reliability levels at least equivalent to current surface-mount applications. New materials with dielectric constants approaching 2.0 and with coefficients of thermal expansion approaching 6.0 ppm/°C will be necessary to meet fine-pitch requirements. Furthermore, uniformity and flatness are critical for organic substrates to ensure uniform joint size and bonding across each die as the interconnect pitch decreases.

The change in substrate materials drives the need for a lower melting temperature solder to replace the high-lead-content Sn-Pb alloys that are processed in excess of 300°C for ceramic substrates. The organic materials in a substrate are typically epoxy-based and char or burn at temperatures in excess of 250°C. The most common solder alloy candidate is near-eutectic Sn-Pb that can be processed at 200°C. However, this could result in a situation where the flip-chip interconnect melts every time an additional solder reflow occurs during subsequent processing. Multiple reflows of the solder interconnects accelerates intermetallic growth between the tin of the solder and the metallized pads for the UBM and substrate. One way to circumvent this issue is to deposit low-melting-temperature eutectic Sn-Pb solder on the substrate lands or on the high-melting-temperature solder bump. Reflow to the substrate can be performed at eutectic Sn-Pb soldering temperatures, but the mixing between the high-lead-content and eutectic solder results in a higher melting temperature composition. Subsequent re-flows at eutectic Sn-Pb processing temperatures will not remelt the composite alloy flip-chip joint. This addresses the low-temperature processing requirements for organic substrates, but does not address the low- dielectric low-temperature processing requirements because the high-lead-content solder must still be reflowed on the silicon die.

Underfill

Flip-chip technology demands that the space between the chip and the substrate be filled with a dielectric organic material to help mechanically interlock the chip to the substrate. The underfill material is a silica-filled anhydride resin polymer. The filler provides the strength in the composite polymer, and the resin bonds with the die and substrate. A scanning electron miscroscopy (SEM) micrograph cross section of a eutectic solder flip-chip joint on an organic substrate with silica-filled underfill is shown in Figure 7.

Figure 7. An SEM micrograph of a flip-chip solder bump with silica-filled underfill. (Photograph courtesy of K.N. Tu at the University of California-Los Angeles.)
Without this underfill material the difference in the coefficient of thermal expansion between the chip (3 X 10-6 mm/mm°C) and the substrate (~17 X 10-6 mm/mm°C) would quickly result in fracture of the solder interconnects when variations in temperature occurred. New underfills must be developed to accommodate increased bump density and shrinking bump height. As bump density increases and pitch decreases, the underfill must flow into smaller and smaller spaces. The currently used 250 µm pitch requires an underfill with a thickness of 110 µm. At a 150 µm pitch, the dimension between the chip and substrate will be less than 50 µm but must result in 100 percent fill for required levels of reliability. Traditional flow dispensing and cure may not be suitable in these thin spaces due to a lack of capillary driving force. For very thin fill spaces, wafer-level deposition of the underfill will need to be developed. This wafer-level deposition would involve spin-on of an uncured polymer to a uniform thickness slightly less than the bump height. The wafer would then be diced and die-mounted on substrates. Reflow would form the solder joints and cure the underfill simultaneously. Materials development is required to create a filled polymer that can be spun to a uniform thickness of 25 µm or less, bonds well with silicon and substrate materials, and is compliant.

Solder Reliability

For surface-mount and area-array applications, the role of solder in the package is increased. In these advanced designs, the solder is an electrical interconnect-a mechanical bond-and must often serve as a thermal conduit to remove heat from the joined device. The interconnects become more critical as chip size, chip carrier size, and the number of I/Os increase, while the solder joint size and cost decrease. Furthermore, some of the use environments are becoming increasingly severe. For example, some automotive electronics are being relocated under the hood with temperature extremes that could range from -55° to 180°C. Portable devices, such as cellular telephones, are expected to withstand severe shock environments caused by dropping a phone.

Under thermal cycling conditions, the solder interconnects undergo microstructural evolution through heterogeneous coarsening and failure that results in electrical opens.9 This situation is alleviated, although not eliminated, through the use of the underfill for the flip-chip interconnects. Currently, BGA interconnects are not underfilled because they are much more massive and typically join materials with more closely matched coefficients of thermal expansion. As the pitch shrinks and the die size increases, the amount of strain imposed upon the solder interconnects also increases. The strain imposed on the solder joints follows the relation

= Ta/h

where is the shear strain imposed, is the difference in coefficient of expansion between the joined materials, T is the temperature change, a is the distance from the neutral expansion point of the joined materials, and h is the thickness of the interconnect.

An example of the change in strain follows: For the 250 µm pitch, a die of 17.3 mm2 has a = 8.65 mm and h = 110 µm; for 100 µm pitch, the die is 22.8 mm2 with a = 11.4 mm and h = 25 µm (numbers are taken from Reference 2). The temperature excursion is 100°C and the die is Si (6 X 10-6/°C) with an FR-4 substrate (17 X 10-6/°C). The resultant shear strains in the outermost solder interconnects at 250 µm and 100 µm pitch are eight percent and 50 percent, respectively. The larger strains result in more rapid fatigue failures in both the solder and potential fractures in the silicon if the underfill transfers the strain to the more brittle component.

This issue may need to be addressed with more compliant interconnects and underfill with more fatigue-resistant interconnects. Another potential solution is to use greater standoff solder interconnects with an aspect ratio much greater than one. This solution has been incorporated with great success on BGA interconnects using column grid arrays10 rather than standard balls.

Radiation-Induced Soft Errors

As silicon-device feature size shrinks, rare and random events have been observed where the state of a logic element changes for no discernible reason. This was observed particularly for complementary metal-oxide seimiconductor memory elements with very small gate dimensions. No physical damage is observed in the device, and after a reset, the device returns to normal operations. The term "soft error" was coined for this phenomenon and was found to be due to extremely low levels of radiation that resulted in charge production (capacitance) in the silicon. High-energy alpha particles caused electron-hole formation. The source of the radiation that caused the errors was found to be due to trace amounts of radioactive materials in the package, particularly the solder.

The most significant emissions of alpha particles was from the lead in the solder interconnects. These radioactive sources are particularly insidious because the solder interconnects are located close to the silicon device and there are no barriers to absorb the particles. Lead and tin-lead alloys (solder) contain radioactive daughter elements of uranium and thorium that decay to form -particles. The principle cause of alpha emission is 210Pb (an unstable isotope found in lead) that decays by - and -particles (-particle energy = 3.72 MeV) to form 210Bi (decays with - and -particles [3.72 MeV]) and then the very unstable isotope of 210Po, which subsequently decays with an -particle with an energy of up to 5.4 MeV.

As an example, a 5 MeV -particle penetrates into 25 µm of silicon and subsequently creates 1.4 X 106 electron/hole pairs that can easily affect the charge state in a silicon device. The half life of 210Pb is 21 years, and 210Bi has a half life of 5.01 days. The half life of 210Po is 138.4 days, but there is a continuing accumulation of 210Po from the decay of the 210Pb. Commonly used lead for solders can have -activity as large as 1.0 /cm2-h.

Soft errors can become more predominant as gate size shrinks and the subsequent charge area in the silicon that results in memory decreases. Therefore, there is a need for a source of lead with a low -radiation rate to limit the possibility of soft errors. Some sources of lead ore have lower -radiation levels, but these are often contaminated during the smelting, handling, and processing steps. Low -Pb ore has an activity on the order of 0.02 /cm2-h.11 Since the half life of 210Po is 22 years, extremely old lead would not be expected to have a large -activity. For example, lead recovered from sunken Spanish galleons and old Roman ship anchors has an -activity of less than 0.001 /cm2-h.11 However, there is a very limited supply of this material. A newly proposed source (Pure Technologies) of low -Pb uses laser-isotope separation to refine lead to a state where -particle generation is drastically reduced to less than 0.0001 /cm2-h.11

It should be noted that cosmic radiation is also a source of alpha particles. At sea level, the cosmic radiation levels are on the order of 0.005 /cm2-h and dramatically increase with altitude. Cosmic radiation issues typically have been addressed through designs that increase the radiation hardness of the device. However, device engineers must be aware that it is difficult, if not impossible, to reduce the -radiation level below cosmic levels through materials changes.

Lead-Free Solder Alloys

Concerns about the presence of lead in the environment and potential exposure scenarios that could result in the ingestion of lead by humans and wildlife has prompted a concerted effort to limit the use of lead in manufactured products (notably gasoline, plumbing solders, and paint). The driving forces of this effort are many and include the U.S. federal government,12 the U.S. Environmental Protection Agency, state and municipal governments, activist groups, and concerned citizens. Japan and Europe have initiated similar efforts in the elimination of lead from the environment. It has been construed by the electronics industry that the use of lead-bearing solders would not be immune from the consequences of these debates and legislative efforts. Furthermore, a lead-free solder alloy would limit the problems created by -particle-induced soft errors. Therefore, efforts have commenced to develop lead-free solder alloys for area-array interconnects.

A number of tin-based solder alloys have been proposed as lead-free substitutes for packaging applications.13 There are no "drop-in" candidate alloys to replace eutectic Sn-Pb solder, but there are three alloys that stand out as strong possibilities for implementation as lead-free alternatives. These lead-free alloys, their melting behavior, and a brief discussion of their microstructure is given in Table II.

Table II. Candidate Lead-Free Solder Alloys
Alloy Liquidus/Solidus Temp. (°C) Processing Temp. (°C) Microstructure
Sn-3.5Ag 221 (eutectic) 20 Tin matrix with precipitates of Ag3Sn
CuSbAgSn (Castin) 211 (eutectic) 240 Tin-rich dendrites with lamellar Sn-Ag-Sb-Cu interdendritic regions
Sn-3.4Ag-4.8Bi 201 (eutectic) 230 Tin-matrix surrounded by tin-rich and bismuth-rich lamella with Ag3Sn intermetallic precipitates

The intermetallics that form upon the reaction of these solders with typical metallizations are the same as are found with eutectic Sn-Pb. On copper, a compound layer of Cu3Sn/Cu6Sn forms. On nickel, a Ni3Sn4 layer forms. The kinetics of intermetallic growth is also similar to eutectic Sn-Pb. The higher melting temperature of these solder alloys makes them good candidates for hierarchical soldering, although the melting points are close to the processing temperature of eutectic Sn-Pb. One possible problem with these alloys is that they do not wet copper or nickel metallizations as well as eutectic Sn-Pb. The presence of lead in solders improves wetting behavior by lowering the liquid surface tension.14

From a mechanical properties perspective, these lead-free alloys exhibit a higher yield strength and slower creep deformation than eutectic Sn-Pb. However, the alloys are susceptible to interfacial delamination at the intermetallic/solder interface under tensile loading conditions.15 These lead-free alloys also have better fatigue performance than eutectic Sn-Pb, with the CastinTM alloy having the longest thermomechanical fatigue life.

FUTURE OUTLOOK

The future development of silicon devices requires the enhancement of electronic packaging for both performance and reliability. The performance requirements include a higher pin count, finer feature size (finer pitch), faster speeds, lower capacitance, and more power. Enhanced reliability includes longer life for larger packages with smaller interconnects in increasingly severe environments. These requirements drive the need for area-array packaging.

In the past, the electronic package did not limit the performance of silicon devices. As the devices become faster and more powerful, the package can be the roadblock to desired functionality. Future device development must be a fully integrated approach between device design, device processing, and packaging to meet advanced performance requirements.

References
1. Electronic Materials Handbook: Volume 1: Electronic Packaging (Materials Park, OH: ASM, 1989).
2. The National Technology Roadmap for Semiconductors Technology (San Jose, CA: Semiconductor Industry Association, 1997).
3. L.S. Goldman, IBM J. Res. Dev., 13 (1969), p. 251.
4. J.H. Lau, IU (New York: McGraw-Hill, 1995).
5. K.J. Blackwell, I.I. Memis, and R.P. Kuracina, High Density Interconnect (HDI), 1 (4) (1998), p. 20.
6. P.C. Andricacos et al., IBM J. Res. Dev., 42 (1998), p. 567.
7. C.-K. Hu and J.M.E. Harper, Mater. Chem. Phys., 52 (1998), p. 5.
8. A.J.G. Strandjord et al., Int. J. Microcircuits and Electron. Packaging, 13 (1996), p. 360.
9. D.R. Frear, D. Grivas, and J.W. Morris, Jr., J. Electronic Materials, 17 (1988), p. 171.
10. T. Caulfield et al., Ball Grid Array Technology, ed. J.A. Lau (New York: McGraw-Hill, 1995), p. 131.
11. R. Brodzinski, "A White Paper on Alpha Activity in Lead" (Richland, WA: Battelle Northwest, 1998).
12. U S. Congress, Senate: Lead Exposure Reduction Act of 1994. DEC. 401 (1994).
13. NCMS Lead-Free Solder Project Final Report (Ann Arbor, MI: National Center for Manufacturing Sciences, 1997).
14. F.G. Yost, F.M. Hosking, and D.R. Frear, eds., The Mechanics of Solder Alloy Wetting and Spreading (New York: Van Nostrand Reinhold, 1993).
15. D.R. Frear and P.T. Vianco, Metall. Tran A, 25A (1994), p. 1509.

ABOUT THE AUTHOR

Darrel R. Frear earned his Ph.D. in materials science at the University of California at Berkeley in 1987. He is currently manager of the low-cost flip-chip project at Sematech. Dr. Frear is also a member of TMS.

For more information, contact D.R. Frear, Motorola, Semiconductor Product Systems, MD-EL725, 2100 East Elliot Road, Tempe, Arizona 85284; telehone (602) 413-6655; fax (602) 413-4511; e-mail r46897@email.sps.mot.com.


Copyright held by The Minerals, Metals & Materials Society, 1999

Direct questions about this or any other JOM page to jom@tms.org.

Search TMS Document Center Subscriptions Other Hypertext Articles JOM TMS OnLine