Click for next page ( 10


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 9
Chapter 1 INTRODUCTION In the 40 years since the invention of the transistor, much attention has been given to silicon devices and integrated circuits (ICs). (See Appendix A for some terms and acronyms.) The impact of silicon electronics triggered the era of information processing and continues to produce amazing progress. The size of the circuit elements, the speed of their operation, the minimal power consumed, the cost per element, and the reliability of the circuits have improved dramatically. Today, silicon integrated circuits (chips), about the size of a fingernail, may contain a million transistors and cost a few dollars or tens of dollars (or hundreds in extreme cases ) , depending on the type and production level. There are good reasons to expect that some chips will contain a billion elements by the end of this century, and other characteristics will be improved as well. The recognition and praise lavished on silicon electronics over the past 4 decades is entirely justified, and the future remains promising. And yet, integrated circuit chips ore not useful until they have been woven into the fabric of interconnections and packaged. Packaging and interconnection (interconnects) provide structural support, mechanical and chemical protection, thermal management, power, ground, and signal transmission, including timing. The package must be durable and manufacturable and allow access for testing and repair. System performance in today's advanced systems is as likely to be limited by packaging and interconnection as by the chips themselves. In this report, high-density electronic packaging and interconnection will be described from the point of view of the materials of construction. The time frame of this report covers the present (1989) and the period in which present practice evolved and extends toward the end of the century. It is probable that reasonable predictions can be made extending for ~ or lO years, but foreseeing system trends in the next century is highly speculative. The many different sectors that encompass the field make it difficult to separate those to include or exclude. Consumer products, computer mainframes, supercomputers, telecommunications switching, personal computers, work stations, automotive components, and various military classes and other subareas all represent modern practice in the context of specific needs. Some conswner products use an aggressive packaging strategy. Very-high- frequency systems are often not high density. Advanced areas (e. g., computer mainframes ~ are inherently expensive and thus do not represent the leading 9

OCR for page 9
10 edge of developments in other applications. Therefore, this report concerns itself with high-density packaging, in the sense of chips with large numbers of I/Os (inputs, outputs, or both) and the need for high-speed communication between chips. Materials properties play a critical role in system performance and in manufacturing process effectiveness. The materials in use have evolved along with the physical design of the electronic systems, and today's structures represent engineering optimization of many factors. It is of little benefit to consider the properties of the materials outside the context of the application. Furthermore, the packaging and interconnection structures are composites containing many different materials in close proximity. Thus, interracial properties and materials compatibility factors are often more important than bulk properties of the isolated materials. The dependence of the materials properties on the manufacturing process is also important. Electronic packaging includes all structures that are designed to protect the integrated circuits, the attendant interconnections and the related circuitry from physical damage and any other impediment to achieving design performance. Interconnections include all means of communication (power, timing, data, results) from one integrated circuit to another or from one system of circuits to another. Interconnections on the chips themselves is not considered in this report, and interconnection by means other than electrical (e.g., photonic) are also excluded. Thus, the subject matter of this report concerns materials issues involved in electric circuits that run from the bonding pad of a chip through the chip package (if any) by means of lead frame (or other fan-out to leads), through strip-line circuits on a substrate that may be connected by "vies" perpendicular to the plane of the substrate, and end at other integrated circuit bonding pads; it also includes other components or connectors that lead to other circuit boards, shelves, or frames. The physical nature of this packaging and interconnection hierarchy is indicated in Figures 1-1 through 1-7. ~ WIRE BONDS ~ PADDLE LEADFRAME: W FIGURE 1-1 Cutaway view of a plastic DIP. F~ r:~7

OCR for page 9
11 ?~ -it /, ///~/li//~/l FIGURE 1-2 Schematic representation of a chip carrier. The chip is die bonded to the package. The signal, power, ground, and timing leads are wire- bonded to bonding pads on the chip. (A TAB inner lead frame could replace the wire bonds shown.) The exterior case may be premolded plastic (as shown) or ceramic. ~ ~,~ \ , ~ A, ~ ~ r\,, 7t FIGURE 1-3 Molded plastic quad design or plastic leaded chip carrier (PLCC). The chip is die-bonded to the paddle, which is connected to one or more ground leads. Signal, power, ground, and timing leads are connected by wire bonding pads. (A TAB inner lead frame can be used for this purpose. ~

OCR for page 9
An,/ ~,~,~/ FIGURE 1-4 Cutaway of a PGA. The chip is die-bonded in the ceramic cavity and wire-bonded to the conductor fan-out, which is plated onto the ceramic. The outer end of each fanout finger is terminated onto a brazed pin, which is inserted into a through-hole on the PWB. This package can then be sealed hermetically with a lid (ceramic or metal). Similar packages with surface- mount leads are also employed. The device is shown inverted, and in use the active side of the chip would face the PWB (face down). The pins can be made to exit the ceramic on the opposite side (face up). The face-down option shown offers a favorable surface for heat-sink attachment. . ~ - FIGURE 1-5 A field of solder bumps with device ready for mounting in C4 attachment (controlled collapse chip connection).

OCR for page 9
13 / /// ~ -A A: \ FIGURE 1-6 Direct chip attachment with protective polymer overcoat (''glob tops. / :_ ~ Hilt - ~ ~ :~': ~/ ,~ ./ FIGURE 1-7 Printed wiring board circuit shown in cutaway to reveal inner connector layers and vies. A surface-mount device (left) and a through-hole device (right) are illustrated. Note that surface-mount devices can be mounted on both sides of the board (as shown) and do not consume valuable inner layer space with the through-hole .

OCR for page 9
14 Integrated circuit chips are encapsulated in three types of packages: dual in-line packages (DIPs), chip carriers (CCs), and pin grid arrays (PGAs). Each option has many variants, and only the most elementary description is attempted herein. In terms of numbers of units manufactured, the DIP is most important. Chip carriers, however, are now being adopted widely and should rival DIPs in numbers within a few years. In monetary terms, DIP and CC packaging are comparable today, owing to the higher average cost per unit of the CC option. PGAs are very expensive and are employed only for chips with large numbers of external connections (I/Os). The magnitude of the packaging market for 1987 is summarized in Table 1-1. The $2.0 billion packaging total is part of a total U.S. IC market of $9.3 billion. The 1987 market for printed circuit boards amounted to $5.1 billion, and edge connectors for boards amounted to $0.8 billion. Thus, the 1987 silicon cost of $7.3 billion was supported by more than $7.9 billion in packaging and interconnection. These numbers are imperfect, owing to the omission of captive production by equipment suppliers and other factors, but nonsilicon costs are comparable to and probably greater than silicon costs. The numbers also omit the hybrid market (U.S. 1987, about $4 billion), a substantial portion of which can be classified as interconnection and packaging. In view of the rate at which silicon costs are dropping, this spread in costs will increase unless packaging technology is given more empha s ~ s . Table 1-1 U.S. Packaging Market, 1987 Estimate Type Units Average Cost Total Cost (billion) (S/unit) (billion $) DIP 7.4 0.13 0.93 CC 0.7 0.68 0.47 PGA 0.05 6.61 0.31 Other 0.7 0.34 0.25 Total (avg.) 8.9 0.22 1.96 The lead frames of DIPs are inserted in holes (vies) that extend through the printed circuit board. These holes are usually drilled on a 100-mil (0.1 in.) grid and each (copper plated) hole passes through each layer of a multilayer circuit. Chip dimensions are conventionally discussed in terms of micrometers (pm). Printed wiring interconnection structures are usually described in terms of mils (i e., 0.001 in.~. Advanced interconnection structures may involve both units; 1 mil is equal to 25.4 ~m. The holes account for an appreciable fraction of the board area and thus limit the number of interconnections that the board can support. Chip carriers (CCs are bonded only to the top surface (surface mount assembly, SMA) of a multilayer circuit, and no grid of holes are needed. With SMA, components can also be mounted on both sides of the board Therefore, CCs with SMA are

OCR for page 9
15 capable of a higher density of interconnection, which is an important driving force for CCs to displace DIPs. Pin grid arrays (PGAs) are also available in surface-mount form. Chips are attached to the lead frame by "die-bonding" on the back side with solder or conducting adhesive, and the bonding pads are connected to the lead frame fingers by means of fine gold or aluminum wires. A "ball" bond is formed at the chip bonding pad and a "wedge" bond is foxed at the inner lead frame finger. The process of wire bond formation is automated and very fast, although they are done serially, one at a time. The bonding pads on the chip are usually restricted to the perimeter. The spacing between bonding pads is, however, limited (about 6 mils) by the size of the wire-handling tool, which is an important design factor affecting chip die size for circuits with a large I/O count. An alternative to wire bonding is tape automated bonding (TAB), in which the lead frame is a thin copper layer supported by polyimide film and all the inner fingers are bonded to the chip pads in a single operation. By supporting the leads on film and eliminating the wire bonding tool, finer lead spacing (about 4 mils) and higher interconnection density are pass ible . This technology has been available for many years, but no large- scale displacement of wire bonding has occurred. Difficulties involved in preparing the pads or tape fingers ("bumping") and in the simultaneous bonding operation are significant. Preformed packages (premolded plastic or ceramic) are sealed by affixing a lid. The chip resides in a cavity. If the package material is ceramic, the seal may be hermetic, a feature required for most military and other high- reliability circuits. All plastic materials are permeable to some degree, and thus hermeticity cannot be achieved today in plastic packages. Even so, plastic packages can be highly reliable. Plastic packages can be premolded, but they are more likely to be transfer-molded after the wire bonds have been formed. In this process, the chip, bonded to the lead frame, is placed in a mold tool and molten plastic is forced in to fill the cavity completely. Good adhesion between the plastic and the chip and the lead frame is achieved, which makes the structure durable. After the chip has been sealed in its package, it can be tested, transported, and mounted on a printed circuit board. An alternative approach involves direct mounting of unpackaged chips on circuit boards. In the soldering process, the chips "float" (face down) on solder balls that form the conducting bridge to the circuit. This approach is sometimes called C4 bonding (for controlled collapse chip connection). The bonding pads are not restricted to the periphery of the chip, thereby offering considerable advantage. Other "bare chip" options exist, including the use of face-down, "beam- lead'' or TAB lead frame arrangements, and also face -up wire- bonded configurations. Following assembly, these chips may be covered with a plastic coat ("glob top" ~ for protection (see Figure 1-61. Chip-on-board (COB) interconnection is not amenable to the burn-in and testing required for military use, and hence this approach has been more popular in consumer-grade systems. Printed circuit boards may have only one or two layers of circuitry, but high-density interconnection usually implies many layers, typically eight but

OCR for page 9
16 possibly as many as 40. Patterns of metal are defined photolithography by means of a resist process on interlayer substrates, whereby t'viat' (i.e., through-sheet) connections are formed. The various layers are then "piled up" and processed to form the registered multilayer structure. For ceramic structures, the metals can be screen-pi-inted in paste form, and the lamination is accomplished by "firing" at a high temperature (about 1600C for alumina). Co-fired ceramics undergo large material shrinkage during processing, and thus present registration difficulties fin multilayer structures. For epoxy-glass boards, copper patterns are etched (or plated) on partially-cured (B-staged) epoxy, and final bonding and curing are carried out under pressure at about 200C. Vias are formed by electroless plating, after lamination for through- hole boards and before lamination for surface-mount boards. These processes are extremely demanding because the circuits are large in area and layer-to- layer registration must be maintained. Figure 1-8 illustrates progress in chip and printed wiring board (PWB) miniaturization as measured by linear dimensions of conductor width or smallest circuit feature. Although there is much detail that is obscured by these lines, the progress is dramatic and real. Clearly, on-chip features have been reduced at a rate that consistently exceeds that for PUB patterns . Where chip features were an order of magni tude smaller in the 1960s than earlier designs, they are now nearly two orders of magnitude smaller than that. This, and other factors, will lead to an increasingly important intermediate level of interconnection. In fact, interconnect structures of intermediate size have existed right along in the form of thin-film hybrids. At this time, however, there is a great proliferation and diversity in the development of multichip modules (MCMs) that are at least a significant advance in hybrid technology and may be considered a new level of packaging. Multichip modules will probably become component-assembled on both hybrids and PWBs. Multichip modules typically consist of a ceramic base section with power and ground planes. Interconnect circuitry is placed on top with fine metal traces, usually copper, separated by organic dielectric interlayers, usually polyimide. Because of the fineness of the signal traces, two layers on a MCM can replace many (>20) signal layer`; of 8 thick-film (screen-defined) structure. In some MCMs, silicon wafers are employed as the substrate in place of ceramic . This form of MCtt should not be confused with wafer- scale integration (WSI ~ technology, in which both chips and interconnections are fabricated on a single silicon wafer to form an independent system or subsystem. Yields are nearer 100 percent, and desk An redundancy i s an inherent feature of WSI . This is a very demanding technology, and success has not been achieved, despite vigorous and wetl-conceived efforts. In the MCM approach, interconnect structures are built up on the silicon wader, and tested chips are attached. The use of pretested chips greatly~reduces yield loss. It seems evident that MCMs will play a central role in t~igh~de~sity electronic packaging over the next decade.

OCR for page 9
17 r~-l-1~-I-r-~--~-l If I ~ I i ' I I I I' ~ ~ I II 500 200 100 50 - '~' 20 ~ 10 loll 5 2 1 0.2 ~ PWB's _' 1 1 ~ I I I I 1 1 1 1 ,, , I I I 1 1 , , , ~ 1 1 1 1970 1980 1990 3/89 Figure 1-8 Progress in IC chip (in development) and PUB miniaturization (line widths expressed in ~m). In Chapter 2 a more conservative extension of the IC line is developed. Solder is employed for attaching chip packages to circuit board substrates. For DIP through-hole mounting, a solder wave is passed across the back side of the board (i.e., on the side with no components), which subjects the board to a brief thermal shock, although it is benign from the point of view of the components. For a surface-mount assembly, the entire structure must be brought to solder reflow temperature, which can place an unusual stress on the components. As the assembly is cooled following the soldering operation, thermal stress can arise if the thermal coefficients of expansion (TCEs) of the chip package and board substrate are not closely matched. This problem is accentuated as package size increases. When the TCEs are not well matched, the chip package must be provided with compliant leads that can adjust to the dimensional mismatches that develop with temperature changes. Conductive polymers, both isotropic and anisotropic, are being developed to replace solder, but widespread use in the next decade is unlikely.

OCR for page 9
18 The materials of electronic systems can be summarized in terms of relatively few basic elements and compounds. The chips are silicon with aluminum metallization, and with silicon dioxide (silica), silicon nitride, or polyimide as the dielectric. A glass (P-glass) or silicon nitride is used for passivation. The chips are packaged in epoxy molding compounds filled with silica powder. Electrical connections are made with gold wires that go from chip bonding pads to lead frames constructed of copper or an alloy of iron and nickel. The packaged chips are attached by solder to printed circuit boards composed of epoxy-glass (or sometimes polyimide) substrates that support the copper interconnection patterns. In some cases, alumina substrates are employed with interconnections of refractory metals. Connections to other boards are made through connectors composed of molded plastic (e.g., diallyl phthalate, polyphenylene sulfide, etc.) with beryllium-copper contact fingers and cables with copper conductors insulated with extruded plastic. This summary, while substantially accurate for the great majority of circuits, fails to reveal the enormous amount of materials engineering work that makes possible the manufacture and use of electronic systems. Each material has to be carefully tailored for specific purposes, and issues of material compatibility are of critical importance. The process sequence, often involving hundreds of steps, must be such that conditions later in the manufacturing sequence do not undermine structures formed earlier. This abbreviated description includes only the materials that remain in the final product. In fact, many other materials are necessary to the processes that give rise to the structures but are not an integral part of the final structure. Process control is increasingly critical. Thus, packaging materials present challenging demands. The package must provide timing information to all parts of all chips with manageable skew. Data must be delivered to and from chips with delays that are consistent with chip circuit switching times, and an increasing number of access points are required as chips evolve. Electrical power and ground must be supplied over leads having small impedance, which translates to extremely short paths. The heat generated by the chip circuitry must be transmitted to the environment effectively, lest the chip temperature rise disastrously. In material terms, the electrical signal paths require high-conductivity metals (e.g , copper, aluminum, gold) and low-dielectric-constant insulation materials (e.g., polyimides, epoxies, hydrocarbons, and fluorinated polymers). The speed of light is a practical limiting factor in advanced circuits. For power delivery, high-conductivity metals are again necessary, but high capacitance that is physically close (i.e., low impedance) to the chip requires high-dielectric-constant capacitor materials, usually ceramic. Thermal conductivity favors metal, whereas other ceramic paths and polymers are thermal insulators. Thermal coefficient of expansion (ICE) is a critical factor in package design. Silicon has a low ICE compared with common metals and organic polymers. Thus, mechanical stresses are created when temperature changes occur as in manufacture, temperature-cycle testing, and use. The problem becomes more difficult as chip size increases. Thinner sections of the molded body are more susceptible to thermal-mechanically induced cracking failure. ICE of polymeric materials can be controlled by the addition of low- TCE fillers, as is conventional for molding compounds and printed wiring board

OCR for page 9
19 resins. Many packaging and interconnection materials are composites designed to improve the TCE match. As electronic equipment density increases, power dissipation becomes an ever greater problem. In general, heat is conducted away from the chips by the thermal conductivity of the materials employed. Metals and ceramics are relatively good thermal conductors, but polymeric materials as a class are poor conductors, a factor that will assume greater importance over the next few years. The tradeoff of dielectric constant and thermal conductivity will have to be met by compromises between materials and physical design. Application of fluid heat-exchange media will also increase, but the cost of this option is high, therefore, integration of this approach into physical design is essential. Thermoelectric cooling offers an alternative to thermal conductivty, a technology that has been applied to some commercial circuits. This report begins with a discussion of system trends and needs (Chapter 2), followed by a description of approaches that have been employed for packaging and an indication of current trends (Chapter 3~. Chapter 4 discusses materials issues, and Chapter 5 gives more specific information and perspective on plastic and ceramic encapsulants, ceramic substrates, and organic printed circuit structures. Chapter 6 presents an analysis of material sources, current trends, and the importance of the field in the context of international competitiveness. These discussions lay the for future actions in terms of specific technical development and research, and the organizational and political factors that distinguish the United States from other countries. Groundwork for the recommended ens ~ ^ his cus s ions lay in terms of

OCR for page 9