US20050002448A1 - Method and apparatus for non-conductively interconnecting integrated circuits - Google Patents

Abstract

A method and apparatus for constructing, repairing and operating modular electronic systems utilizes peripheral half-capacitors (i.e., conductive plates on the outside of the modules) to communicate non-conductively between abutting modules. Such systems provide lower cost, improved testability/repairability and greater density than conventional modular packaging techniques, such as printed circuit boards and multi-chip modules. The non-conductive interconnection technique of the invention can be applied to all levels in the packaging hierarchy, from bare semiconductor dies to complete functional sub-units. Numerous exemplary systems and applications are described.

Description

FIELD OF THE INVENTION
• Packaging Technologies
• The present invention relates generally to the field of electronic and microelectronic packaging and, more particularly, to a multichip package, a method for assembling, testing and repairing systems so packaged, and a method for communicating between circuits so packaged, via capacitive coupling. Of particular interest are digital systems, meaning systems which contain important constituents that operate according to the rules of multistate or binary logic.
BACKGROUND OF THE INVENTION
• Electronic systems are usually implemented as hierarchical packages of components. Passive or active electronic elements, such as resistors and transistors, and their wiring are typically combined into memory or logic units, which are then combined into circuits and devices, which are combined into larger functional units, and so forth up to the level of a system.
• Each higher level of hierarchy grants the designer greater productivity, but compounds costs for connecting packages together logically and physically. Communication of data and timing among devices at each of these levels requires signal interconnection means, which the package provides. The package also provides powering means and fulfills other requirements such as physical support, heat removal and protection.
• By convention, there are five hierarchical packaging levels 0-4, although these levels are not rigidly defined. An exemplary scale in the hardware system hierarchy is the naked semiconductor wafer, sometimes called an undiced “Level 0” package. Many components are formed simultaneously on a common substrate during fabrication stages, even if the substrate is subsequently separated into subunits. For instance, wiring, memory or logic gates may be assembled into integrated circuits on the surface of a semiconductor wafer and then cut into individual dies. Examples of dies include microelectronic devices implemented on semiconductor materials, superconductors bearing Josephson Junctions, and materials bearing other quantum interference devices.
• Individual dies are typically mounted into “Level 1” packages, which provide mechanical stability, protection, cooling and heat dissipation, power and grounding, and interconnection of signal lines (including clocking) to other packages. Examples include DIP, ceramic, surface mounted and socketed packages.
• A “Level 2” package is a module carrying one ormore Level 1 orLevel 0 packages and interconnecting their signal and power wiring. It typically comprises a printed circuit board (PCB), a printed wiring board (PWB), or a thermal conduction module, and may cluster one or more interconnected packages for these purposes. “Level 3” assemblies further organize theLevel 1 andLevel 2 packages, typically with backplanes, but do not differ conceptually fromLevel 2 orLevel 1 packages. The “Level 4” package canonically ties together the lower level packages with power supplies, environmental systems, mechanical systems, peripherals, and so forth to provide system functionality.
• A multichip module (“MCM” or “Level 1½” package) provides modular functionality as a Level 2-like or Level 3-like package for holding and interconnecting multiple dies and/or associated interconnections. At a minimum, an MCM provides the signal distribution, and power is usually distributed by way of the MCM as well. The MCM may also, or merely, encapsulate its constituent dies as anerstwhile Level 1 orLevel 0 package, thereby providing protection. It may also communicate the dies to a heat sinking substrate, thereby providing heat dissipation.
• Strictly speaking, an MCM could be treated as a package at any level of hierarchy, as defined in practice by its interconnection topology. Note that the term “chip” is used interchangeably in the industry both in reference toLevel 0 dies andLevel 1 packages (e.g., a multichip module is usually in fact a multidie module). As used herein, the term chip refers to aLevel 0 package or die unless context indicates otherwise. The term “module” or “submodule” as used herein is intended to be general, and can refer to any package level, for example one ormore Level 0 dies, one ormore Level 1 orLevel 0 chips (packaged or not), and of course higher order ensembles.
• Present MCM Technologies
• An MCM involves two or more dies, whether bare or encapsulated, mounted and conductively coupled to it. It provides power and inter-die signal wiring. In some MCM technologies, the dies are physically bonded to a substrate, and leads that are wire-bonded to peripherally positioned contacts (e.g., pins) supply the conductive connections between the dies and a multichip substrate. Other technologies utilize a “flip-chip” configuration in which the leads of the dies are positioned either peripherally or over much of the die area (such as a pin grid array (PGA) or solder bumps) and are soldered or otherwise bonded to respective contacts on the multichip substrate.
• Several families of multichip packaging technology are standard at present. The so-called MCM-L technology utilizes a laminated, organic board substrate to which dies are bonded by flip-chip, tape automatic bonding (TAB), or wire-bonding. In the MCM-C technologies, dies are attached either directly by flip-chip or indirectly in prepackaged carriers to a ceramic thick-filmed substrate. The ceramic substrate is formed either sequentially, by a print and fire process, or by lamination and sequential co-insertion of screened green sheets. The MCM-D technology utilizes deposited thin-film substrates to which dies are then attached as in MCM-C. There are also variations of these three basic MCM technologies. For example, a variation using plastic packages and involving molding compounds and lead frames is the so-called multichip plastic quad packs (MCM-P) technology. Another recent variation involves the use of deposited thin films on a ceramic multichip substrate, referred to alternatively as MCM-DC or MCM-CD, and typically provides inter-die signal wiring in the deposited polymer-metal thin-film layers and power/ground wiring in the co-fired ceramic thick-filmed substrate. All of these technologies are subjects of intense research and invention in industry and universities. (See Rao R. Tummala, “Multichip Packaging—A Tutorial.”Proc. of the IEEE, December, 1992.)
• Many approaches to the construction of high density multichip modules have been proposed. The IBM C4 technology attaches dies to the multichip module in a flip-chip face-down configuration. The arrangement minimizes the parasitic inductance of the package leads, and allows pad location at any point on the interior of the die. Typically, dies are attached to the module using a reflow-solder approach. Dies are bumped by bonding to each pad several layers of protective metalization followed by a 10-200 micron diameter solder ball. A plurality of dies are then accurately positioned on the multichip module, and reflow-soldered into place. Inspection of the solder joints can be done with thermographic or radiographic techniques, but may be difficult otherwise. Pad location is no longer limited to the die periphery, but is often constrained by thermal coefficient of expansion mismatch between silicon and module to lie within some radius of the center of the die. Repair is carried out by module heating, die removal, and reflow-soldering of replacement die.
• “Chips-first” face-up wire-bonding of silicon dies to high density silicon, ceramic, or copper-polyimide modules is similar to conventional hybrid manufacturing technologies, and shares difficulties in rework and bonding yield. The GE/TI process forms a planar wafer-like module from collections of selected loose dies placed face down on a flat surface, and then encapsulates it in a polyimide carrier. After curing, this carrier is flipped over, planarized, and used as a module for further (possibly multilayer) metalization. The major advantages of metalizing on top of planar ensembles of dies include the fine lithography achievable and very small interconnect parasitics.
• The process can also be generalized to stacks of dies in 3D. In the Irvine Sensors approach, dies with electrically insulated backs are designed with leads fanned-out to contact pads lined up along an edge, and stacked with those side pads aligned precisely. The exposed side is then cleaned and polished, and interconnect lines are metalized onto it. The package is extremely dense, although heat dissipation can become limiting, but leads must still travel out to the edge and back to route to other die.
• In the Cray Computer approach, holes are drilled into a stack of dies in a grid, plated to form a conductive contact with die vias, and then threaded with twisted gold wires, separating the die slightly and supporting them. The package is extremely dense and conceptually trivial, but extremely difficult to manufacture.
• Need for MCM Technology
• In the present art, it costs much more (in terms of power, latency, performance and circuit real estate) to signal off-die than to stay on-die. Off-die signalling designates communicating between two points, on the same die or different dies, using off-die wiring. A major design goal with the present art is therefore to increase the number of circuits on each die, increasing the ratio to expensive off-die interconnections. However, as the size of a die approaches some economically and technologically feasible limit, the probability that randomly occurring manufacturing defects will produce an unacceptable die rises exponentially in a Poisson distribution. Since slightly larger die yield at significantly lower rates, this so-called “surface-to-volume” (communication-to-computation) ratio severely constrains fabrication yield, hence cost per functioning die.
• The need for multichip technologies arises in large part from this inability to produce arbitrarily large semiconductor dies with acceptable yield. Practical limits on die size for a given technology force system designers to partition large digital systems among multiple dies. Unfortunately, such partitioning dramatically impacts system performance since inter-die communication typically inflates packaging costs by tens to many hundreds of percent.
• Multichip technologies also offer the possibility of reducing the cost of intra-die communication in modern CMOS technologies. If off-die signalling became somewhat less expensive than at present, at some point it would become preferable to on-die wiring for certain intra-die communication. The implications of this need to be recognized by practitioners of today's MCM art: Dies should be made smaller, at the expense of more interconnects per non-communication gate. Smaller dies have significantly higher yield and lower cost per area, so if designers could employ cheap inter-die communication on MCMs, they could transparently treat a multichip module as if it were a single enormous ensemble of electronic devices (including their interconnections), using small, very high yielding monolithic integrated circuit dies (and modules of them) as the subunits. This would be tremendously useful for designers and cost-effective for semiconductor foundries.
• Use of MCMs for intra-die signalling should also have important attractions for low impedance—hence low dispersion-signalling between points on the same die. Low dispersion broadcast is essential for clock distribution, and low dispersion point-to-point is useful in general.
• Another application for multichip technologies is in customizing on-die interconnects or engineering change pads. It is valuable to customize interconnects in the manner of a breadboard, within dies up to full wafer-size or among ensembles of dies, in order to rig together specialized functions, test performance, edit out defects, redefine a system's basic connectivity, or perform the functions of engineering change pads. At present, customization generally requires a lengthy (circa six week) logic array masking process, unreliable laser fusing/breaking of wiring junctions, significant expense in microfabrication of wiring, macro-scale (e.g., wire-wrap) assemblages, lack of durability (e.g., hand-wired jury-rigs), or a combination of these. The enormous value of customization can be exploited with properly designed MCMs.
• Inter-die signalling technology trends demand MCM technology. Shorter signal paths directly contribute to higher performance, since they allow shorter delays and faster clock rates (i.e. more operations per second) and wider synchronous instructions (i.e. more operations per clock tick). Dies are generally planar, so the longest signal path will scale as the diameter, or roughly as the square-root of the area. Planar MCMs can be constructed more densely with smaller dies since they ask less wiring overhead for excursions to the edge and back. In principle, the electronic devices a die bears that make it useful could be disposed volumetrically, but in the current art that is uncommon, except for on-die wiring and stacked capacitors. MCMs can in principle be constructed in a space-filling manner, so that the volumetric packing is denser than planar packing, hence signal paths are kept much shorter (e.g., proportional to the cube root of the chip package volume, which is small if the package is thin compared to its area).
• The cost of interconnection networks for signal lines generally scales as a function of area (i.e. a proxy for system real estate) and number of layers (i.e. density and layout efficiency), with linear scaling as a goal. The number of leads or number of chips within an area may inflate cost by a significant factor in many packaging technologies, such as those using wire-bonds. Even if a technology avoids scaling as the number of leads or die, if it requires post-processing steps to form a package, as with solder bump conductive couples, it may still be expensive due to yield losses from handling, amortization of costly test/repair cycles, and of course operating and capital costs.
• Even if a large and a small system have identical numbers of wires starting and stopping in each average square inch, the wires in the larger system will travel further on average, so account for a greater proportion of system density than in the smaller system. This is the well-known “Law of Numbers,” and is a dominant consideration in constraining layout to a minimum number of metalized layers, which generally rises (geometrically) faster than real estate as a chip design grows. Runaway wiring density can in principle choke off the manufacturability of large high performance systems. The requirement for high interconnection density makes MCMs almost inevitable, and rewards volumetric packing as well as the shorter Manhattan distances that accrue to volumetric interconnection.
• A further need for multi-chip technology arises from the cost and complexity of combining hybrid materials to exploit properties of each. For instance, an arbitrarily large, cheap silicon CMOS chip would still lack the speed and optical properties of GaAs, while growing one material on the other is inherently more complicated than forming them separately.
• A further need arises from the enduring value of packaging hierarchies, wherein standardized packages with well-described components, such as microprocessors, are available commercially with various advantages compared to components or full system-level packaging. Important advantages to using such hierarchies may include lower cost, modular upgradeability, well-characterized behavior, and multiple sourcing. The designer balances competing benefits and costs of a hierarchy: necessarily limiting the degrees of freedom of system design, while commensurately reducing complex system interactions and failure modes.
• A further need arises from the difficulty of package manufacture per se. The engineering or manufacturing complexity, process requirements, and cost of the package approach or exceed those of the die, so the cost and turnaround time for the package can become as formidable as those of the die.
• Problems of Present MCM Technologies
• Despite the intensive research efforts of the past several years, present day MCM technologies still have significant problems in terms of cost, performance, design, manufacturability, reliability and reparability, as well as shortcomings with respect to the needs enumerated above.
• Present MCM technologies require significant retooling and/or expensive reorientation of existing integrated circuit (“IC”) fabrication lines. High volume is needed to realize cost advantages of MCM packaging, but re-implementing an existing production system (for which high volume demand already exists) to utilize MCM technology typically requires extensive system-wide redesign. System vendors rationally resist such efforts, choosing instead to implement certain clusters of ICs as application specific Ics (ASICs), which generally involves only local redesign ofLevel 2 boards andLevel 1 chips. Accordingly, the relatively high up-front cost of MCM implementation discourages use of MCM technology for systems where large volume cannot be predicted a priori. (See Balde, J. W., “Crisis in Technology: The Questionable U.S. Ability to Manufacture Thin-Film Multichip Modules.”Proc. of the IEEE, December, 1992.)
• While the electrical performance of TAB, fine-line, or solder bump conductive interconnect can be significantly better than wire-bonding for a conductive interconnect, and pad count is somewhat less constrained, prior art MCM processes often require special processing of wafers or dies prior to assembly to place solder balls or construct metal bonding locations for assembly. The processes also require custom tooling and substantially more sophisticated wafer manufacture post-processing than would be required for standard wafer or packaging lines.
• Reparability and die attach yield issues also arise with current MCM technology, principally because of the difficulty in die removal and replacement. Testing the dies in the MCM before it is fully assembled (and paid for) typically accounts for tens of percent of the delivered MCM cost, due to the need for sacrificial test rigs or time-consuming intermediary connections as well as the cost of compensating for parasitics in order to test at operating speeds. Making physical contact with microscopic probes or rigs of probes is slow, and exposes the probe points to mechanical forces leading to misalignments, fatigue, and wearing. If the dies are packaged beyondLevel 0, only a fraction of the interconnects may even be visible. Nevertheless, working dies should be selected perfectly before module assembly, due to the expense (or impossibility) of reworking the module. Repair methods are very difficult-essentially a tear down and rebuild—and usually require essentially the same elaborate, expensive assembly technology as used to build the MCM originally.
• The difficulties, mechanical constraints and costs of multichip module designs are largely driven by the drawbacks of conductive signalling per se. Conductive signalling is almost universally understood to be inevitable among practitioners of the present art (See Daryl Ann Doane and Paul D. Franzon,Multichip Module Packaging Technology and Alternatives. Van Nostrand Reinhold, 1993). Conductors in general must expose a face in contact to one another; they cannot hide behind shielding or a passivation layer. If there are (re)movable constituents in the electrical path, there will be exposed surfaces, and unless certain materials (e.g., noble metals) are used on exposed surfaces, conductors may be susceptible to oxidation, although some non-metallic conductors avoid the oxidation problems. Troublesome surface chemistry complications include mechanical stability (e.g., whiskering), finite conductance (e.g., charge carrier saturation), and time-dependent material or phase changes (e.g., intermetallic compounds). Non-metal conductors have other chemistry problems. If the entire path is bonded continuously, repair and testing may require physically severing material. Removing conductively mated dies, chips or modules for repair/replacement may entail cutting mechanical linkages and removing solder, metal-metal bonds, pins from sockets, or the like. Such breaks introduce metal fatigue (i.e. increase the likelihood of future failure modes) and contamination by conductive dust. The conductor-conductor junctions will later need to be realigned and restored. Intense localized thermal stresses from (de)soldering may also be involved. Methods for replacement pose severe constraints on the design and manufacture of conductively-coupled components in electronic systems.
• Even if these or other approaches were practical in a manufacturing sense, all suffer the series inductance performance costs of conductive signalling. The performance improvement achievable with solder-bumped or wire-bonded dies is fundamentally limited by the excessive series inductance of the solder bump or wire-bond interconnection. State-of-the-art MCM technologies endure much the same performance limitation as current surface mounted IC technologies. Both families of technology require bonding wires, solder bumps, TAB, or their equivalent, to couple signals conductively between the dies and the substrate (for MCMs) orLevel 2 package (for surface mounted ICs), which impose parasitic series inductance unavoidably. Parasitic inductance of 50-1000 pH is typical, and introduces significant latency, frequency limitations, and power requirements for signalling off-chip. While the best currently available MCM technologies reduce the disparity between on-die and off-die communication compared to the conventionally quoted ten-to-one ratio for surface mounted IC technologies, there remain significant penalties associated with inter-die signalling on conventional MCMs. Current MCM technologies suffer from mechanical and thermal problems, design and fabrication limitations, power costs, complexity, and expense.
• With the present state of the art, technology for conductive signalling is itself directly responsible for many problems with current MCMS. It imposes mechanical and cost limitations on the density, number, and arrangement of signal leads attached to a MCM package or constituent dies in carriers, and generally requires further expense and yet more volume devoted to mounting and interconnecting the leads from chip packages ontoLevel 2 circuit boards or multichip carrier modules and packaging at all higher levels.Typical Level 1 packages holding a die and its conductive leads are generally much larger than the active area of the containedLevel 0 die, due to requirements for spacing out contact pads and attached leads, and MCMs similarly inflate real estate requirements. On each die, fan-out of conductive leads (in order to simplify testing or wiring contact pads) takes up chip real estate. The practical need for a sufficient number of conductive pads, adequately large and well-separated, typically accounts for several tens of percent of the chip's real estate, but can range from a fraction of a percent to essentially all of the chip's real estate. With some MCM technologies, pad bonding locations cannot overlay active circuits, since applied contact probe pressure or other process steps risk damaging the circuits, so pads occupy real estate at the expense of useful electronic devices. That wasted real estate costs money (about $10-1000/square inch). Since present approaches emphasize placing the pads at the perimeter of the die's active area to minimize wiring lengths (or at the center, to minimize thermal expansion mismatch), and pad count is often limited by the available perimeter bonding density of the die, many designs are “pad limited,” wasting chip real estate, and/or “pin-limited,” leading to large, expensive packages. Other leads which can be used as pin-substitutes include solder, eutectic, or soft metal (e.g., gold) bumps, conductive wires (e.g., copper, silver, or aluminum) lithographed on polyimide or tape, and sharpened probe needles.
• Current MCM technologies require special circuitry for interconnecting dies operating at different voltage levels. Dies fabricated from different materials generally use different voltage levels, such as CMOS versus GaAs. Dies designed with different circuit technologies, even in the same material, generally use different voltage levels, such as Si TTL versus Si ECL. Even dies fabricated from the same technology may still use different voltage levels, such as CMOS in silicon at 5.0 volts versus 3.3 volts.
• The circuitry supporting signalling is sensitive to manufacturing variations during wafer fabrication. Manufacturing processes produce variation in basic physical parameters (e.g., impedance, capacitance, inductance) absolutely and spatially, across a wafer or between wafers. This variation causes circuitry to produce waveforms of differing spectra, affecting shape and skew, and differing amplitude, affecting thresholds, noise susceptibility, power requirements and termination characteristics. Unreliable waveform spectra and amplitude means that the yield of dies follows a bell curve distribution, with very few able to operate at higher frequencies. Designers have therefore adopted conservative design rules to compensate for performance ranges.
• Power must still be provided to the electronic devices in a system, whether the devices are packaged in MCMs or any other level. Nearly all leading MCM designs treat power leads and signal leads the same way atLevel 1, which wastes space and loses important opportunities for increasing the density of signal lines.
• In short, there remains a significant need for an improved method and apparatus for coupling signals among modules in a modular electronic system which mitigates one or more of the above-explained problems with the present MCM technology. While MCM technology represents an important application domain, the invention, as defined in the succeeding sections, applies broadly to all levels of electronic packaging and interconnection, such as die, wafer, board, MCM, system, etc. It is therefore the inventors'intent that the invention not be viewed narrowly, or only in the context of the preferred MCM embodiment, except where the context inarguably indicates otherwise.
• Objects of the Invention
• One object of the invention relates to a method and apparatus for interconnecting modules which mitigates one or more of the previously identified deficiencies in the prior art.
• Another object of the invention relates to a method and apparatus for coupling signals among modules which mitigates one or more of the previously identified deficiencies in the prior art.
• Another object of the invention relates to a method and apparatus for coupling signals among modules at very high speeds.
• Another object of the invention relates to a method and apparatus for providing improved inter-module and intra-module signal coupling without the need for costly or exotic fabrication techniques.
• Another object of the invention relates to a modular electronic system having shorter signal paths, and to a method for constructing such a system.
• Another object of the invention relates to a modular electronic system wherein at least some of the signals between two dies, between two substrates or between a die and a substrate communicate via capacitive coupling.
• Another object of the invention relates to a modular electronic system wherein at least some of the signals between two dies, between two substrates or between a die and a substrate communicate via magnetic coupling.
• Another object of the invention relates to a modular electronic system wherein a module couples both capacitively and conductively to other modules.
• Another object of the invention relates to a modular electronic system wherein a module couples both magnetically and conductively to other modules.
• Another object of the invention relates to a module adapted to couple capacitively or magnetically to other modules in a modular digital system.
• Another object of the invention relates to a method and apparatus for integrating modules of physically incompatible materials into a modular electronic system wherein signals couple between nearby, physically incompatible modules via capacitive or magnetic means.
• Another object of the invention relates to a method and apparatus for coupling signals between two or more modules without need for additional means to compensate for differences between the chemistry, conducting state, wave phase, fabrication technology, clock rates, voltage levels, numbers of logical states, thermal expansion properties, operating temperature, ambient pressure, or environmental requirements between components.
• Another object of the invention relates to a method and apparatus for coupling signals from or to a superconducting module.
• Another object of the invention relates to a module comprising a plurality of dies formed by several different fabrication technologies, and a method for assembling and testing such a module.
• Another object of the invention relates to a module comprising more technologies or devices than can be supported economically on a single die, and to a method for designing and laying out such a module.
• Another object of the invention relates to a module wherein the connections between dies, or between a die and a substrate, are easily reversible, and to a method for reversibly forming and breaking such connections.
• Another object of the invention relates to a method of repairing or replacing submodules without tearing down interconnection wiring.
• Another object of the invention relates to an MCM wherein neither the process of installing a die nor the process of removing a die damages either the die or a substrate.
• Another object of the invention relates to a method and apparatus for identifying and replacing undesirable modules in a modular electronic system.
• Another object of the invention relates to a method for improving the performance of a modular electronic system by full speed parametric testing of modules within the system.
• Another object of the invention relates to a method for improving the yield of a modular electronic system by full speed testing of modules within the system.
• Another object of the invention relates to a method for improving yield of an MCM by full speed testing of chips and larger modules prior to assembly.
• Another object of the invention relates to a method for improving the performance of a modular electronic system by selective replacement of modules within the system.
• Another object of the invention relates to an MCM comprising a plurality of known-good modules selected by testing in a test fixture having electrical properties substantially identical to those of the MCM substrate.
• Another object of the invention relates to an MCM wherein modules are selectively replaced in order to correct or improve system performance.
• Another object of the invention relates to modules adapted for full speed testing prior to installation in an MCM, and to a method for testing and installing such modules.
• Another object of the invention relates to an MCM wherein modules are attached to a substrate without use of solder or wire-bonding
• Another object of the invention relates to an MCM or other modular digital system wherein a clock signal is distributed with extremely low dispersion, and to a method of distributing clock signals with extremely low dispersion in such systems.
• Another object of the invention relates to an MCM wherein digital signals are communicated among dies via transmission lines on the substrate, the transmission lines being capacitively or magnetically coupled to the dies.
• Another object of the invention relates to an apparatus and method for coupling a transmission line to a means for capacitively signalling, and to a method and apparatus for terminating such couples.
• Another object of the invention relates to a transmission line connected between two means for capacitively signalling and adapted to provide a capacitive path between two separate modules.
• Another object of the invention relates to a method and apparatus for connecting two separate modules capacitively or magnetically by coupled transmission lines.
• Another object of the invention is a modular electronic system having improved signal-to-noise and higher speed transmission of digital signals between modules.
• Another object of the invention relates to a die wherein a substantial number of intra-die connections are routed via off-die signal paths, and a method of designing such a die.
• Another object of the invention is a modular electronic system having improved routing, busing, networking, or switching of signals among constituent modules.
• Another object of the invention relates to methods for laying out modular electronic systems—entailing optimizing interconnections among devices, circuits, blocks, and modules of any size—and to systems so implemented.
• Another object of the invention relates to design of a modular electronic system partitioned into a number of small, high-yielding dies and low cost, high performance inter-die connections.
• Another object of the invention relates to a module wherein the negative effects of misalignments from design, manufacture, assembly, or operation are substantially mitigated, and a method of isolating those effects.
• Another object of the invention relates to a method of designing, manufacturing, assembling, or operating a modular electronic system which removes the need for long-range physical alignment.
• Another object of the invention relates to a method and apparatus for self-aligning a modular electronic system electrically or logically.
• Another object of the invention relates to a method and apparatus for providing low-cost, defect-tolerant wafer-scale interconnections in a modular electronic system.
• Another object of the invention relates to a method for operating a wafer-scale module in spite of fabrication defects.
• Another object of the invention relates to a method and apparatus for reducing the number of packaging hierarchy levels needed for a modular system of a given complexity.
• Another object of the invention relates to a modular electronic system supporting many hierarchical layers of modules.
• Another object of the invention relates to a modular electronic system using standardized interconnections to serve as a building block for larger systems.
• Another object of the invention relates to a method and apparatus for standardizing the assembly of, and the interfaces between, modules in a hierarchical modular electronic system.
• Another object of the invention relates to a method and apparatus for reducing the cost of package engineering and manufacturing for modular electronic systems.
• Another object of the invention relates to a method and apparatus for reducing the turnaround time and cost for breadboarding or prototyping a modular electronic system.
• Another object of the invention relates to a method and apparatus for assembling a modular electronic system from a standardized kit of parts.
• Another object of the invention relates to a method and apparatus for reducing the turnaround time and cost for engineering and manufacturing the package for a modular electronic system.
• Another object of the invention relates to methods for designing, manufacturing, assembling, and operating systems which are faster, lower-powered, cheaper, or have more logical elements, and to systems so implemented.
• Another object of the invention relates to methods for designing, manufacturing, assembling, and operating systems which have preferable packages (e.g., denser, smaller, lighter, tolerant of planar or volumetric constraints), and to systems so implemented.
• Another object of the invention relates to modular electronic systems with a high density or number of signal interconnections, and to methods for constructing such systems.
• Another object of the invention relates to manufacturing and design methods for scavenging more real estate on wafers for electronic devices than results from methods used in the prior art, and to systems so implemented.
SUMMARY OF THE INVENTION
• One or more of the above objects, as well as other objects and advantages, are provided by the modular electronic system of the invention and the many uses thereof.
• Briefly, one aspect of the present invention relates to a modular electronic system comprising at least two modules, at least one of which is powered, and a means for non-conductively (e.g., capacitively or magnetically) communicating between the modules.
• Another aspect of the invention relates to an MCM wherein two or more dies communicate via non-conductive signalling means. The MCM preferably comprises a substrate having a plurality of dies affixed thereto. The means for non-conductively signalling preferably comprises coupled half-capacitors, one half-capacitor plate located on a die and the other located on the substrate or another die, the die and the substrate or other die positioned so as to couple the half-capacitor plates.
• Yet another aspect of the invention relates to a method and apparatus for assembling, testing and repairing modular electronic systems or MCMs based on the non-conductive signalling techniques of the invention.
• Still a further aspect of the invention relates to a modular electronic system or MCM wherein interconnections between elements are non-conductive and, therefore, do not require direct physical contact. The modular system or MCM preferably utilizes the flexibility of the non-conductive junctions to accommodate or dissipate thermal stresses within the system.
• A further aspect of the invention relates to a die, chip or module adapted for assembly in a non-conductively interconnected modular system, and including an externally accessible half-capacitor plate and, preferably, a plurality of electronic devices.
• Yet another aspect of the invention relates to an MCM comprising a plurality of heterogeneous elements, each affixed to a substrate and having means for non-conductively coupling signals to transmission lines or wires in the substrate.
• Still other aspects of the invention relate to means for non-conductively signalling between electronic elements and the use of various dielectric, adhesive, elastomeric and other materials in combination therewith.
• Yet another aspect of the invention relates to a method for increasing the manufacturing yield and reducing the cost of a modular electronic system by assembling high yield known-good units to form a modular system utilizing non-conductive interconnections and, if necessary, to repair or optimize the system by replacing selected elements.
• Further aspects of the invention relate to methods for differentially signalling between elements in a modular system via non-conductive means, and to receiver and transmitter circuits for practicing such signalling methods.
• According to a still further aspect of the invention, very high frequency signals are communicated between modules by transmission line coupling. Such coupling is preferably implemented with parallel or perpendicularly overlapping microstrip or stripline segments. Modular microwave and millimeter wave systems are preferably assembled to communicate via transmission line coupling.
BRIEF DESCRIPTION OF THE FIGURES
• The above, as well as other, aspects of the invention are explained in the detailed description below, which describes the various preferred embodiments of the invention and is intended to be read in conjunction with the set of figures, in which:
• FIG. 1 depicts an exemplary portion of one embodiment of a modularelectronic system1 in accordance with the invention;
• FIG. 2 depicts an exemplary portion of a modular electronic system wherein a signal is differentially coupled from a pair of transmission lines on a substrate, across a pair of coupled half-capacitors, and to a pair of transmission lines on a die;
• FIG. 3 is a partially exploded view of an exemplary portion of a modular electronic system which includes both capacitive and conductive interconnections between a die and a substrate;
• FIG. 4 depicts an exemplary portion of an MCM constructed in accordance with the present invention;
• FIG. 5 depicts an exemplary portion of an MCM similar to that ofFIG. 4, but including a power substrate, distinct from the signal substrate, as well as a heat sink;
• FIG. 6 depicts exemplary waveforms for a digital (or multi-state) signal capacitively coupled from a die to a substrate in accordance with the invention;
• FIG. 7 depicts an exemplary portion of a modular electronic system in which a die capacitively interconnects (“bridges”) two substrates;
• FIG. 8 depicts an illustrative portion of a non-planar modular electronic system that includes multiple dies and substrates;
• FIG. 9 depicts and exemplary portion of a large-scale modular system constructed in accordance with the invention which utilizes a so-called “leapfrog” geometry;
• FIG. 10 depicts an illustrative portion of a “leapfrog” integrated system constructed entirely from dies;
• FIGS.11A-B depict the effect of misalignment between capacitor plates in a capacitively coupled modular system;
• FIGS.12A-B depict the use of an oversized half-capacitor plate to minimize the impact of misalignment in a capacitively coupled system;
• FIG. 13 depicts a modular electronic system wherein a plurality of (optionally superconducting) electronic devices on implemented on a flexible substrate;
• FIG. 14 depicts an illustrative portion of a large-scale, heterogeneously integrated modular electronic system constructed in accordance with the invention;
• FIG. 15 depicts the use of a uniform capacitive interface to couple modules of varying sizes and packaging levels in accordance with the invention;
• FIG. 16 depicts a modular system in which a plurality of modules capacitively couple to a backplane;
• FIGS.18A-C depict several steps in the assembly of an MCM (such as those ofFIGS. 4-5) in accordance with the invention;
• FIG. 19 depicts a capacitive signal path including means for adjusting overall gain;
• FIGS.20A-C depict an exemplary application specific module (ASM) constructed in accordance with the invention;
• FIG. 21 depicts a cable interface utilizing the capacitive coupling techniques of the invention;
• FIG. 22 depicts an exemplary portion of a module electronic system wherein a substrate provides both inter-chip and intra-chip capacitive signal paths;
• FIG. 23A depicts an exemplary apparatus for testing modules constructed in accordance with the invention;
• FIG. 23B depicts a wafer-scale modular system constructed in accordance with the present invention;
• FIGS. 24-25 compare the fabrication of MCMs in the present art (FIG. 24) and the fabrication of wafer-scale modules in accordance with the invention (FIG. 25);
• FIG. 26 depicts the general flow of assembly, testing and rework of a wafer-scale or similar modular system in accordance with the present invention;
• FIG. 27 shows the yield of good dies in a typical semiconductor fabrication process as a function die size;
• FIG. 28 depicts several plots of system cost as a function of die size for an exemplary capacitively interconnected system;
• FIG. 29 depicts the flow of a design cycle for a modular system constructed in accordance with the present invention;
• FIG. 30 depicts a block diagram of an off-die capacitive signal path in accordance with the invention;
• FIGS.31A-B depict a final stage of a preferred embodiment of a transmitter;
• FIG. 32 depicts a preferred embodiment of a switching means adapted to drive differential capacitive signal paths;
• FIG. 33 depicts a simplified electrical model of the interface between an exemplary switching means and a terminated substrate transmission line;
• FIG. 34 depicts a simplified electrical model of the interface between an exemplary switching means and an unterminated substrate wiring line;
• FIG. 35 depicts a block diagram of a preferred, differential off-die signal path;
• FIG. 36 is a schematic diagram of an exemplary single-ended receiver compatible for use in the signal path depicted inFIG. 30;
• FIG. 37 is a schematic diagram of an exemplary differential receiver compatible for use in the signal path depicted inFIG. 35;
• FIGS.38A-B illustrate non-conductive signalling between modules by means of parallel and perpendicular coupled transmission lines.
DETAILED DESCRIPTION OF THE INVENTION
• Reference is now made toFIG. 1, which shows an exemplary portion of one embodiment of a modularelectronic system1 in accordance with the invention. As depicted,system1 comprises asubstrate10, adie11 and ameans13 for capacitively signalling, which provides a capacitive signal path betweensubstrate10 and die11. Means13 for capacitively signalling comprises two electromagnetically communicating regions illustratively depicted as “half-capacitors”14 and15. A dielectric17 is preferably used to partly or totally fill the gap between half-capacitors14 and15.Dielectric17 may be employed to increase the capacitance of capacitive signalling means13, to provide passivation fordie11 orsubstrate10, to enhance the thermal conductivity betweendie11 andsubstrate10, and/or to mechanically bond or supportsubstrate10 and die11.
• As depicted inFIG. 1, die11 also illustratively includes a plurality ofelectronic devices12, implemented on an active surface of the die, and a means for powering the die, shown as aconductive contact16 on the surface of the die.
• Die11 is preferably a digital integrated circuit fabricated by a conventional, low cost process.System1 preferably includes additional means for capacitively signalling and additional dies, as depicted in later Figures. For example,system1 could comprise an MCM. In such a system, die11 might be a CMOS silicon die withactive components12 fabricated on its surface, while thesubstrate10 might be a die fabricated of GaAs.
• As will be discussed at length below, use of capacitive signalling means13 in such a system offers considerable advantages over conventional alternative approaches. Conventional approaches, such as packaging the dies independently and wiring them together conductively, or growing a crystal of one material on a region of the other, are significantly more expensive for comparable performance than comparable systems constructed in accordance with the present invention. Moreover, such conventional systems generally suffer low yields and low operational reliability, due to thermal expansion stresses.
• Die11 preferably includes one or moreconductive contacts16 for powering thedevices12. For a conventional silicon CMOS die11, contact16 illustratively is a metallic aluminum region fabricated lithographically on the silicon CMOS die. In some applications, it may be desirable to gold-plate the outer surface of the aluminum region to protect against corrosion. Illustratively, power could be delivered by means of a metallic fuzz button conductively connected tosubstrate10 or to a different substrate. Fuzz button power connections, in combination with coupled half-capacitor signal connections14 and15, allow die11 to be successively mounted and removed fromsubstrate10 non-destructively, which affords significant advantages in terms of testing, repair or performance optimization, as compared to destructive approaches. Such non-destructive mount/remove cycles entail substantially zero forces, torques, or changes in energy across themeans13 for capacitive signalling.
• Substrate10 is preferably constructed from a low dielectric factor material in order to facilitate high speed signalling. The substrate can be either passive or active. An example of a passive substrate is a printed wiring board wherein half-capacitor15 would preferably comprise a lithographically defined copper region ofsubstrate10. An example of an active substrate is a collection of active and/or passive semiconductor devices and wiring. Typical active elements are set forth below.
• Superconducting substrates are particularly well-suited to use coupled half-capacitors, due to the absence of any metallic contact betweendie11 and thesuperconducting substrate10. Materials problems which might otherwise arise in soldering or bonding to brittle high temperature superconducting ceramic layers are eliminated. All of the propagation advantages of superconducting transmission lines (especially low dispersion) are retained. While termination resistor fabrication is potentially challenging, this can be done with localized heating of a line above the transition temperature, or by application of sufficiently intense DC magnetic fields to force the superconductor normal in the resistive region. Such techniques may also provide a means for tuning the value of the termination resistance in a post-fabrication step by adjusting the current which heats or induces the magnetic field.
• Illustratively, half-capacitors14 and15 comprise conductive plates. Die half-capacitor14 may comprise an aluminum metal pad lithographically defined on a CMOS silicon integratedcircuit11. Adequate capacitance can be realized using the upper level of the integrated circuit die metalization to form half-capacitor14, and a matching plate on ahigh density substrate10 as the other half-capacitor15.
• Coupled half-capacitors can be formed in arrangements, like13, which couple one-to-one, or alternatively in structures which couple one-to-many or many-to-many. The circuitry for sending and receiving signals via coupled half-capacitors can advantageously fit underneath or beside the corresponding half-capacitor; the plates themselves therefore need not be lithographed at the expense of usable chip real estate. Using present technology, the invention provides significant advantages in terms of the area required to implement high performance inter-die signalling circuits.
• Coupled half-capacitors are preferably formed by affixingdie11 tosubstrate10 such that half-capacitors14 and15 effectively overlap. Importantly, use of coupled half-capacitors instead of conventional chips first, soldered, or wire-bonded connections substantially relaxes the need for high precision alignment. Indeed, the size or shape of half-capacitors14 and15 need not be identical, and can be advantageously optimized to mitigate the effects of anticipated misalignments, such as those from manufacturing tolerances, assembly errors, or thermal mismatch.
• Circuitry for coupled half-capacitors can cost substantially less chip real estate per bit than connections for conductively coupled off-die signalling. The driver and receiver circuits are small enough to be fabricated underneath the corresponding half-capacitor plate. The plates are smaller than contact pads and fan-out leads in the present art, and can overlay other active circuitry, including circuitry unrelated to I/O. Plates can be packed much more densely than pads, and can occupy any part of the surface of the die rather than predominantly the perimeter.
• A coupled half-capacitor of approximately 1 picofarad can be realized with aplate 70 microns by 70 microns metalized on top of the die, assuming the dielectric factor of the separator is K=112 (TiO₂ceramic) and the plate separation is 3.5 microns. This represents an I/O surface density of approximately 10⁵equivalent leads per square inch. Alternative dielectric materials, such as oriented barium titanate crystals which have a dielectric factor more than 100 times larger than titanium dioxide, may further increase the I/O density. However, once the half-capacitor plate size becomes smaller than the driver/receiver circuitry, and the die is circuit-limited rather than I/O pad-limited, there is little point in further reducing the size of the plate except in certain scenarios: where they comprise multiple platelet regions ganged together electrically and logically; where they are used for AC powering; or where other surface accessible functions like optical transmitters/detectors compete for surface space.
• Standard semiconductor processes allow active circuitry to be implemented beneath a metal half-capacitor plate. In contrast, implementation of active circuitry beneath conductive contact pads is undesirable because mechanical forces imposed on the contact pad can irreversibly alter the electronic behavior of the underlying domain.
• Half-capacitors14 and15 need not be constructed of the same material and can even be formed from chemically incompatible materials. The half-capacitors can be formed by any means from any uniform or composite material(s)—including metals, superconductors (including High T_c), semi-conductors, semi-metals, impedance-varying or conductance-varying materials—in single or multiple layers or domains. They can be built in any state or form of matter-including gases, gasified material, liquids, liquified material, solids, solidified material, composite, slurry, gel, suspension, matrix, or others.
• Since the conductance, or charge-carrying capability, of the half-capacitor plates is typically so high, the effective plate-to-plate overlap and separation, and the properties of dielectric17, largely determine the performance of the coupled half-capacitors13. Methods for forming capacitor plates are well-known in the microelectronics industry, although their application to MCMs and other modular interconnection schemes is not recognized in the prior art.
• The inter-plate gap between half-capacitors14 and15 may be filled with a high dielectric material or left empty (with air or vacuum). Illustratively, the dielectric17 is a high dielectric factor fluid, such as ethylene glycol.Dielectric17 can be disposed between a series of passivation layers on each plate, if any, and intervening insulator, if any. Such passivation layers can be formed cheaply and reliably from oxides of the metalization layer and/or from applied emulsion coatings such as titanium dioxide, crystalline barium titanate oriented along the high-dielectric axis, and other materials which are well-known in the electrochemical community and commercially available.
• The known corrosion properties, cleanliness, and viscosity as a function of temperature of the dielectric are crucial, as is its dielectric factor. Colloidal suspensions of high dielectric factor solids (e.g., ceramics like titanium dioxide applied in a slurry) can be used as the filler material, as well as a corrosion-free low dielectric factor liquid, such as the Fluorinerts. The dielectric may be an insulating liquid, such as, ethylene glycol, propylene glycol, glycerin, or nitrobenzene; a gas, such as uranium hexafluoride; or a mixed-phase suspension or gel, in order to exploit heat transport properties, cleanliness, or mechanical stability. The hygroscopic nature of these materials remains a major concern, and some (lower dielectric factor) hydrophobic materials, such as 1,1,1 trichloroethane may therefore be preferable. Hygroscopy and temperature-dependent sensitivity to vibration and to changes in viscosity or dielectric factor complicate implementation, but their treatment is well-known to those skilled in the art. Some processes for applying coatings also require a polishing step.
• Liquid dielectrics may be useful in certain systems, in part because no sheer force is transmitted if the chips have to be moved, and in part because the dielectric is scratch proof. Spacers on the chip surface could assure constant separation in the presence of a high dielectric liquid like glycerol but are not required unless shorting is a risk. In the case of liquid dielectrics, surface tension may be sufficient to hold the die and substrate against one another.
• Intervening insulator(s), if used, should generally slide on at least one side to accommodate thermal expansion, and may either be bonded to one of the dies, substrates or modules, as with an adhesive, or maintained as a monolithic material bonded to neither, such as Cellophane. Bonding to both surfaces should be avoided, except with liquid dielectrics, elastomers, or systems carefully engineered to accommodate thermal expansion.
• Temperature and frequency dependence of the dielectric factor remain a significant engineering consideration. Nearly all known materials exhibit complicated frequency dependence on their dielectric factor, which can effectively limit their range of use in coupled half-capacitor gaps. Of course, dry air or a vacuum represent.exceptions to this rule, but their low dielectric factor and compressibility make them less attractive.
• In the MCM design community, high dielectric materials are shunned in general, since signals propagate through wiring at less than the speed of light by a factor of the square-root of the surrounding dielectric factor. (See, for example, Robert Johnson, Robert Teng, and John Balde, “Multichip Modules Systems Advantages, Major Constructions and Materials Technologies.”IEEE Press,1991.) Contrary to the teachings in the MCM community, high dielectric materials have a central role in the present invention.
• One disadvantage with high dielectric materials, however, is that they cause the electric field to spread outside the region immediately between opposing half-capacitors, necessitating greater spreading out of the placement of neighboring plates. In this regard, lower dielectric material would allow tighter packing of neighbors, but at the cost of a commensurate increase in voltage or plate area, or a decrease in plate separation, with accompanying complexities. For a system with a uniform dielectric and a plate dimension much greater than the separation, neighboring plates can be packed with a separation approximately twice the vertical separation without encountering significant cross-talk problems. A patterned dielectric, with lowered dielectric factor outside of the gaps between opposing plates, would allow slightly tighter areal packing of plates, but increase the manufacturing complexity. Dielectrics may also be engineered to enable/disable capacitance above a threshold between paired half-capacitors. Extremely low cost processes from the painting/copying industry (e.g., lithography) are well-known for fabricating printed wiring boards, and can analogously be employed to implement the dielectric pattern, effectively defining interconnections.
• While a pair of coupled half-capacitors represents the preferred means for capacitively signalling, other such means are available. For example, multi-storey capacitively coupled structures can be constructed wherein one or more intervening conductive regions are disposed (preferably in the dielectric region) between two terminal half-capacitors, thereby effectively forming a series connected chain of two or more capacitors. Importantly, such multi-layer structures do not require any overlap between the two terminal half-capacitors. For example, in a three-layer structure, overlap only between the bottom plate and a middle conductor and between the top plate and the middle conductor suffices to implement a means for capacitively signalling between the top and bottom plates. Printing/copying processes (e.g., xerography) may be applicable for forming such structures cheaply.
• Reference is now made toFIG. 2, which depicts an exemplary portion of a modular electronic system wherein a signal is differentially coupled from a pair oftransmission lines32 and33 onsubstrate10, across a pair of coupled half-capacitors13, to a pair oftransmission lines34 and37 ondie10. A means for terminatingtransmission lines32 and33 is provided onsubstrate10 bytermination resistors31band31c, connected to a ground plane byline31a.
• Advantageously,transmission lines34 and37 connect to on-die driver and/or receiver circuits, implemented as some of theactive elements12 ofdie11.Transmission lines32 and33 may couple to driver or receiver circuits implemented onsubstrate10, to other chips, modules or substrates via additional coupled half-capacitors, or to external conductive leads.Transmission lines32 and33 are preferably implemented as microstrip, stripline or slotline transmission lines. Of course, ordinary unterminated metal or polysilicon interconnect will suffice in certain applications.
• Reference is now made toFIG. 3, which shows a partially exploded view of an exemplary portion of a modular electronic system which includes both capacitive and conductive interconnections betweendie11 andsubstrate10.Conductive contact pads45,46 and47 illustratively receive power, ground and a plurality of I/O signals fromsubstrate10. Conductive connection is achieved by affixingdie11 tosubstrate10 such that conductive connection means41,42 and43 touch theirrespective contacts45,46 and47. Importantly, this permits non-destructive removal of die11 fromsubstrate10, thereby providing substantial fabrication, testing and repair advantages for modular systems based on the present invention.
• Conductive connection means preferably comprise metallic fuzz buttons, such as the Cinapse™ contacts manufactured by Cinch, Inc. The holes insubstrate10 which accommodatemetallic fuzz buttons41,42,43 are preferably conically tapered to ensure that the buttons do not interfere with seating of die11 againstsubstrate10. Alternative conductive means include solder balls, wiring on tape or films, bonded wires, contact probes, liquid metals such as mercury, low melting point metals such as gallium, and other high conductance media such as may be appropriate for particular engineering situations.
• Advantageously, the conductive connections can be customized by impairing (e.g., removing, eroding or destroying) or adding selected fuzz-buttons, thereby disabling or enabling conductive connection between the selected fuzz-buttons and their respective contact pads.
• Typically, theconductive connections45,46,47 can be much larger than the capacitive couplings, since power feeds are rarer than signal feeds and benefit from the lessened resistance of bigger leads. Conductive contacts are however relatively very expensive and failure prone compared to coupled half-capacitors. It is therefore advantageous to employ the relatively coarse technology of conductive contacts only for DC power leads, while retaining the invention's advantages of many high bandwidth capacitively coupled paths for signalling.
• Contact pads for fuzz buttons occupy 0.3 mm²each in the preferred embodiment. In contrast, circuitry to drive the capacitive couplings occupies about 0.04 mm²per half-capacitor, based on a 1 micron CMOS process. Multiply-redundant contact pads may advantageously compensate for the relatively low reliability of mechanical contacts.
• Reference is now made toFIG. 4, which depicts an exemplary portion of an MCM constructed in accordance with the present invention. The MCM comprises asubstrate10, afirst die11 and asecond die61. (First die11 and second die61 could also represent different parts of a single die.) A terminatedtransmission line33 interconnects half-capacitors15 and65 onsubstrate10, thereby providing a signal path between half-capacitor14 on thefirst die11 and half-capacitor64 on thesecond die61.
• Transmission line33 provides efficient communication between half-capacitors65 and15. Importantly, the power required to couple a signal from half-capacitor64 (on second die61) to half-capacitor14 (on first die11), viatransmission line33, is substantially independent of thelength51 oftransmission line33. For communications beyond a reasonably short distance, the off-die route (via transmission line33) offers potentially superior performance to an equivalent on-die routing of the same signal. By contrast, in prior art MCMs based on conductive signal interconnections between die and substrate, the off-die coupling of signals via the substrate involves substantial performance penalties due to the excessive parasitic inductance of the conductive die/substrate connections. Transmission lines and unterminated wiring both, however, suffer limitations from the integrated net series resistance over a path and the present invention cannot avoid this problem unless a shorter wiring path can be achieved. Modem signal coding techniques such a those discussed below are useful in confining this problem to one of power attenuation rather than signal dispersion.
• First die11 and second die61 also includeconductive contact pads52 for, among other things, receiving power fromsubstrate10.
• Advantageously, dies11 and61 can be formed from different or incompatible materials. For example, first die11 could comprise a low cost CMOS integrated circuit while second die61 could comprise a GaAs integrated circuit, including optical I/O devices. Thespacings50 and60 between the dies and substrate can be the same or different, as needed to accommodate the electronic or physical requirements of dies11 and61. Likewise, similar ordifferent dielectrics17 and67 can be used in the half-capacitors that couple dies11 and61, respectively, tosubstrate10.
• The tolerance of the capacitive interconnections to thermal expansion represents an important advantage of the invention over conventional MCMs. Where dies11 and61 represent different portions of a large (e.g., wafer-scale) die, the misalignment errors due to thermal expansion mismatch accrue and can become particularly troublesome. In conductively coupled prior art MCMs, thermal expansion problems often require that the substrate material be selected to match, as closely as possible, the thermal expansion properties of a particular die material, typically silicon. In contrast, the present invention permits integration of heterogeneous dies or modules on a common substrate with substantially relaxed thermal expansion constraints, at least for dies of modest size. To increase the tolerance of the capacitive interconnection betweendie61 andsubstrate10 to misalignments and/or thermal expansions, half-capacitor65 is advantageously sized larger than half-capacitor64.
• For large die size, matching the thermal expansion of thesubstrate10 to the thermal expansion of the die becomes a major consideration. Sheer stresses will be induced in the dielectrics unless the large size die and the substrate can slide along their adjacent surfaces or have well matched coefficients of thermal expansion. In addition, unsupported standard wafers bow as much as 50 microns with a 70° C. rise in temperature, suggesting a requirement for edge-to-edge thermal control to within a few degrees Celsius, unless the mating of the substrate, the dielectric, and the large size die is compliant.
• Reference is now made toFIG. 5, which depicts an exemplary portion of an MCM similar to that ofFIG. 4, but including apower substrate70, distinct from thesignal substrate10, as well as aheat sink81.
• Power substrate70 preferably includes aground plane74,power plane75 anddecoupling capacitance79 coupling therebetween. A plurality of terminals represented aselements76,77 and78 onpower substrate70 are conductively connected to respective terminals on die11 and provide ground, power and a plurality of external I/O leads, respectively, to die11.
• Ground plane74,power plane75 and the power/ground wiring associated therewith are preferably fabricated using a low resolution lithographic process.Power substrate70 might alternatively be formed from materials, such as metals, ceramics, organic polymers, silicon, inorganic polymers, glasses, amorphous solids, crystalline solids, polycrystalline solids, composite materials, heterogeneous multilayer materials, or others.Power substrate70 may be made of rigid or flexible material, depending upon the application, and may include holes or similar patterned features to accommodate, for example, an external optical I/O path83 from/to an optical I/O device82 onchip85. Advantageously, external power, ground and I/O terminals71,72 and73 onpower substrate70 can be wire-bonded to appropriately configured pins or other external leads, thereby providing aconventional Level 1, 1½, 2, or 3 package lead-out as appropriate.
• The conductive connections betweenterminals71,72 and73 onpower substrate70 and similar terminals onchip11 can be fabricated by reflow-solder, liquid metal or other conductor, metallic fuzz buttons, pin & socket, other mechanically compliant reversible or irreversible contacts, or by wire-bond or other welding techniques.
• Alternatively, power could be provided to die11 or other dies using integrated batteries or high frequency alternating current instead of direct current, thereby dispensing with any requirement for conductively coupled contact points altogether. Use of optical or near-optical frequencies to couple power is preferable since it increases transmission efficiency, and polarizing the radiation decreases crosstalk interference with other signal interconnections, notably including on-chip circuitry. Advantageously, essentially all of a die's surface, except that employed for I/O (e.g., signal coupling half-capacitors, semiconductor lasers, photodetectors and the like), can be used to receive power.
• FIG. 5 also depicts a means for dissipating heat from dies11 and85 comprising aheat sink81 and athermal elastomer80 positioned betweenheat sink81 and dies11 and85.Thermal elastomer80 preferably comprises a material having high thermal conductivity, such as organic elastomers filled with boron nitride. Alternatively, other known methods for effecting good thermal contact between dies11 and85 andheat sink81 can be used, including maintaining the dies and heat sink in intimate physical contact, employing thermally conductive greases or epoxies, soldering or eutectic bonding of the heat sink and dies, and the like.
• It ordinarily does not matter whether the thermal elastomer is an electrical insulator or conductor, unless dies operating at different bulk potentials share the same heat sink. But for this or other reasons an electrical insulator may be preferred. Ceramic loaded elastomers containing aluminum oxide or (preferably) boron nitride are one such approach.
• Heat sink81 can be cooled with any standard cooling technique, including conductive, convective, or radiative transfer, or by phase change of a suitable material, as in a heat pipe.
• Preferably,heat sink81 also comprises apressure plate84.Force84 applied throughpressure plate81 assures intimate contact between dies11 and85 andsubstrates10 and70, as well as dies11 and85 andheat sink81. Positioning the dies and substrates relative to one another by means of pressure offers important advantages with respect to building and unbuilding the MCM assembly reversibly (e.g., for repair, inspection, testing, performance enhancement, and so forth). Advantageously, use of areformable elastomer80 further contributes to these reversibility advantages.
• If twodistinct substrates10 and70 are used to carry wiring and power, thermal expansion should be matched. However, this constraint is no different from that faced in conventional MCM designs, which commonly employ distinct power and signal substrates. Accordingly, appropriate techniques for matching or compensating for the thermal expansions ofsubstrates10 and70 are known to those ordinarily skilled in the art.
• Construction of the interconnects onsignal substrate10 is extremely rapid and amenable to mass production. The signal lines insubstrate10 may advantageously be fabricated as microstrips, striplines or slotlines. In general, more layers will be required onsubstrate10 as the size of substrate10 (and the number of or size of die(s)) grows to compensate for the typical geometric increase in wire density. Advantageously, the superior electrical characteristics of the lithographically formed transmission lines onsubstrate10 allows use of longer-than-minimal routing paths in order to relieve layout wiring congestion. Assuming that the resistors terminating a transmission line are lithographically fabricated, the cost ofmanufacturing substrate10 is approximately linear as a function of area or at worst is a slowly growing geometric function. Importantly, the manufacturing cost is not proportional to the number of signal interconnections between substrate and die(s), as is the case for the many conventional contact junctions needed for conductive signalling. The cost of the remaining components-power substrate70,heat sink81,pressure plate84, andthermal elastomer layer80—scale approximately linearly with area, also independent of the number of die/substrate signal interconnections.
• FIG. 5 also shows two exemplary passivation layer configurations for dies11 and85. Passivation layer86 (covering die11) and layer87 (covering die85) are preferably thin, protective layers of silicon dioxide, silicon nitride, polyimide, borosilicate glass or similar materials. Illustratively,passivation layer87 covers the entire active surface ofdie85, whilepassivation layer86 is patterned and does not cover half-capacitor plate14. Common passivation materials have significantly lower dielectric factors than many of the materials contemplated for use in the preferred embodiment ofdielectric17. Accordingly, the passivation materials preferably should not be used on the half-capacitor plates and, if used, should have as high dielectric factors as practical.
• Passivation layer86 is patterned by lithographically defining regions of the passivation layer to be removed during the chip fabrication process, then selectively removing such regions. Advantageously, the lithographically defined region can be smaller than the half-capacitor plate, so as to provide a border region surrounding the half-capacitor wherein the passivation layer partially overlaps the plate. With only a portion of the half-capacitor plate14 exposed without passivation, the influx of chemical contaminants is effectively discouraged.
• In contrast,passivation layer87 covers half-capacitor plate64. This approach typically reduces the available capacitance of the coupled half-capacitors, but eliminates possible contamination problems at the exposed plate boundary.
• In either passivation layer configuration, theconductive contacts52 will be unpassivated, as in current fabrication technologies. Advantageously, noble metal contacts may be useful for avoiding oxidation or corrosion.
• FIG. 5 also illustrates why manufacturing and construction of a modular system in accordance with the invention is simpler than current techniques for forming a modular system with a comparable number of signal interconnections and similar density. The exemplary fabrication of half-capacitor plates and passivation layers employs standard semiconductor fabrication techniques, and the dielectric can itself be the adhesive holding the modular system together, while the heat sink and thermal elastomer are largely standard components. Similarly, the substrate (and power substrate, if employed) can be easily mass-produced with current techniques. While some changes will be required in the CAD layout tools, the flexibility to locate half-capacitor interconnect plates directly above other signal lines on the dies simplifies the layout problem.
• Reliability of the interconnects is extremely high if theFIG. 5 embodiment is employed. Fuzz buttons are comparable to pressed gold wires and TAB in assuring reliable electrical contact, although neither is as mechanically robust as solder or a lithographed metal junction. Fuzz buttons are at present, however, the easiest and most reliable of those conductive contact means to handle through repair/replacement cycles. The plurality of means for capacitively signalling avoids, by design, significant amounts of thermally induced sheer stress, assuming the dielectric is properly selected. In any case, the heat stress and forces perpendicular to the surface will likely be significantly lower than in conventional modular systems in which conductive bonds are pushed or soldered together.
• Using capacitive couplings for signalling means that many fewer conductive contacts are needed (e.g., only for power, ground, or off-module I/O). Therefore, the Poisson statistics for anticipating failure need only account for a fraction of the number of mechanical components, as compared to conductively coupled signalling systems. Generally, important failure modes in MCM-level modular systems include solder deformities, electromigration, and metal-metal corrosion. Typical processes for assembling MCMs involve repeated installation and removal of dies from the module. During these steps, failure modes associated with the conductive connections to dies represent the dominant factor diminishing the yield or reliability of the manufacturing process. Such yield loss correlates strongly with the number of conductive connections. Therefore, the present invention, which employs dramatically fewer conductive connections than conventional MCM technologies, enjoys considerable advantages in terms of manufacturing/assembling yield, reliability, reparability and cost. These simplifications also benefit the design and operating cycles.
• Reference is now made to FIGS.6A-D, which depict exemplary waveforms for a 3-state digital signal coupled from die11 to die85 ofFIG. 5 and appropriately supported by 3-state logic.
• FIG. 6A shows a typical digital waveform in a high speed digital system. The waveform consists of a series of states, the states being defined by the level of the waveform during aparticular clock period29. For example, duringperiod21, the waveform represents a digital “1”. Inperiod22, it represents a digital “0”. In the case of a 3-state system, the waveform duringperiod23 represents a state designated “−1”. The waveform is generated by a digital circuit constructed fromdevices12 ondie11.
• Before transmission across capacitative signalling means13, a transmitter circuit on die11 converts the digital waveform ofFIG. 6A into a pulsed waveform as depicted inFIG. 6B. Note that in this example, the 3-state transmitter suppresses the trailing edge transition, which a 2-state (binary) transmitter advantageously would not bother doing. The exemplary pulsed waveform generated by the transmitter includes apulse24 representing the logic state during a particular portion of the clock period. The pulsed waveform is applied to half-capacitor14, from which it travels to half-capacitor64 by way of a first half-capacitor15, terminatedtransmission line33, and second half-capacitor65 onsubstrate10.
• FIG. 6C shows the pulsed waveform received on half-capacitor64, lagged in time due to the finite speed of transmission over the electrical distance from half-capacitor14, and attenuated by the non-zero series impedance of the electrical path. At high speeds, the transmission of pulsed waveforms between half-capacitor plates14 and64 achieves much better signal-to-noise ratio than would a conventional conductive interconnection operating at the same power.
• The signal received by half-capacitor64 is coupled to a receiver circuit ondie85. This receiver circuit generates the digital output signal depicted inFIG. 6D. Note that in this example, the 3-state receiver has timing circuitry or access to a clock signal, so can discern and restore “0” states22 as well as the trailing edges suppressed by the transmitter. As is apparent, the waveform ofFIG. 6D represents a delayed and restored version of the original digital input waveform depicted inFIG. 6A.
• The waveforms shown in FIGS.6A-D are merely exemplary. A myriad of signalling schemes are possible, many of which are preferred over the simple, un-encoded, 3-state single-ended scheme depicted in FIGS.6A-D. In particular, differential binary signalling offers significant advantages in terms of noise immunity.
• The signalling across coupled half-capacitors need not utilize the pulsed digital mode depicted in FIGS.6A-D, but can instead employ any modulation of some combination of phase, frequency, amplitude and/or signal-to-noise ratio. The signals can have any form that does not require coupling of the DC component across the coupled half-capacitor. The signals may comprise binary or multistate codings, and, in the case of a ganged half-capacitor junction, may represent a multibit encoding encompassing the signals from one or more half-capacitors as channels.
• The presently preferred embodiment employs pulses, each representing a bit, which are sent across a “differential pair” of coupled half-capacitor junctions. This is essentially a binary differential version of the technique illustrated in FIGS.6A-D. Many of the telecommunication industry's techniques for communicating digital information over bandwidth-limited communication links have no DC component, and thus are advantageous in connection with the present invention. Notable such methods include alternate mark inverse and10B8 codes. In addition, any coding scheme based on (approximately) constant numbers of zero & one bits, such as Knuth's methods, will have particularly advantageous signal-to-noise properties for a given power budget. Modem techniques designed for limited bandwidth environments, including frequency key shifting, phase encoding, and amplitude coding, are also useful as are hybrid coding schemes, such as QPSK. Multidimensional sphere packing, Reed-Solomon codes, Trellis codes, or other algorithms for coding states across one or more signal channels can be used to maximize inter-chip signalling bandwidth, up to the Shannon limit for the system, which may be higher than the clock rate. Of special interest is the class of coding schemes with limited bandwidth, since they can be used to transmit signals through exponentially tapered stripline transmission line transformers, allowing the matching of signals with high voltage swing but low current to low impedance transmission lines as part of the coupled half-capacitor Limiting the bandwidth can advantageously avoid difficulties arising from dispersive lines. Major causes of such dispersion include the frequency-dependence of the line's dielectric factor, frequency dependent dielectric losses, skin effect and losses from the source resistance.
• Reference is now made toFIG. 7, which depicts an exemplary portion of a modular electronic system in which a die bridges two substrates. Bridging die89 includes at least two half-capacitors14a-b, each of which couples to half-capacitors15a-bondifferent substrates10a-bviadielectrics17a-b. Bridging die89 preferably includes a plurality of active electronic devices implemented on anactive surface12 thereof.Substrates10aand10bmay be of similar or dissimilar construction. Likewise, half-capacitors14a-band15a-bare advantageously sized to accommodate anticipated thermal expansion and/or mechanical misalignment.
• Either a passive or active signal path on bridgingdie89 connects half-capacitors14aand14b. Advantageously, an active signal path may be used to convert between incompatible signal waveforms, timings and/or codings onsubstrates10aand10b.
• Bridging die89 preferably comprises a part of a very large system wherein a multiplicity of substrates are tied together with bridging modules, in this example a die. Systems formed from a multiplicity of substrates bearing microelectronic devices and/or wiring need not be primarily planar like those depicted inFIGS. 5 and 7, and can be extremely large: Barn Door Scale Integration is possible using the invention.FIGS. 9, 10, and13 extend the teaching to exemplary extensible non-planar and planar arrangements.
• Buckling modes may arise at the edges overlapped by bridging modules, particularly in vibration-rich environments. Such bending or expansion modes may effectively limit the diameter of buildable systems. Fortunately, these issues are well-understood by practitioners, and can be ameliorated in the preferred embodiment by a combination of compliant, mechanically damping dielectric or mounting materials, and use of compensatingly larger half-capacitors and circuit thresholds near the affected regions.
• Reference is now made toFIG. 8, which depicts an illustrative portion of a modular electronic system that includes multiple dies11,92,95 andsubstrates10,90,93. InFIG. 8, die11 is capacitively coupled tosubstrate10. Note that the multi-storey structure of the capacitive signalpath coupling chip11 to substrate10 (by way ofconductor14y, illustratively depicted as having been fabricated monolithically amidstdielectrics17xand17y) creates an “effective overlap” between half-capacitors14xand15x, which are not “substantially overlapped” plates. Substantially overlapped half-capacitors, such as94aand15, are effectively overlapped ipso facto.
• Substrate10 is also capacitively coupled tosubstrate90. Withinsubstrate90, wiring91 connects a plurality of half-compactiors94ato a plurality of half-capacitors94b, thereby providing an AC signal path betweensubstrate10 andsubstrate93 and betweensubstrate10 and die92.Substrate93 also connects conductively to die95 via a plurality of conductive leads97, bonded to a plurality ofconductive pads96 onsubstrate93.
• Illustratively,substrate10 may comprise a printed wiring board, whilesubstrate90 may comprise a multilayer ceramic substrate.Substrate93 may be encapsulated within aplastic carrier package98, which package also advantageously encapsulates die95.Modules92 and11 may represent separate dies or different portions of the same die or wafer.
• Those ordinarily skilled in the art will recognize numerous alternative arrangements of single or multiple dies coupled to or between single or multiple substrates by means of the invention. Such alternative arrangements are advantageously selected to accommodate particular physical, mechanical, electrical, thermal, and economic constraints of a particular system application. The distribution and costs of all components—on dies or substrates, as higher level modules of them, the thermal and environmental systems, the mechanical chassis, and so forth—can be considered in the overall system-wide Level 0-4 package optimization. For example, a substrate bearing capacitive contacts at right angles to the interconnection orientation may facilitate construction ofLevel 3 packages akin to backplanes or card cages, as inFIG. 16, and the given form factor may therefore influence the choice of circuitry in detail and assignment of circuits to modules. Flexible or articulated substrates may be used to further extend the number of feasible physical arrangements. Abutting or partially overlapping substrates may be used to couple signals, clocks, etc. throughout a large system, as inFIGS. 9-10. Using means for capacitive coupling to link regions bearing powered microelectronic devices to others or to substrates bearing wiring, numerous alternative regular or irregular topologies follow obviously from the present invention, including tilings, lattices, or cluster configurations. Arrangements of organized congregations of such regions, such as modular ensembles of dies and/or substrates, likewise follow directly from the invention. To the extent that such alternative system arrangements employ various claimed aspects of the invention, such arrangements shall be considered within the scope of the present invention.
• Reference is now made toFIG. 9, which depicts an exemplary portion of a large-scale modular system constructed in accordance with the invention.FIG. 9 illustrates a so-called “leapfrog” geometry: Signals “leap” from a first module to a second via a bridging module which partially overlaps the first and second modules. Using this leapfrog assembly technique, large-scale systems can be constructed very economically.
• InFIG. 9, a portion ofdie115overlaps substrate123 such that half-capacitors111 and113 align to provide a capacitive signal path betweensubstrate123 and die115. The leapfrog arrangement continues withsubstrate124, which partially overlaps and capacitively couples to die115 via coupled half-capacitors125 and116. Similarly,substrate140 partially overlaps and capacitively couples withsubstrate124, thus extending the leapfrog configuration. Advantageously, the leapfrog technique can also be used to distribute power from one die/substrate to the next through, for example, the previously mentioned fuzz-button connections.
• Also depicted inFIG. 9 is a non-bridging die119 (not part of the leapfrog arrangement) having a plurality of active devices and coupled tosubstrate124 via a plurality of coupled half-capacitors117. Similarly, non-bridging die136 illustratively employs an off-die signal path (in substrate140) to couple circuits onactive surface138 proximate to half-capacitors133 and135.
• Several advantages of the leapfrog geometry are worth noting. First, the negative effects of thermal expansion are greatly reduced as compared, for example, to a similarly sized system in which a single substrate and die mate. Thus, if die115 runs very hot and causes a slight misalignment of half-capacitors116 and125, the alignment of half-capacitors117 and128 and half-capacitors120 and130 will remain largely unaffected.
• Another advantage of the leapfrog configuration is its easy reparability. Thetop level modules123,124 and136 can be individually replaced, if defective, without further disassembly of the system. Even replacement of thebottom modules115,119 and140 requires relatively little disassembly in a system having a large number of modules.
• Still another advantage of the leapfrog configuration is the high performance of its inter-die connections. As previously noted, coupled half-capacitors (or coupled transmission lines) have lower parasitic inductance than similarly sized conductive connections. Additionally, since the modules in the leapfrog configuration overlap, the lengths of signal paths will be shorter than, for example, conventional technologies which utilize peripherally distributed pins, such as PC boards or MCMs with surface mounted dies. The combination of shorter signal paths and lower parasitic inductances can lead to a substantial increase in overall system performance.
• The leapfrog configuration can also be denser than conventional planar packaging techniques. Therefore, clock skew problems are significantly mitigated. Advantageously, a clock signal can be broadcast, preferably from a centrally located module in a leapfrog integrated system, via a plurality of clock distribution paths, each of which comprises one or more transmission line segments linked by coupled half-capacitors. Such clock distribution paths preferably do not include restoring logic (e.g., transmitter and/or receiver circuits) and thus avoid introducing unnecessary gate delays into the clock signal.
• In conventional MCM systems, great effort is expended to reduce the dielectric factor of the substrate material, since the speed of signal propagation scales inversely with the square root of the dielectric factor. Advantageously, the leapfrog configuration of the present invention reduces the lengths of signal paths, which provides a linear improvement in signal propagation time. Accordingly, low latency can be achieved using the invention without need for exotic, low dielectric factor substrates.
• A further advantage of the leapfrog configuration is that it enables integration of heterogeneous, incompatible modules. The voltage levels, materials, etc. can be different among semiconductor die115, die119, and die136, thereby permitting a system designer to exploit the particular advantages of various technologies within a densely integrated system. Likewise, the properties ofsubstrates123,124 and140 can be selected to enhance performance, reliability, reparability, cost, and other factors. For example, die115 might be consumer-grade silicon CMOS, die119 might bear microelectronic devices on diamond, and die136 might be a high-temperature superconducting material bearing Josephson Junctions, whilesubstrate123 might be a laminate material such as FR4,substrate124 might be an active substrate formed with a thin-film such as lines lithographed onto a chips-first plane, andsubstrate140 might be a thick-film fabricated xerographically.
• The cheapness of interconnect and amenability to heterogeneous integration means that small dies or modules may be used to support partial or specialized functions. Conventional technology, in contrast, generally attempts to bundle multiple functional subsystems into large-scale integrated circuits due to the cost of off-die signalling, often compounding manufacturing costs and sacrificing performance (due to use of suboptimal technologies for particular components, e.g., analog, etc.). Some important examples of components which a capacitively coupled multichip system might carry include logic devices (e.g., Field Programmable Logic Arrays), (field-programmable) gate arrays, processors, arithmetic units, bitwise manipulation units (e.g., cellular automata), permutation units, interconnection networks, memory, nanomechanical actuators, sensors, arrays of fractional analog or digital components (e.g., resistors of various sizes), or others.
• Reference is now made toFIG. 10, which depicts an illustrative portion of a leapfrog integrated system constructed entirely from dies. Advantageously, replacement ofsubstrates123,124 and140 (ofFIG. 9) with dies163,164 and180 permits denser integration of logic and other circuits, as well as perfect matching of thermal coefficients of expansion. Otherwise, the properties and advantages of theFIG. 10 die-die-die leapfrog configuration are analogous to those of theFIG. 9 die-substrate-die leapfrog assembly orFIG. 7 substrate-die-substrate assembly, and therefore need not be reiterated. Power is advantageously delivered through the interstices between dies in such a system, rather than through either perforations or diffused contacts through dies, and may advantageously employ a dedicated power substrate and pressure plate.
• Reference is now made toFIGS. 11a-b, which depict the effect of misalignment between capacitor plates in a capacitively coupled modular system.FIG. 11ashows a portion of a capacitively coupled system wherein dies251,252,253 and254 are perfectly aligned so as to define overlappingregions255,256 and257 identical to the size of the half-capacitor plates. In this idealized configuration, capacitive signal coupling is maximized and unwanted crosstalk is minimized.
• FIG. 11bshows the same portion wherein die252 (ofFIG. 11a) has been replaced by amisaligned die262. The misalignment ofdie262 decreases the area ofregions265 and256 which represent overlap of the misaligned die's half-capacitors. Therefore, the magnitude of the signal coupled betweendie262 and dies251 and253 will diminish; crosstalk betweendie262's half-capacitors and other proximate conductive signal lines or half-capacitors may increase as well. Importantly, however, the misalignment ofdie262 does not affect the coupling or alignment between other dies not directly coupled to262, for example, die254.
• If misalignment of particular dies can be anticipated (e.g., the dies run very hot) the size of the half-capacitors associated with those particular dies can be advantageously increased to compensate. Importantly, only those particular half-capacitors associated with the troublesome dies need be enlarged. Therefore, there is no significant decrease in the overall density of off-die connections in the system. The low cost or inherent tolerance to misalignment of coupled half-capacitor junctions makes these junctions useful in many different applications, not just in multichip modules. For example, cables utilizing half-capacitor connections potentially provide cost, reliability, size and performance advantages over conventional cables (SeeFIG. 21). Likewise, high pin-count modular consumer products, such as dockable laptop computers, will realize similar benefits by use of the invention. Pad-limited chips in general can advantageously be implemented using the invention. Notable such chips typically compose signal routing networks, binary integrated circuits whose size is limited by the number of permissible simultaneous 1-0 or 0-1 transitions, and many very small chips.
• Reference is now made toFIGS. 12a-b, which depict the use of an oversized half-capacitor plate to minimize the impact of misalignment. InFIGS. 12a-b, half-capacitor272 is intentionally made larger than half-capacitor273. This allows die274 to expand substantially (as depicted inFIG. 12b) without affecting the area of half-capacitor plate overlap (i.e.270 versus271).
• Another use for an oversized half-capacitor plate such as272 is to terminate some of the fringing field lines from the smaller half-capacitor plate273 which would otherwise terminate on other nearby structures. This advantageously reduces the amount of crosstalk received by the smaller half-capacitor plate273.
• Reference is now made toFIG. 13, which depicts an exemplary portion of a heterogeneously integrated system and illustrates various techniques for integrating modules in accordance with the invention.
• The system ofFIG. 13 illustratively depicts two hierarchical levels arranged in a substantially space-filling package. However, the integration techniques exemplified therein are extensible systems utilizing deeper levels of hierarchical encapsulation in 2D or 3D. The system comprisesmodules200,211 and219.Module200 comprises capacitively coupledchips201 and204.Module211 comprises capacitively coupledchips208 and210.Module219 comprises capacitively coupledchips215 and219a.
• FIG. 13 exemplifies a variety of means for couplingmodules200,211 and219 in accordance with the invention.Cable214 conductively couples chip204 ofmodule200 to chip208 ofmodule211. A capacitive signal path is provided betweenchip202 ofmodule200 andchip210 ofmodule211 via half-capacitor203, half-capacitor205,interconnection substrate206, half-capacitor207 and half-capacitor209.
• The mounting ofchips215 and219ainmodule219 is angled against aspacer212 to accommodate an illustratively depicted externally imposed form factor requirement.Flexible interconnection substrate216 completes acapacitive signal path217 betweenchip208 ofmodule211 andchip215 ofmodule219, and is preferably fabricated from a flexible material so as to accommodate non-planer form factor constraints.Chip215 ofmodule219 receives further input from external conductive oroptical connection218.
• Although depicted in partially exploded view, it is apparent that the partial system depicted inFIG. 13 makes up a substantially space filling assembly. Thus, the space filling assembly depicted onFIG. 13 may itself be a modular part of a larger modular electronic system. Systems implemented in volumetrically dense 3D space filling packages generally have shorter average path lengths and faster clock rates than 2D implementations.
• Heat sinks are not explicitly shown inFIG. 13, but may be easily positioned as disclosed elsewhere herein.Spacer212, in addition to providing mechanical support to meet externally imposed form factor requirements, also advantageously conducts heat away fromchips210 and219a.
• Reference is now made toFIG. 14, which depicts an embodiment of the invention wherein a plurality of optionally superconducting devices are implemented as a layer on a flexible or contoured substrate. Prefabricated die231 and239, each bearing a plurality of microelectronic devices, are affixed to opposing faces of asingle substrate230. Capacitive signalling betweendie231 and239 is provided by coupled half-capacitors232 and238, also implemented on or otherwise affixed to opposing faces ofsubstrate230 and advantageously need not be planar.
• When material constraints permit, both the microelectronic devices and the half-capacitors are advantageously implemented insubstrate230 by monolithic fabrication techniques. For example, regions of material may be deposited ontosubstrate230 or otherwise processed into it (e.g., by ion implantation). Advantageously, Josephson Junction devices may be implemented in a (preferably high-T_c) superconducting thin film painted or otherwise deposited ontosubstrate230.
• Substrate230 is contoured so as to provide adequate capacitive coupling between half-capacitors232 and238. Although illustrated as a simple U-shape, other shapes forsubstrate230, such as spirals or coaxial cylindrical shapes, may also be used.
• Dies231 and239, as well as half-capacitors232,238,optical sensor240 andpower contact pad241 are preferably fabricated prior to deformation ofsubstrate230.Substrate230 may be formed either from two or more distinct pieces bonded or otherwise melded during manufacture, from a flexible material, or from a single piece of rigid material deformed during manufacture by heat or other appropriate treatment.
• Power and ground may be provided to the devices by conductive or radiative contact topower pad241, or by on-board batteries. Advantageously, particularly in superconducting systems, AC powering means235 may communicate power to pad241 without need for mechanical contact.
• External signalling is advantageously provided by alightpipe234 optically coupled, throughdielectric237, to an optical sensor (or driver)240. Alternatively, capacitive or magnetic means, as detailed elsewhere herein, may be used.
• Reference is now made toFIG. 15, which depicts the use of a uniform capacitive interface to couple modules of varying sizes and packaging levels. InFIG. 15, the use of auniform capacitive interface276 onmodule274 and auniform capacitive interface275 on modules277a-cpermits interchangeable connection of modules277a-ctomodule274.
• Capacitive interfaces275 and276 are illustratively depicted as a single half-capacitor and a half-capacitor with bonded dielectric portion, respectively. In actual systems, such capacitive interfaces will preferably comprise a plurality of half-capacitors (and optional conductive contacts) disposed in a standardized footprint. Advantageously,capacitive interface276 may more half-capacitors than are coupled tointerface275, the additional half-capacitors oninterface276 being used to accommodate coupling to test-modules, engineering changes, etc.
• InFIG. 15, alternatively connectable modules277a-cillustratively represent different levels in the packaging hierarchy:module277aillustratively comprises a Level-0 unpackaged die;module277billustratively comprises a Level-2 packaged module; and,module277cillustratively comprises a Level-2 or Level-3 packaged module.
• The use of standardcapacitive interfaces275 and276 in accordance with the present invention facilitates improved methods for product and/or system evolution. In terms of product evolution, a given module is preferably implemented initially utilizing a semicustom assembly of standard or semicustom parts (e.g.,modules277bor277c). As the product matures and production volume increases, the module may be advantageously reimplemented as a custom ASIC die (e.g.,module277a) and installed into the product without need for any higher-level system modifications. Other sequences of product evolution, such as replacement ofmodule277cwith277bor replacement of either with a Level 1½ MCM implementation, are also possible within the scope of the present invention.
• To illustrate the system evolution aspect of the present invention, one may view modules277a-cas different implementations of a functional block of a massively parallel computer. Depending on the requirements of a particular application, the computer manufacturer could alternatively install alow performance module277a, amedium performance module277b, or ahigh performance module277c. Similarly, such functional modules can be advantageously upgraded as the overall requirements of a particular system in the field evolves.
• A pervasive problem in hierarchically decomposing a design in the prior art encompasses the partitioning of high level functions, followed by the discovery that a low level module cannot be implemented effectively in a given technology, followed by the need to repartition at a high level again to avoid the suboptimal (e.g., very expensive) technology mix. Underlying both the improved product evolution and system evolution provided by the invention is the fact that the invention permits implementation of common dense, high performance interfaces uniformly to all the levels in the packaging hierarchy. The hierarchical decomposition design process can exploit a uniform interface and functional specifications for each submodule in accordance with the invention. The invention's support for common interfaces assures that a submodule specified early can be implemented late in the process, even if multiple dies or different technologies are needed to accomplish the function. Thus, the invention greatly enhances a designer's ability to exploit hierarchical decomposition, since the implementation of particular modules in a hierarchical system is not necessarily constrained by the interfaces between modules. If, as the implementation proceeds, the anticipated technology proves inappropriate for implementing a submodule, an alternative technology or MCM can generally be substituted without reworking higher-order partitions.
• Reference is now made toFIG. 16, which depicts a modular system in which a plurality of modules capacitively couple to a substrate mounted to a supporting chassis, illustratively depicted as a backplane.Substrate278 comprises the backplane and includes a plurality of module mounting supports279a-c. Supports279a-care preferably fabricated from a material of high thermal conductivity, so as to enhance heat dissipation from themodules280aaffixed thereto. Eachbackplane slot281 preferably accommodates amodular subsystem282, illustratively mounted onsupport279aand depicted in an exploded view adjacent to support279b.
• Modular subsystem282 includes a plurality of dies280a-band a bridgingsubstrate280c.Die280acapacitively interconnects to die280b. Conductive power connections are not shown; these can be implemented as shown inFIGS. 4-6. Bridging substrate provides a capacitive signal path between die280aandsubstrate278.Substrate278 preferably includes a plurality of terminated transmission lines (not depicted, but generally as shown inFIG. 5) which interconnect the variousmodular subsystems282.
• Reference is now made toFIG. 17A, which depicts a perspective of a prior art face-up level 1 package. Asingle die289m, typically formed of silicon, is supported by asubstrate289n, typically formed of an epoxy or ceramic material. A plurality of power leads289gand signal leads289eare typically formed from aluminum struts. Power leads289gconductively contact power/ground contact pads289ivia wire-bonds289h; ground andpower rails289kdistribute voltage to circuits implemented ondie289m. Signal leads288econductively couple to I/O circuitry288aby means of low-resistance fan-out288b,contact pads288c, and wire-bonds288d.
• FIG. 17A illustrates a fundamental disadvantage of the prior art: useful real estate ondie289mis wasted to support conductive signalling to other chips and/or other package levels. InFIG. 17A, such real estate for signalling includescontact pads288c, low-resistance fan-out288bto the pads, and drivers/receivers288a. In addition to the costs of excess real estate utilization, the prior art also requiresbond wires288dand the package itself288n. Ifpads288care limited to the periphery of the die289m(thereby shortening fan-outs288b), additional real estate may be needed to provide enough perimeter area to accommodate the pads. Thus, designs requiring a high density of I/O terminals further increase the fraction wasted chip real estate. Finally, the presence ofsubstrate289nincreases the size of the overall package assembly to many times of that thebare die289m.
• FIG. 17A also depicts some of the sources of reliability and performance problems commonly encountered in the prior art. The series inductance of acontact junction288b-288candbond lead288dis typically very high (e.g., 0.05-10 nH), and serves as a low pass filter, thereby necessitating more power to drive signal lines at higher clock rates. Conductive contacts are large relative to logic gates, and do not scale desirably as clock rates increase and lithographic dimensions decrease. Finally, conductive contact junctions entail mechanical contact, thereby limiting ability to assemble and disassemble the connections to die289mand decreasing the reliability of such connections.
• Reference is now made toFIG. 17B which depicts an exploded view of illustrative flip-chip Level 1 modular system implemented in accordance with the present invention.Substrate289fmates withdie289e, and advantageously provides signallinginterconnections289b, preferably implemented as terminated transmission lines. Power/ground contact pads289imate reversibly withconductive fuzz buttons289cso as to provide a reversible conductive interconnection betweensubstrate289fand die289e. Alternatively,substrate289fcan be adapted such thatcontact pads289imate with conventional conductive contacting means, such as solder bumps, wire-bonds, pins or ball grid arrays. Fan-outs289kdistribute power to various circuits implemented ondie289e. One or more drivers and/orreceivers288a, preferably utilizing pulse signalling and implemented underneath half-capacitor plates289a, signal between matched half-capacitor plates289a-aaacross a gap. The gap is preferably filled with a material having a high die electric factor.
• Comparing the illustrative single-die system of the invention (FIG. 17B) to the typical single-die system of the prior art (FIG. 17A), it is apparent that the invention frees up most of the real estate devoted to fan-outs288bandcontact pads288c, making this area available for the implementation of useful circuitry inFIG. 17B. Further, the invention eliminates the costs associated with elements withbond wires288eand external signal leads288d; and the cost for the conductivepower contact pads289cin the invention is likely to be less than the sum of the costs ofpower bond wires289gand leads289hin the prior art.
• The cost per square inch of substrate of289fand the associated dielectric is roughly comparable to that of substrate of289n. However, sincesubstrate289nis likely to be smaller thansubstrate289f, the invention provides reduced substrate cost as well. Since the system of the invention (FIG. 17B) is generally denser than the prior art system (FIG. 17A) the invention will require shorter signal lines, thereby decreasing latency and attenuation loss.
• Certain prior art flip-chip technologies, such as solder bump, avoid the real estate needed forbond wires288band leads288c. However, such prior art technologies still introduce the undesirable series inductance associated with conductive interconnections. Moreover, the small size of coupled half-capacitor junctions allows large numbers of junctions to be fabricated cheaply, densely and at high yield. Further, the simple mating process provided by the invention allows assembly of the die into MCMs at the foundry, thus avoiding the need to ship die elsewhere in order to manufacture MCMs or boards.
• Reference is now made to FIGS.18A-C, which depict several steps in the assembly of an MCM (such as those ofFIGS. 4-5) in accordance with the invention. While a number of methods could be used, FIGS.18A-C illustrate an approach which uses a vacuum hold-down during assembly.
• Referring now toFIG. 18A, a plurality ofdie503 are aligned with ahigh resolution substrate503, coated with a dielectric502, using optical feedback and fiducial marks on both die and substrate. Such fiducials can be tracked through holes in the substrate under the die, or by infra-red imaging through silicon wafers. Registration of thedie503 can be verified, and alignment refined, by measuring and maximizing the capacitance between selected pairs of die and substrate half-capacitors. Interferometry by the pick'n'place apparatus may also be used.
• Once all die are aligned, a vacuum is drawn undersubstrate501—byvacuum chamber505,vacuum hose attachment506 and O-ring504—so as to hold the die503 firmly in place.
• Referring now toFIG. 18B, an elastomericthermal conductor511 and pressure plate/heat sink510 are then positioned and locked into contact with the back side of thedie503 bybolts512 andnuts513, thereby forming a rigid assembly.
• Referring now toFIG. 18C, the rigid assembly is inverted andconductive contacts521 are placed into thehigh resolution substrate501. Thepower substrate520 is placed over the contacts. Assembly of the module is completed by compression of the power substrate against the pressureplate using bolts522 and nuts523.
• An advantageous alternative assembly technique involves the use of ultra-violet curable adhesives. In this method, the substrate is coated with a dielectric material, a UV curable adhesive is applied, the die are then individually positioned and aligned, and UV radiation is applied to cure the adhesive.
• Reference is now made toFIG. 19, which depicts a capacitive signal path including means for adjusting overall gain. The need for gain adjustment may arise from a number of factors, including lateral and/or angular misalignments of the half-capacitor plates (from construction or thermal expansion), processing variations affecting transmitters or receivers, variations or non-linearities in the dielectric, and so forth.
• Onmodule286,transmitter286a, in response to an input signal onsignal line286b, generates signals on half-capacitor286c. Onmodule287,receiver287a, in response to the signals received at half-capacitor287, generates an output signal onsignal line287b.
• FIG. 19 depicts several means for adjusting the overall gain of the capacitively coupled signal path in order to optimize the signal-to-noise ratio betweeninput286bandoutput287b: (1) the gain oftransmitter286amay be adjusted; (2) a trim plate (a/k/a “platelet”286dmay be ganged in parallel with or disconnected from half-capacitor286c; (3)trim plates287d-ddmay be ganged in parallel with or disconnected from half-capacitor287; and/or, the gain ofreceiver287amay be adjusted.
• Transmitter configuration means286econtrols the gain oftransmitter286aand/or the connections betweentrim plate286dand half-capacitor286c. Similarly, receiver configuration means287econtrols the gain ofreceiver287aand/or the connections betweentrim plates287d-ddand half-capacitor287c. The use of such configuration means generalizes easily to differential signal paths as well.
• Adjustment of capacitive signal path gain can be either static or dynamic. Static adjustment is preferably performed prior to assembly (i.e., mating of286 and287) and compensates for variances in the manufacture of dies286 and287. The goal is to avoid a situation in which a particularly weak transmitter is paired with a particularly weak receiver. Static adjustment ofdie286 involves temporarily affixing die286 to a well characterized test die. The test die receives pulses fromdie286 and provides feedback used to program transmitter configuration means286e, thereby configuring the gain oftransmitter286awithin the ordinary tolerance. Static adjustment ofdie287 involves an analogous procedure wherein a temporarily affixed test die applies well characterized pulses to die287, so as to enable appropriate programming of receiver configuration means287e.
• Dynamic configuration compensates for gain variances that arise either during manufacture ofmodules286 and287 or during operation of the assembled system. Dynamic configuration is preferably performed during system operation by adjusting the configurations of transmitter configuration means286eand/or receiver configuration means287eso as to optimize the signal-to-noise ratio across the capacitive signal path. Various techniques are known in the telecommunications industry for performing such configuration. The periodicity with which such dynamic adjustments need to be performed will vary anywhere from once at power-up to many times per second, depending on surrounding environmental conditions such as temperature fluctuations and mechanical stresses, among other things.
• Information may advantageously be relayed among configuration means286eand287e, and devices in the larger system. The larger system preferably provides storage means, analysis & data reduction means, diagnostic means, and reporting & auditing means, as well as other services. Preferably, the larger system may co-optimize the settings in configuration means286eand287etogether, advantageously as a function of data from both, advantageously including sensory data.
• Reference is now made to FIGS.20A-B, which depict an exemplary application specific module (ASM) constructed from application specific submodules (ASSMs) in accordance with the invention. As is well known, advances in CAD technology have precipitated explosive growth in the use of application specific integrated circuits (ASICs) wherein a stored description of a system is implemented automatically by CAD system customization of an integrated circuit mask, and the mask is then fabricated and packaged in astandard Level 1 package. In conventional technology, the high costs and performance limitations of packaging and interconnecting multiple die make application specific implementation of multiple die systems (e.g., MCMs) much more costly than single die ASICs. The invention, however, eliminates many of the costs and performance limitations of multi-die interconnections, thereby enabling economically advantageous implementation of application specific multi-die systems, ASSMs, or ASMs.
• FIG. 20B depicts an exemplaryASM comprising ASSMs186aand186, conductively coupling power throughsolder bump188 and capacitively signalling throughdielectric189. Use of the assembly/testing techniques of the present invention permits very low cost and high yield, high reliability fabrication ofASM190.
• FIG. 20A depictsASSMs186aand186 prior to mating. Only a portion ofASSM186 is depicted;ASSM186 may advantageously mate with more than one other ASSM186a. Either or both ofASSMs186aand186 may be custom fabricated utilizing ASIC-like techniques and/or otherwise personalized by the techniques used to customize (field) programmable logic parts.
• Structurally,ASSMs186aand186 each preferably include a plurality ofelectronic devices184aand184b, respectively.Conductive contact pads181aand181bare configured to accommodate asolder ball188 or the like, so as to provide a means for powering ASSM186afromASSM186. Capacitive signalling betweendevices184aand184bis provided viawiring185b, driver/receiver183b, coupled half-capacitors182band182a, driver/receiver183aandwiring185a.ASSM186 optionally includesadditional wiring185c, driver/receiver(s)183cand half-capacitor(s)182cto accommodate additional ASSM(s). Use of the ASM design paradigm in accordance with the present invention effectively extends the reach of application specific implementation technology to systems too large to implement with acceptable yield on a single die.
• Reference is now made toFIG. 21, which depicts a cable interface utilizing the capacitive interface of the invention.Cable410 comprises anouter sheath418 encapsulating a plurality oftransmission lines413 disposed in an appropriately selected dielectricinner medium419.Transmission lines413 each preferably connect to half-capacitors414 atinterface412 so as to form a standard interface adapted to mate with another capacitively terminated cable or socket.Interface414 preferably includes means415 for affixing410 to another cable or connector socket, which may entail mechanical guides, optical positioning, registration marks, or other means for positioning and holding thecable410 to another device's interface.Interface414 may be flat or patterned to enhance alignment stability.Cable410 may additionally incorporate one or more lines adapted for conductive interconnection atinterface414.
• Use of half-capacitor cable interfaces in accordance with the invention provides numerous advantages over conventional cable connections. Coupled half-capacitors allow substantially better impedance matching at the interface than conventional conductive interfaces. In addition, the invention permits substantially denser interconnections at the interface. Also, capacitive cable interfaces are simpler to fabricate and offer higher reliability (in terms of resistance to contamination corrosion) than conventional conductive cable connections.
• Reference is now made toFIG. 22, which depicts an exemplary portion of a modular electronic system wherein a substrate provides both inter-chip and intra-chip capacitive signal paths.
• Inter-chipcapacitive signal path402interconnects chips401aand401b. Intra-chipcapacitive signal path403binterconnects two points onchip400avia a transmission line and two half-capacitors onsubstrate404. Intra-chipcapacitive signal path403asimilarly interconnects two points onchip400b. It should be noted that signal paths need not be simply inter-chip or intra-chip. For example, the pattern of transmission lines onsubstrate404 may include branches which implement a hybrid signal path, providing both inter-chip and intra-chip interconnections.
• Chips400aand400bare illustratively depicted as being powered byconductive power connections401aand401b, respectively, fromsubstrate404. Alternatively, a powering means such as401bmight provide connectivity to a battery supply comprised by thechip400b, thesubstrate404, or elsewhere in the system, preferably eliminating the need for external power connections. Numerous means for powering electronic devices are well-known in the art and may be employed advantageously alongside the invention.
• Reference is now made toFIG. 23A, which depicts an exemplary apparatus for testing modules constructed in accordance with the invention. A module-under-test296 couples capacitively to a test-module297, which drives test vectors and receives (and optionally evaluates) test signals. Importantly, from the perspective of module-under-test296, the electrical interface to test-module297 can be made essentially identical to the interface that module-under-test296 will see when installed in an actual system. Therefore, module-under-test296 can be tested at full speed without any need to over-design the drivers on it to accommodate the higher parasitics of a testing fixture. The apparatus permits circuits being tested inmodule296 to be operated at full speed with essentially no parasitics. Except for latency, the test signals can be made substantially the same as those seen when coupled to a second “real” module, and the latency can be made arbitrarily small.
• Conventional test procedures involve actual contact between each contact point on a die and a corresponding probe on a test fixture. This typically requires specialized microscopes or similar optical alignment means, and a fragile, expensive “bed-of-nails” or sacrificial test rig. In contrast, the invention preferably utilizes capacitive alignment in which a tuned circuit on the module-under-test296 or ontest module297 is used to servo a capacitance extremization (and thus alignment) between certain half-capacitors on module-under-test296 and test-module297. Such capacitive alignment advantageously obviates any need for expensive optical alignment equipment. Of course, other circuitry can instead be used to facilitate the alignment process, although tuned circuitry is advantageous and capacitive means may be preferable since already available.
• During testing, module-under-test296 is positioned relative to test-module297 and mated such that signals can pass between a plurality of respective coupled half-capacitors296aand297a. Respectiveconductive connections296band297ballow test-module297 to provide power to module-under-test296 during the testing cycle, and possibly during a fine alignment cycle.
• The test assembly296-297 is durable and may advantageously be constructed by a process of building orunbuilding modules296 and297, without requiring any positive or negative insertion force or energy changes across thesignal junctions296a-297a. It preferably does not require sacrificial test-rigs, nor the severing or joining of conductor-conductor signal junctions.
• Referring again toFIG. 23A, either or both module-under-test296 and test-module297 preferably comprise undiced wafers. Advantageously, test-module297 includes a plurality of testing circuits, each associated with one or more particular half-capacitors297aand adapted to generate and/or evaluate test signals at the particular half-capacitor(s). Such testing circuits implemented on the active surface of test-module297 (preferably near or beneath half-capacitors297a) substantially increase the bandwidth of the testing process. In order to compensate for the presence of defective testing circuits, one may sequentially test a given module-under-test296 using two or more different test-modules297, thereby ensuring that each circuit (or die) on the module-under-test is tested by at least one operational test circuit on at least one of the test-modules.
• A wafer-scale test-module297 may also be advantageously used as a micrometer caliper when many components are assembled into a multichip module, thereby effectively providing long-distance fiducials.
• Module-under-test296 need not be passivated. Indeed, a high dielectric factor material may advantageously be disposed on the plurality of half-capacitors297a, and not on half-capacitors296a, to lower manufacturing costs and simplify testing. Use of a lower-than-designed dielectric factor or greater separation can advantageously help identify weak drivers or receivers. Likewise, a higher-than-designed dielectric factor or smaller separation can advantageously simulate over-strong communications circuitry. Varying separation and/or alignment between module-under-test296 and test-module297 during parametric testing may be used to simulate various environmental extremes. As a result, full parametric testing can be done easily, exercising permutations of extreme variance in clock rate, temperature, voltage, and so forth.
• Numerous means for powering module-under-test296 are known in the art. Preferably, module-under-test296 is powered through a conductive connection between respective contact points296band297bformed from reversible fuzz buttons, compliant conductive material, or AMPSTAR™—like springy contacts.
• Although the testing protocols discussed in regard toFIGS. 23A, 25 and26 are exemplified with reference to the testing of an undiced semiconductor wafer, these testing protocols are applicable to modules at any level in the packaging hierarchy, to the extent that such higher-level modules employ the capacitive interfaces associated with the invention.
• Testing of modules constructed in accordance with the invention affords significant advantages, compared to testing of conventional, conductively interconnected modules. One advantage is the substantially reduced cost of the test fixture (e.g., test-module297). The cost of conventional testing depends largely on the number of conductive connections between fixture and module-under-test. A conventional system requires conductive signal connections for each node being stimulated and each node being observed, as well as conductive connections for powers and grounds. The present invention eliminates the need for all these conductive connections, except for powers and grounds (e.g.,296band297b). Advantageously, power and ground may be applied to module-under-test296 via a network on theback side296c, thereby completely eliminating the need for conductive connections between module-under-test296 and test-module297. In either case, the cost of a non-conductively coupled test-module297 will generally be substantially less than the cost of a comparable conductively coupled fixture.
• In cases requiring a very high density of interconnections between fixture and module-under-test, use of thenon-conductive interconnections296a-297aof the invention permits a substantially greater number of signal connections between module-under-test296 and test-module297 to be realized. Therefore, the present invention provides substantially increased controllability and observability of module-under-test296. This capability to provide very high interconnection density between module-under-test296 and test-fixture297 is particularly valuable in the testing of very small modules (e.g., silicon dies) and makes economically feasible the manufacture of larger and/or denser circuitry on such modules because of the increased capability and reduced cost of the testing process.
• It is well known that identification of known-good die is critical to the economics of high-yield assembly of MCMs. (See Hagge et al., “High-Yield Assembly of Multichip Modules Through Known-Good IC's and Effective Test Strategies.”Proc. of the IEEE, December 1992, and “Known Good Die.”, 1993 Proceedings,International Conference on Multichip Modules, International Society for Hybrid Microelectronics, Reston Va. 1993.) Deficiencies in the ability of present testing techniques to identify defective die prior to assembly into an MCM contribute approximately ten to fifty percent of the cost of present day MCMs. (See Tai et al., “Exploration and Advances in MCM Technology.” IEEE Multi-Chip Module Conference MCMC-93, Santa Cruz, Calif.) The invention provides novel testing methods which dramatically reduce these costs. These methods reduce the problem of identifying Known Good Die[s] (which in the prior art typically entails packaging a die in a sacrificial test rig and tolerating poorly matched or poorly characterized parasitics) into the Known Good Chip problem whose treatment is well-known by practitioners in semiconductor fabrication. Moreover, packaged chips can still be tested after assembly into a module, just by opening the module and removing some of the high resolution signal substrate or other chips overlaying the suspect chip. A specialized test rig can then be applied directly to the chip, either in situ or after removal from the module.
• Still referring toFIG. 23A, test-module297 advantageously incorporates test-circuitry (not depicted) operable to apply and/or receive and evaluate test data from module-under-test296 via half-capacitors297a. Such incorporated test-circuitry provides in effect a “built-in self-test” (“BIST”) capability, but without the disadvantages of conventional approaches. With conventional built-in self-test, significant real estate on the module-under-test must be used to implement the testing circuits, which increases the cost, decreases the performance and decreases the yield of the module-under-test. Because the invention allows test circuitry to be implemented on the test-module (as opposed to the module-under-test), it eliminates these problems of conventional built-in self-test, while retaining the advantages of low cost for test equipment and procedures, high observability/controllability, high testing bandwidth, and so forth. Thus, the incorporation of test-circuitry into test-module297 in accordance with the invention enables a variety of novel approaches for testing, verifying and repairing modular electronic systems. Self-test circuits can still be built in, but do not need to be built in. The test method does not disadvantage or rule out the use of boundary scan methods, examination by electron beams, or the use of plasmas for testing, but it makes them less necessary by comparison.
• In the prior art, changes to built-in self-test circuitry require expensive mask changes to the module-under-test. Such mask changes often introduce costly new failure modes and subtle side effects, which eventually may require still further engineering changes and increase the time-to-market. The present invention permits changes to the “built-in” testing hardware (i.e. test-circuitry on test-module297) usually without requiring any changes to the part being manufactured (i.e. module-under-test296), thereby substantially reducing the cost and risk associated with changes in the testing process of a production part. Further, different test-modules297 may be advantageously employed to qualify a module-under-test296 for different specifications, such as military, high-speed commercial, or low-cost commercial.
• The method and apparatus can test or validate anymodule296, so any module which can be built can be tested. In the prior art, particularly at higher clock rates, it is possible to build a die or submodule which cannot be tested until fully assembled into a system, which may be combinatorily prohibitive. (DEC's 36-bit Jupiter machine proved untestable hence unbuildable for this reason, ending that company's line of 36-bit machines.) In accordance with the invention, testing is advantageously performed before spending money to package the module-under-test, which is often infeasible in the prior art.
• Reference is now made toFIG. 23B, which depicts the use of mated, undiced wafers to implement fault-tolerant, wafer-scale modular electronic systems in accordance with the invention.
• Important examples include applications for a breadboard (optionally including a routing network), for wafer-scale integration per se, and in multi-die systems of wafer-scale or smaller.
• Breadboards serve purposes well-known to practitioners in the field, such as providing a readily customized routing network.
• Using the invention, the off-die connectivity of a wafer-scale module396 may be defined by customizing the interconnections oninterconnection substrate397, capacitively coupled to the wafer-scale module. Such personalized interconnections can be made durable enough to be “shipped,” not just handled in a laboratory. Most of the uses and benefits of wire-wrap coupling can be exploited by breadboard-like interconnection substrates397 using capacitive coupling, but without paying wire-wrap's penalties for size, power, reliability, durability, parts count, or other factors. The interconnections oninterconnection substrate397 are advantageously made electrically configurable by use of programmable logic devices (PLDs).
• Personalizing wafer-scale systems is straightforward using the invention, as outlined above, since the matrix of interconnections oninterconnection substrate397 can be lithographed or assembled in one piece in a fixed amount of time, rather than wired in a time proportional to the number of leads. Unlike conventional breadboards, such interconnection substrates are mechanically stable and compact enough for commercial production runs. In addition, submodules may readily be swapped in or out by replacing, reconfiguring, or otherwise altering the interconnection network represented on397.
• Unneeded good dies on wafer-scale module396 can be powered or unpowered, depending on the maintenance strategy. Extra good functional units, so called “hot spares,” may be advantageously exploited through adaptive configuration or targeted compiler output. For instance, Very Long Instruction Word (VLIW) compiler techniques can be used to generate object code optimized to the custom configuration of circuits found to be good on the wafer-scale module296.
• Referring again toFIG. 23B, personalization of such a wafer-scale modular system may also be achieved by use of aninterconnection substrate397 comprising a routing network to programmably interconnect half-capacitors397aonmodule397. Routing networks offer a general means for redefining pairedcouples297a-397a. In testing, routing networks may be advantageously used to reconfigure the configuration ofmodule297 by rerouting signals from the half-capacitors297aassociated with defective test circuitry to other functioning test circuitry during a second testing phase; the second phase tests the circuits on module-under-test296 missed during the first testing phase.
• The interconnections oninterconnection substrate397 can be personalized either through hardware configuration or software programming of an incorporated routing network. On the hardware side, customization of a coarse resolution substrate entails opening or closing connections among leads, using any of a variety of techniques. Cutting connections is the simplest. Customization at the high resolution substrate could be done in a number of ways, including metallizing leads, opening or closing fuses with a laser, or voting. Various xerographic or mundane photoresist processes might be employed to create custom multilayer high resolution interconnect substrates rapidly.
• On the software side, there exists a large body of literature analyzing how to use routing networks or buses to arbitrate communication among dies when the exact pattern of faults will not be known until after fabrication. If the individual dies on a wafer have to be large, which is desirable when signalling is expensive, their yield collapses and connecting good die together can become a prohibitively general problem. Worse, the real estate given over to editing out faults with conductive connections can easily swamp the gains in density won by wafer-scale fabrication, as has been demonstrated by the many failed wafer-scale integration efforts in the prior art.
• Buses or routing networks can be made fully general, independent of the detailed fault map, for a bounded fault density. They can also be reconfigured dynamically, allowing hot restart with spare dies. The fault tolerant routing network or bus may be a significant, even dominant, portion of the overall real estate of a wafer such as296. Fault tolerant interconnection mechanisms become important in systems with a large number of components, at the system level if not at the module level. Implementation details of such schemes are available in published literature. One can avoid global interconnect schemes by developing locally fault tolerant clusters, leaving longer distance interconnects possibly fault intolerant. For example, nine adders might be fabricated into a cluster requiring eight, with entirely local circuitry concealing a bad unit. No external customization of interconnect wiring would then be needed.
• Reference is now made toFIGS. 24 and 25, which compare the fabrication of MCMs in the present art (FIG. 24) and the fabrication of wafer-scale modules in accordance with the invention (FIG. 25).
• Referring now toFIG. 24, prior art assembly methods typically entail an initial sequence ofsteps420a,420b,420cand420d, the output from which is a collection of individually separated, known good die; some methods swap the order ofsteps420band420c. Assembly into the MCM typically proceeds in either of two ways. In a chips-first method (steps420e-f), good die are assembled into a planarized wafer-like module, whereupon interchip wiring is implemented using a photolithographic process. Alternatively, the more common chips-last MCM assembly method (steps420g-h) involves implementing wiring on an MCM substrate onto which the known good die are bonded, by flip-chip, wire-bond or TAB. In either case, the known good die are subjected to substantial high-temperature and/or mechanically stressful processing steps after theinitial step420btest that identified the die as good. Therefore, complete functional test of the assembled system (step420i) is advisable to ensure a working multi-die system.
• Reference is now made toFIG. 25, which shows the assembly of a wafer-scale system in accordance with the invention proceeds initially, in steps421a-b, along the same course as the conventional manufacturing methods. (The testing instep421b, however, advantageously utilizes the methods discussed in reference toFIG. 23 above.) Wafer-scale module396, however, is not separated into individual die. Rather, instep421e,interconnection substrate397 is personalized, in response to the results ofstep421b, so as to realize a pattern of interconnect, half-capacitors and/or power connections that will interconnect and power the known good die to form the desired system. Instep421f, the unsawed wafer-scale module396 is mated with thepersonalized interconnection substrate397 to realize the wafer-scale modular system. Step421ioptionally performs a final system test.
• Advantageously, the wafer-scale assembly method depicted inFIG. 25 does not subject the known good die to high-temperature or other physically stressful processing steps. Therefore, the likelihood that a die identified instep421bas good will fail at thefinal step421itest is dramatically reduced, as compared to the conventional methods ofFIG. 24. Moreover, even if such failures occur, rework in accordance with the invention is substantially simpler since it advantageously confines changes to theinterconnection substrate397 and does not require changes to the unsawed wafer-scale module396 and since the mating to the unsawed wafer-scale module can be easily reversed without damage to either.
• Personalization of the interconnection substrate is advantageously performed by selectively deleting power and/or ground contacts, preferably fuzz buttons likemetallic protrusions397k, corresponding to known bad dies from a coarse resolution power substrate portion of the interconnection substrate. Bad dies will therefore not be powered, so shorts or other defects in them can be electrically isolated from the system. This process of editing out defects by deleting connections to power and/or ground isolates bad dies as effectively as sawing them out would, for any point defect or region defect of the unsawed wafer; equivalently, power connections could be enabled only for known good die.
• Referring again toFIG. 23B,interconnection substrate397 can be fabricated by a number of methods, including those taught earlier. It may be the same size as theunsawed wafer396 or composed of regularly or irregularly tiled smaller sections. It may also be built of one or more wafers or dies itself. Leads can be brought outside through fuzz buttons at the same cost as in conductively coupled systems, and breadboarded or connected to other components in a traditional fashion.
• Leavingwafer396 undiced saves manufacturing steps and wastes less real estate. While electrically empty gutters are still needed to separate dies, the neighbor-to-neighbor die separation need only be about 10 microns in the current art, which is realistic in light of present technology for stepping cameras and is comparable to the separation of neighboring half-capacitor plates on a die. This amounts to less than one percent of the real estate for a 2 mm₂die. (N.B.: If the capacitive coupling were designed with much smaller separation (e.g., h=1 micron), the line width would then have to be considered in the optimization, since the gutter must also be wider than several standard deviations of the linewidth.) In comparison, the gutters needed when die are to be sawed are typically 300-900 microns wide, and restricted to having common edges abutting chords across the wafer, typically in a rectangular grid. The embodiment taught above removes the requirement for colinearity, which makes available the real estate padded out to form rectangular dies from non-rectangular layouts, and facilitates the placement of fabrication process control cells, which, in the prior art, are typically lithographed in the gutters. Leaving the wafer unsawed also avoids stress-induced failures from handling, sawing, or drilling, which increases yield. Also, since all of the active die used to form the wafer-scale module advantageously derive from a single wafer, wafer-to-wafer variances in device parameters are not relevant.
• Reference is now made toFIG. 26, which depicts the general flow of assembly, testing and rework of a wafer-scale or similar modular system in accordance with the present invention. Amodule431 is initially provided from existing prefabricated stock (e.g., a configurable array or the like) or manufactured by manufacturing means430.Module431 is characterized by testing means433 to produce adescription434 of its actual performance. In response toperformance description434, layout generation means utilizes well-known CAD techniques to generate ageometric description436 of an interconnection substrate. In response togeometric description436, substrate generation means437 personalizes aninterconnection substrate442.
• Assembly means432 then assemblessubstrate442 and thewafer431 into a (potentially operable)module443. System testing means440 compares the behavior ofmodule443 with a description of the desired system behavior. Ifmodule443 tests acceptably, system testing means440 designates the module as a knowngood module441. If, on the other hand, it identifies problems in the performance ofmodule443, system testing means440 generates reworkinstructions438 for substrate generation means437, in which case a new or reworked interconnection substrate is created and tested.
• Various novel methods for testing and assembling wafer-scale systems have been discussed previously. The invention further provides advantages in terms of repair, optimization and/or rework of such modular systems. Since wiring is advantageously disposed on mated modules, a die or submodule identified as bad can be identified and swapped out with much less tear-down and rebuild than would be required in test-replace cycles for conductively coupled systems in the prior art. Non-destructive assembly and disassembly in accordance with the invention advantageously allows full systems to be built for testing and torn down for repair with substantially no damage to submodules. Reparability is facilitated by the lack of impact on neighboring die, tolerance of alignment mis-registration, absence of metal fatigue, and simplicity of testing, among other factors.
• Rework involves disassembling a modular system and repairing or replacing a module so as to correct a performance deficiency. Importantly, rework may be used not only to repair or replace inoperable modules, but also to repair or replace slow or otherwise suboptimal modules so as to improve the performance of an already operable system.
• In accordance with the invention, the forming or breaking of capacitive signal junctions advantageously entails substantially zero force, sheer, torque, or change in energy across any pair of coupled half-capacitors. Such assembly or disassembly advantageously proceeds with: no rise or fall in temperature at the signal junctions; no application or removal of heat at the signal junctions; no severing or joining of conductor-conductor leads at the signal junctions; no melting, solidifying, or other phase changes at the signal junctions; and/or no use of conductive liquids, solids, pastes, gases, plasmas, solvents, suspensions, or combinations of these at the signal junctions. Those skilled in the art will appreciate the cost and complexity advantages afforded by the avoidance of each of these requirements of various prior art disassembly methods.
• Reference is now made toFIG. 27, which shows the yield of good dies in a typical semiconductor fabrications process as a function of die size. For an average of 1 fault per square centimeter, a Poisson distribution of point-localized defects predicts defect-less yields of approximately 90% for 0.10 cm²die, 37% for 1.0 cm²die, and 5% for 3 cm²die, which are realistic figures for certain mature VLSI processes. Under these conditions, a 60 cm²wafer would on average yield approximately 1good die of 3 cm²size, or 21 of 1 cm²size, or 540 of 10 mm²size. A good 10 cm²die will be found, on average, once every 22 ft²of wafer, and thus is not a practical design target.
• Given the foregoing yield constraints, it is readily apparent that a system much larger than approximately the average fault density (1.0 cm²in this example) may advantageously be fractionated into subunits (e.g., die) which can themselves be manufactured with significantly better yield.
• At a high level of abstraction, the choice of die size can be viewed as essentially a tradeoff between system cost and system performance. While techniques for analyzing these system partitioning issues are known for conventional, conductively interconnected systems, the analysis of such tradeoffs for the capacitively interconnected systems of the invention requires different yield and performance models. An exemplary yield model for a capacitively interconnected system appears below.
• Reference is now made toFIG. 28, which depicts several plots of system cost as a function of die size for an exemplary capacitively interconnected system.FIG. 28 shows the value of fractionating large dies into smaller ones and coupling them together using the invention. The cost for a substrate to interconnect signal lines is assumed to be much, much smaller than the cost of the dies themselves, so can generally be ignored to a first approximation. A system requiring a certain amount of silicon real estate (e.g., 10 cm²for useful circuitry) might variously be designed as one 10 cm²piece, ten 1.0 cm²pieces, one hundred 0.1 cm²pieces, or so forth. In addition to the 10 cm²of useful circuitry, additional real estate must be taken up in every case for power and ground, and also taken up on each of the fractionated dies for “overhead” circuitry for inter-die communication that would not otherwise have been necessary in a single die, including drivers, level-to-pulse converters, pulse-to-level converters, and/or receivers. InFIG. 28, the communication real estate on each die is estimated for illustrative purposes to scale in several ways: intrace299aas a constant (i.e. no dependence on the overall number of pieces), intrace 299e as a constant times the number of pieces (i.e. a unit of circuitry for communicating with every other piece), and as a constant times the number of pieces raised to the intermediate powers of ¼, ½, and ¾, intraces299b,299cand299d, respectively. Tradeoffs for real systems will roughly resemble one of these canonical curves. Furthermore, contact pads for DC powering, typically occupying 0.5 mm²for every 25 mm²of die, need to be provided even if the die is not a multiple of that size, so introduce a step-function of real estate, which is observable as a segmentation of the curves. The example actually depicts a system with 10.0 cm²for useful circuitry plus 0.2 cm²for power pads.
• Consider, for example, trace299c, in which the communication circuitry is assumed to scale as 0.5 mm²times the square root of the number of die into which the 10 cm²of system is packaged.
• • In a system built from two hundred dies with 5 mm²of useful circuitry each, each die would need one power-ground pair (0.5 mm²) and 7.1 mm²(=0.5×200^1/2) of communication circuitry, so would actually be a 12.6 mm²die. The yield for 12.6 mm²die is 95 percent, so the full system would “cost” about (200×12.6/0.95)=26.5 cm²of wafer, or 2.6 times as much as the 10.2 cm²system described above. Very small dies are dominated by the “surface-to-volume” size of their power pads and the circuitry for inter-die communication, so are inefficient even though their yield is high.
  • In a system built with fifty dies with 20 mm²of useful circuitry each, each die would still need one power-ground pair and 3.6 mm²(=0.5×50^1/2of communication circuitry, so would actually be a 24.1 mm²die. The yield for 24.1 mm²die is 82 percent, so the “cost” of the full system would be about 14.7 cm²of wafer, or 1.4 times as much as the 10.2 cm²system described above. For the particular model selected to simulate the cost of power pads, signal circuitry, and useful circuitry, dies of about 20 mm²provide the most cost effective building blocks.
  • In a system built with ten dies with 1.0 cm²of useful circuitry each, each die would need four power-ground pairs (2.0 mm²) and 1.6 mm²(=0.5×10^1/2) of communication circuitry, so would actually be a 103.6 mm²die. The yield for 103.6 mm²die is 37 percent, so the “cost” of the full system would be about 28.0 cm²of wafer, or about 11/4 times as much as the 10.2 cm²system described above. Large dies lose in fabrication yield more than they gain in design efficiency, and very large dies yield at close to zero, even though they have an even more attractive ratio of useful circuitry to overhead.
• Reference is now made toFIG. 29, which depicts the flow of a design cycle in accordance with the present invention. Asystem description451 is provided to CAD means450, which advantageously utilizes performance, yield, reliability and cost models for available or manufacturable submodules to produce an optimized design for the modular system, which design is provided to stock or manufacturing means452 as a list of parts461, to means for packaging and assembling456 as a definition ofgeometry454, and to testing means459 as a definition of the expected behavior of thesystem453.
• Parts455 are provided to means for packaging and assembling456 from stock or means for manufacturing452. Themodule458 is assembled and provided to testing means459, which decides whether to designate the module as a knowngood module460 or initiate a rework orcontinuation cycle457.
• CAD means450 advantageously optimizes the partitioning of the modular system into modules. The present invention permits considerable flexibility in selecting the size of system partitions (e.g., die sizes) since it largely obviates the need to rely on standard pad fan-out and wire-bonding configurations. Therefore, yield modelling, as exemplified inFIG. 28 above, is advantageously incorporated into CAD means450 and forms an integral part of the design process for semi-custom systems based on the present invention.
• Since the invention facilitates low-cost integration of heterogeneous systems comprising a plurality of modules (each potentially implemented by a different manufacturing technology), such yield models are advantageously derived by curve-fitting data from all available fabrication processes, in order to provide a library of yield models that CAD means450 can use in optimizing the implementation of a capacitively interconnected system. In addition to such yield models, CAD means450 advantageously utilizes models of the electrical performance characteristics and other physical characteristics of the various technologies available to implement the modules of a capacitively interconnected system. Equipped with models of the cost, yield, performance and physical characteristics of various technology resources, the CAD means450 employs well-known algorithms to optimize the partitioning and implementation of a semi-custom, capacitively interconnected system. Such optimization preferably includes: (1) the selection of the sizes and types of modules utilized to construct the system, (2) the assignment of logical subcomponents of the system to particular modules for implementation, and/or (3) the design of patterned substrates or like means for capacitively interconnecting the modules to realize the system.
• Reference is now made toFIG. 30, which depicts a block diagram of an off-die capacitive signal path in accordance with the invention. In the signal path shown inFIG. 30,digital data302 is: (1) input totransmitter300; (2) coupled to die half-capacitor15avia transmission line orwire32a; (3) coupled from substrate half-capacitor14ato substrate half-capacitor14bvia transmission line orwire34; (4) coupled from die half-capacitor15btoreceiver301 via transmission line orwire32b; and (5) provided asdigital output304 fromreceiver301.
• Transmitter300 andreceiver301 may be implemented on the same die10 (an intra-die off-die signal path) or on different dies (an inter-die signal path).Transmitter300 andreceiver301 are preferably implemented on the active surface ofdie10, preferably directly beneath half-capacitors15aand15b, respectively.
• Transmitter300 transformsdigital input data302 to a form appropriate for transmission on the capacitive signal path. Such transformations might include, for example, level shifting, edge rate modification, phase, frequency or amplitude modulation of an appropriate carrier, coding of data using Non Return to Zero (NRZ) or Alternate Mark Inverse (AMI) techniques and the like.
• Receiver301 preferably performs an inverse transformation on the signal received from die half-capacitor15bto generatedigital output signal304. Preferably,receiver301 converts modulated signals coupled from substrate half-capacitor14binto digital data by detection, discrimination or pulse slicing techniques, or by recovery of AMI or NRZ data. In the preferred embodiments, the waveform coupled to the receiver online32bis either a pulse or a level waveform.
• FIG. 30 depicts a single-ended capacitive signal path for illustrative purposes. It should be noted, however, that differential signalling, as depicted in later figures, is considered preferable. It should also be noted that a receiver or driver could be connected to more than one half-capacitor, and that both a receiver and a driver could be connected to a single half-capacitor or collection of capacitive coupling means. It should be further noted thattransmission line34 is merely exemplary. Receivers and/or transmitters could be implemented directly onsubstrate11, preferably beneath or near half-capacitors14aand/or14b.
• Reference is now made toFIGS. 31A and 31B, which depict a final stage of the preferred embodiment oftransmitter300. InFIG. 31A, a transmitter includes a means300afor rapidly switching theoutput signal32abetween twovoltage levels312 and313, the means for rapidly switching comprisingswitches310 and311, controlled bycontrol signals314aand314b, respectively. Control signals314aand314bare determined by thedigital data302 entering the transmitter.
• FIG. 31B depicts a preferred embodiment of the means for rapidly switching comprising aCMOS inverter300b.Inverter300brequires only asingle control signal314, which controls bothPMOS switch315 andNMOS switch316.
• The switching means300aoftransmitter300 imparts a waveform on substrate half-capacitor14a(and transmission line34) whenoutput32aswitches voltage levels. Iftransmission line34 is terminated, the imparted waveform will be a pulse. If wiring is unterminated, the waveform will be a level voltage.
• Reference is now made toFIG. 32, which depicts a preferred embodiment of a switching means adapted to drive a differential capacitive signal path. Differential switching means300ccomprises acontrol signal inverter321 and four switches322-325. Depending on the state ofcontrol signal314, one ofvoltage levels312 and313 is selectively coupled tooutput line32awhile the other voltage level is coupled tooutput line32aa.Die half-capacitors15aand15aaimpart a differential waveform to their respective substrate half-capacitors (not shown) wherein each substrate half-capacitor sees a waveform of opposite polarity. The waveforms are pulse waveforms if the substrate transmission lines are terminated and level waveforms if the substrate transmission lines are unterminated.
• Reference is now made toFIG. 33, which depicts a simplified electrical model of the interface between an exemplary switching means300 and a terminatedsubstrate transmission line330. Alevel waveform333 is applied to switching means300a, which couples a similar waveform to die half-capacitor15a. In response thereto, apulse waveform334 is imparted onsubstrate transmission line330. Atermination resistor331 connected betweentransmission line330 and anAC ground332 terminatestransmission line330.
• Reference is now made toFIG. 34, which depicts a simplified electrical model of the interface between an exemplary switching means300 and unterminatedsubstrate wiring line330b. Alevel waveform333 is applied to switching means300, which couples a similar waveform to die half-capacitor15a. In response thereto, alevel waveform335 is imparted onsubstrate wiring330b.Capacitor336 represents the parasitic capacitance betweenunterminated wiring330bandAC ground332. A voltage divider is formed bycapacitor336 and the coupled half-capacitors15aand14a. Therefore,waveform335 represents a reduced swing version ofwaveform333.
• Reference is now made toFIG. 35, which depicts a block diagram of a preferred, differential off-die signal path in accordance with the invention. In the path shown inFIG. 35,digital data302 is: (1) input totransmitter300a; (2) coupled differentially to die half-capacitors15aand15aavia wiring32aand32aa; (3) coupled differentially from substrate half-capacitors14aand14aato substrate half-capacitors14band14bbvia wiring34aand34b, respectively; (4) coupled differentially from die half-capacitors l5band15bbtoreceiver301aviatransmission lines32band32bb; and, (5) provided asdigital output304 fromreceiver301a.
• Differential transmitter300aandreceiver301amay be implemented on the same die10 (an intra-die off-die signal path) or different dies (an inter-die signal path).Transmitter300aandreceiver301aare preferably implemented on the active surface ofdie10, preferably directly beneath half-capacitors15a-15aaand15b-15bb, respectively. Advantageously, one side of the differential signal path can be employed as a “reference line,” for example, by enlarging half-capacitors14aa,15aa,14bb, and15bb.
• As with the single-ended capacitive signal paths, pulse or level waveforms ontransmission lines34aand34bpreferably convey digital information betweentransmitter300aandreceiver301a. However, other differential signalling techniques, such as those based on modulation of amplitude, phase, frequency or signal-to-noise ratio, may also be used to transmit data via the differential capacitive signal paths of the invention. Certain combinations of pulse and level waveforms offer advantageous implementations, such as a configuration wherein a transmitter drives a single-ended pulse and the receiver reads a differential signal, taking a level waveform, preferentially ground, as the other side of the difference.
• Reference is now made toFIG. 36, which is a schematic diagram of an exemplary single-endedreceiver301 compatible for use in the signal path depicted inFIG. 30.Receiver301 converts a pulse waveform received on half-capacitor15band coupled viawiring32bto adigital output304.Receiver301 comprises a biasing means360 coupled to inputline32b, a slicing means361 also coupled to inputline32b, and a pulse-to-level converter362 coupled to the slicing means.
• Biasing means360 comprises a resistive means363 coupled to inputline32band a voltage generator means364 coupled to the resistive means. Resistive means363 preferably comprises anMOS transistor365. Voltage generator means364 preferably comprises a feedback-connected CMOS inverter wherein the relative sizes ofPMOS transistor366 andNMOS transistor367 are selected to provide the desired bias voltage. Preferably, this ratio of transistor sizes is the same as that for an ordinary inverter in the given process, so as to provide a bias voltage approximately equal to the switching threshold of an ordinary inverter. Resistive means363 and voltage generator means364 are designed to track process variations. Advantageously, a single voltage generator means364 may be used in conjunction with multipleresistive means363 to bias multiple half-capacitor plates15b.
• Slicing means361 illustratively comprises a high-threshold inverter368 and a low-threshold inverter369, both coupled to inputline32b.High-threshold inverter368 preferably comprises a CMOS inverter having a PMOS-to-NMOS size ratio larger than that ofvoltage generator364. Similarly, low-threshold inverter369 preferably comprises a CMOS inverter having a PMOS-to-NMOS size ratio smaller than that ofvoltage generator364. Slicing means detects deviations of the voltage oninput line32bimparted by signals coupled from half-capacitor15b, as either pulse or level waveforms. The switching thresholds ofinverters368 and369 advantageously track process variations.
• Pulse-to-level converter362 comprises an R-S flip-flop, includinginverter372 and cross-coupled NAND gates373-374, all preferably implemented in CMOS. Pulse-to-level converter362 effectively reconstructs the digital waveform driven bytransmitter300 on half-capacitor15a(SeeFIG. 30).Output370 from slicing means361 drives the Set input of pulse-to-level converter362, thereby drivingdigital output304 low when a positive pulse or level change is imparted oninput line32b. Similarly,output371 from slicing means361 drives the Reset input of pulse-to-level converter362, thereby drivingdigital output304 high when a negative pulse or level change is imparted oninput line32b. Pulse-to-level converter362 is preferably designed such that the timings in the Set and Reset paths of the R-S flip-flop are substantially identical, so as to provide adigital output waveform304 having substantially the same timing as thedigital input waveform302 attransmitter300.
• Reference is now made toFIG. 37, which is a schematic diagram of an exemplarydifferential receiver301acompatible for use in the signal path depicted inFIG. 35. Differential receiver comprises a biasing means380, a signal conditioning means384, a slicing means361 and a pulse-to-level converter362.Receiver301areceives a differential pulse or level waveform input from half-capacitors15b-15bband provides adigital output304.
• Biasing means380 comprises a voltage generator means383, first resistive means381 and second resistive means382. Voltage generator means383 preferably comprises an ordinary sized, feedback-connected CMOS inverter. First resistive means381 comprises an appropriately sized NMOS transistor connected between voltage generator means383 andinput line32bb. Second resistive means similarly comprises an appropriately sized NMOS transistor connected between voltage generator means383 andinput line32b. Biasing means380 advantageously tracks process variations.
• Signal conditioning means384 receives differential input fromlines32bversus32bband provides a single-endedoutput390, thereby enhancing the rejection of common-mode noise or other signals oninput lines32band32bb. Signal conditioning means384 illustratively comprises a DMC differential amplifier, wherein an NMOScurrent source385 and the gates ofload devices388 and389 are self-biased by one output of the differential amplifier, and differential input fromlines32band32bbdrives an NMOSdifferential pair386 and387. (See Chappel et al.,IEEE J. of Solid State Circuits, February 1988.) This differential amplifier design advantageously provides excellent tracking of theoutput voltage390 with the input threshold of a nominal CMOS inverter across process, temperature and supply voltage variations, as well as good common mode and power supply rejection. Thus, signal conditioning means384 provides anoutput390 approximately equal to the switching threshold of an ordinary CMOS inverter when itsinputs32band32bbare approximately the same.
• The single-ended output from signal conditioning means384 is provided to slicing means361, which drives the Set and Reset inputs370-371 of pulse-to-level converter362 to providedigital output304. Slicing means361 and pulse-to-level converter362 have been previously described in regard toFIG. 26.
• Reference is now made to FIGS.38A-B.FIG. 38A shows an exploded view of a modular system including atransmission line471 onmodule470 and aparallel transmission line474 onmodule473. Coupledtransmission lines471 and474 provide a means for non-conductively signalling betweenmodules470 and473.FIG. 38B shows an exploded view of an alternative arrangement, whereintransmission line471 onmodule470 andperpendicular transmission line477 on module476 couple magnetically, thereby providing a means for non-conductively signalling betweenmodules470 and476. Signalling means472 comprises a transmitter and signalling means475 comprises a receiver, and/or vice versa.
• While the invention has been illustrated with reference to preferred embodiments utilizing capacitive coupling, other modes of non-conductive interconnection, as described inFIG. 38, can be used to construct modular systems in accordance with the invention. U.S. Pat. No. 5,012,321, entitled INTERCONNECTION DEVICE BETWEEN THE CELLS OF A PRE-IMPLANTED HYPERFREQUENCY INTEGRATED CIRCUIT, incorporated herein by reference, teaches the use of transmission line coupling between overlapping microstrips as a technique for implementing a custom interconnect layer on a pre-implanted hyperfrequency integrated circuit. The '321 patent differs from the present invention in at least two important respects: (1) the '321 patent does not teach the use of transmission line coupling to signal inter-chip between different modules in a modular system, or intra-chip via an off-chip path such as between two units of a chip via a distinct capacitively coupled substrate; and (2) the '321 patent expressly excludes “integrated circuits on silicon . . . operating at a low frequency.”
• We have found that with a sufficiently high dielectric factor, transmission line coupling as depicted inFIG. 38 may be used in place of previously described half-capacitor coupling methods to implement a means for non-conductively signalling in accordance with the present invention, and that such alternative signalling means may thus be used in silicon-based systems such as CMOS. In particular, modular microwave and millimeter wave systems can advantageously employ transmission line coupling as a means for non-conductively signalling between modules in the system. Additionally, it may be advantageous to distribute clock signals and the like through coupled transmission lines. In these and similar circumstances, transmission line coupling and capacitive coupling represent alternative means for non-conductively signalling. Thus, the present invention encompasses modular electronic systems in which at least some of the modules communicate via coupled half-capacitors, coupled transmission lines, or both.

Claims (17)

1. A modular electronic system comprising:

a substrate including a transmission line and a ground plane for the transmission line;

a monolithic integrated circuit chip including digital circuitry;

means for powering said chip; and, means mounted in part on the substrate and in part on the chip for capacitively coupling digital signals between said chip and said substrate.

2-26. (Cancelled).

27. A modular electronic system as defined inclaim 1 wherein said transmission line comprises a stripline, a microchip or a slotline.

28-29. (Cancelled).

30. A modular electronic system as defined inclaim 1 wherein said transmission line is coupled to said means for capacitively coupling.

31. A modular electronic system as defined inclaim 30 wherein said transmission line is further coupled to a second means for capacitively coupling.

32-205. (Cancelled).

206. A modular electronic system as defined in claim20 wherein said transmission line is terminated.

207. A modular electronic system as defined in claim20 wherein said transmission line connects a plurality of points together electrically.

208. A modular electronic system as defined in claim20, wherein said plurality of points includes a plurality of listeners.

209. A modular electronic system as defined in claim20, wherein said plurality of points includes a plurality of senders.

210. A modular electronic system as defined in claim20, wherein said plurality of points includes a plurality of transmission lines.

211. A modular electronic system comprising a first module, including a first array; and

a second module, including a second array; and

a third module, including a third array and a fourth array, said third array capacitively coupled to said first array, and said fourth array capacitively coupled to said second array;

each array comprising a plurality of half-capacitors.

212. A modular electronic system as defined inclaim 211 further comprising a fourth module, including a fifth array; and

said second module further including a sixth array;

said fifth array capacitively coupled to said sixth array.

213. A modular electronic system as defined inclaim 211 wherein said modules are tiled in a regular geometric pattern.

214. A modular electronic system as defined inclaim 211 wherein said modules include a monolithic integrated circuit.

215. A modular electronic system as defined inclaim 211 further including structures for facilitating mechanical alignment.

Images