3D chippackageboard modeling: advanced computing languages will help to develop flexible solver architectures and efficient simulation methods.The rapidly growing needs of microelectronics integration have resulted in a mindboggling variety of innovative packaging concepts. Some of these package types can be seen in FIGURE 1. These technologies create an increasing technical challenge for related layout and design tools, and more critically, the associated electromagnetic and electrical modeling challenges are rising exponentially and creating design bottlenecks and potential design gaps. These challenges are exacerbated by the growing need for integration across chippackageboardsystem hierarchies in terms of design, modeling and verification. The concept of an SoC includes multiple chip functionality, as well as digital, analog and mixedsignal functions in a single chiplevel system with all associated integration and packaging. In FIGURE 1A, an SoC component comprised of two bondwire connected dies on one package and a BGAconnected die on another package, stacked in a PoP configuration are shown. Such geometries and connectivities lead to complex electromagnetic situations required to obtain broadband electrical models. In FIGURE 1B, a different approach to stacked die, often used in memory applications, is depicted. In this case, the bondwires are situated on the top and bottom layers, and a cut is created through the package/printed circuit board to route the bond wires. This can create electromagnetic issues. In FIGURE 1C, multiple dies and stacked dies are connected to a single package layer, potentially creating chiptochip crosstalk if not accounted for appropriately. Finally, in FIGURE 1D, 3 DICs are stacked together and routed using TSVs, a recentlydeveloped innovation for WLCSPs and 3 DICs applications. While Figure 1 depicts systemlevel integration, electromagnetic issues are found at every level of hierarchy. For example, the inclusion of chip effects, necessitated by recent requirements by designers of integrated chippackageboard codesign, drastically changes the requirements on electromagnetic solvers. Several scenarios arise for onchip electromagnetic modeling. These vary from standard resistancecapacitance extraction for onchip nets to the inclusion of inductance to modeling of fullwave electromagnetic effects for analog onchip integrated passives. The presence of digital and analog components and subsystems necessitates modeling the effect of mixedsignal coupling from highspeed digital sections to sensitive analog subsystems. In addition, the connection of chiptopackage via bond wires, or BGA (Ball Grid Array) A popular surface mount chip package that uses a grid of solder balls as its connectors. Available in plastic and ceramic varieties, BGA is noted for its compact size, high lead count and low inductance, which allows lower voltages to be used. , creates its own coupling challenge. Redistribution layers (RDL RDL  Requirements and Development Language. ["RDL: A Language for Software Development", H.C. Heacox, SIGPLAN Notices 14(9):7179 (Sep 1979)]. ) are required to appropriately connect onchip signal and power nets to bond wires. Similarly, the locations of onchip I/O (Input/Output) The transfer of data between the CPU and a peripheral device. Every transfer is an output from one device and an input to another. See PC input/output. I/O  Input/Output pads for BGA connections need to be optimized to minimize certain metrics such as inductive coupling In electronics, inductive coupling refers to the transfer of energy from one circuit component to another through a shared magnetic field. A change in current flow through one device induces current flow in the other device. in power nets. Integrating these geometric intricacies and multiscale features with all the associated electromagnetic effects from package and boardlevel interactions is an added challenge for codesign. [FIGURE 1 OMITTED] Electromagnetic Modeling Challenges FIGURE 2 depicts instances of the varied electromagnetic (EM) effects that are encountered in chippackageboard codesign. Whether the ultimate goal is power integrity (PI), signal integrity (SI) or electromagnetic interference See EMI. (EMI (ElectroMagnetic Interference) An electrical disturbance in a system due to natural phenomena, lowfrequency waves from electromechanical devices or highfrequency waves (RFI) from chips and other electronic devices. Allowable limits are governed by the FCC. ) prediction, these EM effects have critical roles to play. The first order effects for signal (and power) propagation are shown by blue arrows. Within the confines of a digital chip, propagation along horizontal (traces) and vertical (via) structures needs to be modeled, including possible reflection at discontinuities and transitions. In an analog, radio frequency (RF) or mixed signal chip, the propagation modeling needs additional detail to model highfrequency effects such as through an onchip inductor inductor, electric device consisting of one or more turns of wire and typically having two terminals. An inductor is usually connected into a circuit in order to raise the inductance to a desired value. or filter. Once the signal leaves the chip, the propagation (and reflection) is effected by discontinuities at the RDL layer or I/O pad where the signal enters a BGA or bond wire. The signal then encounters another discontinuity on the other side of the connection, at the package (or PCB PCB: see polychlorinated biphenyl. PCB in full polychlorinated biphenyl Any of a class of highly stable organic compounds prepared by the reaction of chlorine with biphenyl, a tworing compound. in a CoB) side. Once the signal enters the package, the dominant propagation through traces and vias is affected by reflections caused by impedance discontinuities including layer changes, antipads and voids and path dimension changes. Finally, impedance discontinuities may be encountered at the solder balls. Multiport impedance, admittance Admittance The ratio of the current to the voltage in an alternatingcurrent circuit. In terms of complex current I and voltage V, the admittance of a circuit is given by Eq. (1), and is related to the impedance of the circuit Z by Eq. (2). or scatteringmatrix models that can be used by circuit simulators are directly affected by the dominant first order path effects. In particular, selfterms that account for reflection and transmission along a specific net (connected metal) are highly dependent on these first order effects. The remainder of the models is populated by crosstalk terms that are typically undesirable but are produced by electromagnetic coupling. In Figure 2, the red arrows The Red Arrows, officially known as the Royal Air Force Aerobatic Team, is the aerobatics display team of the Royal Air Force, based at RAF Scampton, United Kingdom. depict EM crosstalk mechanisms. Within the chip, cross coupling between traces is an example of crosstalk. There may also be crosstalk between different metal layers on the same chip or even between two stacked dies, depending on the orientation and the silicon thicknesses. When the signal emerges from the chip, there is possible crosstalk at the I/O pads, RDL layer and between C4 bumps. Crosstalk between bond wires presents itself both for signals from the same chip and from different chips, as in a stacked die case. There are significant crosstalk mechanisms in the package and PCB that must be accounted for. Tracetrace coupling, viatrace coupling, viavia coupling, coupling paths through voids and antipads, current return path spreading (for example with plane discontinuous discontinuous /dis·con·tin·u·ous/ (dis?kontin´uus) 1. interrupted; intermittent; marked by breaks. 2. discrete; separate. 3. lacking logical order or coherence. as shown in the CoB case) and discontinuities are all contributive. Loss is a critical factor for both propagation and crosstalk. All metal that has excited or induced current In`duced´ cur´rent 1. (Elec.) A current due to variation in the magnetic field surrounding its conductor. impressed upon it will be susceptible to ohmic ohm n. A unit of electrical resistance equal to that of a conductor in which a current of one ampere is produced by a potential of one volt across its terminals. See Table at measurement. loss. The loss will also increase with frequency (skin effect) and with the vicinity of other metal (proximity effect Proximity effect may refer to:
imaginary part of a complex number complex number, complex quantity, imaginary, imaginary number  (mathematics) a number of the form a+bi where a and b are real of its electrical permittivity Permittivity A property of a dielectric medium that determines the forces that electric charges placed in the medium exert on each other. If two charges of q_{1} and q_{2} coulombs in free space are separated by a distance r , which will increase with frequency to the point where it may even become dominant in the GHz range. As the number, type and functionality of wireless components, systems and protocols has increased, the need for minimizing throughtheair EM interference has also increasedwhether through the demands of regulatory bodies such as the FCC (1) (Federal Communications Commission, Washington, DC, www.fcc.gov) The U.S. government agency that regulates interstate and international communications including wire, cable, radio, TV and satellite. The FCC was created under the U.S. (EMC (1) (EMC Corporation, Hopkinton, MA, www.emc.com) The leading supplier of storage products for midrange computers and mainframes. Founded in 1979 by Richard J. Egan and Roger Marino, EMC has developed advanced storage and retrieval technologies for the world's largest companies. or electromagnetic compatibility (hardware, testing) Electromagnetic Compatibility  (EMC) The extent to which a piece of hardware will tolerate electrical interference from other equipment, and will interfere with other equipment. ) or by actual interferencecausing, selfinterference or crossinterference with other systems (EMI or electromagnetic interference). Along with SI and PI, the need to also account for EMI is achieving critical levels. EMI mechanisms of radiation are represented by green arrows. Toplayers of chip, bond wires, package layers and traces are all potentially strong sources of EMI. This is particularly true if systems are either not designed to have differential excitation and ports (which tend to have low radiation due to cancellation of fields) or have excessive asymmetry resulting in common mode generation even under differential design (tends to happen due to a variety of geometrical effects including boundaries, voids, path lengths, impedance profiles and mismatches, nonparallel paths, etc.). Edge effects are created by abrupt boundaries and may give rise to impedance mismatches (critical in power integrity), reflections (critical in signal integrity) and radiation (critical in EMI). It is important that these effects also be modeled accurately. The graywhite arrows show some examples; external boundaries of chips, packages and PCBs cause edge effects. Boundaries of metal planes within these structures also give rise to the same effect. Abrupt edges of material layers, such as the epoxy layer shown, can also cause edge effects. Importantly, internal discontinuities within planes, such as large slots or separate power planes, also yield strong edge effects. These effects can range from fringe capacitance through impedance mismatches to radiation, and they need to be accounted for in generating models for SI, PI or EMI. [FIGURE 2 OMITTED] We may be approaching a limit on the clock frequencies of individual digital cores. With the advent of multicore systems and the mushrooming of mixedsignal, analog and RF systems at multigigahertz frequencies, the throughput of signals through these systems is continually on the rise. The everincreasing speed and frequencies of chippackageboardsystems, unfortunately, has a deleterious effect on the electromagnetic scenario. Practically every EM phenomenon gets more involved as frequencies and speeds increase. Skin effect forces current to travel through smaller crosssections and rapidly increases the loss and resistance with frequency. Dimensions of geometries become electrically larger as frequencies increase, and more sections of the system become efficient radiators, thereby increasing EMI, crosstalk and radiation loss. Simple approximations, such as transmission line models, transverse EM (TEM TEM 1. transmission electron microscope. 2. triethylenemelamine. 3. transmissible encephalopathy of mink. ) models and lumped models, also become drastically inaccurate. Dielectric materials Dielectric materials Materials which are electrical insulators or in which an electric field can be sustained with a minimal dissipation of power. Dielectrics are employed as insulation for wires, cables, and electrical equipment, as polarizable media for start behaving in an increasingly lossy See lossy compression. (algorithm) lossy  A term describing a data compression algorithm which actually reduces the amount of information in the data, rather than just the number of bits used to represent that information. and frequencydependent manner, causing signal distortion and loss through dispersion and attenuation Loss of signal power in a transmission. Attenuation The reduction in level of a transmitted quantity as a function of a parameter, usually distance. It is applied mainly to acoustic or electromagnetic waves and is expressed as the ratio of power densities. . Even for power integrity applications, with power signals near DC, switching effects on multiGHz frequency rise times are now common and necessitate highfrequency modeling even for PI problems. It is clear that SIPIEMI is moving into the arena where fullwave, 3D electromagnetic modeling is a necessity and simpler, more efficient approximations are becoming irrelevant. Designers need to be aware that such approximations may produce results that have no basis in reality for design or verification. Electromagnetic Modeling Approaches 3D Resistance Extraction. This entails the solution of either a resistive resistive /re·sis·tive/ (rezis´tiv) pertaining to or characterized by resistance. network problem involving digital chip layers or modeling substrate losses. While this requires modeling of full 3D geometry, it is a relatively simple computation. 3D RC Extraction. The next level of complexity involves creating 3D RC models for chip metal layers and substrates, obtained by solving Poisson's equation In mathematics, Poisson's equation is a partial differential equation with broad utility in electrostatics, mechanical engineering and theoretical physics. It is named after the French mathematician, geometer and physicist SiméonDenis Poisson. . The models are suitable for timing and delay computations and for onchip signal integrity and power distribution networks at relatively low clock frequencies. 3D RLC RLC Residual lung capacity Extraction, Onchip modeling at GHz frequencies necessitates the modeling of inductance in addition to RC modeling. The potentially distributed nature of inductance loops, as well as frequency dependence due to skin and proximity effects, makes this a significantly more involved problem, but very relevant for onchip modeling. 3D QuasiStatic Modeling. This technique of modeling has at one end RLC modeling at the package level. In a more complex version, the R, L and C are coupled and derived from the same solution. The phase term (occurring in Green's functions) is ignored during fullwave modeling. Physically, this is equivalent to assuming all influences (currents, voltages, fields) travel instantaneously through the system. Quasistatic models are widespread for fullport electrical models for packages. These typically have validity up to the low GHz range and are used for both power integrity and signal integrity. 2D Transmission Line Modeling. This is traditionally a very useful technique when there are significant symmetries and structures present in the problem. For example, welldesigned parallel PCB traces over a ground plane can be modeled very well with perunitlength (2D cross section) RLGC parameters. These have restricted validity for more general propagation paths and need bookkeeping and corrections when traces pass over voids, bend, connect to vias, change their width, have impedance discontinuities, etc. This technique primarily focuses on differentially excited, symmetric structures. 2.5 DTEM DTEM Division of Toxicology and Environmental Medicine (US CDC) Approximations. When electromagnetic fields are localized to flow between parallel metallic plates with small gaps relative to wavelength, transverse electromagnetic approximations are valid. Such an approach is termed to be of 2.5 dimensions and has been used successfully, primarily for power integrity modeling. 2.5D Layered Approximations. For onchip RF, analog and mixedsignal components such as embedded passives (filters, inductors, transformers, etc), the relative sizes of the components are small when compared to the lateral extent of the dielectric layers. In this case, layered infinite (in transverse directions) approximations make the solution efficient and accurate, and 2.5Dlayered methodofmoments techniques can be used. This also has applicability on some subsets of PCB modeling, but it is generally inapplicable in·ap·pli·ca·ble adj. Not applicable: rules inapplicable to day students. in·ap there and on packages where 3D edge effects are important. 3D FullWave Modeling. This represents the most accurate approach to EM modeling of chippackageboardsystems. There are significant challenges to achieving this goal in terms of scale, memory and time while preserving the expected accuracy of 3D fullwave modeling. There are three main techniques that are considered to be the workhorses of this area, and most successful commercial solvers are built around highly efficient versions of these techniques. The finite element See FEA. method (FEM FEM Female FEM Finite Element Method FEM Feminine FEM Finite Element Model FEM Fédération Européenne des Métallurgistes (European Metalworkers' Federation) FEM Faculdade de Engenharia Mecânica (Brasil) ), initially used in civil engineering, is a staple of 3D fullwave modeling. It is a general technique that uses volumetric volumetric /vol·u·met·ric/ (vol?umet´rik) pertaining to or accompanied by measurement in volumes. vol·u·met·ric adj. Of or relating to measurement by volume. meshes to convert the partial differential equation partial differential equation In mathematics, an equation that contains partial derivatives, expressing a process of change that depends on more than one independent variable. (PDE PDE Pennsylvania Department of Education PDE PlugIn Development Environment PDE Partial Differential Equation PDE Phosphodiesterases PDE Personal Digital Entertainment PDE Pulse Detonation Engine PDE Product Data Exchange PDE PresentDay English ) form of Maxwell's equations Maxwell's equations Four equations, formulated by James Clerk Maxwell, that together form a complete description of the production and interrelation of electric and magnetic fields. in the frequency domain to a large, sparse matrix In the mathematical subfield of numerical analysis a sparse matrix is a matrix populated primarily with zeros. Conceptually, sparsity corresponds to systems which are loosely coupled. equation. Challenges of Inputs to EM Solvers, Chippackageboardsystem models need to be prepared for analysis by EM solvers. This typically means integrating several different file formats. Net identification, port definition and boundary condition specification (especially in FEM and FDTD FDTD Finite Difference Time Domain FDTD From Dusk 'Til Dawn where this is not automatic) are also needed prior to solution. Challenges of Outputs from EM Solvers. Even with complex EM solvers that may solve part of or the entire simulation problem, the appropriate outputs are critical. In particular, Sparameter, impedance or admittance matrices are useful to understand propagation and coupling in systems. However, for circuitlevel simulation, circuit models need to be generated from these matrices. These models need to be passive and causal in order to ensure that the circuit outputs are accurate and stable. Impedance profiles and timedomain reflection/transmission plots are also required. Recent Advances Advances in computing platforms, coupled with recent breakthrough in algorithms, have dramatically changed the picture in 3D EM simulation for chippackageboardsystem codesign. In particular, MoM techniques, that typically were restricted to 2.5Dlayered solvers, have recently become extremely powerful due to advanced multiscale algorithms. These techniques accelerate the iterative solution of MoM matrices by replacing the quadratic quadratic, mathematical expression of the second degree in one or more unknowns (see polynomial). The general quadratic in one unknown has the form ax^{2}+bx+c, where a, b, and c are constants and x is the variable. cost of matrixvector products with nearlinear costs in the size of the system matrixwhich in turn is dependent on the number of mesh elements. This linear scaling, in terms of solution time and memory, is particularly attractive for large structures such as integrated chippackageboardsystems. In addition, these multiscale algorithms are also amenable to parallelization on shared memory systems; the entire matrixvector product is parallelized over multicore CPUs. These three techniques will continue to be the mainstay of 3D EM simulation for chippackageboardsystems. Emerging Needs and Future Directions With rapid advances in solver technology, focus is now shifting towards true design tools. These tools will enable fast parametric simulation in a large number of parameters enabling whatif analysis and early design. Tunable accuracy and precision will also enable early design optimization. As systems continue to be designed at the edge of technological capabilities, manufacturing tolerances and design for manufacturing (DfM) will be critical at all levels of hierarchy. Statistical modeling, meanvariance and higherorder variability measures, probability distributions and yield estimates will all be criticalboth in a standalone manner and in the context of robust optimization. Optimization will rely on multiple objectives and multiple parameters, requiring extremely efficient solvers as well as visualization and graphing techniques. Eventually, advances in computing languages will allow the development of more flexiblesolver architectures that will determine the most efficient simulation method for a specified accuracy or a subset of the entire problem. These needs and directions are sure to keep designers, design automation tool developers and researchers in excited and busy for years to come. DR. VIKRAM JANDHYALA is CEO (1) (Chief Executive Officer) The highest individual in command of an organization. Typically the president of the company, the CEO reports to the Chairman of the Board. and founder of Physware, Inc. He is also an associate professor in the department of electrical engineering, University of Washington, Seattle and director of the applied computational engineering lab there. He can be reached at vj@u.washington.edu. DR. DIPANJAN GOPE GOPE Geodetic Observatory Pecny is VP of research and development, DR. FENG LING is VP of engineering, DR. XIREN WANG is a research and development engineer, DR. SWAGATO CHAKRABORTY is the VP of products, DEVAN WILLIAMS is a senior research and development engineer and JAMES PINGENOT is a senior research and development engineer, all working for Physware, Inc. 

Reader Opinion