Frequency-Domain Tensor Extraction for Multilayer Glass Core Build-Up Packages

Glass

High-density interconnect packaging relies on inorganic core materials to maintain spatial dimensional stability across multi-chip modules. Synthetic fused silica and specialized aluminosilicate formulations deliver thermal expansion coefficients near 3.2 ppm per kelvin, matching silicon dies. By contrast, organic laminate cores such as FR-4 or high-performance polytetrafluoroethylene exhibit out-of-plane thermal expansion exceeding 30 ppm per kelvin, causing severe micro-cracking and pad displacement during thermal reflow.

Glass panel processing enables line width and spacing features below 5 micrometers on core thicknesses spanning 100 micrometers to 800 micrometers. Coating these rigid cores with organic build-up layers ~ like Ajinomoto Build-up Film or silica-filled epoxies ~ creates a composite structure for high-density signal routing.

Material datasheets often report dielectric constant and dissipation factor as scalar values measured at a single frequency. A single scalar, however, cannot describe high-frequency signal propagation through composite packaging. While synthetic fused silica has an isotropic dielectric constant of 3.8 at 10 gigahertz and a low loss tangent of 0.0004, functional glass cores are rarely pure bulk silica.

Woven E-glass or low-loss L-glass fibers inside composite cores introduce structural anisotropy: field lines along the weave pass through alternating regions of glass filaments and organic resin. Vacuum lamination of build-up dielectric films adds mechanical stress, driving asymmetric polymer chain alignment and non-uniform silica filler distribution along the z-axis.

Sub-terahertz and millimeter-wave electromagnetic fields interact directly with these directional dielectric variations. As high-frequency transmission lines navigate multilayer build-up packages, signal velocity and wave impedance change with field orientation. In-plane fields pass through a dielectric environment shaped by horizontal polymer chains and aligned glass fibers; out-of-plane fields between vertically stacked reference planes see a different effective permittivity altogether.

High-density interconnect routing depends on characterizing these spatial variations to avoid impedance discontinuities, phase skew across differential pairs, and signal degradation.

Drilling and plating through-glass vias introduces localized stress around every vertical conductor. Laser drilling generates micro-fractures, heating, and residual stress radiating from hole edges, while copper electroplating inside high-aspect-ratio vias creates thermal expansion mismatches against the glass matrix. When tens of thousands of vias are present, this mechanical distortion alters local material density and shifts local tensor properties across the active routing area.

Test methods that treat the substrate as a uniform, isotropic block cannot accurately predict how real package layouts perform.

Adopting IPC-TM-650 Method 2.5.5.13 Clause 3.1 requires raw substrate coupon verification under controlled relative humidity prior to metallization release.

Measuring dielectric properties before building up the stack poses major practical challenges. Unplated glass cores lack conductive reference planes, forcing tests to rely on external cavity fixtures or free-space optical systems. Furthermore, properties change once the package undergoes lamination at pressures up to 3 megapascals and temperatures reaching 180 degrees Celsius.

Resin flows around metallic features and settles into configurations far different from raw film datasheets, meaning extraction routines must characterize the fully processed stack rather than unlaminated stock.

The operational protocol for preparing raw glass panels for frequency-domain extraction involves precise mechanical and chemical controls:

1. Substrate surfaces undergo micro-abrasive liquid jetting to strip organic contaminants without disturbing surface micro-roughness parameters.
2. Optical laser profilometry maps macro-level warpage and baseline surface profile variation across the full working area of the cleaned panels.
3. Sputtered copper seed layers are deposited under argon plasma to a target thickness of 200 nanometers, forming low-stress ground planes.
4. Photolithography defines calibration structure footprints using negative photoresist exposed at a wavelength of 365 nanometers.
5. Electroplating builds copper thickness to 3 micrometers within photoresist channels, keeping current density below 1.5 amperes per square decimeter to limit stress.
6. Photoresist stripping and chemical flash etching remove the bare copper seed layer, isolating individual frequency-domain test structures across the panel field.

Extracting accurate dielectric tensor properties requires evaluating material behavior across both frequency and physical axes. Ignoring directional variations results in miscalculated signal propagation delays and impedance mismatches. Sourcing practices must require dielectric data derived from completed multilayer structures rather than raw material datasheets.

Anisotropy

Spatial variation in dielectric constant and dissipation factor across orthogonal axes defines the complex permittivity tensor of high-density package substrates. For an orthotropic material, the relative permittivity tensor matrix contains non-zero diagonal elements representing the principal physical axes of the substrate panel. Horizontal axes within the substrate plane are designated as the x-axis and y-axis, while the vertical build-up axis represents the z-axis.

The relative permittivity tensor is expressed as a three-by-three matrix containing real and imaginary components for each axis:

barbarvarεr = beginbmatrix varε’rx – jvarε”rx & 0 & 0 \ 0 & varε’ry – jvarε”ry & 0 \ 0 & 0 & varε’rz – jvarε”rz endbmatrix

The real components represent energy storage capacity along each physical axis, setting wave phase velocity and characteristic line impedance. Imaginary components quantify dielectric attenuation mechanisms within the material matrix. Off-diagonal tensor elements remain near zero when coordinate axes align with the principal mechanical axes of the packaging substrate.

Routing transmission lines at arbitrary angles across a woven glass substrate induces off-diagonal tensor coupling, creating cross-polarized field components that radiate into adjacent signal traces.

Build-up packaging geometries amplify the functional impact of tensor directional variance. Microstrip lines on outer package layers concentrate electric fields within both the upper air volume and the underlying build-up film. Striplines embedded between internal reference planes confine fields entirely within the dielectric layer, exciting primarily out-of-plane z-axis field components.

Vertical via transitions excite radially symmetric z-axis electric fields while inducing in-plane azimuthal magnetic fields. Calculating single-ended impedance with an in-plane permittivity figure introduces systematic calculated impedance errors up to 8 percent on stripline features. Differential pair routing encounters additional phase velocity imbalances when individual traces align differently relative to the underlying glass weave pattern.

Evaluating glass build-up package panels across lamination batches quantifies tensor directional variance.

Across all samples at 28 gigahertz, out-of-plane permittivity exceeded in-plane permittivity by 5.4 percent. This discrepancy stems from structural alignment within composite packaging layers. Silica filler particles added to reduce thermal expansion migrate during vacuum lamination, forming layered distributions parallel to the substrate surface.

Polymer chains in thin build-up films compress vertically during high-pressure cure cycles, aligning molecular dipoles predominantly within the horizontal plane. Glass fiber filaments running along x- and y-axes contribute high-permittivity glass directly along in-plane paths, whereas vertical z-axis fields cross alternating layers of resin and glass, altering effective dielectric path lengths.

Failure mechanisms driven by uncharacterized tensor anisotropy degrade system performance in predictable ways:

• Impedance Mismatch Cascade occurs when stripline dimensions designed using scalar in-plane dielectric constants yield lower characteristic impedance than intended, resulting in signal reflections at package ball interfaces.
• Differential Mode Skew manifests when individual conductors within a high-speed differential pair experience unequal effective permittivity due to local glass weave positioning and asymmetric tensor distribution along trace length.
• Resonant Cavity Shift degrades power distribution network decoupling performance because internal power-ground plane resonant frequencies shift under anisotropic z-axis permittivity variations.
• Cross-Polarization Coupling arises when high-frequency signal currents on angled traces excite unexpected field modes, transferring energy into adjacent non-terminated conductor loops.

Accurate frequency-domain tensor extraction demands mathematical models that isolate directional vector components from measured scattering parameters. Standard single-line microwave characterization techniques cannot separate in-plane dielectric loss from out-of-plane loss. Multi-line extraction schemes alter conductor field geometry by varying trace widths and dielectric height ratios across test vehicles.

Narrow traces produce tightly confined fringing fields that sample both in-plane and out-of-plane components. Wide conductor traces generate uniform parallel-plate field configurations dominated almost entirely by out-of-plane z-axis tensor components. Systematically comparing S-parameters across varying conductor aspect ratios isolates individual tensor coefficients through iterative field solver inversion.

Extracting these complex tensor properties at millimeter-wave frequencies requires strict control over sample fabrication tolerances. Copper trace etching variations, conductor sidewall slope angles, and dielectric layer thickness non-uniformities distort extracted tensor values if omitted from electromagnetic inversion models. Physical cross-sectioning of test coupons after electrical measurement provides exact structural dimensions to seed back-extraction algorithms.

Precision extraction relies on separating manufacturing geometry variations from true fundamental material tensor responses. Whether planar laser-induced stress maps can reliably predict out-of-plane permittivity shifts across sub-millimeter package regions remains unproven over extended thermal cycling.

Resonator

Specialized test fixtures isolate target field orientations to measure tensor dielectric components in the frequency domain. Split-cylinder resonators operating in transverse electric modes yield precise characterizations of in-plane permittivity and loss tangent. A thin, unplated substrate sample slides into a narrow gap separating two cylindrical metallic cavity halves.

Exciting the fundamental TE011 mode establishes a circular electric field parallel to the substrate surface, eliminating out-of-plane z-axis electric field interactions. Measuring shifts in cavity resonant frequency and quality factor allows direct calculation of in-plane tensor components (varεrx, varεry, tanδxy) at discrete modal frequencies.

Characterizing out-of-plane z-axis tensor components requires alternative resonant structures that enforce dominant vertical electric fields. Balanced circular disk resonators build multi-layer resonant cavities directly within the substrate stack. Circular conductor disks etched on internal layers form parallel-plate resonant structures bounded by ground vias.

Exciting dominant transverse electromagnetic or transverse magnetic cavity modes establishes uniform vertical electric fields between the conductive disk and adjacent reference planes. Frequency shifts across multiple resonant modes yield z-axis tensor values across broad frequency bands.

The total cost per working substrate panel increases by 14 percent when custom resonator test coupons occupy edge routing real estate.

Fabry-Perot open resonators extend tensor characterization into the millimeter-wave regime up to 110 gigahertz. An open resonator uses two polished spherical mirrors to focus electromagnetic energy into a tight Gaussian beam waist. The flat substrate sample sits at the central beam waist position.

Rotating the sample relative to the polarized beam vector maps in-plane permittivity variations along arbitrary angles. Open resonators avoid physical contact with fragile packaging materials, while high quality factors exceeding 50,000 enable precise determination of minute loss tangent variations down to 0.0001.

Can Split Ring Resonators Isolate in Plane Permittivity?

Split-ring resonators printed directly onto build-up dielectric layers offer localized tensor extraction within active package areas. Micro-scale split-ring structures induce intense localized magnetic resonances, driving strong electric fields across narrow capacitive ring gaps. Aligning these capacitive gaps along specific substrate axes isolates localized in-plane dielectric properties within regions as small as 200 micrometers square.

This enables spatial mapping of tensor shifts near high-density through-glass via arrays, capturing physical stress influence zones created during assembly.

Executing accurate frequency-domain tensor extractions using physical resonators requires strict operational controls:

• Thermal Stabilization mandates maintaining sample test environments at 23 degrees Celsius within plus or minus 0.5 degrees to eliminate temperature-induced permittivity drift.
• Environmental Humidity Control requires storing test coupons in dry nitrogen cabinets below 5 percent relative humidity for 24 hours prior to testing to evacuate absorbed atmospheric moisture.
• Sample Edge Preparation obligates precision diamond saw dicing of substrate edges to eliminate delamination burrs that disrupt cavity electromagnetic field boundaries.
• Fixture Calibration Baseline demands performing empty-cavity baseline sweeps prior to every sample insertion to compensate for ambient atmospheric pressure and temperature shifts.

Discrete resonant methods provide exceptional measurement precision but operate only at isolated modal frequencies. Continuous frequency coverage across wide bandwidths requires combining discrete resonator data points with broadband transmission line extraction techniques. High-frequency package modeling relies on continuous frequency-dependent dielectric functions to evaluate broad-spectrum digital pulses.

Resonator measurement results serve as fundamental anchor points to validate continuous transmission line extraction models from 1 gigahertz to 110 gigahertz.

Micro-cracking along panel edges renders the multiline reference line phase standard non-linear on quartz probe calibration substrates.

Deembedding

Transmission line methods extract broadband tensor dielectric properties by measuring scattering parameters of planar structures fabricated directly on target packaging substrates. Vector network analyzers measure composite S-parameters including coaxial cable transitions, probe pads, launch vias, and test fixtures. Isolating intrinsic propagation characteristics requires mathematical de-embedding algorithms to remove fixture parasitics.

The multiline Thru-Reflect-Line calibration procedure provides an absolute baseline for frequency-domain dielectric extraction on packaging panels.

Multiline Thru-Reflect-Line uses transmission lines of varying physical lengths alongside short or open reflection standards. The algorithm calculates the complex propagation constant of uniform line geometry without prior knowledge of probe pad impedances or line characteristic impedances. The complex propagation constant γ(f) contains attenuation constant α(f) and phase constant β(f) expressed as:

γ(f) = α(f) + jβ(f) = sqrt(R + jω L)(G + jω C)

Extracting tensor dielectric parameters from γ(f) requires decoupling conductor loss from dielectric loss. Conductor attenuation αc(f) dominates total line loss at low frequencies due to series resistance R(f) driven by skin depth reduction. Dielectric attenuation αd(f) increases linearly with frequency, dominating total loss above 20 gigahertz.

The total attenuation constant is modeled as:

α(f) = αc(f) + αd(f) = c0 sqrtf · Kr + fracπ f sqrtvarεeff(f)c · tanδeff(f)

Conductor roughness increases series resistance beyond classic skin effect predictions. Copper foil applied to organic build-up layers displays complex surface topography with root-mean-square roughness parameters (Rq) ranging from 0.1 micrometers to 0.8 micrometers. Sputtered copper seed layers on polished glass cores achieve smooth surfaces with Rq values below 0.05 micrometers.

Modeling conductor loss requires advanced surface roughness correction formulations, such as the Hemispherical Snowball or Gradient Model, to prevent copper roughness attenuation from falsely inflating extracted dielectric loss tangent figures.

Automatic Fixture Removal offers a faster alternative to multiline calibration when panel surface area limits the placement of long reference lines. Automatic Fixture Removal measures a single 2x Thru structure, mathematically splitting the fixture into two symmetrical single-ended error networks. Time-domain gating isolates main reflections at probe-to-line transitions, generating two-port scattering matrix models for launch networks.

Cascading inverse matrix representations of these launch networks removes fixture effects from measured device S-parameters, exposing intrinsic package trace transmission behavior.

Evaluating extracted characteristic impedance requires transforming frequency-domain scattering matrices into time-domain reflectometry profiles. Impedance profiles reveal spatial discontinuities along the transmission line path. Non-uniform trace widths, localized glass weave windows, and adjacent metal voids create distinct impedance steps.

Comparing phase velocity extractions across narrow and wide transmission lines allows mathematical inversion of tensor permittivity components. Fringing field factors derived from 2.5D electromagnetic field solvers link measured effective permittivity varεeff(f) back to pure spatial tensor components (varεrx, varεry, varεrz).

Evaluating vendor material data requires verifying the exact de-embedding standards used during characterization. Datasheets citing simple Short-Open-Load-Thru calibrations often retain residual probe pad parasitic capacitance, artificially elevating loss tangent figures at millimeter-wave frequencies. Sourcing documentation must mandate multiline calibration execution directly on test vehicles matching target packaging stackups.

Adopting IPC-TM-650 Method 2.5.5.14 Clause 4.2 obligates substrate manufacturers to supply unplated reference coupon sets from the identical glass melt lot, transferring the cost of calibration non-linearity back to the fabricator.

Yield

Fabricating high-density packaging panels requires strict alignment between electrical performance specifications and manufacturing capability limits. Glass core build-up substrate processing occurs on large panel formats, standardizing at 515 millimeters by 510 millimeters to maximize manufacturing throughput. Dielectric constant and loss tangent variations across panel formats alter trace impedance, driving dimensional yield losses when impedance tolerances exceed tight specified limits of plus or minus 5 percent.

Characterizing frequency-domain tensor maps across full panel fields identifies high-risk zones near panel edges where resin flow variations alter effective dielectric thickness.

Manufacturing process windows impose hard boundary conditions on allowable tensor variation limits across production lots:

1. Line width control must maintain plus or minus 0.5 micrometers on 2 micrometer trace features to prevent geometric variation from obscuring dielectric tensor shifts.
2. Dielectric layer thickness variations across 515 mm by 510 mm panels cannot exceed plus or minus 3 percent to maintain consistent out-of-plane field distribution.
3. Laser-drilled through-glass via registration must hold within plus or minus 2 micrometers relative to target capture pads to prevent asymmetrical field distortion.
4. Glass core panel warpage must remain below 0.2 percent of total panel diagonal length to ensure uniform contact during wafer-level high-density probing.

Integrating frequency-domain tensor extraction structures into production panel margins enables real-time process monitoring without sacrificing active routing areas. Edge coupons located within panel trim zones carry compact multiline transmission structures and split-ring resonators. Automated wafer-level probing stations execute high-speed S-parameter sweeps across these margin structures immediately following final copper metallization.

Extracted tensor properties pass through statistical process control filters to flag lamination temperature non-uniformities or resin cure anomalies before panels advance to high-cost die-attach packaging stages.

Substrate yield optimization requires balancing material performance against raw panel manufacturing cost steps. Ultra-low loss dielectric films featuring loss tangents below 0.002 at 28 gigahertz carry raw material prices three to five times higher than standard build-up films. Specifying high-cost materials does not compensate for yield loss caused by poor conductor line control or uncharacterized z-axis tensor mismatch.

Stackup selection must balance physical dielectric thickness, glass fabric style, and conductor plating roughness against target impedance tolerances to achieve optimal cost per working substrate unit.

Panel utilization efficiency dictates landed unit price for advanced packaging substrates. Placing high-frequency characterization coupons across panel center regions reduces available package layout area, increasing effective unit cost. Sourcing agreements must define minimum panel usage thresholds while retaining sufficient extraction coupon density to guarantee lot-to-lot electrical consistency.

When package substrate thickness drops below the core drill aspect limit, panel yield follows copper roughness rather than dielectric constant tolerance.