Dielectric Constant Measurement and Prepreg Glass Weave Selection in Multilayer Board Design

Bench

Datasheet dielectric constants reflect test setups and frequencies that rarely mirror real operating conditions. Quoting a relative permittivity of 3.8 at 10 GHz, for instance, often relies on a clamped stripline test under IPC-TM-650 Method 2.5.5.5, where an unclad sample sits loosely between fixtures. That arrangement leaves air between substrate and foil, depressing the measured permittivity compared to a fully cured, copper-clad laminate on a finished PCB.

On an actual panel, a microstrip signal sees a complex composite of glass filaments, cured epoxy, and surface-treated copper with microscopic roughness peaks.

Evaluating high-speed laminates requires distinguishing raw material test methods from operational extractions. Resonant cavity techniques, such as the split-post dielectric resonator in IPC-TM-650 Method 2.5.5.13, measure bulk dielectric properties along the transverse electric plane of unclad slabs. Here, the resonator excites a TE011 mode inside a cavity bisected by the sample, using shifts in resonant frequency and quality factor to calculate in-plane relative permittivity and loss tangent.

This isolates intrinsic resin-glass properties at discrete spot frequencies ~ typically 1.1 GHz, 2.45 GHz, 5.0 GHz, or 10.0 GHz. Because field vectors remain parallel to the sample surface, split-post fixtures miss out-of-plane permittivity entirely, even though it governs z-axis electric field propagation between reference planes in multilayer stackups.

The z-axis relative permittivity measured via clamped stripline at 10 GHz systematically underestimates real-circuit effective permittivity by three to six percent due to surface roughness and fixture air entrainment.

Clamped stripline testing under IPC-TM-650 Method 2.5.5.5 uses mechanical pressure to hold two unclad dielectric sheets around a central resonator card. Electric field vectors run perpendicular to the sample faces, matching the z-axis orientation in multilayer boards. Pressure alone, however, leaves microscopic air voids at the interface.

With air sitting at a dielectric constant of 1.0006, those voids act as a low-permittivity series capacitance, yielding a value lower than what a copper-clad trace experiences after its metal foil bonds tooth-to-resin during high-temperature lamination.

Inversion methods pull effective dielectric constants directly from fabricated test structures using vector network analyzers. By measuring scattering parameters over a broad sweep, phase velocity and propagation delay yield frequency-dependent permittivity. Time-domain reflectometry tracks the arrival time of a fast voltage step down a precision trace of known physical length to resolve phase velocity.

Testing physical traces captures the complete system ~ resin content, glass weave geometry, foil profile, and solder mask coverage. Teams working at high frequencies rely on short-pulse propagation or multiline Thru-Reflect-Line calibration patterns to untangle copper losses from the bulk dielectric dissipation factor up to 50 GHz.

Copper foil topography directly alters the effective dielectric constant measured on high-speed lines. Standard electrodeposited copper uses a pronounced tooth structure to grip the epoxy matrix during pressing. Modern low-loss laminates opt for smoother foils ~ classified as Very Low Profile or Very Very Low Profile ~ with root-mean-square roughness below 1.5 micrometers and 0.7 micrometers, respectively.

High-frequency currents concentrate within the conductor’s shallow skin depth, which drops below the height of those copper teeth as frequency climbs. Field lines then loop around peaks and valleys, extending the wave’s path and storing extra energy in surrounding resin. The resulting phase delay reads mathematically as a higher effective dielectric constant: a material with an intrinsic resin-glass value of 3.6 can shift above 4.1 at 28 GHz under standard reverse-treated electrodeposited copper.

Temperature shifts cause additional drift across a board’s operating range. As epoxy resin approaches its glass transition point, it expands and alters structurally. This movement is tracked by the thermal coefficient of dielectric constant in parts per million per degree Celsius.

Standard FR-4 exhibits positive thermal coefficients between 100 and 300 parts per million per degree Celsius, dropping impedance and stretching propagation delay as temperatures rise. By contrast, hydrocarbon, PTFE, and specialized polyphenylene ether resins stay below 40 parts per million per degree Celsius, holding timing margins steady across industrial environments from negative 40 to positive 125 degrees Celsius.

Bare laminate panels stored in ambient air absorb moisture, impairing dielectric stability. Water carries a static dielectric constant around 70 at room temperature, and epoxy matrices absorb up to 0.3 percent atmospheric moisture by weight under humid conditions. Introducing polar water molecules into non-polar resin raises both the relative permittivity and dissipation factor of the cured prepreg.

Pre-baking inner layers or finished boards before assembly drives off trapped water and restores nominal dielectric values prior to testing or reflow.

High-speed system architects remain divided on whether standard coupon screening can reliably predict full-wave performance across 112 gigabit per second pulse amplitude modulation links, or if custom TRL test vehicles are necessary for every batch of incoming material.

Pitch

Woven glass fabric provides mechanical rigidity and dimensional stability to cured cores and prepreg bonding sheets. Standard E-glass uses filaments drawn from silica, alumina, and lime, resulting in a solid glass permittivity of 6.1 at 10 GHz. Surrounding resins ~ epoxy, cyanate ester, or polyphenylene ether ~ sit lower, between 2.8 and 3.4.

The cured prepreg ends up as a composite of alternating high-dielectric glass bundles and low-dielectric resin windows.

Glass fabric styles differ in yarn count, filament diameter, linear density, and weave tightness. Style 106 uses fine filaments in an open grid with wide gaps between warp and fill yarns, resulting in low glass volume and an average pressed dielectric constant near 3.3. Style 7628 takes the opposite approach, using heavy, tightly packed yarns for a dense composite near 4.4.

Intermediate styles such as 1080, 2116, and 3313 land between those endpoints based on their glass-to-resin ratio.

Glass Weave Architecture and Physical Parameters

Filament bundle geometry determines the pitch of dielectric variations across the board plane. Standard weaves leave square or rectangular resin windows where warp and fill yarns cross. A trace running over a resin window sees a local permittivity dominated by epoxy, while its partner over a glass bundle sees a value weighted toward glass.

That difference introduces skew between the two legs of a balanced differential pair.

Spread-glass styles address these variations by flattening yarn bundles before weaving. Using air jets or ultrasonic processing, round bundles are drawn into broad, thin ribbons. Styles 1035, 1067, 1078, and 3313 close up open resin windows, creating a nearly continuous sheet of glass across the substrate so differential pairs see a uniform background wherever they lie.

1. Style 106 Prepreg is a lightweight open weave using 5-micrometer filaments, with nominal resin content near 68 percent and a pressed dielectric constant of 3.31 at 10 GHz.
2. Style 1078 Prepreg uses mechanically spread yarns in a tight geometry, reducing warp-and-fill bundle pitch while maintaining 64 percent resin content and a dielectric constant of 3.42 at 10 GHz.
3. Style 2116 Prepreg offers intermediate thickness using 7-micrometer filaments, with 54 percent nominal resin content and a dielectric constant of 3.78 at 10 GHz.
4. Style 7628 Prepreg provides maximum structural rigidity using heavy 9-micrometer filaments, supporting 43 percent resin content with a bulk dielectric constant of 4.15 at 10 GHz.

Resin content sets both the final pressed thickness and average dielectric constant of a prepreg layer. Heat and pressure during lamination force resin to flow into inner-layer copper clearances. High-resin glass styles flow extensively, filling tight copper features while thinning out over solid planes.

The cured layer’s permittivity scales linearly with resin volume fraction following a simple parallel mixing rule ~ more resin drops the overall dielectric constant, though it elevates z-axis thermal expansion above Tg.

Low-dielectric glass formulations substitute E-glass filaments with modified chemistries known as NE-glass or Low-Dk glass. NE-glass carries an intrinsic permittivity near 4.6 at 10 GHz, well under the 6.1 of standard E-glass. Combining NE-glass fabric with low-loss polyphenylene ether resin produces laminates with aggregate dielectric constants under 3.0 and dissipation factors below 0.002.

Closing the gap between glass and resin permittivity suppresses skew, even across long runs on unspread fabrics.

Matching the spatial pitch of the glass weave pattern against differential trace spacing eliminates local dielectric velocity variations without forcing multi-degree angle routing routines.

Fabricators estimate pressed thickness with empirical flow models based on copper weight and layout density. A nominal 1080 prepreg sheet rated at 75 micrometers unpressed often compresses to 62 micrometers over solid copper, or down to 52 micrometers over sparse signal layers with large clearance voids. Calculating impedance accurately demands using permittivity values matched to the net resin content left after flow.

Relying on raw datasheet figures leads to impedance errors up to eight ohms on standard 50-ohm lines.

Specifying spread glass across all prepreg and core layers prevents localized timing skew on differential pairs operating above 10 gigabits per second.

Transit

Propagation velocity along a trace scales inversely with the square root of effective dielectric constant. Microstrip waves travel at speeds governed by a combination of laminate, soldermask, and air. Striplines, fully buried between reference planes, see a uniform environment dictated entirely by the resin-glass matrix.

If local permittivity fluctuates along a route, phase velocity shifts, introducing timing uncertainty across parallel signal lines.

Differential skew occurs when the legs of a balanced pair experience unequal phase velocities. In serial links running at 28 or 56 gigabaud, rise times drop under 15 picoseconds. A mismatch of just two picoseconds between traces closes the differential eye, converts signal energy into common-mode radiation, and degrades receiver jitter tolerance.

Unspread E-glass fabrics create local permittivity swings from 3.2 to 4.2 within less than a millimeter; routing parallel to the weave axis with one trace on glass and the other on resin builds up maximum skew over distance.

Propagation Mechanics in Anisotropic Substrates

Phase velocity variations across heterogeneous weaves generate measurable skew over standard panel dimensions. Total skew depends on interconnect length, the Dk delta between glass and resin, and conductor orientation relative to yarn pitch. The phase delay delta across a pair follows directly from basic transmission line relations:

Delta_t = (L / c) ( sqrt(Er_max) – sqrt(Er_min) )

Here, L represents total routing length, c is the speed of light in vacuum, Er_max is effective dielectric constant over a dense glass bundle, and Er_min is effective dielectric constant over a resin window. For standard E-glass with a Dk delta of 2.7, Delta_t reaches 7 picoseconds per inch. On a 20-inch backplane run, cumulative skew can exceed 140 picoseconds, shutting the eye completely at 10 GHz.

Design teams break spatial alignment with glass patterns through routing layout choices. Setting traces at an off-axis angle relative to panel edges forces conductors to cross glass bundles and resin windows evenly. Angles between two and five degrees average out delay differences along the line.

Alternatively, zig-zag routing introduces small periodic bends to achieve the same weave-averaging effect without rotating panel artwork in CAD.

Fabricators can also tilt artwork on the production panel, rotating gerber files by 5.5 degrees against the panel border so straight traces run diagonally across the weave. However, panel-level rotation wastes 8 to 12 percent more material around margins, raising board costs. Selective spread-glass or CAD-level zig-zag routing achieves equivalent timing without the material penalty.

Dual-ply prepreg configurations reduce dielectric variation by stacking two thin glass layers instead of one thick sheet. Combining two plies of style 1078 or 1035 creates an overlapping weave where window alignment through the entire thickness is statistically negligible. Average permittivity smooths out across the dual layers, holding peak differential skew below 1.5 picoseconds per inch without needing angled routing rules.

Outer-layer solder mask loads microstrip traces unevenly at high frequencies. Liquid photo-imageable masks exhibit dielectric constants between 3.3 and 3.8, alongside higher dissipation factors near 0.02 at 10 GHz. Thickness variations over trace corners and narrow gaps alter permittivity and impedance.

Moving critical routes to internal stripline layers shields conductors from soldermask tolerances, embedding signals entirely within controlled dielectric core and prepreg layers.

Under IPC-6012 Class 3 performance specifications, high-speed differential signal pairs must maintain phase skew limits below five picoseconds across the total routed length as verified through TDR coupon testing on finished production panels.

Ledger

Material selection directly governs bare-board production costs, panel utilization, and supplier options. Advanced low-loss laminates with NE-glass and high-Tg polyphenylene ether resin cost three to five times more than standard high-temperature FR-4. Because raw laminate represents 30 to 45 percent of the manufactured cost on a complex 16-layer board, over-specifying non-critical layers drives up expenses without offering signal performance benefits.

Hybrid stackups manage total panel cost by mixing material grades within one board. High-speed signals run on thin, low-loss cores and low-Dk spread prepreg, while power and control lines use standard high-Tg FR-4 cores and economical E-glass. Building hybrid panels requires matching z-axis expansion, cure temperatures, and lamination pressure profiles across distinct materials ~ mismatched cure cycles lead to delamination, micro-voids, or warping during reflow.

Panel utilization dictates how effectively board footprints fit master manufacturing sheets. Fab shops typically run 18 by 24 inch or 21 by 24 inch panels. Tooling borders, test coupons, and optical targets take up roughly 1.5 inches along each edge, shrinking net usable space.

A board measuring 4.2 by 6.1 inches hits 72 percent utilization on an 18 by 24 inch panel; increasing a dimension by just 0.2 inches drops nesting efficiency to 54 percent, boosting unit cost by 25 percent.

Cost Mechanics of Laminate Grades and Weave Choices

Bare-board pricing reflects material selection, layer count, copper weight, and surface finish. Baseline models compute cost per working panel, spreading fixed overhead across net usable boards per sheet.

Total Panel Cost = ( C_mat Area ) + C_lam + C_drill + C_plate + C_finish + ( C_yield / Y_panel )

In this model, C_mat is unit material cost per square metre, Area is panel surface area, C_lam covers lamination labor and energy, C_drill accounts for hole density and aspect ratio penalties, C_plate covers chemical copper deposition, C_finish adds surface plating fees such as ENIG or Immersion Silver, and Y_panel is the final electrically verified panel pass rate.

1. Standard High-Tg FR-4 carries a base material index of 1.0, using E-glass and standard epoxy for non-critical, low-frequency digital assemblies.
2. Mid-Loss High-Tg Laminate carries a material index of 1.8, using modified epoxy resin with 1080/2116 E-glass for signals up to 10 gigabits per second.
3. Low-Loss Hydrocarbon/PPE carries a material index of 3.2, using spread E-glass and low-Dk resin systems rated for 28 gigabit per second links.
4. Ultra-Low Loss NE-Glass Laminate carries a material index of 4.8, using flat NE-glass filaments and specialized resin matrices engineered for 112 gigabit per second PAM4 channels.

Foil selection adds further increments to the quote. Standard electrodeposited copper acts as the baseline. Upgrading to Very Low Profile foil adds 5 to 8 percent to core costs, while Ultra-Low Profile rolled-annealed copper adds 12 to 18 percent.

High-aspect microvias and heavy copper power planes lengthen plating cycles, generating extra processing surcharges.

Hybrid stackups combining low-loss signal cores with standard FR-4 power planes reduce master panel material costs by up to 35 percent compared to full homogeneous low-loss builds.

Fab shops pass yield losses to buyers via lot charges or adjusted unit pricing. Pairing tight-weave spread glass with line-and-space widths below 75 micrometers pushes manufacturing limits. If panel yield drops from 95 down to 78 percent due to layer registration shifts or micro-voids, per-unit pricing increases to absorb scrap panels and lost capacity.

Extended lamination press dwell times required to fully flow high-viscosity low-loss resin into dense inner-layer power copper clearances drive up panel manufacturing rates.