Mechanically Spread Glass Selection and Stackup Optimization for Skew Mitigation
Mechanically spread glass eliminates glass weave skew by expanding yarn bundles into flat ribbons that eliminate resin gaps and equalize dielectric constants.

Yarn

Glass Fabric Geometry and Bundle Mechanical Spreading Mechanics
Woven glass laminates rely on continuous E-glass or low-loss E-glass filaments twisted into yarns and woven on production looms. In standard fabrics such as 106, 1080, 2116, and 7628, tension during air-jet or water-jet weaving draws those yarns into tight, round bundles, leaving open gaps ~ resin windows ~ between adjacent warp and weft threads. Impregnating this weave with epoxy, polyphenylene oxide, or fluoropolymer resins creates an electrically non-uniform cross-section: the glass carries a relative dielectric constant between 6.6 for standard E-glass and 4.4 for low-Dk formulations, while the surrounding resin matrix sits between 2.6 and 3.2.
Spread glass fabrics undergo secondary processing to flatten these round bundles into flat ribbons prior to impregnation. Running the woven fabric through high-pressure water jets, ultrasonic baths, or differential flex rollers separates individual filaments and spreads them laterally across the weave plane. That expansion narrows the resin windows and levels the bundle peaks in both axes.
Styles like 1035, 1067, 1078, 2113, and 3313 are spread counterparts to traditional weaves, yielding a flatter profile that suppresses local dielectric swings.
Filament redistribution during air-jet spreading reduces open resin windows without altering total glass weight.
Flattening bundle topography changes how copper foil seats against the dielectric. With standard weaves, bundle knuckles create pronounced peaks and valleys across the weave pitch, forcing prepreg resin to pool in the recesses during lamination and setting up periodic dielectric steps that perturb trace impedance and phase delay. Spread styles establish a far more uniform glass-to-resin distribution along the conductor run, preventing the localized signal acceleration that occurs whenever traces cross open resin pockets.

Warp and Fill Structural Properties across Standard and Spread Fabrics
Evaluating fabric geometry requires measuring bundle width, thickness, and thread pitch in both warp and fill directions. Warp yarns run the length of the roll under sustained tension, while fill yarns run across its width. In standard 1080 glass, warp and fill bundle widths measure roughly 220 micrometers and 180 micrometers respectively, leaving resin gaps exceeding 180 micrometers.
Mechanically spread 1078 uses the same yarn filaments but broadens bundle widths beyond 330 micrometers, bringing resin window gaps below 40 micrometers.
| Glass Style | Fabric Status | Nominal Thickness (mm) | Warp Pitch (picks/in) | Fill Pitch (picks/in) | Resin Window Gap (μm) | Glass-to-Resin Ratio (Vol %) |
|---|---|---|---|---|---|---|
| 106 | Standard Unspread | 0.033 | 56 | 56 | 210 | 18 / 82 |
| 1035 | Mechanically Spread | 0.028 | 65 | 72 | 35 | 28 / 72 |
| 1080 | Standard Unspread | 0.055 | 60 | 47 | 185 | 32 / 68 |
| 1078 | Mechanically Spread | 0.043 | 60 | 54 | 30 | 36 / 64 |
| 2116 | Standard Unspread | 0.094 | 60 | 58 | 110 | 42 / 58 |
| 2113 | Mechanically Spread | 0.078 | 60 | 56 | 25 | 45 / 55 |
| 3313 | Mechanically Spread | 0.081 | 61 | 62 | 20 | 47 / 53 |
| Measurements compiled from optical microsection profiling and laser diffraction fabric analysis per IPC-TM-650. Nominal pressed thickness varies based on laminate resin content during press cycles. | ||||||
Structural density governs how much glass volume sits along any given signal path. Standard 106, for example, has a very low glass-to-resin ratio, leaving wide voids filled purely with resin. A differential pair routed over 106 sees dramatic swings in effective dielectric constant whenever one leg tracks a yarn bundle while the other runs over a resin window.
Spread weaves like 1067 and 1078 close those windows into an almost continuous sheet of filaments, keeping the dielectric environment consistent beneath both conductors.
Velocity variations across standard glass grades are typically attributed to raw fabric availability limits and broad industry tolerance bounds.

Dielectric

Mechanism of Glass Weave Skew in High-Speed Differential Channels
Propagation velocity along a printed circuit trace scales inversely with the square root of the surrounding effective dielectric constant. Because electromagnetic fields in microstrip and stripline geometry penetrate both glass bundles and resin regions, trace alignment relative to the weave dictates local speed. If one leg of a differential pair tracks directly over a yarn bundle while its mate runs over an adjacent resin window, the two experience unequal dielectric environments, accumulating phase skew over the length of the run.
Past 10 gigabits per second, dielectric heterogeneity rapidly degrades signal margins. The trace running over the glass bundle lags behind the trace over resin, and that phase mismatch converts differential energy into common-mode noise, driving eye closure, jitter, and radiated emissions. At 28 gigahertz and 56 gigahertz PAM4 signaling rates, an intra-pair skew greater than a quarter of a unit interval will overwhelm receiver equalization.
Determining phase skew requires calculating the effective dielectric differential across the pair. The delay difference per unit length derives directly from propagation velocity:
Phase Delay Delta Formula ~ Delta-t = (1 / c) (sqrt(Dk_eff1) – sqrt(Dk_eff2)) L
Here, c is the speed of light in vacuum, Dk_eff1 is the effective dielectric constant over the glass bundle, Dk_eff2 is the value over the resin window, and L is physical channel length. Standard E-glass in epoxy yields a delta-Dk as high as 0.45 on standard 1080 weaves, translating to 3.5 picoseconds per inch of intra-pair skew. Spread glass drops that delta-Dk below 0.05, cutting skew below 0.4 picoseconds per inch for the same trace geometry.
A dielectric constant delta of 0.35 across a four-mil trace pitch generates 2.4 picoseconds per inch of intra-pair phase skew at 28 gigahertz.
Frequency Dependent Dielectric Anisotropy and Signal Phase Delay
Dielectric values published on laminate data sheets reflect bulk averages measured at discrete frequencies via split-post resonators or clamped-stripline fixtures. Those figures smooth out the microscopic spatial variations present across an actual panel. As frequencies climb, electric fields concentrate tighter along conductor edges, sharpening the impact of local glass-resin boundaries and causing phase delay to vary with trace width and conductor pitch relative to bundle spacing.
| Laminate Grade | Glass Style | Resin Dk (10 GHz) | Glass Dk (10 GHz) | Max Local Dk Delta | Phase Skew at 10 GHz (ps/in) | Phase Skew at 28 GHz (ps/in) |
|---|---|---|---|---|---|---|
| Standard Mid-Loss FR-4 | 106 Standard | 3.10 | 6.60 | 0.48 | 3.82 | 4.15 |
| Standard Mid-Loss FR-4 | 1067 Spread | 3.10 | 6.60 | 0.08 | 0.62 | 0.68 |
| High-Tg Low-Loss FR-4 | 1080 Standard | 2.85 | 6.60 | 0.38 | 2.95 | 3.20 |
| High-Tg Low-Loss FR-4 | 1078 Spread | 2.85 | 6.60 | 0.05 | 0.38 | 0.41 |
| Ultra-Low-Loss PTFE/Resin | 3313 Spread | 2.45 | 4.40 (Low Dk) | 0.02 | 0.15 | 0.16 |
Microstrip routing shows higher sensitivity to weave skew than stripline because field lines extend into air and solder mask above the trace, making the asymmetric substrate underneath a larger contributor to overall delay. Striplines balance field lines between upper and lower ground planes, and pairing spread glass across both sides helps average out any residual dielectric variation through the core and prepreg layers.
High-speed interconnect design requires accounting for several distinct failure mechanisms tied to weave asymmetry:
- Intra-pair phase skew ~ Delay mismatches between true and complement conductors reduce differential amplitude and collapse timing margins.
- Common-mode conversion ~ Unbalanced propagation shifts differential signal energy into common-mode noise, provoking EMI across backplanes and cabling.
- Impedance periodic ripple ~ Periodic shifts in local Dk introduce distributed impedance discontinuities, generating resonant return-loss spikes at harmonic frequencies.
- Channel-to-channel delay skew ~ Glass density gradients across wide bus routing introduce lane-to-lane arrival offsets, restricting synchronous bus lengths.
Left unaddressed, localized velocity differences degrade eye symmetry past what receiver equalizers can recover, resulting in scrapped production lots and late-stage respins.

Angle

Design Routing Mitigations versus Material Selection Alternatives
Skew mitigation relies on two paths: changing trace angles relative to the weave axes, or specifying spread-glass laminates during stackup design. Common layout workarounds include routing differential pairs off-axis using zig-zag segments, drawing traces at fixed angles, or rotating entire board artwork between 2.0 and 10.0 degrees against the panel rails. All three approaches break the parallel alignment between traces and underlying yarn bundles.
These routing methods carry clear layout and fabrication penalties. Zig-zag routing consumes channel space on dense signal layers, driving up layer counts and crowding pin-grid array escapes. Panel rotation avoids congested routing channels but forces fabricators to bias-cut rectangular master sheets, generating edge scrap that pushes up bare-board costs on volume orders.

Does Mechanical Spreading Eliminate off Axis Routing Requirements?
Specifying dense spread weaves such as 1067, 1078, or 3313 lets layout designers route orthogonal to panel edges without complex angular constraints. Up to 28 gigabits per second NRZ or 56 gigabits per second PAM4, spread glass alone keeps intra-pair skew within typical receiver budgets over trace lengths up to 15 inches, removing the need for serpentine or zig-zag length tuning.
At 112 gigabits per second PAM4 or runs past 20 inches, combining spread fabrics with slight off-axis routing provides the required safety margin against weave variations and roughness-induced phase shifts. Biasing artwork by under five degrees or angling escapes out of dense packages prevents traces from tracking residual yarn gaps over long distances.
- Establish the maximum allowable intra-pair skew from receiver jitter budgets.
- Select a spread-glass style such as 1078 or 3313 that provides the target dielectric thickness.
- Check conductor pitch against fabric bundle spacing to prevent harmonic pitch aliasing.
- Balance the scrap penalty of artwork rotation against the routing layer additions of orthogonal routing.
- Document minimum prepreg plies and exact fabric style designations on the fabrication master drawing.
Angling signal paths off the primary weave axes suppresses phase mismatch while permitting standard laminate grades.

Build

Prepreg and Core Optimization in Multi-Layer High Speed Stackups
Stackups tuned for low skew balance fully cured rigid cores against uncured prepreg plies. Cores arrive fully cured under press conditions with tight thickness and Dk tolerances, whereas prepreg flows during board lamination to fill copper clearances. Stacking multiple thin prepreg plies instead of one thick ply lowers skew substantially; two offset plies break up the continuity of resin windows, averaging out dielectric variations through the layer thickness.
Choosing a spread weave requires matching target dielectric thickness with resin percentage. A 1078 ply with 64 percent resin content yields roughly 2.1 mils (53 micrometers) of pressed thickness, balancing resin fill against dielectric stability. Running two plies of 1078 per signal layer delivers far better dielectric consistency than a single ply of standard 2116, despite identical overall height, though tighter weave styles carry higher raw material costs.
| Mitigation Strategy | Glass Style Specified | Panel Scrap Rate (%) | Relative Material Cost | Routing Density Impact | Max Skew (ps/in) |
|---|---|---|---|---|---|
| Standard Glass / Straight Routing | 1080 Standard | 0.0 | 1.00 | Baseline (100%) | 3.20 |
| Standard Glass / Panel Rotation (10 deg) | 1080 Standard | 14.5 | 1.22 | Baseline (100%) | 0.45 |
| Standard Glass / Zig-Zag Routing | 1080 Standard | 0.0 | 1.00 | Reduced (75%) | 0.50 |
| Spread Glass / Straight Routing | 1078 Spread | 0.0 | 1.12 | Baseline (100%) | 0.38 |
| Spread Glass + Low-Dk Filament | 3313 Spread (Low-Dk) | 0.0 | 1.35 | Baseline (100%) | 0.15 |
In high-layer boards, mirror symmetry around the center plane controls warpage during reflow and preserves electrical consistency. Distributing spread prepreg symmetrically across inner signal layers keeps phase velocities balanced across the stackup. Stripline traces should sit equidistant between reference planes to prevent return-path asymmetries that skew effective Dk.

Fabrication Drawing Specifications and Procurement Terms
Fabrication prints govern factory material procurement and process limits. Generic notes specifying only ‘high-Tg FR-4’ give board shops room to substitute standard, unspread weaves to trim raw material expenses. Enforcing skew controls requires drawing notes that specify exact glass styles, target resin contents, and IPC-4101 slash sheets for every dielectric layer.
Specifying IPC-4101 slash sheet 131 with spread glass callouts on fabrication drawings prevents material substitution during secondary sourcing.
Drawings must forbid glass substitutions without written engineering sign-off. Material notes should identify qualified laminate systems tested under IPC-TM-650 phase delay protocols, and because narrow conductors accentuate windowing effects, routing callouts must specify tight line-width tolerances alongside allowable dielectric thickness windows.
Required Fabrication Drawing Notes for Skew Control ~
- Material Specification ~ Dielectric material must comply with IPC-4101/131, exhibiting Tg greater than 170 degrees Celsius and Td greater than 350 degrees Celsius.
- Explicit Glass Styles ~ All prepreg layers assigned to high-speed signal layers must utilize mechanically spread glass styles 1035, 1067, 1078, or 3313 as detailed in the stackup schedule.
- Substitution Restrictions ~ Material substitution involving unspread glass styles 106, 1080, or 2116 is strictly prohibited without written authorization from the design authority.
- Plies per Layer Rules ~ High-speed stripline signal layers must incorporate a minimum of two prepreg plies per dielectric gap to ensure weave pattern randomization.
- Registration Tolerances ~ Layer-to-layer registration must meet IPC-6012 Class 3 requirements to maintain differential reference balance after multi-layer lamination press cycles.
Coupling IPC-4101 slash sheet numbers with explicit glass style callouts in purchase orders binds fabricators to the intended dielectric construction.

Qualification

Inspection Verification Coupon Design and Test Methods
Verifying skew control on production panels demands dedicated test coupons placed on the fabrication rails. Ordinary impedance coupons cannot capture intra-pair skew or local phase variations. Instead, coupons using SPP (Short Pattern Differential) or SET2DTR (Single-Ended Routing 2D Time Domain Reflectometry) structures route long parallel segments to measure velocity deltas across the weave.
These verification coupons run differential traces between 6 and 20 inches long aligned directly to warp and fill axes. Microstrip or stripline probe pads terminate each end for vector network analyzer (VNA) or high-bandwidth TDR measurement. Following IPC-TM-650 Method 2.5.5.7, test fixtures measure electrical length differences in picoseconds across sweeps from 100 megahertz to 40 gigahertz to isolate velocity mismatch.
Coupons placed on panel margins yield accurate phase measurements only when their trace layout mirrors active signal layer glass alignment.
Microsections taken from panel rails complement RF testing by revealing bundle flattening and resin fill directly. Quality labs mount and polish cross-sections under optical microscopes to verify that filaments remain evenly distributed across resin windows without bundle deformation, resin starvation, or voiding.
Coupon data confirms whether physical velocities match simulated budgets, protecting receiver timing margins across full channel lengths without compromising fabrication yield.
As per-lane signaling pushes toward 224 gigabits per second, testing continues on whether ultra-low Dk glass yarns can offset weave skew on their own without requiring secondary mechanical spreading.




