Glass type and glass style are two separate choices
Glass cloth is the woven fiberglass reinforcement in every CCL and prepreg, and it is specified by two independent parameters: the glass type (E, NE, L, T, quartz), which sets the cloth's dielectric constant, loss and thermal expansion, and the glass style (1027, 1078, 2116, 7628 and so on), which sets its thickness and weave density. A 1078 cloth can be woven from E-glass or NE-glass; a T-glass cloth can be a thin or a thick style.
Most fabrication drawings name the style, because it sets dielectric thickness and impedance, and leave the type at the laminate supplier's default, which for most FR-4 and mid-loss materials is E-glass. That default is correct for a large share of boards. It stops being correct on two kinds of boards: high-speed boards where the cloth's Dk and weave create loss and skew, and large or thick boards where the cloth's expansion creates warpage and barrel stress. Those are exactly the AI and aerospace boards Tecsply builds.
Four glass types compared
E-glass has a fiber Dk of 6.8 at 1 GHz and a CTE of 5.6 ppm/°C, while T-glass brings CTE down to 2.8 ppm/°C and NE-glass brings Dk down to 4.8, according to Nittobo's published values. The table compares the four types we specify.
| Glass type | Dk (fiber) | Df (fiber) | CTE ppm/°C | Modulus | Relative cost | Design role |
|---|---|---|---|---|---|---|
| E-glass | 6.8 | 0.0035 | 5.6 | 75 GPa | 1× | Baseline for FR-4 and most mid-loss laminates. Fine for power, control and aerospace boards where loss is not the limit. |
| NE-glass (low-Dk; L-glass is a comparable family) | 4.8 | 0.0015 | 3.3 | 64 GPa | 2–3× | The reinforcement inside most very-low-loss and ultra-low-loss CCL. Lowers cloth Dk toward the resin Dk, which also reduces fiber-weave skew on differential pairs. |
| T-glass (low-CTE, high-modulus) | 5.4 | 0.0043 | 2.8 | 86 GPa | 3–5× | Warpage control on large-BGA accelerator boards and substrate-like PCBs; IC substrate (ABF/BT) cores; thick boards where z-axis expansion threatens plated through-holes. |
| Quartz | ≈ 3.7 | ≈ 0.0002 | ≈ 0.5 | ≈ 70 GPa | 10×+ | Millimeter-wave and space RF only. Difficult to drill; specified layer by layer, never for a full stackup. |
T-glass is not low-Dk glass
T-glass is specified for its 2.8 ppm/°C CTE and 86 GPa modulus, not for its dielectric properties: Nittobo lists its Df at 0.0043 at 1 GHz, higher than E-glass at 0.0035, so choosing T-glass is an expansion and stiffness decision, not a loss decision. For low Dk and low loss the reinforcement is NE-glass, at Dk 4.8 and Df 0.0015 against 6.8 and 0.0035 for E-glass at 1 GHz. The two families are often confused because both are "premium" cloths that cost two to five times E-glass and both appear in AI board discussions.
The distinction matters when reading a stackup. A signal layer that needs low Dk and low loss should carry NE-glass or L-glass; putting T-glass there buys expansion control the layer does not need and gives up Dk and loss margin. A core under a large package that needs to stay flat should carry T-glass; NE-glass there comes close on expansion (3.3 against 2.8 ppm/°C) but its 64 GPa modulus, against 86 GPa, lets the core bend more. Some suppliers now offer hybrid cloths that combine low CTE with low Dk, and we evaluate them case by case against availability.
Why reinforcement CTE matters: warpage and barrel stress
Copper expands at roughly 17 ppm/°C while E-glass expands at 5.6 ppm/°C and T-glass at 2.8 ppm/°C, and the board's in-plane expansion is a compromise among copper, resin and glass. Silicon packages expand far less than the board does. The larger the package, the larger the absolute mismatch across its footprint when the assembly goes through reflow, and the more the board bows or twists relative to the package.
Two failure modes follow. In-plane, a board that bows by even 0.2 mm across a large BGA field opens or bridges solder joints during reflow. Through the thickness, resin expansion stretches plated through-hole barrels on every thermal cycle. A stiffer, lower-CTE reinforcement helps with both: it restrains in-plane expansion and warpage directly, and its higher modulus resists bending.
Why AI boards now ask for both low-Dk and low-CTE glass
A current-generation AI accelerator baseboard carries packages larger than 80 mm on a side, 20 to 30 copper layers, and 112G or 224G lanes, so it needs low-Dk glass for its signals and low-CTE glass for its packages at the same time. The lanes want NE-glass for loss and skew. The package wants T-glass, because a board that bows by even 0.2 mm across the BGA field opens solder joints during reflow.
The practical answer is a mixed stackup: low-Dk cloth on the signal-pair layers, low-CTE cloth in the outer cores under the package, and a fabricator who can laminate both in one press cycle without registration drift. The AI compute page covers how we qualify that combination.
Why aerospace boards care about the same glass for a different reason
Avionics and satellite boards rarely run 112G lanes, but at 2.4–4.0 mm thick and cycling through wide temperature ranges they face the same expansion problem through the board's thickness. Here T-glass or low-CTE E-glass constructions cut z-axis expansion, which is the main driver of plated through-hole barrel cracking in IPC-6012 Class 3 thermal-shock testing. On these boards we treat the glass decision as a reliability decision and document it in the first-article report.
Our default rule is to consider T-glass once total board thickness passes 3.0 mm, and to use low-CTE E-glass constructions below that unless a package or program requirement says otherwise. The stackup calculator flags any stackup over 3.0 mm for this reason.
Glass style and fiber-weave skew
On differential pairs at 56G PAM4 and above, the glass weave itself causes skew, and Tecsply specifies spread-glass styles by default on any board above 25 Gbps. A trace running over a glass yarn sees a higher local Dk than one running over resin between yarns. Spread-glass styles (1067, 1078, 3313) flatten the yarn so the Dk is more uniform; low-Dk glass narrows the gap between glass and resin Dk; rotated-panel or zig-zag routing adds a further margin.
Glass type and style work together here. Spread NE-glass is the usual choice for 112G and 224G signal layers because it attacks skew from both sides: a flatter weave and a smaller Dk contrast between glass and resin.
Quartz cloth: layer by layer only
Quartz cloth has the lowest fiber Dk (about 3.7), Df (about 0.0002) and CTE (about 0.5 ppm/°C) of any reinforcement we specify, at more than 10× the cost of E-glass. It is hard to drill and is reserved for millimeter-wave and space RF layers, specified layer by layer and never for a full stackup.
T-glass supply and lead time
Production-grade T-glass cloth is dominated by a single Japanese supplier, with Taiwanese producers recently qualified, which makes T-glass the reinforcement most exposed to allocation during the 2026 glass-cloth shortage. Electronic-grade glass cloth prices rose by more than 100% year on year in 2026. We check T-glass lead time before a stackup is frozen and, where a design allows it, identify a low-CTE E-glass fallback construction in the same review so a program is not held on a single cloth.
How to call out glass on a drawing
A stackup drawing needs 2 glass entries per dielectric layer, type and style, plus ply count, so that the fabricator cannot substitute a default cloth on the layers where it matters. On Tecsply stackups we list, for every core and prepreg: glass type (E, NE/L, T, quartz), glass style, number of plies, and whether a spread-glass variant is required. Layers where the type is not critical are marked "supplier standard" so that the fabricator keeps flexibility where it costs nothing. See the thickness guide for style thicknesses.
Not sure whether your board needs T-glass?
Send the package size, board thickness and thermal environment. We return a stackup with glass type and style called out layer by layer.
References
- Nittobo, NE-glass and T-glass product pages (nittobo.co.jp): Dk/Df at 1 GHz, CTE and tensile modulus for E-, NE- and T-glass.
- TrendForce, "Nittobo reportedly plans 2028 next-gen T-glass" (Feb 2026): T-glass supply concentration.
- Panasonic Megtron and Isola Tachyon/Astra supplier datasheets: glass reinforcement by product.
- IPC-6012 (rigid PCB qualification and performance), IPC-TM-650 2.6.7 (thermal shock).
- NCAB Group PCB Supply Chain Outlook, September 2026; AtlasPCB material cost and lead-time reporting, July–August 2026.
Last reviewed . Values are typical published figures for stackup planning.