PCB Glass Transition Temperature: What Tg Decides, and What It Does Not

Glass transition temperature is the one thermal number almost every PCB quotation carries, and the one most often read as a reliability score. It is not one. Tg marks where a cured resin stops behaving like a glass and starts behaving like a rubber — nothing more. Below it sits the number that decides whether a board survives lead-free assembly (Td, and the T260/T288 times), and beside it sits a measurement problem: the same laminate can honestly be sold as Tg 150 or Tg 175 depending on which test produced the figure. This article separates the five thermal numbers that get bundled together, works out what actually fails when a board goes through reflow, and then explains why the question barely applies to the PTFE laminates used above 10 GHz.

A cured epoxy does not melt. Heat it and the resin passes through a second-order transition: the amorphous phase loosens, the modulus drops by one to two orders of magnitude, and the coefficient of thermal expansion steps up by roughly a factor of five. There is no latent heat and no liquid state — push further and the resin decomposes chemically instead of flowing. That distinction is the whole reason a single “Tg 170” line on a data sheet cannot describe a laminate’s thermal capability.

Five separate numbers are usually compressed into that one line. They answer five different questions, and only one of them is Tg:

Note what is missing from the first row: Tg says nothing about how long the resin lasts at temperature. It is the point where mechanical behaviour changes, and on a standard FR-4 that point is passed during every single reflow pass without the board failing. If Tg crossing were a failure criterion, no lead-free assembly would ever have worked.

Three test methods are in routine use, and they do not return the same number on the same coupon:

• DSC finds the step in heat capacity. It reports the lowest of the three, and it is the method named in IPC-4101 and in most laminate data sheets.
• TMA finds the knee in the expansion curve — where the CTE jumps. Typically a few kelvin above DSC.
• DMA finds either the onset of the storage-modulus drop or the tan δ peak. The tan δ peak is the highest of the three, and on the same board it can sit 20–30 K above the DSC figure.

The commercial consequence is straightforward. A mid-Tg laminate with a DSC Tg of 150 °C can be advertised as “Tg 175” with a DMA curve in the appendix, and a buyer comparing two quotes sees a 25 K advantage that does not exist. The fix costs one line in the fabrication note: specify the value, the method and the standard together — Tg ≥ 170 °C by DSC per IPC-TM-650 2.4.24 — and ask for a lot certificate rather than a brochure.

There is a second, quieter reason to compare on something other than Tg: the figure is easy to quote and hard to police. Td and T260 are less flattering to a cheap laminate and are harder to restate in marketing language, which is exactly why they are the better basis for comparing two suppliers.

When the board goes through reflow, the laminate expands in all three axes. X and Y are constrained by the glass cloth, so the movement there is small — 11–17 ppm/°C on FR-4, close to copper. Z is not constrained by anything, and Z is where the damage happens:

The plated barrel is copper, anchored at both ends by surface pads, and it must absorb that expansion as strain. Each excursion loads the same three locations, and they fail in a predictable order: a radial crack in the barrel wall; a corner crack where the barrel meets the surface pad; and pad cratering in the resin just beneath a BGA land, which is a laminate-and-copper-adhesion failure rather than a plating one.

Two details make this worse than the single-pass picture suggests. First, a double-sided assembly is not one excursion: top reflow, bottom reflow, a wave or selective solder pass, plus rework on the units that fail test — four or more full cycles to the same joint, on parts of the board that were already at temperature. Second, thick boards and high layer counts multiply the absolute expansion while adding more presses and more vias per unit area, which is why the layer count correlates with the Tg recommendation more strongly than the operating temperature does.

The design levers are the ones that show up in stack-up design: keep the aspect ratio inside the shop’s qualified range, keep the dielectric thickness under the via field as thin as the impedance allows, and treat the plating elongation as a strain budget rather than a process detail. A plating that reaches 20 % elongation will follow a PTFE stack through reflow; a brittle one will crack at 45 µm of expansion, which is a figure standard FR-4 reaches on its own.

FR-4 is not one material. The IPC-4101 slash sheets divide the family by Tg and Td, and the price steps are modest compared with the process steps they trigger:

Two notes on reading that table. The T260/T288 column is what actually gates lead-free assembly, and the industry floor for a lead-free-compatible laminate is generally T260 ≥ 30 min with T288 ≥ 15 min; a laminate can clear a Tg 170 requirement and still miss that. And the right-hand column is material cost only — a high-Tg resin also needs a hotter, longer press cycle and a harder drill recipe, so the delivered price gap is wider than the laminate gap.

Each of these shows up in real specifications, and each one costs either money or a field failure:

“Operating temperature must be 20 °C below Tg.” This is a reasonable floor, not a thermal rating. It gives 110 °C for a Tg 135 board and 150 °C for Tg 170, but continuous high-temperature operation is limited by long-term resin aging (the UL RTI figure, typically 130 °C for standard FR-4) and by Td — not by the softening point. A product that must run at 125 °C ambient is an RTI and resin-chemistry question before it is a Tg question.

“A higher Tg means lower loss.” There is no causal link. Within one Tg band, Df at 1 GHz spans roughly 0.012 to 0.020 depending on the resin system, the flame-retardant chemistry and the filler; a Tg 170 laminate can be lossier than a Tg 150 one. If loss matters, quote Df at the operating frequency and stop reading the Tg line.

“A higher Tg gives better CAF resistance.” Conductive anodic filament growth depends on resin chemistry (halogen-free systems are usually better), glass cloth type, hole-to-hole spacing, drilling quality and moisture history. It is assessed separately, under IPC-TM-650 2.6.25. Nothing about the Tg value predicts it.

“Tg affects heat transfer.” Thermal conductivity is set by the resin and the inorganic filler — 0.25–0.40 W/(m·K) across the FR-4 family, and 0.20 for PTFE against 0.71 for a ceramic-filled thermoset. Tg does not enter the calculation. The thermal path is a copper-and-via problem, worked through in high-frequency PCB thermal management.

Here the conventional FR-4 reading of Tg breaks down completely, and it is worth being blunt about why, because it explains a whole family of wrong specifications.

PTFE has a glass transition around −100 to −120 °C. Every PTFE-based laminate — RT/duroid 5880, Taconic TLY-5, the RO3000 series — is therefore above its Tg at room temperature, at storage temperature, and at every temperature it will ever see in service. If crossing Tg were the mechanism behind dielectric instability, these materials would be unusable, and they are the best low-loss laminates available.

The reason is that the RF properties come from the C–F bond and the symmetric, highly crystalline chain, not from segmental mobility in an amorphous phase. What changes Dk with temperature on PTFE is described by the temperature coefficient of Dk, and that number is small and often negative: −125 ppm/°C on 5880, against +40 on RO4003C and +50 on RO4350B. On a 100 mm line at 10 GHz, a 100 K excursion moves phase by about 15° on 5880 and by 4.4° on 4003C — the PTFE part moves more per kelvin and still wins on absolute loss, which is the actual trade being made.

For the RF family the useful questions are different: what is the TCDk, how tight is the Dk tolerance, what is the Z CTE, and how much elongation does the plating have. The general trade between these materials and FR-4 is set out in high-frequency PCB versus standard PCB, and the 5880 figures are collected in the 5880 laminate parameters.

Remove Tg from the discussion and the real mechanism is easier to see. A PTFE board expands about 5.1–5.6 % through a reflow excursion on 1.6 mm — 82 µm for 5880, 75 µm for TLY-5 over a 215 K swing — against 11 µm for a 4350B board. That is the number the fabrication note has to survive, and every PTFE process rule follows from it:

• Plating elongation ≥ 20 % and a qualified aspect ratio, typically 6:1 for PTFE against 10:1 or better on FR-4. A barrel that has to stretch 5 % must be ductile, not just thick.
• Hole activation by sodium naphthalene or plasma only. PTFE has no polar surface for a conventional desmear chemistry to grip, and the standard permanganate line will produce a hole that plates but does not adhere.
• No laser drilling on PTFE. Above roughly 400 °C the polymer depolymerises and releases hydrogen fluoride; the mechanical and CNC routes are the ones covered in drilling high-frequency PCB.
• Cold flow under mechanical load. PTFE is a thermoplastic with essentially no cross-linking, so a clamped connector or a bolted heat sink thins the dielectric permanently. A 4 % thickness loss moves a 50 Ω line to 48.8 Ω and does not come back. This is the real “Tg problem” of PTFE, and it is why a thermoset like 4003C appears in coil formers and clamped brackets — see the 4003C medical device article.

The plating thread continues in plating high-frequency PCB, and the TLY-5-specific expansion figures are in the TLY-5 process notes.

Everything above collapses into a short block that belongs in the drawing or the purchase specification, not in a conversation. It is deliberately written so that a supplier cannot satisfy it with a brochure figure:

The reflow-count line is the one most often left out and the one that changes the material. A board designed for two passes and then reworked twice sits at 2× the thermal dose its qualification assumed, and the failure shows up as a corner crack after a field thermal cycle rather than at assembly.

Choosing the higher rung is not a drop-in substitution, and the differences belong in the quote discussion rather than in the surprise column:

• Press cycle. The lamination temperature has to sit above the resin’s Tg for the prepreg to flow and cure, so a Tg 170 system runs a hotter and longer cycle than a Tg 135 one. Multi-lam boards absorb that cost several times over. The sequence is described in lamination in high-frequency PCB manufacturing.
• Drilling. A harder, more highly filled resin wears bits faster, so the drill recipe and the hit count per bit change. On very high Tg and polyimide materials this is the dominant yield factor, not the laminate price.
• Desmear. Higher cross-link density resists permanganate attack, so the desmear window narrows and the risk of either insufficient cleaning or excessive resin removal rises.
• Cost and lead time. Expect 20–35 % on material for a high-Tg grade, wider on polyimide, plus a longer press queue because fewer panels fit a cycle.

There is also a subtlety that matters on thick, multi-laminate builds: thermal history is cumulative. Each press and each reflow pass nudges the resin a little further along its cure and aging path, so a board that measures Tg 175 on the incoming certificate can measure a few kelvin lower after three press cycles. The allowance for that belongs in the margin, not in the specification.

In order, and stopping as soon as one question settles it:

• How many thermal excursions will this board see? Two reflows plus wave plus rework means the T260/T288 and Z CTE columns decide, and standard FR-4 is out at any layer count.
• What is the maximum continuous ambient? Above 105 °C, move to mid or high Tg and check the RTI, not the Tg.
• What is the board thickness and aspect ratio? Over 2 mm with small vias, the absolute Z expansion is the constraint even at moderate temperature.
• Is the board also an RF board? Then Tg is the wrong first question: start from Dk tolerance, Df at the operating frequency and TCDk, and pick the family from those.
• Is it PTFE? Then skip Tg entirely: specify Z CTE, plating elongation, aspect ratio, hole activation and the cold-flow clamp load.

Ten parameters ranked by how often they decide a design are collected in ten critical parameters for high-frequency PCBs, which is the companion checklist to this one.

Tg answers exactly one question: at what temperature the resin stops holding mechanical dimensions. Everything that gets attached to it — continuous operating temperature, dielectric stability, thermal performance, service life — belongs to other numbers: Td and T260/T288 for thermal survival, RTI for long-term operation, TCDk for dielectric stability, Z CTE and plating ductility for the failures that actually appear in the field.

Two habits prevent most of the mistakes. Specify Tg together with its method, so a DMA figure cannot be sold as a DSC one. And on RF laminates, start from Dk tolerance, Df and TCDk — on PTFE the Tg question does not apply, and asking it in an RFQ tells a capable supplier that the specification was copied rather than written.

XCE PCB laminates FR-4 and high-frequency materials from 2 to 72 layers, holds impedance to ±10 %, presses panels to 600×1200 mm and has documented 10,000+ builds since 2009 — with DSC verification on incoming laminate and a press recipe per material rather than per family. The wider process context is in the twelve-step manufacturing sequence.