Most material-selection articles answer “what are the key indicators” with a glossary: Dk, Df, CTE, Tg, thermal conductivity. A glossary does not help anyone choose, because it never says where each number starts to matter. What a selection decision needs is the opposite: for every indicator, the failure mode it controls, the threshold at which that failure appears, and the test that produces the number. Ten indicators carry almost all of the decision, and four of them decide nine out of ten real projects — Df if the link is loss-limited, TCDk if it is phase-limited, dielectric thickness tolerance if it is impedance-limited, and UL94 if the product has to pass a flammability review.
The Indicator List Is Not the Decision
A laminate datasheet is a table of measured properties; a selection decision is a ranking of which of those properties will fail first on a particular board. The two documents have different shapes. The datasheet lists forty rows in no order of importance; the decision usually hangs on three or four, and which three depends on the application rather than on the material.
That is why the useful form of the question is not “what are the key indicators” but “which indicator is binding here”. A 77 GHz radar board and a 100 G optical module use the same shortlist of laminates and are decided by different rows: the radar board by thermal drift of Dk and by z-axis expansion, the optical module by Df and dimensional stability. Selecting on the wrong row is the most common expensive mistake in this part of the industry — it buys a property nobody needed and leaves the one that fails unmeasured.
The table below is the whole shortlist in one view: ten indicators, the failure each one controls, and how the number is produced. The sections that follow give the thresholds. The same ten are cross-referenced in the broader guide to ten critical parameters, which covers the physics; this article covers the accept-or-reject line.
| # | Indicator | Failure mode it controls | How the number is produced |
|---|---|---|---|
| 1 | Dk (process and design) | Wrong impedance, wrong electrical length | Clamped stripline or split-post resonator near 10 GHz; design value from a field solver that includes glass and roughness |
| 2 | Df | Insertion loss, link budget | Same resonator fixture; must be quoted at the frequency of use, not at 1 MHz |
| 3 | TCDk | Phase drift over temperature | Resonator measurement over −40 to +150 °C, quoted in ppm/°C |
| 4 | Dk tolerance, batch to batch | Impedance spread between lots | Lot qualification data; rarely on the datasheet |
| 5 | Dielectric thickness tolerance | Impedance, phase | Micrometer on the finished core, or cross-section |
| 6 | CTE, x-y and z | Layer registration, PTH barrel strain | TMA per IPC-TM-650 2.4.24 |
| 7 | Td and Tg | Decomposition and drill smear across reflows | TGA for Td, DSC or TMA for Tg |
| 8 | Peel strength | Copper delamination after thermal cycling | IPC-TM-650 2.4.8, quoted in pli or N/mm |
| 9 | Thermal conductivity | Hot spot temperature in high-power RF | Steady-state or laser flash, W/m·K |
| 10 | Water absorption and UL94 | Loss drift in humidity; flammability review | IPC-TM-650 2.6.2.1; UL listing per grade and thickness |
Indicator 1: Dk Has Two Values and You Design With the Wrong One
Every high-frequency laminate is quoted with two dielectric constants, and confusing them is a five per cent impedance error before the first panel is etched. The process Dk is what a resonator measures on a bare laminate. The design Dk is what the laminate supplier recommends for field solving, and it is higher because it includes two effects the resonator does not see: the glass weave averaging across the trace width and the copper foil roughness that slows the field at the conductor surface.
| Laminate | Process Dk (10 GHz) | Design Dk | Gap | Consequence of using process Dk |
|---|---|---|---|---|
| RO4003C | 3.38 ±0.05 | 3.55 | +5.0 % | Line too wide, impedance about 2.5 % low |
| RO4350B | 3.48 ±0.05 | 3.66 | +5.2 % | Same direction, same magnitude |
| RO3010 | 10.2 ±0.30 | 11.20 | +9.8 % | Line too narrow for the same 50 Ω, thinner-etched, higher loss |
| RO6010.2LM | 10.20 ±0.25 | 10.7 | +4.9 % | Smaller gap, same trap |
The gap widens with Dk, which is why the 3000-series high-Dk grades are where this error costs most: at Dk 10 the same 5 % mistake moves a 50 Ω line by more than 3 Ω and changes its width by 12 %. The physics and the measurement methods behind the two values are set out in the function of different dielectric constants.
Dk also sets electrical length, and the sensitivity is only half as large as the Dk change: a 1 % increase in Dk lengthens the guided wavelength by about 0.5 %, which at 10 GHz on a Dk 3.38 stack is 0.09 mm of phase per wavelength, or 1.8 °. On a single 50 mm run that is 5 °; across ten wavelengths it is 18 °. For a phase-critical design, Dk tolerance is not an impedance issue at all — it is a phase budget line.
Indicator 2: Df, and the Only Formula You Need to Rank Grades
Df is the one indicator whose consequence can be calculated before anything is bought. The dielectric part of attenuation in a stripline or microstrip follows closely enough for ranking purposes:
αd ≈ 2.6 · fGHz · √Dk · Df dB/inch
Two things follow from it. Loss rises linearly with frequency, so a grade chosen at 10 GHz is being asked to do 7.7 times the dielectric loss at 77 GHz. And loss rises only with the square root of Dk, so a high-Dk material is not automatically lossy — the Df term dominates everything.
| Laminate | Dk | Df | Dielectric loss at 10 GHz | At 77 GHz |
|---|---|---|---|---|
| Standard FR-4 | 4.15 | 0.020 | 1.06 dB/in | 8.2 dB/in |
| RO4350B | 3.48 | 0.0037 | 0.18 dB/in | 1.39 dB/in |
| RO4003C | 3.38 | 0.0027 | 0.13 dB/in | 1.01 dB/in |
| RO3010 | 10.2 | 0.0022 | 0.19 dB/in | 1.48 dB/in |
| RO3003 | 3.00 | 0.0010 | 0.045 dB/in | 0.35 dB/in |
| RT/duroid 5880 | 2.20 | 0.0009 | 0.035 dB/in | 0.27 dB/in |
| Taconic TLY-5 | 2.20 | 0.0009 | 0.035 dB/in | 0.27 dB/in |
The ranking is what matters, and it is not linear with Df: 4350B at 0.0037 is 4.1 times lossier than 5880 at 0.0009, while FR-4 at 0.020 is 6 times lossier again. On a 200 mm (7.9 in) run at 10 GHz the difference between 4003C and 4350B is 0.39 dB — inside the tolerance of most link budgets. The same comparison at 77 GHz is 3.0 dB, which is the difference between a working and a non-working radar front end. That is exactly why a material that is perfectly adequate at 10 GHz is a mistake at 77 GHz, and why the loss behaviour of an ultra-low-loss grade deserves its own treatment, as in the analysis of the 5880 laminate.
One caveat on the input number: Df must be quoted at the frequency of use. A Df of 0.0009 at 1 MHz on a PTFE grade and a Df of 0.0009 at 10 GHz are not the same material claim, and mixing the two when comparing suppliers produces tables that look decisive and mean nothing.
Indicator 3: TCDk, the Number That Decides Phase Over Temperature
TCDk is Dk drift per degree Celsius, quoted in ppm. It is the indicator most often left out of a selection table and the one that most often explains a radar or phased-array problem that survived every other check.
| Laminate | TCDk | Dk change over 100 °C | Phase drift on 100 mm at 10 GHz | At 77 GHz |
|---|---|---|---|---|
| RT/duroid 5880 | −125 ppm/°C | −1.25 % | 16 ° | 125 ° |
| RO4003C | +40 ppm/°C | +0.40 % | 4.0 ° | 32 ° |
| RO4350B | +50 ppm/°C | +0.50 % | 5.1 ° | 39 ° |
| RO6035HTC | +66 ppm/°C | +0.66 % | 6.6 ° | 52 ° |
| RO6010.2LM | +425 ppm/°C | +4.25 % | 43 ° | 340 ° |
The 6010.2LM row is not a defect — it is a very high-Dk grade, and that is the price of the Dk. But it shows why TCDk has to be read together with the application: on a single-channel sensor the drift is common mode and can be calibrated out in software, while on a multi-channel phased array the drift is common mode only if every channel sits at the same temperature, which is never quite true. A 5 °C gradient across a 100 mm aperture on 5880 is 6 ° of phase between the ends of the array.
The sign matters as well. A negative-TCDk PTFE and a positive-TCDk hydrocarbon drift in opposite directions; in a hybrid stack that mixes the two, the phase error at temperature is the difference between the two materials’ behaviour along the same line, and it is larger than either number suggests.
Indicator 5: Dielectric Thickness Tolerance Is the Biggest Single Error Term
Impedance error is almost never dominated by Dk. On a 50 Ω microstrip the sensitivity to dielectric thickness is about 0.6 % of impedance per 1 % of h, and a laminate specified at ±10 % on thickness therefore contributes about ±6 % on its own — most of a ±10 % window before Dk and etch are counted. (Indicator 4, batch-to-batch Dk spread, is dealt with in the section on datasheet gaps: it is a lot-history question rather than a specification, and the number that matters is the spread between two deliveries, not the window on one.)
| Error source | Typical window | Effect on 50 Ω, h = 0.254 mm | Who controls it |
|---|---|---|---|
| Dielectric thickness | see below | ±6 % | Laminate supplier, press cycle |
| Dk tolerance | ±0.05 on 3.38 | ±2.4 % | Laminate supplier |
| Etch width | ±25 µm | ±2.8 % | Fabricator, CAM compensation |
| Solder mask | −1 to −3 Ω | −2 to −6 % | Fabricator, coupon practice |
| Bondply in the stack | 0.10 mm omitted | 50 → 61.1 Ω (+22 %) | Designer, stack-up arithmetic |
The ±10 % on thickness is not a supplier excuse; it is what a press does to a reinforced composite. It can be tightened to ±5 % on a specific grade with a dedicated press cycle and a thickness callout, and on high-Dk grades it is often quoted as an absolute figure such as ±0.05 mm rather than a percentage, because the percentage would be unachievable.
Two process effects move thickness after the laminate leaves the supplier. PTFE cold-flows under press: a 4 % reduction in dielectric thickness takes a 50 Ω line to about 48.8 Ω without any change to the artwork. And prepreg shrinkage of 0.02–0.05 % moves features across a 600 mm panel by 0.12–0.30 mm, which matters for registration long before it matters for impedance.
Indicator 6: CTE in Three Axes, Because They Fail Differently
CTE is always quoted as three numbers, and the three have nothing to do with each other. In-plane CTE (x and y) governs how far features move during lamination and reflow; z-axis CTE governs how far the hole walls are stretched. Selecting on a single “CTE” figure is how boards pass impedance and fail in the press.
| Laminate | CTE x / y | CTE z | Expansion of a 1.6 mm board over a 215 °C reflow | Plating requirement |
|---|---|---|---|---|
| RO3003 | 17 / 16 ppm/°C | 25 ppm/°C | 8.6 µm | Standard ductility sufficient |
| RO3010 | 13 / 11 ppm/°C | 16 ppm/°C | 5.5 µm | All three axes below copper — the selling point |
| RO4350B | 10 / 12 ppm/°C | 32 ppm/°C | 11 µm | Standard |
| RT/duroid 5880 | 31 / 48 ppm/°C | 237 ppm/°C | 82 µm | Elongation ≥20 %, aspect ratio ≤6:1 |
| Taconic TLY-5 | 26 / 15 ppm/°C | 217 ppm/°C | 75 µm | Elongation ≥20 %, aspect ratio ≤6:1 |
The 8- to 15-fold difference in z-axis expansion between a hydrocarbon and a PTFE grade is the single most consequential mechanical number in this list, because it decides two manufacturing rules at once. PTFE boards need plating with at least 20 % elongation so the barrel can stretch 75–82 µm without cracking, and they need a via aspect ratio of 6:1 or less so that the plating bath can deliver that ductility down the hole. A 3.2 mm thick 5880 board with 0.3 mm vias is a 10.7:1 aspect ratio, and no plating chemistry will save it.
In-plane CTE also decides whether a hybrid works. Over a 200 °C excursion and a 600 mm panel, free expansion of a 17 ppm laminate is 2.0 mm against 1.2 mm for a 10 ppm grade; the press constrains most of it, but the residual is what causes layer-to-layer movement in a mixed stack, and what the ±25 µm registration spec has to absorb.
Indicator 7: Td and Tg Answer Different Questions
Tg is the temperature at which a polymer stops being glassy; Td is the temperature at which it starts to decompose, measured as 5 % weight loss. They are routinely used interchangeably in selection documents and they should not be, because they predict different failures.
For thermoset hydrocarbon laminates the two are quoted and meaningfully different: RO4003C decomposes at 425 °C, RO4350B at 390 °C, while the ceramic-filled PTFE grades in the 3000 series are quoted at 500 °C. For PTFE the Tg concept barely applies — the polymer is semi-crystalline and its useful mechanical transition is far below any soldering temperature, so a high Tg claim on a PTFE datasheet should be read as a claim about the ceramic filler system rather than about the polymer.
What Td controls in practice is cumulative exposure: a board that sees three reflows and a rework pass spends perhaps 15 minutes above 250 °C, and each excursion above Td degrades the resin a little. It matters most where the same board is reflowed repeatedly, which is why the number is quoted alongside Tg in high-layer-count work rather than on its own. The definitions and the measurement methods are covered in glass transition temperature in PCB manufacturing.
Indicators 8 and 9: Peel Strength and Thermal Conductivity
These two are the mechanical indicators, and both are quoted in units that invite mistakes.
Peel strength is adhesion of copper to dielectric. Datasheets use either pli (pounds per linear inch) or N/mm, and the conversion is 1 pli = 0.175 N/mm — a factor of 5.7 between the numbers. A laminate quoted at 1.05 pli is 0.18 N/mm, which is a normal figure for a thermoset; a PTFE grade quoted in N/mm will carry a different-looking number for the same physical adhesion. Never put the two in one comparison column without converting.
| Laminate | Peel strength | Converted | Thermal conductivity |
|---|---|---|---|
| RO4003C | 1.05 pli | 0.18 N/mm | 0.71 W/m·K |
| RO4350B | 0.88 pli | 0.15 N/mm | 0.69 W/m·K |
| RO3003 | quoted in N/mm | — | 0.50 W/m·K |
| RO3010 | quoted in N/mm | — | 0.95 W/m·K |
| RO6035HTC | 7.9 pli | 1.38 N/mm | 1.44 W/m·K |
| RO6010.2LM | 12.3 pli | 2.15 N/mm | 0.86 W/m·K |
| RT/duroid 5880 | quoted in N/mm | — | 0.20 W/m·K |
Thermal conductivity spans a factor of seven across this shortlist, from 0.20 W/m·K for 5880 to 1.44 W/m·K for the ceramic-heavy 6035HTC. On a receive-only front end that number is irrelevant. On a 10 W power amplifier it sets the junction temperature: with 10 W/cm² into the same footprint, the 0.254 mm dielectric of a 1.44 W/m·K grade rises 17.6 K where a 0.69 W/m·K grade rises 36.8 K — a 19 K difference that is decided entirely by one datasheet row.
Indicator 10: Water Absorption and UL94, the Two That Disqualify
Most indicators trade off against each other. These two do not: they either pass or they do not, and either one can remove a laminate from the shortlist regardless of how good its Df is.
Water absorption on the PTFE grades is around 0.02 %, which is why their loss figures stay put through a damp-heat test. A thermoset absorbs an order of magnitude more, and the absorbed water raises Df and lowers Dk — a reversible change that reappears every humid season. For a link that is designed to 1 dB of margin, that is a qualification test, not a footnote.
UL94 is the trap. Some widely used high-frequency grades carry a V-0 listing and some carry none at all — RO4003C is in the second group, while the same family’s RO4350B is listed V-0, as are the 5880 PTFE grades. A design that needs a V-0 board and specifies a laminate with no rating has a problem that no amount of electrical performance fixes, and it usually surfaces at the safety review rather than at the material review.
Four Numbers the Datasheet Does Not Contain
The gaps in the datasheet are as decision-relevant as its contents, and four of them are worth asking for explicitly.
Batch-to-batch Dk spread. The ±0.05 on a 3.38 grade is a specification window, not a lot history. Two deliveries within that window produce a 4 % impedance difference on the same artwork, which is the mechanism behind “the second batch does not tune”.
Dk after lamination. Resin flow and bondply change the effective Dk of a bonded stack relative to the bare core, and the change is stack-specific. The only reliable number is a coupon measured on a panel built from the actual stack.
Copper roughness contribution. The Dk/Df pair describes the dielectric. At 10 GHz on a 0.0027 material, conductor loss and dielectric loss are the same order of magnitude, so the foil profile moves total insertion loss by tenths of a dB and appears in no laminate datasheet at all.
Process compatibility. Whether the material can be desmeared with permanganate or needs sodium-naphthalene or plasma, whether it can be laser drilled, whether it cold-flows in the press — these are capability questions rather than material questions, and they decide which fabricator can build the design. The comparison of material families in the types of high-frequency and high-speed materials is organised along exactly those lines.
Ranking the Indicators by Application
The same ten indicators produce different shortlists because different applications fail differently. The table below is the practical form of the whole article: find the application, and the binding indicator and its threshold follow.
| Application | Binding indicator | Threshold that decides | Typical grade |
|---|---|---|---|
| 77 GHz automotive radar | TCDk, CTE z, Dk tolerance | ≤100 ppm/°C, thickness ±3 %, Dk ±0.02 | 5880, TLY-5, RO3003 |
| 5G massive-MIMO antenna | Df, cost per area | ≤0.0027 at 10 GHz, Dk ±0.05 | RO4350B, RO4003C, RF-30 |
| Power amplifier, 10 W/cm² | Thermal conductivity | ≥0.9 W/m·K | RO3010, RO6035HTC |
| 100G/400G optical module | Df, dimensional stability | ≤0.003 at 10 GHz, 0.05 % stability | RO4835, ultra-low-loss digital grades |
| High-layer backplane | Td, CTE x/y | Td ≥390 °C, CTE ≤17 ppm/°C | RO4000 series, high-Tg hybrids |
| Consumer product with safety review | UL94, then Df | V-0 at the finished thickness | RO4350B, 5880 |
Two further comparisons are worth reading alongside this one: the Rogers material guide for the grade-by-grade figures, and the brand and laminate guide for how the Taconic and Panasonic families map onto the same indicator set. For optical-module work specifically, the model and performance summary for high-speed optical modules applies the Df and stability thresholds to that application in detail.
Conclusion: Ten Numbers, Four That Usually Decide
High-frequency material selection is decided by ten indicators, and the useful version of the question “what are the key indicators” is the one that also asks which of them binds. Df sets dielectric loss through αd ≈ 2.6·f·√Dk·Df, which spans 0.035 dB/in for 5880 to 1.06 dB/in for FR-4 at 10 GHz. TCDk converts that into phase drift over temperature, from 32 ° to 340 ° per 100 mm at 77 GHz across the grades in this article. Dielectric thickness tolerance contributes ±6 % of a 50 Ω window on its own, more than Dk and etch together. And UL94 quietly removes some grades from a product before any of the electrical numbers are read.
The four indicators that most often go unmeasured — batch Dk spread, Dk after lamination, copper roughness, and process compatibility — are also the four that most often explain a board that met every datasheet row and still missed its specification. Asking for them is part of selecting a material, not an extra step after it.
XCE PCB builds to those thresholds on both conventional and high-frequency stacks: 2 to 72 layers, panels to 600×1200 mm, impedance held to ±10 %, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. Coupons are measured on the panel that ships, because a datasheet is a claim about a laminate and a coupon is a statement about a board.
Frequently Asked Questions
Which indicator matters most in high-frequency material selection?
Should I design with process Dk or design Dk?
Is a Df quoted at 1 MHz usable for an RF design?
Why does dielectric thickness tolerance cause so much impedance error?
How much z-axis CTE can a plated through hole tolerate?
Does a laminate with a better Df always cost more?
Need the Indicators Verified on Your Stack, Not Just Quoted?
Send us your Gerber files and requirements for a manufacturability review and a tailored quotation.
sales1@xcepcb.com
+86 180 3341 4657