1. Why comparing parameters row by row costs two review cycles
Open the SKPCB materials library and you get a table a few dozen rows deep: type, manufacturer, product, Tg, Td, Dk, Df, application. Almost everyone's first instinct is to read it horizontally — sort the Tg column high to low, or circle the rows with the smallest Df.
That instinct is where the rework starts. The reason is concrete: an entire group of materials in that table has N/A in the Tg column. Rogers TMM10, TMM10i, TMM13i, RT/duroid 6002, RO3003, RO3006, RO3010, RT/duroid 6006, RT/duroid 6010, RT/duroid 5870/5880 and Taconic TLY-5 all show N/A there. That is not a gap in the page. Those families are not described by a glass transition temperature in the first place.
So "sort by Tg" quietly pushes the whole ceramic/PTFE space to the bottom of the list or off it. If your board happens to be a microwave board, you just discarded the only correct answer. Run it the other way and sort by lowest Df, and what you circle — TLY-5 at 0.0009, RT/duroid 5870/5880 at 0.0012, RT/duroid 6002 at 0.0012 — is the PTFE space. For a six-layer digital board that only has to run DDR, that trades cost and process difficulty for a figure of merit you will never use.
That is how two review cycles appear. The first one takes a part number picked by cross-family comparison to the manufacturing side and hears that this board's structure neither needs it nor suits it. Only on the second one does the question become "which column should I have been reading".
Parameters rank candidates inside a family. They do not select between families. Pick the family first, then read the numbers. Reverse that and you rework.
2. Filter one: the application decides the family
The leftmost column of that table is type; the rightmost is application. The first step of selection is not in the four numeric columns in between — it is in how those two outer columns map onto each other.
The page sorts base materials into six families: FR-4, Polyimide, Ceramic Polymer, Ceramic/PTFE, Hydrocarbon/Ceramic and PTFE Composite. The page text itself puts it this way: "Your application should influence your material choice — for RF/microwave, signal integrity or temperature cycling." Those three application words are exactly the recurring values in the Application column, plus Standard.
Those four values are the entire scale of the first filter:
- Standard — general digital, power and control boards. In the table: Isola 370HR, ITEQ IT180A, Ventec VT-42 / VT-47 / VT-481, Kingboard KB-6167F, and halogen-free VT-447. All FR-4.
- Signal Integrity — high-speed digital links where loss has to be controlled. In the table: Isola FR408HR / FR408 / Itera MT40, Panasonic Megtron 4 / 6 / 7, Nelco N4000-13 EP SI, halogen-free Ventec VT-464(D). Still mostly inside FR-4, just the high-speed branch of it.
- RF / Microwave — actual RF and microwave circuits. This is where Isola Astra MT77 and Tachyon 100G appear, along with the Rogers TMM series, the RO3000 / RO4000 series, the RT/duroid series and Taconic TLY-5 — the home ground of Ceramic Polymer, Ceramic/PTFE, Hydrocarbon/Ceramic and PTFE Composite.
- Temperature Cycling — boards that see large swings repeatedly. In the table: Arlon 85N, Ventec VT-901 / VT-90H, Isola P95/P25. All polyimide.
The output of filter one is not a part number. It is one or two families. Once it is done, most rows in the table no longer concern you, and the remaining three filters only operate on what is left.
Scope note: this article covers the cross-family order only. If filter one lands you in RF/Microwave, the trade-offs between part numbers from there are in high-frequency PCB material selection and Isola material selection by frequency. If it lands you in the ceramic or metal-core thermal direction, see the ceramic substrate PCB guide and aluminum MCPCB thermal management.
3. Filter two: Tg/Td answers whether the board survives the process
With the family fixed, the second filter is thermal stability — Tg (glass transition temperature) and Td (decomposition temperature).
One thing has to be said first: this is a gate to pass, not a column of scores. And it is a within-family comparison, never across families. As above, the ceramic/PTFE space shows N/A for Tg in the materials library, so comparing an FR-4 Tg against those rows means nothing.
Inside FR-4 the typical-value spread the library gives is clear: Ventec VT-42 at Tg 140 / Td 310, Ventec VT-481 at 155 / 345, ITEQ IT180A at 175 / 345, Kingboard KB-6167F at 175 / 350, Panasonic Megtron 4 at 176 / 360, Isola 370HR at 180 / 340, Ventec VT-47 at 180 / 355, Ventec VT-447 at 180 / 380, Isola FR408HR at 190 / 360, Panasonic Megtron 6 at 185 / 410, with Megtron 7, Isola Itera MT40 and Astra MT77 in the Tg 200 band and Nelco N4000-13 EP SI at 210 / 350.
The polyimide family sits in a different segment entirely: Ventec VT-901 and VT-90H at Tg 250 / Td 390, Arlon 85N at 250 / 407, Isola P95/P25 at 260 / 416. That is the physical origin of the Temperature Cycling application value — not that these materials are "better", but that their thermal stability window is somewhere else.
⚠️ The library's own qualifier applies to every number above and has to travel with it: "Values are typical; consult the manufacturer datasheet for complete and accurate data." Use these figures to build a sense of magnitude and of relative position within a family. Anything entering a design freeze or an acceptance document goes back to the manufacturer datasheet.
In practice, filter two means matching the thermal history this board will actually see — layer count and how many lamination passes it implies, high-temperature assembly steps downstream, or a service environment that cycles hard — against that within-family spread. You are choosing a segment, not the top row. And read Td with Tg: the table contains rows at the same Tg with a 40°C difference in Td (VT-47 at Td 355 and VT-447 at Td 380, both at Tg 180). Watching one number alone hides the difference in decomposition behaviour.
4. Filter three: Dk/Df only means something after the electrical target exists
The third filter is dielectric constant (Dk) and loss tangent (Df). It sits third not because it matters less, but because without a stated electrical target neither number can be called good or bad.
Dk sets impedance and propagation velocity for a given geometry; Df sets dielectric loss. The spread in the library is wide: Dk runs from Taconic TLY-5 at 2.20 and RT/duroid 5870/5880 at 2.33 up to Rogers TMM13i at 12.85, with RO3010 and TMM10i at 10.2 and 9.8; Df runs from TLY-5 at 0.0009 to Kingboard KB-6167F at 0.017 and Isola P95/P25 at 0.0179.
High Dk is not a defect and low Dk is not a virtue — they serve different circuit forms. A high-Dk material like RO3010 (Dk 10.2) exists so RF structures can be made physically smaller at a given frequency. A low-Dk, low-loss material like RT/duroid 5880 (Dk 2.33, Df 0.0012) exists so loss stays controlled over long traces. Ranking them against each other to find "the best Dk" answers a question nobody asked.
By the same logic, if filter one put your board in the Standard band, the difference between Df 0.015 and Df 0.0031 has no observable effect on it, and switching material for it just imports process difficulty and material constraints together. Df is worth paying for once dielectric loss is a line item in the link budget.
This filter is also within-family and under one fixed electrical target, and it carries the same page qualifier: values are typical, consult the manufacturer datasheet. Note too that several values are given as ranges — Isola 370HR at Df 0.015–0.025, Ventec VT-464(D) at Dk 3.0–3.6 and Df 0.004–0.008, Nelco N4000-13 EP SI at Dk 3.3–3.7 and Df 0.007–0.008. The range is itself a signal that these parameters depend on construction and test conditions rather than being single constants.
5. Filter four: manufacturability is last, and the only one that overrules the first three
The first three filters all happen inside the materials table. The fourth happens outside it: can the manufacturing side actually build the structure this board has to become.
This filter has to exist because base material choice never stands alone — it is locked in together with layer count, stack-up, hole size and trace width. The SKPCB capabilities page states the bare-board manufacturing windows: 2–64 layers, maximum panel 1250×570 mm, board thickness 0.2–6.0 mm, copper from 0.5oz to over 12oz, minimum trace/space 2/2 mil, minimum hole 0.075 mm, mechanical drill from 0.15 mm and laser drill from 0.075 mm, hole diameter tolerance ±0.05 mm, positional accuracy ±0.025 mm, aspect ratio ≤10:1, back-drill control ±0.05 mm; special processes include impedance control ±5% (±10% standard), blind and buried vias, any-layer, resin plugged vias, copper paste filled vias, castellated holes and gold fingers.
⚠️ That page carries its own qualifier, which has to travel with any citation of it: "The listed windows are not universal commitments for every product; materials, stack-up, dimensions, processes and acceptance criteria are confirmed from project data."
In practice filter four asks three questions:
- Can the material selected by the first three filters support the layer count and board thickness this board needs? Layer count and thickness together decide how the dielectric is distributed and where the hole aspect ratio lands — the ≤10:1 aspect ratio window is a hard edge, and it acts jointly with your material choice rather than being a property of the material alone. The structural trade-offs of multilayer construction are on the multilayer PCB product page.
- Does the impedance requirement demand Dk consistency from this material? The capabilities page lists impedance control ±5% as a special process, with ±10% as standard. Once a design reaches the ±5% band, material choice is no longer a matter of reading a typical Dk value; it has to be engineering-confirmed together with the structure of this specific board.
- Has this combination been done before? This is the one item no table on any page can answer, and the one most worth raising before design freeze. The capabilities page is explicit that its windows are confirmed from project data. Read the other way round, that sentence says: submit material, stack-up, dimensions, processes and acceptance criteria together, and you get a definite answer.
Filter four can overrule the first three. The reverse never happens. That is why it sits last but carries the most weight: the first three filters make a choice on paper, the fourth confirms it on a real line. And whether the choice was right is ultimately answered by observable results on the board — which is why material confirmation ends up under a microscope.
6. The four filters as one funnel
Pulling the order together, this is a one-way funnel, not four scoring criteria you can shuffle:
| Order | Filter | Input | Output | Why it cannot move earlier |
|---|---|---|---|---|
| 1 | Material family | Application: Standard / Signal Integrity / RF-Microwave / Temperature Cycling | 1–2 families | Without a family you compare parameters across families and discard the N/A rows by accident |
| 2 | Tg/Td | The thermal history the board will see | A thermal band inside the family | Tg is not comparable before the family is fixed |
| 3 | Dk/Df | An explicit electrical target | Candidate part numbers inside the family | Without an electrical target, Dk/Df have no better or worse |
| 4 | Manufacturability | Layer count, thickness, hole size, trace width, impedance band | A buildable combination, or a return to filters 1–3 | Nothing to verify against until the structure is defined |
Doing filter one first removes most of the invalid comparisons in the other three. Keeping filter four last but never skipping it removes the "selected on paper, rejected on the line" cycle. That is the whole mechanism behind saving two review cycles — there is no shortcut, you just stop running the filters backwards.
To confirm a material and structure combination for a specific board type, submit layer count, thickness, hole size, trace width and impedance requirements together through the engineering contact on the SKPCB materials library.
FAQ
Q1: Can you just tell me the single best base material?
No, and the question itself means filter one has not been applied. The materials library sorts base materials by application into Standard, Signal Integrity, RF/Microwave and Temperature Cycling, and the best answer in each band lives in a different family. With no application input, any part number recommendation is a guess.
Q2: Why is Tg N/A for some materials? Is data missing?
Nothing is missing. The Rogers TMM series, RO3000 series, RT/duroid series and Taconic TLY-5 all show N/A for Tg in the library because those ceramic/PTFE systems are not characterised by a glass transition temperature. That is the direct reason Tg cannot be sorted across families.
Q3: Is higher Tg more reliable?
No. Tg is a gate to pass, not a score. It has to match the thermal history the board actually sees, and it has to be read together with Td — the library contains cases with the same Tg and a 40°C difference in Td (both at Tg 180, VT-47 at Td 355 and VT-447 at Td 380). All values in the table are typical; consult the manufacturer datasheet for complete and accurate data.
Q4: My board is ordinary digital. Should I move to a low-Df material?
First confirm whether dielectric loss is actually a line item in the link budget. If filter one placed this board in the Standard band, going from Df 0.015 to 0.0031 brings no observable improvement while importing material constraints and process difficulty together.
Q5: Is high Dk better, or low?
Neither is "better". The library spans Dk from TLY-5 at 2.20 to TMM13i at 12.85; high Dk is used to make RF structures smaller, low Dk with low loss is used to control loss on long traces. They serve different circuit forms. The values remain typical; consult the manufacturer datasheet.
Q6: Once the material is chosen, can manufacturing still overrule it?
Yes, and that is what filter four is for. Structural constraints like layer count, thickness, hole size and the ≤10:1 aspect ratio limit which material combinations work; once impedance reaches the ±5% band (±10% is standard), material and structure have to be engineering-confirmed together. The capabilities page states its own position: the listed windows are not universal commitments, and materials, stack-up, dimensions, processes and acceptance criteria are confirmed from project data.
Q7: How does this article divide up with the materials page and the high-frequency articles?
The materials library page provides the data (families, part numbers, Tg/Td/Dk/Df, application). This article provides the order in which to use that table. The high-frequency and single-family articles cover how to settle on a part number once filter one is done.