What the 1,650 °C hydrogen co-fire is actually doing

What HTCC means

HTCC stands for high-temperature co-fired ceramic. The sequence is: cast a ceramic powder into a flexible green tape, screen-print the conductor pattern onto individual sheets, stack and laminate them, then fire the whole stack once so the ceramic densifies and the metal sinters in the same cycle. The result is a monolithic body with the circuit inside it, not on it. That is the structural difference from a thick-film heater, where the resistor is printed on an already-fired substrate and stays exposed. Everything that follows — the 4,500 V / 1 s withstand, the absence of power drift, the 700–1,000 °C ceiling — traces back to this one decision to bury the circuit and fire it together with the alumina. Read the datasheet as a description of a co-fired body and the numbers stop looking arbitrary.

• Check: the sheet describes a co-fired monolithic body, not a surface-printed thick-film element.
• Check: substrate line reads 96% Al₂O₃ multilayer HTCC and resistor reads tungsten / Mo-Mn paste.
• Criterion: the circuit is buried; a surface track means a different process and different limits.
• Criterion: one firing cycle densifies ceramic and sinters metal — no post-fire resistor print.

96% alumina green tape

The green tape is alumina powder with a glass phase and an organic binder, cast into a sheet that is flexible enough to punch, via and print before it is fired. 96% Al₂O₃ is the workhorse composition: it gives thermal conductivity of 20–25 W/m·K, a coefficient of expansion of 8.1 × 10⁻⁶/K over 20–1,000 °C, flexural strength around 300 MPa and Young's modulus near 300 GPa. Those four numbers are not catalogue decoration — they are what the buried tungsten has to be compatible with. A tape with more glass lowers the firing temperature but also lowers conductivity and strength; a purer alumina is harder to co-fire with a metal. Ask for the composition and the green-tape thickness, because both set the fired thickness of 0.7–2.0 mm.

• Check: composition stated as 96% Al₂O₃, with the glass / binder system named.
• Check: green-tape thickness and the number of layers behind the fired 0.7–2.0 mm body.
• Criterion: thermal conductivity 20–25 W/m·K — it sets how small the through-thickness gradient is.
• Criterion: CTE 8.1 × 10⁻⁶/K must be matched by the buried metal, or the body cracks on cooling.

Tungsten / Mo-Mn paste printing

The circuit is screen-printed as a paste of tungsten, often with Mo-Mn and glass frit, onto the green tape. Because the resistor is a printed pattern, its value is a geometry problem: R = R_sheet × N × (L / W), where R_sheet is the sheet resistance in ohms per square and N is the number of squares in the meander. A 60 W, 24 V part needs 9.6 Ω from R = V² / P; at 20 mΩ/sq that is 480 squares, for example 24 passes of 20 squares each. The line width that fits those squares into the footprint is what fixes the real watt density. Tungsten is chosen because its expansion — 4.5 × 10⁻⁶/K — is lower than alumina's 8.1 × 10⁻⁶/K, so the ceramic ends up in compression around the circuit after firing.

• Check: paste system named (tungsten / Mo-Mn with glass frit) and sheet resistance in Ω/□.
• Check: the printed meander satisfies R = R_sheet × N × (L / W) for the target resistance.
• Criterion: tungsten CTE 4.5 × 10⁻⁶/K vs alumina 8.1 × 10⁻⁶/K leaves the ceramic in compression.
• Criterion: line width and square count fit the footprint without pushing watt density past 45 W/cm².

Stacking, lamination and warm pressing

Individual printed sheets are aligned, stacked and laminated under heat and pressure so the stack behaves as one body before it ever sees the kiln. Vias punched through the tape and filled with the same paste connect the resistor to electrodes on another layer; alignment here is what decides whether the termination lands where the drawing says. Lamination also decides whether there are voids between layers — a void inside the body is a lamination defect, and a crack that follows a printed layer points back to this step rather than to the material. Ask how layer-to-layer registration is held and how the stack is checked before firing. The goal is a green body dense enough that the single firing produces uniform shrinkage rather than a warped or delaminated plate.

• Check: via fill uses the same tungsten paste and lands on the intended electrode layer.
• Check: layer-to-layer registration tolerance and how the green stack is inspected before firing.
• Criterion: no inter-layer voids — a crack following a printed layer is a lamination defect.
• Criterion: uniform green density, so the later shrinkage is repeatable across the whole plate.

Why the atmosphere has to be hydrogen

This is the step that gives the process its name and its constraint. Tungsten and molybdenum oxidise readily in air at elevated temperature, forming oxides that are neither conductive nor adherent; at co-firing temperature an air atmosphere would turn the buried circuit into an insulating, voluminous oxide and the part would be scrap before it densified. A reducing hydrogen atmosphere keeps the metal as metal while the alumina sinters, which is why the furnace runs under hydrogen rather than air. The same atmosphere is why the paste can be a refractory metal at all — a metal that survives 1,650 °C is exactly the metal that air would destroy. Ask what atmosphere the kiln runs, how it is monitored, and whether the dew point is controlled; a reducing atmosphere that drifts is a lot-wide resistance shift, not a single bad part.

• Check: the firing atmosphere is reducing hydrogen and is monitored continuously through the cycle.
• Check: the supplier states why air is excluded — tungsten / Mo-Mn would oxidise instead of sinter.
• Criterion: no oxide at the metal–ceramic interface; an oxidised circuit is an open or a drift.
• Criterion: atmosphere control is a lot-level parameter — a drift shifts resistance across the batch.

Co-firing at about 1,650 °C

The green stack goes into the kiln once, at roughly 1,650 °C, and two things happen together: the alumina particles densify into a nearly fully dense body and the tungsten paste sinters into a continuous conductor. Doing both in one cycle is the whole point of co-firing — the metal and the ceramic shrink together, so the circuit ends up locked inside a monolith with no adhesive layer and no interface that can peel. It is also the hard part: the metal, the glass frit and the alumina must all have compatible shrinkage and must all survive the same temperature. The fired body is dense enough that the surface is gas-tight, which is what lets the next steps — metallisation, brazing, lead attachment — be done on the outside without ever exposing the circuit.

• Check: the stated co-firing temperature is about 1,650 °C, one cycle for ceramic and metal together.
• Check: metal, glass frit and alumina shrinkage are matched, or the part warps or delaminates.
• Criterion: fired body is dense and gas-tight — the buried circuit is not reachable from the surface.
• Criterion: no second resistor-printing step after firing; the circuit exists only because of this cycle.

Shrinkage and dimensional tolerance

A green tape part is always made larger than the finished part. If the linear shrinkage of the stack is s, every fired dimension is the green dimension times (1 − s), so the screen, the punched vias and the tape outline all have to be scaled by 1 / (1 − s) at the design stage. That scaling is only as good as its repeatability: a shrinkage that varies across the plate produces warp, and a shrinkage that varies between lots produces parts that miss the pocket. This is why dimensions are quoted as an envelope — L ≤ 100 mm, W ≤ 100 mm, fired thickness 0.7–2.0 mm — and why the drawing carries a tolerance rather than a nominal. Ask for the shrinkage value used for tooling and the lot-to-lot spread; the second number is the one that sets your incoming tolerance.

• Check: green tooling is scaled by 1 / (1 − s) for the stated linear shrinkage s.
• Check: the lot-to-lot shrinkage spread, which is what actually sets the dimensional tolerance.
• Criterion: shrinkage is uniform across the plate — a gradient shows up as warp, not as size error.
• Criterion: finished dimensions sit inside L ≤ 100 mm · W ≤ 100 mm · T 0.7–2.0 mm with tolerance.

Metallisation and brazing after firing

Firing produces the circuit, but the outside of the part still needs a way to attach a lead. Electrode pads exposed at the surface are plated — typically nickel — and the lead is brazed or welded to that metallisation, which is why the datasheet lists nickel leads of Ø0.25 / 0.3 / 0.5 mm, Ni tabs, wire, FPC or connectors. This is a metal-to-ceramic joint, and it is where expansion mismatch has to be managed: tungsten at 4.5 × 10⁻⁶/K sits comfortably against alumina at 8.1 × 10⁻⁶/K, while a stainless fitting at roughly 17 × 10⁻⁶/K pulls the ceramic surface into tension on cooling. That is the reason lead pull strength is specified as ≥ 2 kgf axial and ≥ 3 kgf lateral, and the reason failures cluster at joints rather than in the middle of a plate.

• Check: pad metallisation is nickel and the lead is brazed or welded, not merely pressed on.
• Check: lead pull strength ≥ 2 kgf axial and ≥ 3 kgf lateral is stated and tested.
• Criterion: keep low-CTE metals at the joint; a 17 × 10⁻⁶/K stainless boss puts ceramic in tension.
• Criterion: the braze fillet has a radius — a sharp fillet is a stress raiser, not a stronger joint.

Why the circuit is sealed inside

Three consequences follow from a buried, co-fired circuit, and all three are datasheet lines. First, the tungsten never sees oxygen in service, so it does not oxidise and the resistance does not drift — which is why an MCH part is quoted with no long-term power drift where a surface element is not. Second, there is no printed layer on the surface to peel or abrade, so the circuit cannot detach; a crack that follows an internal printed layer is a firing defect, not wear. Third, the surface is ceramic, not conductor, so it is electrically isolated: 4,500 V for 1 s with no breakdown, insulation resistance > 100 MΩ, and leakage < 0.5 mA on assembled parts. Those numbers are not added by a coating; they are the direct result of the circuit being inside a dense monolith.

• Check: no exposed resistor track — the circuit is inside the fired body, not under a glaze.
• Check: dielectric 4,500 V / 1 s, IR > 100 MΩ, leakage < 0.5 mA stated at part level.
• Criterion: stable resistance over life; a drifting power figure points to an exposed or oxidised element.
• Criterion: the surface is ceramic and not live, so the safety case does not depend on a coating.

The spec ceilings this process sets

Read the datasheet backwards from the process and the limits stop being arbitrary. The 700–1,000 °C maximum is set by the tungsten–alumina interface and the glass frit, not by the control loop. The 45 W/cm² ceiling is set by how hot the buried element runs for a given surface temperature and by how fast the alumina can carry that flux away — with a 0.635 mm plate, 3.15 W/cm² per kelvin, 45 W/cm² needs only about 14 K across the ceramic, so the limit is the interface, not the plate. The 200 °C in about 10 s heat-up comes from low thermal mass. The ±10% cold resistance at 25 °C comes from paste printing and firing uniformity. The 96% alumina body sets L ≤ 100 mm, W ≤ 100 mm and 0.7–2.0 mm thickness through shrinkage and flatness.

• Criterion: 700–1,000 °C is an interface limit — do not specify above it for a co-fired tungsten part.
• Criterion: 45 W/cm² is a continuous-duty ceiling; q″ = k · ΔT / t with 0.635 mm gives 3.15 W/cm² per K.
• Criterion: 200 °C in ~10 s assumes low thermal mass; added spreaders or clamps change it.
• Criterion: ±10% cold R at 25 °C reflects paste and firing control, not measurement noise.

How to verify a supplier really runs it

The process description is only useful if it is actually the supplier's process. The questions that separate a co-firing factory from a trading intermediary are about control and records, not about the catalogue: is sintering done in-house, will a resistance–temperature curve and dielectric test record be supplied for the part, what is 100% tested at end of line, and who owns the screen and fixture tooling. A module datasheet lists 100% dielectric, power and function testing plus lot number and test record traceability — ask to see that record for a pilot lot. Ask for a pilot lot before releasing production tooling, and ask which of the six stages — brief, DFM and circuit design, sampling, validation, mass production, delivery — the supplier itself controls. A supplier who can name the atmosphere, the shrinkage value and the EOL test plan is describing a process they run.

• Ask: are the co-firing kilns in-house, and what atmosphere and temperature do they run?
• Ask: for an R–T curve, a dielectric test record and a pilot lot before tooling is released.
• Criterion: 100% end-of-line dielectric, power and function test, with lot number and test record.
• Criterion: the supplier can state the shrinkage value, the EOL test plan and who owns the tooling.

Specifications

Questions buyers ask