Equipment-ready materials recipes for your exact process — from first principles.

What does LabForge actually deliver?

A complete, equipment-ready deposition recipe for your specific chamber — power, pressures, gas flows, ramps, bias, temperature, expected film properties, failure-mode warnings, and QC acceptance criteria. Not a textbook range; a spec calculated for the machine on your floor.

Why are the free recipes arcing, doming, and gold TiN — and what does the paid version add?

Because they are the three failures that eat the most chamber time in reactive PVD, and each has a known mechanism — so a general recipe can actually solve them:

• Arcing is target poisoning. Reactive-sputtering a metal (Ti, Al, Cr) in O₂ or N₂ grows a dielectric skin — Al₂O₃, TiO₂, TiN — on the cathode, and charge builds on the insulating patches until it breaks down. The general fix is asymmetric pulsed-DC: a reverse pulse that discharges the surface before it arcs, run on the metallic side of the reactive-gas hysteresis loop.
• Doming is compressive-stress buckling. Low-pressure, high-energy ion peening drives the intrinsic film stress compressive; past a critical σ·t the film delaminates and domes. The general fix trades energy for pressure — raise the working pressure and trim bias to walk σ back toward neutral, holding ΔT so the film/substrate CTE mismatch doesn't finish the job.
• Gold TiN is a stoichiometry window. Ti + ½N₂ → TiN is only gold and hard near 1:1 Ti:N; N-poor runs brown, N-rich runs violet and soft. The general fix is closed-loop control of the N₂ partial pressure — by discharge voltage or optical-emission feedback — to pin the operating point inside the narrow gold band.

That is the mechanism and the canonical operating regime — what a general recipe can give you. What it cannot give you is the exact number, because the pulse frequency and duty, the O₂/N₂ flow in sccm, the pressure, and the power all scale with your chamber: pump speed, target size, magnetron, throw distance, gas manifold, and power supply. The free recipe is general by design. The paid Single Recipe is computed for your exact gear — which is why the order asks for your chamber, source, power supply, and gas lines. Give us the hardware and the general regime collapses to the setpoints that work on the machine on your floor.

Why does my TiO₂ come out as absorbing, sub-stoichiometric TiOₓ instead of a clear dielectric?

You're running the Ti target in metallic mode — not enough O₂ reaches the growing film, so you build oxygen-deficient TiOₓ: blue-grey, absorbing, semi-conductive, with Ti³⁺ and O-vacancy states in the gap. Clear, stoichiometric TiO₂ (n ≈ 2.4) needs the film fully oxidized, which pushes toward the poisoned regime where the rate collapses and arcing returns. The real operating point is the transition zone, held by O₂ partial-pressure feedback (plasma-emission or λ-probe), often delivering more O₂ near the substrate than near the target so the cathode stays metallic while the film finishes oxidizing. Where that O₂ setpoint lands in sccm is a function of your pump speed and chamber volume — the general recipe gives you the control strategy; the exact flow comes from your gear.

My CrN coating delaminates off steel (Fe) — how do you fix adhesion?

Delamination is almost always the interface, not the film. Steel carries a native Fe/Cr oxide and adsorbed H₂O and hydrocarbons; nitride nucleates poorly on it and the compressive CrN stress peels it off. The fix is a clean, graded start: an Ar⁺ sputter-etch to strip the oxide in-vacuo, then a thin metallic Cr strike (no N₂) for a metallurgical bond to Fe, then ramp N₂ to grade Cr → CrN with no abrupt modulus or stress step, plus substrate bias to densify and a stress budget that keeps σ·t below the delamination threshold. The etch time, bias, and N₂ ramp all scale with your target power, bias supply, and pumping — the general recipe is the sequence; the numbers are yours.

Even with pulsed-DC, reactive Al → Al₂O₃ still arcs — how do you stabilize the hysteresis?

Al₂O₃ is one of the worst offenders: it's a near-perfect insulator, so charge on a poisoned Al target re-accumulates between pulses, and the metallic-Al ↔ Al₂O₃ reactive loop is steep and bistable at high pump speed. Pulsed-DC alone treats the symptom. Stabilizing the loop means fast reactive-gas feedback — optical-emission or discharge-voltage control with bandwidth above the loop's own time constant — usually with high pumping to steepen and pin the transition, plus a dual-magnetron mid-frequency AC pair so each target is alternately anode and cathode, which cures the disappearing-anode problem that makes a single Al₂O₃ cathode arc. Your loop's time constant is set by chamber volume, pump speed, and gas-injection geometry, so the general recipe specifies the control architecture; the tuning constants come from your hardware.

Is this a chatbot or AI guessing?

No. LabForge is a proprietary first-principles physics engine that computes every parameter from the mechanism up. It does not scrape ranges or interpolate folklore, and it is not a language model.

Does LabForge actually reason to new answers, or just retrieve known recipes?

It reasons — and we stress-tested exactly that. We handed the engine a problem pinned at the true frontier, with a stack of hard constraints that no established method satisfies all at once (real-world operating conditions, passive and reagent-free regeneration, earth-abundant materials, long-cycle durability). Instead of reaching for the nearest known material, it produced a non-obvious reconciliation — turning one of the problem's own constraints, normally treated as a liability, into the mechanism that makes the solution work. An independent AI review then recomputed the load-bearing thermodynamics from scratch, confirmed they close, could not dismiss the result as retrieval, and honestly flagged the one claim that still needs a bench measurement. That is the standard: a physics engine that derives the answer, states its confidence, names its own weak point, and hands you a single falsifiable test — not a chatbot matching to the nearest paper.

What do I need to provide?

The film's function and a measurable target (hardness, sheet resistance, refractive index, stress, adhesion), plus your hardware: chamber and base pressure, source and target material, power supply (DC, pulsed-DC, RF, or HiPIMS), process gases and MFC ranges, substrate, max temperature, and bias.

What if the recipe doesn't work on my machine?

Then it tells us something useful. The engine computes the physical requirement exactly; what no spec sheet can tell it is how far your specific tool has drifted from its nameplate — an eroded target changing plasma density, a mass-flow controller reading high after hundreds of hours, a thermal sensor offset by buildup. The first run doubles as a measurement: you report what the film actually did, and the engine baselines your hardware's real state against the computed ideal and locks the spec to that offset. Every recipe includes 3 revisions and 1 deep analysis for exactly this. There are no cash refunds; the calibration commitment is the assurance.

Is my chamber and process data safe?

Yes. Your configuration is treated as your confidential trade secret — used only to build your recipe, never sold, never shared, and never used to train any public model. The engine runs privately and offline; only your finished recipe ships back.

What does it cost?

A single recipe is $599. Recipe packs are $2,495 for 5 and $8,980 for 20. Professional is $2,299 per month for 4 recipes with rollover. Enterprise is custom volume and white-label.

How do I pay?

After you place an order, we email you a secure Stripe payment link.

Can consultants resell LabForge recipes?

Yes. Approved partners buy recipe packs at partner pricing and deliver recipes to their own clients under their own brand.