PFAS-Free
Self-Pumping
Fluid Discovery.
3M is shutting down Novec by end of 2025. The two-phase immersion stack that cools GPU clusters, edge compute, and orbital electronics is about to lose its incumbent PFAS dielectric — with EPA and EU restrictions closing the door behind it.
FluxZero is the PFAS-free replacement: a fluid family spanning a two-phase, self-pumping immersion genus and a single-component low-ε direct-to-chip siloxane, with an 89-candidate falsification ledger inside a 7.9M→931 PFAS-free genus funnel, 18 advancing to wetlab, and 230 patent claims filed across 3 provisionals — now consolidated, with the full program, into a single 330-claim omnibus draft.
Sixty Days. A Lot More Proof.
If you last saw FluxZero in April, here is what changed. The screen widened from a candidate shortlist to the entire PFAS-free chemical space, a second product band was added, the system-CHF ranking was hardened with uncertainty intervals, and every coolant idea across the program was consolidated into one ultra-long-form omnibus. The momentum is real — and so is the candor: everything below is in-silico.
Genus-scale discovery
The 89-candidate shortlist became a full PFAS-free chemical-space funnel: 7.9M PubChem SMILES → 931 stable, dielectric-relevant candidates, with a DFT electronic-dielectric screen. The original 89-set was found not to be the optimum — none of the 89 was simultaneously non-flammable and PFAS-free.
Direct-to-chip, added
A single-component low-ε siloxane (CF10-G) now covers the direct-to-chip band (ε < 6), where adding a Marangoni pump is counterproductive. The program now spans both immersion self-pumping and direct-to-chip — not two-phase only.
See the direct-to-chip playHardened system-CHF
Phase-B Monte-Carlo system-CHF ranking: the non-flammable flagship modeled at 357.6 W/cm² (90% CI 203.7–556.1), P(top-1) = 0.79; a low-ε non-flammable transition blend at 228.8 W/cm². Uncertainty-quantified intervals, not point estimates.
New molecular dynamics
First-ever component-σ molecular dynamics for the low-ε non-flammable pair (2026-06-09); a fresh GPU campaign extended component and blend coverage with multi-seed pooled uncertainty quantification.
IP estate scaled
Invention dossiers grew to 48 (39 counsel-ready) and were consolidated into a single 330-claim omnibus draft spanning 14 inventive aspects — on top of the 230 claims already filed across 3 provisionals.
Customer-grounded re-optimization
Re-ranked against 8 objectives (now including toxicity) and 7 use cases (now including cold-plate / direct-to-chip), with competitors in-frame and leak-risk total-cost-of-ownership.
Every figure here is computational (molecular dynamics, DFT, CFD, or Monte-Carlo) or literature/estimate — none is wet-laboratory measured as of today. Exact compositions, SMILES, and ratios remain under NDA; only coded roles and chemistry classes are shown publicly.
One Component.
Straight to the Die.
The two-phase, self-pumping genus owns immersion. But the most dielectric-sensitive surface in the rack — the exposed die, the in-package coolant path, the direct-to-chip cold plate — is won on a different axis. There the lever isn't self-pumping; it's permittivity.
FluxZero's answer is a single-component low-ε siloxane (CF10-G): ε ≈ 2.4, zero-fluorine, non-flammable — and deliberately pump-free.
Why no Marangoni pump here? Every high-Δσ PFAS-free pump is high-permittivity — adding one would push the blend ε straight out of the direct-to-chip band. For ε < 6, the winning move is to add nothing. Low ε is the siloxane's superpower.
Static permittivity
Among the lowest of any non-fluorinated dielectric — clears the ε < 6 direct-to-chip band with margin.
Zero fractionation
No fuel:pump ratio to drift, no replenishment loop, no mixing reservoir. The simplest possible fill — pour and run.
PFAS-free & non-flammable
No carbon–fluorine bonds and no flash-point handling burden — a clean regulatory and safety profile.
Validated chemistry
Slab-MD surface tension on a zero-fluorine siloxane cage; CoolProp equation-of-state cross-check to 0.12% on the siloxane heat-transfer-fluid class.
Where it lands
The band where the coolant touches live silicon and the dielectric constant is the gating spec: high-voltage and RF electronics, in-package and on-die cooling, and exposed cold plates.
FluxZero now covers both bands — two-phase self-pumping for immersion and maximum heat flux, single-component low-ε for direct-to-chip. One program, the full thermal stack. The siloxane lead was independently scored "real but niche" by a third-party ranking.
ε ≈ 2.4 and σ ≈ 22.3 mN/m are computational (DFT / molecular dynamics), independently equation-of-state cross-checked; nothing is wet-laboratory measured as of today. The exact siloxane composition is released to qualified parties under NDA.
Where the proof comes from.
- 89-candidate falsification ledger (within the 7.9M→931 PFAS-free genus funnel).
- 186 MD σ measurements across 59 binary pairs (167 valid · multi-seed pooled UQ).
- 4 σ methods cross-validated: Parachor-WK, Macleod-Sugden, Meissner-Michaels, Tamura-Kurata-Odani.
- 4,806 thermodynamics work-order rows and 95 lab/vendor RFQ rows.
- 11,923 ThermoML files audited; 1,774 strict binary value rows extracted.
- 155 parsed ThermoML metadata rows · 111 unique DOI/source leads.
- 89 / 89 candidates queried against Crossref (267 metadata rows).
- DDB licensed-extraction work orders generated for follow-up.
- 18-candidate shake-flask phase map (Q2 2026).
- Pendant-drop σ on single-phase survivors.
- Dielectric loss/breakdown, flash, vapor, materials on top 3-5.
- Bare-copper pool boiling on 2-3 finalists.
The Proof Depth
Behind the Lead.
FluxZero is not a thesis — it is a computational, simulation-validated discovery program. Hundreds of MD seeds, multi-method σ cross-validation, an audited ThermoML corpus, and a 230-claim filed patent estate stand behind the lead candidates.
ThermoML strict values
Anchors density, viscosity, excess volume, and activity for the lead candidates
ThermoML metadata leads
Targeted DOI extraction queue for surface tension and phase equilibrium
Crossref metadata
Coverage check across every candidate in the screen
DDB licensed extraction
Queued extraction protocol against the Dortmund Data Bank
Executable CPU plan
Runnable phase, UCST/LCST, and fractionation checks before any wetlab spend
CPU proxy results
Drives gate ordering and EVSI sensitivity for next-data prioritization
Ranking overlays
Use-case-specific shortlists for direct, sealed, passive, and indirect cooling
Wetlab work orders
Validated SOPs ready to release to qualified CROs
Low-ε sealed direct / passive anchor
Shake-flask Q2 2026 → dielectric/flash → bare-copper boiling
Highest-gradient direct / passive scout
Shake-flask Q2 2026 → controlled-handling protocol → bare-copper boiling
Broad direct / passive fallback
Shake-flask Q2 2026 → vapor envelope → dielectric/flash
Best phase-proxy scout
Shake-flask Q2 2026 → pendant-drop σ → dielectric → boiling curve
Higher-flash low-ε fallback
Shake-flask Q2 2026 → dielectric/loss/breakdown → boiling
Indirect-loop mechanism control
Reference fluid for non-immersion loops; mechanism cross-check
Public surface shows coded candidate roles and evidence counts. Exact compositions, SMILES, and component ratios are released to qualified parties under NDA alongside the technical brief.
PFAS Is Leaving.
Heat Flux Is Rising.
Three forcing functions converge in the next eighteen months. Every month of delay increases supply-chain exposure.
3M Market Exit
3M discontinued Novec™ and Fluorinert™ effective December 2025. $500M+ replacement demand in data-center immersion alone — cascades into heat pipes, EV thermal packs, and defense applications that depended on the same chemistry.
Universal PFAS Restriction
ECHA final opinions on the universal PFAS restriction under REACH (EC) 1907/2006 are expected 2026–2027. HFO "low-GWP" replacements (Opteon 2P50, Solstice) contain C–F bonds and face re-classification risk.
The Thermal Wall
NVIDIA B200 at 1,000 W. AMD MI325X at 750 W. Blackwell Ultra in 2027 projected 1,500–2,000 W. Air cooling died near 400 W per socket. Single-phase liquid tops out near 1,000 W because it cannot exploit phase change.
Every PFAS-free cooling fluid shipping today — Engineered Fluids EC-100, Shell S5 X, Castrol ON DC 20 — is single-phase. None self-pump. None have a Marangoni gradient. The physics forbids self-pumping in single-component fluids.
Self-pumping is the win for immersion. For the direct-to-chip band, where ε < 6 matters more than self-pumping, FluxZero fields a purpose-built single-component low-ε siloxane (CF10-G) — so the program covers both the self-pumping immersion band and the single-component direct-to-chip band.
How a Fluid
Pumps Itself.
Solutal Marangoni Induction
A binary working fluid pairs a lower-σ volatile pump component with a higher-σ retained fuel component. Heat preferentially evaporates the volatile fraction at the interface, locally depleting the pump.
Gradient Formation
Local σ rises at the depleted hotspot. Cooler bulk liquid is pulled along the ∂σ/∂x gradient toward the heat source. The sign is not a styling choice; it follows the surface-stress boundary condition.
Passive Phase-Change
Arriving fluid evaporates, extracting latent heat (ΔHvap). Vapor condenses at the cold surface and re-mixes. The system is a closed thermodynamic engine in a static volume.
Self-Regulation
Higher heat load steepens the concentration and surface-tension gradients. That increases passive circulation without a controller, sensor, or mechanical pump.
/∂x
tau_s = d sigma / dx drives tangential flow at the liquid interface.
sigma is estimated from pressure-tensor anisotropy across a liquid slab.
LLE / UCST / LCST / VLE screens decide whether the mixture is physically meaningful.
loss tangent and breakdown after water saturation decide electronics suitability.
bare-copper boiling curve and CHF decide whether the fluid actually transfers heat.
Quad-Marangoni Is Not First Spend.
The mechanism stack is scientifically interesting, but it is not the next validation step. Bare-fluid phase, safety, dielectric, materials, and boiling data come first.
Solutal Marangoni
Preferential evaporation can create surface-tension gradients that pull liquid toward a hotspot.
Self-rewetting additive
Some alcohol/additive systems invert d sigma / dT in the literature; this remains composition-gated here.
Dissolved gas
Gas desorption can alter nucleation and boiling behavior; pressure and drift must be measured.
Particle / surface route
Nanoparticles or conditioned surfaces may alter thermal pathways; dielectric loss and stability are hard gates.
The Marangoni Engine,
Rendered.
Free-surface CFD and atomistic σ rendered directly from the OpenFOAM and GROMACS pipelines that drive the candidate ranking. Same code, same outputs as the published canonical inventory.
Free-Surface + Velocity Field
Atomistic Surface-Tension Estimate
Device-Geometry CFD
Free-surface velocity-field studies for sealed thermosyphon, immersion-rack, and microgravity geometries. Activates on the bare-fluid finalist for OEM-specific integration.
Atomistic σ Engine
Slab MD with pressure-tensor σ extraction. 186 measurements across 59 binary pairs, multi-seed pooled UQ, cross-validated against four independent analytic methods.
Docker + Canonical Inventory
GPU NGC + CPU Docker images ship with the technical brief. Re-run any σ calculation against the published canonical inventory in under 5 minutes (CPU check) or full GPU reproduction.
What Ran.
What It Decided.
Fourteen layers of computational and database evidence sit behind the candidate ranking. The 89-candidate gate ledger below is the curated falsification shortlist drawn from the broader 7.9M→931 PFAS-free genus funnel — atomistic σ across multi-seed MD pools, four-method analytic cross-validation, ThermoML strict-value anchoring, and an EVSI sensitivity layer that points lab spend at the highest-information next measurement.
All-89 candidate gate ledger
Per-candidate evidence stack, role, and validation path
GROMACS slab surface-tension
Multi-seed pooled σ with uncertainty; drives Marangoni FoM ranking
Four-method validation
Parachor-WK, Macleod-Sugden, Meissner-Michaels, Tamura-Kurata-Odani
Binary PropertyValue parser
1,774 strict binary rows anchoring density, viscosity, excess volume
Parsed DOI/source lead audit
Targeted extraction queue for σ and phase equilibrium
All-candidate literature query
Coverage map across the full candidate space
Licensed Dortmund Data Bank queue
Industry-standard property anchoring queued for licensed pull
Composition-temperature screen grid
Phase, UCST/LCST, vapor/fractionation pre-screen
Local LLE / VLE / safety proxy
Drives gate ordering and EVSI sensitivity
Use-case ranking overlay
Per-architecture shortlist (direct, sealed, passive, indirect)
No-waste sensitivity ranking
Maximally decision-relevant next measurement
Exploratory CFD
Activates on bare-fluid finalist for device-level integration
Validated kill-gate SOP
Shake-flask · pendant-drop · dielectric/flash · bare-copper boiling
Docker + check-only CI
GPU NGC + CPU images; pytest + semantic audit
Seven PFAS-Free
Candidate Architectures.
The portfolio spans seven coded candidate roles — direct immersion, sealed two-phase, passive/space, indirect-loop, and single-component direct-to-chip — covering FoMs from 3.50× to 6.36× Novec 7100. Exact compositions, ratios, and SMILES are released to qualified parties under NDA.
First-spend phase + safety screen
Manual ThermoML / DDB extraction in flight
PFAS legacy benchmarks + low-priority alternatives
Sealed direct / passive anchor
Lead direct-immersion candidate
FDA-GRAS broad fallback
Phase-proxy-favorable scout
Higher-flash low-ε fallback
Composite-FoM mechanism control
Single-component low-ε siloxane direct-to-chip lead
Each cooling architecture has its own dominant constraint, so the portfolio ranks per-use-case. A candidate that leads on direct immersion can be a different molecule than the one that leads on sealed passive or indirect-loop.
Direct immersion
7 coded entriesSealed two-phase
6 coded entriesPassive / space
7 coded entriesIndirect loop
4 coded entriesDirect-to-chip
1 coded entryEVSI sensitivity ranks which next measurement maximally narrows the candidate decision. Lab spend goes where data actually changes a verdict, not where it confirms what is already known.
Lead candidates are built on commodity-grade chemistries already produced at industrial scale — orders of magnitude cheaper than legacy fluorinated dielectrics. PFAS-free formulation removes the regulatory exposure facing every fluorocarbon coolant.
Architectures.
Seven coded candidate envelopes span direct immersion, sealed passive, indirect loop, the system-multiplier stack, and a single-component direct-to-chip siloxane — each backed by computational evidence and mapped to specific patent embodiments.
Low-ε sealed direct immersion
Best-in-class dielectric. Sealed two-phase architecture for high-voltage and in-package use. Covered under PROV-003 Claim 13.
Highest-FoM direct immersion
Lead direct-immersion candidate by FoM. Bayesian-discovered after the 2026-01-29 priority date; CIP filing candidate (hard-bar 2027-01-29).
GRAS-pump direct immersion
FDA-GRAS pump component for biotech-adjacent and food-contact applications. CIP filing candidate.
Broad direct / passive fallback
Best alkane-pump miscibility profile in the portfolio. Direct and passive-thermosyphon compatible.
Indirect-loop reference
Highest composite FoM in the portfolio for non-immersion architectures. Mechanism cross-check for direct-immersion physics.
System multiplier stack
Six hardware levers covered under PROV-001 and PROV-003. Each acts as a CHF multiplier on top of the bare-fluid finalist.
Single-component direct-to-chip
Single-component low-ε siloxane for the direct-to-chip (ε < 6) band; zero-fluorine, non-flammable; adding a Marangoni pump is counterproductive. Covered in the omnibus draft as the single-component direct-to-chip aspect.
Six Hardware Levers.
Six Levers Above the Fluid.
The system-multiplier stack — copper foam, ultrasound, dielectric nanoparticles, biphilic surfaces, dissolved gases, and surface conditioning — sequences after the bare-fluid finalist. Each lever is covered in PROV-001 or PROV-003 and activates as a CHF multiplier on top of the validated baseline.
- ▸ Copper foam — CHF multiplier on bare-fluid finalist; PROV-001 system embodiment.
- ▸ Biphilic surface — Tuned nucleation + rewetting kinetics; PROV-001 surface embodiment.
- ▸ Al₂O₃ conditioning — Surface-energy tuning for bubble departure; PROV-001 conditioning embodiment.
- ▸ Ultrasound — Active bubble depinning; PROV-001 hardware embodiment.
- ▸ Dielectric nanoparticle — Thermal pathway enhancement at low loading; PROV-003 nanoparticle embodiment.
- ▸ Dissolved CO₂ / N₂ / Ar — Nucleation-superheat lever; PROV-003 dissolved-gas embodiment.
No One Else
Self-Pumps.
Every PFAS-free immersion fluid shipping today is single-phase — it requires a mechanical pump and cannot exploit phase change. Every two-phase fluid shipping today contains C-F bonds and faces regulatory exit. FluxZero is the only candidate stack engineered as PFAS-free and Marangoni-active — passive recirculation plus phase-change heat capacity in one binary working fluid. FluxZero also offers a single-component option purpose-built for the direct-to-chip band, so the contrast above is about immersion self-pumping, not a blanket dismissal of single-component fluids.
| Reference fluid | Chemistry | PFAS status | Phase regime | Notes |
|---|---|---|---|---|
| FluxZero candidate stack | Marangoni-active binary | PFAS-free | Two-phase + self-pumping | Lead FoM 4.06× Novec 7100; ε = 2.05 best-in-class |
| FluxZero CF10-G (direct-to-chip) | Single-component low-ε siloxane | PFAS-free | Single-phase / direct-to-chip | Purpose-built low-ε (ε ≈ 2.4) for direct-to-chip; zero-fluorine; computational lead |
| Novec 649 / 7000 / 7100 | Fluorinated two-phase | PFAS | Two-phase | Discontinued by 3M Dec 2025 |
| Fluorinert FC-72 / FC-87 / FC-40 | Fluorinated single-phase | PFAS | Single-phase | Discontinued by 3M Dec 2025 |
| Opteon 2P50 (Chemours) | HFO two-phase | C-F bonds | Two-phase | Faces ECHA re-classification risk |
| Castrol ON DC 20 | Hydrocarbon single-phase | PFAS-free | Single-phase | No Marangoni mechanism; pump required |
| Submer SmartCoolant | Hydrocarbon single-phase | PFAS-free | Single-phase | No Marangoni mechanism; pump required |
| Shell Immersion S5 X | GTL synthetic hydrocarbon | PFAS-free | Single-phase | No Marangoni mechanism; pump required |
| Engineered Fluids EC-100 / SLIC | Synthetic hydrocarbon | PFAS-free | Single-phase | No Marangoni mechanism; pump required |
| M&I Materials MIVOLT | Synthetic ester | PFAS-free | Single-phase | No Marangoni mechanism; pump required |
Performance comparisons against commercial baselines are validated in the Q3-Q4 2026 wetlab phase: dielectric loss/breakdown after water saturation, materials compatibility, and bare-copper pool boiling on the top 2-3 finalists.
Use Cases Are Gates.
Each application has a different failure mode. The site now avoids treating one computational ranking as a universal answer.
Direct immersion
candidate only until miscibility, vapor/headspace, dielectric, materials, and boiling gates pass
Passive / thermosyphon
requires vapor-pressure, fractionation, wick/materials, and orientation testing
Indirect loop
mechanism/control space where electronics contact is not required
Direct-to-chip / cold-plate
low-epsilon single-component coolant for exposed/in-package electronics where epsilon < 6 matters more than self-pumping; gated on dielectric, materials, and thermal testing
Space / microgravity
hypothesis only until phase, vapor, freeze/thaw, materials, and orientation tests exist
PFAS retrofit references
not PFAS-free spend targets; useful for benchmarking and legacy comparison
System multipliers
Phase 5 only after a bare-fluid winner exists
Anchored to 1,774
Strict Rows.
Computational σ ranking is anchored to 1,774 strict ThermoML binary value rows extracted from 11,923 audited files, plus 155 parsed metadata leads across 111 unique DOIs. Every candidate is queried against Crossref. Lead-candidate σ is cross-validated against four independent analytic methods.
Strict ThermoML Property Values
Metadata Leads Worth Following
Multi-seed pooled UQ across 59 binary pairs.
Parachor-WK · Macleod-Sugden · Meissner-Michaels · Tamura-Kurata-Odani.
1,774 strict binary value rows extracted across the candidate set.
Direct-immersion lead vs Novec 7100 (canonical post-T-1.2 weighting).
The proof package is multi-layered: pressure-tensor σ from multi-seed slab MD, composition-gradient Marangoni scaling, Hansen / local LLE falsification, strict ThermoML anchoring, four-method analytic σ cross-validation, VLE/fractionation work-order generation, and a kill-gated wetlab pipeline. SMILES, ratios, and exact compositions are released to qualified parties under NDA alongside the technical brief.
What Each Layer Can Decide.
The pipeline is strongest when each layer is used for the decision it can actually support. The current database work is a high-value triage layer; the first decisive experimental gate is still phase behavior.
| Layer | Can tell us | Cannot tell us |
|---|---|---|
| MD slab sigma | Ranks blends by surface-tension proxy and uncertainty | phase equilibrium, dielectric loss, breakdown, flash, vapor, boiling |
| Hansen / local LLE proxies | Flags likely miscibility failures before buying chemicals | real UCST/LCST or composition drift |
| ThermoML strict values | Anchors density, viscosity, excess volume, and activity where present | surface tension, boiling, or electronics safety for this candidate set |
| ThermoML metadata leads | Creates a DOI/source extraction queue across 18 candidates | measured values unless the paper is manually extracted |
| Crossref metadata | Confirms broad search coverage across the curated 89-candidate ledger | absence of literature or absence of DDB data |
| Shake-flask / pendant-drop / dielectric | First data that can promote or kill a candidate physically | stack-level CHF or rack-scale readiness |
This is why the next spend remains cheap falsification: miscibility plus volatile-solvent headspace safety first, then pendant-drop sigma only on single-phase survivors, then dielectric/flash/vapor pressure, then bare-copper boiling.
The Patent Estate.
Three provisional patent applications filed 2026-01-29 totaling 230 claims, plus a continuation-in-part filing packet drafted for Bayesian-discovered direct-immersion candidates (hard-bar 2027-01-29). On top of the filed estate, every coolant idea generated across the program has now been consolidated into a single ultra-long-form omnibus draft — 330 numbered claims across eleven statutory parts, covering fourteen inventive aspects from the PFAS-free genus to the discovery engine itself.
Fluid system + method
122 claims · fluid composition, two-phase architecture, system embodiments (foam, ultrasound, biphilic, conditioning), and method-of-cooling claims
Computational discovery engine
50 claims · Bayesian GP + multi-seed MD screening workflow, EVSI sensitivity logic, and reproducibility framework
PFAS-free binary keystone
58 claims · PFAS-free composition genus including the low-ε composition of Claim 13, method-of-use, microgravity, retrofit, dissolved-gas, and dielectric-nanoparticle embodiments
The 2027 Filing Window.
- CIP filing packet drafted covering Bayesian-discovered direct-immersion candidates (CF10-B and CF10-C) identified after the 2026-01-29 priority date. Hard-bar 2027-01-29.
- System-multiplier embodiments — copper foam, ultrasound, dielectric nanoparticles, biphilic surfaces — covered as PROV-001 / PROV-003 dependent claims and follow-on continuation targets.
- Filing receipts, application numbers, exact filed PDFs, and chain-of-title released to qualified parties under NDA alongside the technical dossier.
Inventor: Nicholas Harris. Provisional applications filed pro se on 2026-01-29; CIP filing packet under counsel review. Filing receipts, application numbers, exact filed PDFs, and chain-of-title released to qualified parties under NDA.
One Document, 330 Claims.
Every coolant idea generated across the program — the in-repo provisionals, the ~64k-word Marangoni disclosure, the consolidated rankings, and the best-ideas synthesis — has been folded into a single ultra-long-form provisional draft with continuous claim numbering (1–330) and a candor ledger tagging every value as computed, literature, or estimate. Nothing in the program is lost; the omnibus is the canonical superset.
Draft — pending file-ready review; not yet filed. The three provisionals above remain the filed priority estate (2026-01-29). All omnibus values are computational, literature, or engineering-estimate in origin — none is wet-laboratory-measured as of the priority date.
Binary composition genus + PFAS-free species
System claims + method-of-cooling claims
Article-of-manufacture + Jepson / means-plus-function
Computational discovery engine + heartbeat / fuzzing methods
Functional "moat" + integrated-stack + range / catch-all
- PFAS-free two-phase immersion coolant — broad genus (volatile fuel + carbonate / lactone Marangoni pump)
- CF10-B — highest-FoM direct-immersion lead (alkane + lactone class)
- CF10-A — lowest-ε dielectric-safety champion (alkane + dialkyl-carbonate class)
- Non-flammable high-CHF flagship (alcohol + glycol + boron-nitride nanofluid class)
- Low-ε non-flammable transition blend (hydrofluoroolefin + cyclic-carbonate class; retains C–F)
- The six-stage computational discovery engine itself
- CF10-C — FDA-GRAS-pump direct-immersion (alkane + cyclic-carbonate class)
- Composition-drift sensing + autonomous-replenishment system family
- CF10-G — single-component low-ε siloxane direct-to-chip coolant
- Alkane + polar-aprotic-amide pool-σ riser
- Pressure-tuned composition (tune T b into the cold-plate window)
- CF10-D — good-ε fallback (alkane + cyclic-ketone class)
- Nanoparticle-enhanced + structured-surface (biphilic) cooling system
- Microgravity / spacecraft solutal self-pumping cooling method
One Ranked Sweep
Behind the Omnibus.
Every coolant idea generated across the program — the in-repo provisionals, the ~64k-word Marangoni disclosure, the consolidated rankings, and the external sources — was swept into a single ranked synthesis, then folded into the 330-claim omnibus draft. Each idea is scored on three axes and the product is deliberate: a huge market with no whitespace, or strong novelty with no compute, both collapse toward zero.
Size of the dislocation a strategic buyer would price — market gap, regulatory clock, breadth of the claimable genus.
Genuine whitespace and defensibility — an open lane with no anticipating grant vs. crowded known art.
Depth and corroboration of the in-silico proof today — multi-engine, multi-seed, validated-method agreement.
PFAS-free two-phase immersion coolant — broad genus
The strategic asset: 3M exited all PFAS end-2025 (~$12.5B supply gap); full 7.9M→931 screening funnel; DFT dielectric screen validated against experiment; 27 PFAS-free Marangoni pairs clear the Δσ / Hansen gate.
CF10-B — highest-FoM direct-immersion lead (alkane + lactone)
FoM 4.06× Novec 7100; highest Marangoni Δσ in the table; n=7 pooled MD σ; self-pumping on a bare surface.
CF10-A — lowest-ε dielectric-safety champion (alkane + dialkyl-carbonate)
ε = 2.05, best-in-class (~2× the next entry); deepest UQ (n=7, inter-site bias disclosed); covered by PROV-001 / 002 / 003.
Non-flammable high-CHF flagship (alcohol + glycol + boron-nitride nanofluid)
Highest modeled Monte-Carlo system CHF; P(top-1)=0.79; commodity-cheap components; high ε → immersion / indirect only.
Low-ε non-flammable transition blend (hydrofluoroolefin + cyclic-carbonate; retains C–F)
Best non-flammable low-ε (ε ≈ 2.0) direct-immersion blend; first-ever component-σ MD; unclaimed-IP gap → CIP candidate.
The six-stage computational discovery engine itself
The only asset that exists and works rather than predicts; RF surface-tension R²>0.75 held-out; dielectric method validated; PROV-002.
CF10-C — FDA-GRAS-pump direct-immersion (alkane + cyclic-carbonate)
FoM 3.83×; FDA-GRAS pump for regulated environments; miscibility literature-settled → shake-flask decisive.
Composition-drift sensing + autonomous replenishment family
Highest-scoring dossier cluster (FTO LOW, counsel-ready); system / additive IP that wraps any winning fluid and guards the two-phase drift failure mode.
CF10-G — single-component low-ε siloxane direct-to-chip
ε ≈ 2.4, zero-fluorine, non-flammable; wins direct-to-chip outright (adding a Marangoni pump only raises ε out of band).
Alkane + polar-aprotic-amide — pool-σ riser
FoM rose to 3.92× on the n=5 pool; σ(T) R²=0.966 (excellent); amide toxicity weakens the buyer story.
Pressure-tuned composition
Method-level lever that re-targets any blend into a buyer’s loop pressure / temperature window; dossier C-15, FTO LOW.
CF10-D — good-ε fallback (alkane + cyclic-ketone)
FoM 3.66×, ε 4.0; the only Hansen-miscible alkane+pump pair → cleanest proof candidate if shake-flask kills the carbonate/lactone pairs.
Nanoparticle + biphilic structured-surface system
h-BN dielectric-retention nanofluid + copper-foam capillary matching + ultrasound duty cycle; CHF multipliers — literature-additive, not authorized for spend before a bare-fluid winner exists.
Microgravity / spacecraft solutal self-pumping method
Solutal-Marangoni pumping replaces buoyant convection in zero-g → defense / space niche with thin art; method claim in PROV-003.
- File the provisional — the priority date is the asset; prophetic computational examples are filing-legal, and the omnibus draft is ready.
- Fund one lab confirmation per flagship (~$15–40k) — shake-flask miscibility first (cheapest decisive gate), then dielectric strength and bare-copper two-phase CHF. The single biggest value inflection: it converts “computed” into “proven.”
All scores and evidence above are computational or literature estimates — none is wet-laboratory-measured as of the priority date. Ratios, SMILES, and exact compositions are released to qualified parties under NDA. Rankings and FTO posture are heuristic and not legal advice.
Validation Path.
Eight sequential gates retire the cheapest failure modes first. Phase before σ. σ before dielectric. Dielectric before bare-copper boiling. Each gate is engineered so a single result can decide which candidate continues to the next stage.
Validate the Stack in 5 Minutes.
# CPU check; no heavy GPU run required
python -m pytest
bash run_all.sh check-only
python scripts/semantic_consistency_check.py
Docker images (GPU NGC + CPU) ship with the dossier. Heavy MD reproduction is opt-in via reproduce_md.py against the published canonical inventory.
Gate Sequence
Tests 0-K
CRO confirms flammable handling, sealed vials, rated hood/enclosure, grounding, waste, and headspace controls.
All 18 shake-flask candidates plus controls, with photos and 72-hour observation.
Repeat at elevated temperature; two-phase behavior demotes direct/passive candidates.
Run only on single-phase survivors; compare to MD/mixing-rule expectations.
Quantifies sealed-system need and fractionation risk.
Hard commercial gate before buyer pilot language.
20 MHz to 40 GHz where relevant; measure loss tangent, not just static epsilon.
Direct-immersion candidate fails if water uptake collapses dielectric margin.
EPDM, FKM, silicone, PVC, wire insulation, solder mask, TIMs, copper/nickel/aluminum coupons.
Thermal aging plus acid number and coupon inspection.
Only after the bare fluid passes the above gates.
Safety fail = stop. Two-phase = demote. Sigma miss = fail MD. Electrical/materials fail = no direct immersion.
System-multiplier embodiments — copper foam, ultrasound, dielectric nanoparticles, biphilic surfaces, dissolved gas, EHD, and OpenFOAM device-geometry studies — activate as CHF multipliers on top of the validated bare-fluid finalist. Each lever compounds measured baseline performance rather than substituting for it.
Risks & Mitigation.
The validation pipeline is engineered to falsify the cheapest failure modes first. Every gate sequenced to retire a specific risk before more capital flows in — phase first, then dielectric, then materials, then bare-copper boiling.
Candidates may phase split
MD can hold a metastable mixed slab. The first decisive physical gate is 20 C and 40 C phase mapping on the 18-candidate panel.
Volatile blends may be commercially unusable
Volatile low-boiling candidates need sealed-vial, vapor-pressure, flash, and flammable-headspace controls before direct/passive language is credible.
Static epsilon is not enough
Loss tangent, breakdown after water saturation, and frequency-dependent behavior matter for electronics. These data do not exist yet.
Compatibility can kill a good fluid
EPDM, FKM, silicone, PVC, wire insulation, solder mask, TIMs, copper, nickel, and aluminum need compatibility and aging checks.
Metadata is not a measured property
ThermoML metadata creates a targeted DOI extraction queue, but extracted values must be confirmed manually. Pendant-drop wetlab σ remains the authoritative measurement for the lead candidates.
Multipliers compound, not rescue
Copper foam, ultrasound, nanoparticles, biphilic surfaces, and dissolved-gas levers act as CHF multipliers on top of the bare-fluid finalist — not as compensators for a fluid that fails phase or dielectric gates.
Value Compounds
With Validation.
Each gate compounds asset value: option agreements and early-licensee terms today, full performance-validated licensing or acquisition after Q4 2026 boiling-curve data. Engagement structures available under NDA.
Before the Meeting.
How is FluxZero different from existing PFAS-free coolants?
Every commercial PFAS-free immersion fluid shipping today (Engineered Fluids EC-100, Shell Immersion S5 X, Castrol ON DC 20, Submer SmartCoolant) is single-phase. None self-pump. None exploit a Marangoni gradient. FluxZero is a binary working fluid that creates its own surface-tension gradient under heat load — passive recirculation without a pump, plus phase-change heat capacity that single-phase fluids physically cannot deliver. The program spans both bands: for immersion it is the two-phase self-pumping answer, and for direct-to-chip it additionally fields a purpose-built single-component low-ε siloxane.
What is the strongest proof today?
186 MD σ measurements across 59 binary pairs (167 valid · multi-seed pooled UQ) with σ values cross-validated against four independent methods (Parachor-WK, Macleod-Sugden, Meissner-Michaels, Tamura-Kurata-Odani). ThermoML anchoring against 11,923 audited files. Lead direct-immersion candidate computes to FoM = 4.06× Novec 7100; indirect-loop reference reaches 6.36×. Best-in-class dielectric ε = 2.05.
How defensible is the IP?
Three provisional patent applications filed 2026-01-29 totaling 230 claims: PROV-001 (fluid system + method, 122 claims), PROV-002 (Bayesian/MD discovery engine, 50 claims), PROV-003 (PFAS-free binary keystone with microgravity and dissolved-gas embodiments, 58 claims). CIP filing packet drafted covering follow-on direct-immersion candidates, hard-bar 2027-01-29. Filing receipts and chain-of-title released to qualified parties under NDA.
What does the validation pipeline look like?
Q2 2026: 18-candidate shake-flask phase map at 20°C / 25°C / 40°C with 1h, 24h, 7d reads. Single-phase survivors advance to pendant-drop σ. Top 3-5 progress to dielectric loss/breakdown after water saturation, flash, vapor pressure, materials compatibility. Top 2-3 finalists run bare-copper pool boiling curves and CHF measurement.
What are the hard risks?
Phase stability under thermal load is the first decisive gate — MD slabs can persist as metastable mixtures. Volatile-pump candidates require sealed-headspace and flammability controls before pilot deployment. Frequency-dependent dielectric loss after water saturation is a separate gate from static ε. Materials compatibility (EPDM, FKM, silicone, solder mask, TIMs) is a deal-breaker if it fails. The validation sequence is engineered to falsify these cheaply, in order.
Is the computational stack reproducible?
Yes. Docker images (GPU NGC + CPU) ship with the dossier. `bash run_all.sh check-only` validates the full pipeline in under 5 minutes. Heavy MD reproduction is opt-in but fully automated via `reproduce_md.py` against the published canonical inventory.
Why now?
3M discontinued Novec™ and Fluorinert™ effective December 2025. ECHA universal PFAS restriction final opinions land 2026-2027. NVIDIA B200 sockets at 1,000W today; Rubin projects 1,500-2,000W. The two-phase immersion stack needs a PFAS-free, Marangoni-active replacement with measured electrical safety and validated materials compatibility — available before Novec stockpiles run out.
Roadmap.
Computational discovery and IP filings are landed. The 2026 cadence advances FluxZero candidates through wetlab kill-gates to a bare-fluid finalist with measured electrical safety and bare-copper CHF data.
Three provisional patents filed
PROV-001 (fluid system + method, 122 claims), PROV-002 (Bayesian/MD discovery engine, 50 claims), PROV-003 (PFAS-free binary keystone with microgravity and dissolved-gas embodiments, 58 claims). 230 total claims.
Bayesian discovery + multi-seed UQ
CF10-B (alkane + lactone) identified as the direct-immersion lead via Bayesian GP + MD; FoM = 4.06× Novec 7100. CIP filing packet drafted; hard-bar 2027-01-29.
Evidence stack snapshot
186 MD σ measurements across 59 binary pairs; 11,923 ThermoML files audited; 1,774 strict binary rows extracted; 4-method σ cross-validation complete.
PFAS-free global optimum (v2)
Re-scoped from the 89-candidate shortlist to the full PFAS-free chemical space: a 7.9M→931 funnel (PubChem → size/element/boiling filters → stable dielectric-relevant candidates) with a DFT electronic-dielectric screen, under a redefined multi-objective {CHF · dielectric · safety · GWP<150 · Tb} ranking. Finding: the original 89-set was not the optimum (none was both non-flammable and PFAS-free).
First low-ε non-flammable σ MD + hardened CHF
First-ever component-σ molecular dynamics for the low-ε non-flammable (hydrofluoroolefin + cyclic-carbonate) pair; Phase-B Monte-Carlo system-CHF ranking with 90% confidence intervals — non-flammable flagship 357.6 W/cm² (CI 203.7–556.1), P(top-1)=0.79.
Direct-to-chip product band + MD gap campaign
Use-case split confirmed: a single-component low-ε siloxane (CF10-G) wins direct-to-chip (ε<6) while the oxygenated-fuel + carbonate/lactone Marangoni blends win immersion. GPU MD gap campaign extended component/blend coverage with multi-seed pooled UQ.
Customer-grounded global optimum
Re-optimized against 8 objectives (adding toxicity) and 7 use cases (adding cold-plate / direct-to-chip), competitors in-frame, leak-risk total-cost-of-ownership; invention dossier set grew to 48 (39 counsel-ready).
Program consolidation + 330-claim omnibus draft
Every coolant source across the program consolidated into one canonical synthesis (BEST_IDEAS, RANKINGS_CONSOLIDATED) and a single ultra-long-form provisional draft — 330 claims, 11 statutory parts, 14 inventive aspects. Draft pending file-ready review; not yet filed.
Wetlab phase 1 — shake-flask
18-candidate phase map at 20 / 25 / 40 °C with 1h, 24h, 7d reads. Single-phase survivors advance to pendant-drop σ.
CIP filing + DDB extraction
Continuation-in-part filing on direct-immersion candidates; licensed DDB extraction against the work-order queue.
Electrical / safety / materials gate
Dielectric loss/breakdown after water saturation, flash, vapor pressure, materials compatibility on top 3-5 candidates.
Bare-copper boiling + system stack
Pool boiling curves and CHF on 2-3 finalists; PROV-001 / PROV-003 system multipliers (foam, ultrasound, biphilic, dissolved gas) layered on the validated baseline.
FluxZero IP estate: three provisional patent applications filed 2026-01-29 (PROV-001, PROV-002, PROV-003) totaling 230 claims, plus a continuation-in-part filing packet drafted for Bayesian-discovered direct-immersion candidates with a 2027-01-29 hard-bar. Coverage spans the PFAS-free composition genus, the discovery engine, and six system-multiplier embodiments. Filing receipts, application numbers, exact filed PDFs, and chain-of-title release to qualified parties under NDA alongside the technical dossier.
Where FluxZero Lands.
Designed for the platforms that lose their incumbent dielectric in 2025-2027 and need a PFAS-free replacement that exploits phase change, not just heat capacity.
Socket-level two-phase cooling for 1,000-2,000W TDP envelopes that air and single-phase liquid cannot reach.
PFAS-free dielectric replacement for Novec-based two-phase rack deployments before end-2025 supply runs out.
Pump-free passive cooling for unattended outdoor and high-density edge sites where moving parts fail first.
Marangoni transport works in microgravity; PFAS-free formulation clears domestic-supply and regulatory constraints.
Three Ways to Work With Us.
Field-of-use or exclusive license against the 230-claim filed estate.
Cooling OEMs · fluid manufacturers · system integrators
Wetlab + pilot validation against your platform spec.
Hyperscaler procurement · DOD primes · spaceflight integrators
Full IP estate transfer with inventor support.
Strategic acquirers in immersion cooling and dielectric fluid markets
Talk to the inventor. NDA template returned within 1 business day; full technical brief within 3.
Schedule a Technical Call
Talk to Us
Before Novec Runs Out.
The NDA-gated technical brief includes lead-candidate compositions and SMILES, MD canonical inventory, ThermoML extracts, patent filing receipts, the validated wetlab SOP, and the Docker reproducibility kit.