Zero-PFAS Cooling.

Two modelled lead blends with no fluorine in them, proposed for a first bench test, as Novec's maker leaves PFAS manufacturing and regulators move to restrict the class.

See the resultsTalk to usTry the Verifier

Section through an immersion tank: one hot spot on the die, vapour leaving it, and the film drawn back in as surface tension rises. It draws the mechanism the coolant is designed around.

Why now

3M committed to stop making Novec and every other fluorinated fluid by the end of 2025 (3M, 2022), and Europe is moving to restrict the entire PFAS class. FluxZero's lead coolants contain no fluorine.

  • 3M committed to exit all PFAS manufacturing — fluoropolymers, fluorinated fluids and PFAS-based additives — by the end of 2025. Novec was its two-phase cooling fluid line.

    news.3m.com

  • The EU's proposed restriction on the whole PFAS class cleared its risk-assessment committee on 2 March 2026; the socio-economic committee's final opinion is due by the end of 2026, then the Commission decides.

    echa.europa.eu

  • Minnesota and Maine prohibit intentionally added PFAS in nearly all products from 2032, unless a use is ruled currently unavoidable.

    pca.state.mn.usmaine.gov

Who it is for

Coolant makers and test labs who would run the first bench test, and then data-centre operators who would use the result.

Who is behind it

Nick Harris, founder. CTO of VivaMed BioPharma; co-founder of MedSim.ai, FastRead.io and Formulai.

FluxZero is one of the companies of VerifyCore Labs, an AI research lab whose AI agents work on the systems AI runs on; the lab's own site links FluxZero's results. VerifyCore Labs

The fluid, the physics, the screen

Two-phase immersion cooling has run on fluorinated fluids. FluxZero's lead candidates contain no fluorine; its screen and model also rank fluorinated blends, two of them among its own four modelled candidates, all before any bench test.

The chemistry

  • Every FluxZero coolant pairs a low-boiling 'fuel' with a 'pump' that has higher surface tension and boils later. The pair is designed so that where a chip runs hot, the fuel boils off first, local surface tension rises, and liquid is drawn toward the hot spot: a self-pumping film with no moving parts. That effect has not yet been simulated with surface-tension-driven flow or measured.

The screen

Read the source

  • 14,559 two-component blends scored on heat flux, dielectric, safety, climate impact and boiling window; 7,020 pass a global-warming-potential ceiling of 150.

  • Ten leading blends are scored against the published properties of Novec 7100, the fluorinated fluid they are intended to replace. The scores are estimates.

Every formula, and what each one does not show

The leads (modelled)

Read the source

  • For immersion tanks: pentane with dimethyl carbonate

    An 80:20 blend meant for closed-tank direct immersion. Modelled heat flux 184.6 W/cm² (90% interval 103.5 to 292.1); modelled, not measured, and it rests on assumed boosts.

    What the model assumes

  • For cooling loops: methanol in triethylene glycol

    Methanol in triethylene glycol with boron nitride comes out ahead most often in our model of four in-house candidates. Modelled heat flux 357.6 W/cm² (90% interval 203.7 to 556.1); modelled, not measured, and it rests on assumed boosts.

    Read the result

The next experiment

  • Four bench tests, in order: whether the two parts mix, surface tension by the pendant-drop method, dielectric strength, then boiling on bare copper. A nine-sample panel is planned, with Novec 7100 as the reference.

1.43%pass bar 5%

The orange bar is the largest gap between the solver’s interface position and the exact answer after start-up; the tick is the typical gap; the line is the pass bar.

The boiling solver we use, checked against an exact answer

Run on its authors’ textbook evaporation problem, the open-source boiling solver we use stayed well inside its pass bar.

All comparisons with outside references

Go deeper

  • 25data last re-run April 2026, 172 d ago multi-seed surface-tension runs

    Each run’s surface tension recomputes exactly from its raw output

    On the lab’s machine, each run’s value and each pair’s pooled mean recompute exactly from the stored raw simulation output.

    How the lab checks its own work

Each result on its own page

How the numbers on this page are checked
Each underlined number is read from a published file when the site is built, and the build stops if the page and the file disagree. A number would be marked if
  • its file were still loading in your browser,
  • the question behind it had not been looked into,
  • it had been looked for and not found,
  • its file held no value for it,
  • its file were missing or had changed,
  • two files disagreed about what it describes,
  • its sample were too small for the claim,
  • two files gave it different values,
  • its file could not be published,
  • the file it is read from had last changed more than thirty days before this build (this applies to 3 numbers on this page),
  • the question did not apply to this page or
  • the program that makes it had stopped with an error.
A number in plain text, without the underline, is not checked this way and carries no mark; the date each file last changed is in the list of published files.

Writing

Plain-English explanations of what the fluorine finding means, how a simulation is checked against datasheets, and what a modelled result is worth.

All posts

Check our files yourself

Drop one of our published files on the panel, or try one of ours, and your browser checks that it is the exact file this page quotes. Nothing is uploaded.

Checked in your browser

$ Drop a .json file anywhere on this panel.

or try one of ours
  • The ten-blend score tableDownload
  • The blend countDownload
  • The blend count with one number altered (should be rejected)Download