PRISMby Mirdyne
Concept visualisation · credits

A closed loop for discovering new materials.

In-silico optimisation. Robotic synthesis. Extreme validation.

PRISM, our first product, finds the material. Then we make it, qualify it and scale it into production.

Register interestThe problem
Hessen Ideen Wettbewerb 2026 – KI‑Sonderpreis von hessian.AI

01The problem = 20,000 possible alloys

Materials decide what we can build.

Finding new materials has become faster. Putting one into service still takes ten to twenty years, the figure in the 2011 US Materials Genome Initiative white paper. This count is an example: refractory alloys.

474,311,376
possible alloys from just five of nine high-melting metals
÷10
alloys a day, an assumption, not a measurement
≈130,000
years, if that assumed rate held every day

Nobody can make them all, so PRISM picks the few worth making.

How we counted

Example: refractory alloys. 126 ways to pick five of nine metals that all melt above 1,650 °C, × 3,764,376 ways to mix five in whole percent. Ten a day is an assumption, not a measurement, so the year count that follows from it is not a result. Each dot is 20,000 alloys; their positions are illustrative. 10 to 20 years: US Materials Genome Initiative, June 2011.

02Run 01 — first proof of the loop (ESA FLPP, PRISM Alpha)

The first loop, run on real metal.

Photos

01 to 05 are drawn. 06 to 10 and 14 are our own photographs. 11 and 12 show the same steps in other labs, until we have our own; 13 is drawn.

  • Diagram: A specification sheet beside a thermometer: what the part must survive.

    01Your requirementWhat the part must survive, and what it has to beat.

  • Diagram: A stack of papers under a magnifying glass: the literature on the requirement.

    02PRISM reads the literaturePapers and data on your requirement, with their sources.

  • Diagram: Four classes of material as tags; the two worth researching are marked.

    03Classes worth researchingThe families of material that could meet it.

  • Diagram: A network of choices: from an empty start, our models build recipes step by step, and more of the flow reaches the better ones.

    04Our generative models propose recipesThey build each recipe step by step; the better a recipe, the more of the flow reaches it.

  • Diagram: Several models score each candidate. Where they agree on a good score, the candidate goes on; where they disagree, an exact calculation decides.

    05The best few are chosenSeveral models judge each one; where they disagree, an exact calculation decides.
  • Small pieces of raw metal spread out on a paper towel before a melt.
    06Raw metals, ready to melt
  • Close-up of a copper hearth: small pieces of raw metal loaded into its hollows.
    07Loaded into the hearth
  • Looking through the viewport of an arc-melting furnace: a small alloy button glows orange inside.
    08Melted with an electric arc
  • A cast alloy button with a crystalline surface pattern, resting in a red lid on a lab bench.
    09An alloy button, as cast
  • A metal block being milled down: bright curled chips cover it, with milled channels beside them.
    10Milled down
  • Metal powder for 3D printing under an electron microscope: tiny, almost perfect spheres.
    11Made into powder
  • Through the tinted safety window of a laser powder-bed printer: small test pieces stand in the metal powder they were printed from.
    12Printed in a powder bedLPBF: research at Fraunhofer IAPT, industrial at Bimo Tech.

  • Diagram: A nozzle blows metal powder into a laser beam and builds the part up, bead by bead.

    13Or built up by DEDDirected energy deposition.
  • A small round test coupon held between two fingers in the lab at IPPT PAN, Warsaw.
    14Coupons, then testsDensity and flaws first, then real conditions.

Why Europe

Europe depends on others for key materials.

For some critical raw materials, the EU relies almost entirely on one country.

65%
By 2030, the EU wants no more than this share of any strategic raw material to come from one outside country.
SX500
No existing alloy could survive inside SpaceX’s Raptor engine, so SpaceX made its own.

We design new alloys in Giessen, and Bimo Tech makes them in Wrocław.

Sources

EU Critical Raw Materials Act, Regulation (EU) 2024/1252, recital 12 and Article 5. SX500: Elon Musk, 23 December 2018 and 25 May 2019.

03How it works

Every result feeds the next design.

A new material class would plug in as new modules. Polymers are planned, and we are seeking funding for that run.

01 / 07RequirementPrototype

Diagram: The PRISM loop: your requirement in the middle; research, design, screen, make, test and learn around it, and round again.

You tell us the heat and loads the part faces, and the material it has to beat.

  1. 01

    Requirement Prototype

    You tell us the heat and loads the part faces, and the material it has to beat.

  2. 02

    Research Prototype

    PRISM reads the literature on your requirement, and we pick the classes of material worth researching.

  3. 03

    Design Prototype

    Our generative models propose a whole batch of promising recipes in those classes, not one best guess.

  4. 04

    Screen Prototype

    Several models judge each one, an exact calculation settles the doubtful ones, and most ideas stop here.

  5. 05

    Make In use

    We melt the best candidates, mill them down, turn them into powder and print them by laser powder bed fusion or DED.

  6. 06

    Test In use

    We measure each sample against your targets.

  7. 07

    Learn Prototype

    Every result goes back into the models and decides what we try next.

Diagram: The PRISM loop: a requirement, then research, design, screen, make and test; every result goes back to design. Evidence is recorded at every step.

What must the part survive?

You tell us the heat and loads the part faces, and the material it has to beat.

Prototype

04Roadmap

Three stages. Today, and next.

Stage 1 runs today. Stage 2 runs today through our partners, and the next step automates it. Stage 3 is the goal.

  1. 01

    Live

    Stage 1 · In silico

    PRISM runs today: literature, generative models, scoring and exact calculations, inside the ESA projects SPARK and PRISM Alpha.

  2. 02

    Today

    Stage 2 · Physical synthesis

    Today the samples are made with partners: arc melting at the WUST lab in Wrocław, laser powder bed fusion at Fraunhofer IAPT, industrial printing at Bimo Tech, and testing at IPPT PAN.

    Next: automated, robotic synthesis.

  3. 03

    Goal

    Stage 3 · Own qualification rigs

    Robot-controlled rigs that test to failure and qualify materials in the environment they are made for, so the loop learns from measurements nobody else has. This is the goal. Testing today uses partner equipment.

05How we compare

Not software. Not a consultancy. A closed-loop accelerator.

What each company in AI for materials does itself, from proposing a candidate to making it at scale.

CompanyWhat they sellProposesMakesTestsCloses the loopQualifiesScales productionLargest reported round
MirdyneGiessenPRISM software access and closed-loop programmesAlloys for extreme environments, first for rocket enginesYesPlanned, in-housePlanned, in-houseYesPlanned, in-houseThrough partnersRaising €4M seed
Radical AINew YorkMaterials discovery in its own self-driving labMaterialsYesYesYesYesNot statedNot stated$55M seed, Jul 2025
alqemMunichA discovery engine with in-house synthesisRare-earth-free magnetsYesYesYesYesNot statedNot stated€8M pre-seed, Jul 2026
AlloyedOxford, UKAlloy-design software (Alloys-By-Design, Architect), its own alloys and 3D-printed partsMetal additive manufacturingYesYesNot statedNot statedNot statedYes£37M Series B, Mar 2025
ExoMatterMunichAn AI platform to screen, simulate and rank materials, plus its scientistsMaterialsYesNoNoNoNoNo€1.7M pre-seed, Oct 2024
  • Yes
  • Through partners
  • Planned, in-house
  • No
  • Not stated

Mirdyne will make and test in-house: its own robotic synthesis cells and test rigs are planned, and until they run, the work is done at partner labs (WUST, Fraunhofer IAPT, IPPT PAN). Production at scale runs with a partner, Bimo Tech, until Mirdyne builds production systems of its own.

See how we compare with eleven companies

06Markets

Where the loop runs next.

  • The Aestus rocket engine in an altitude test stand: its dark nozzle hangs between tanks and pipework.01 · Run 01

    Space propulsion

    Alloys for the hottest parts of rocket engines. First with ESA and ArianeGroup.

  • The afterburner of a Eurojet EJ200 jet engine, its ring of metal nozzle petals seen from behind.02 · Next

    Defence

    Hot-section materials for aircraft engines, with export rules built in.

  • Inside the Wendelstein 7-X fusion experiment: rows of carbon wall tiles being fitted.03 · Next

    Fusion and energy

    Walls that face the plasma in fusion reactors, and parts for turbines. Bimo Tech already supplies ITER.

  • Tungsten rods grown over with metal crystals in a colourful tarnish.04 · Next

    Supply independence

    Alloys that do the job with metals Europe can get.

  • A white PTFE block with drilled holes and two rolls of PTFE tape on a wooden table.05 · Planned

    Replacing harmful substances

    PFAS-free polymers, the next material class. Seeking funding.

  • A silicon wafer covered in rows of chips, lit so that it shimmers in colour.06 · Next

    Key technologies

    Materials for semiconductors, and lightweight polymers.

07Explore

More about PRISM and Mirdyne.

NO HUMAN IN THE LOOP. JUST THE LOOP.

The loop runs itself. A named engineer signs off what leaves it.

Start

Bring us a requirement. We’ll run the loop on it.

We’ll tell you what PRISM can find, and how we’d prove it.

Register interestGitHub