Technology
Replacing trial and error with computational certainty.
SuperMatics is an AI-driven platform that screens millions of superconductor candidates in silico. We turn a decade of guesswork into a verifiable data pipeline, sending only the highest-probability targets to the bench.
A representative run of the engine: a brief is read, millions of candidate materials are written against it, the gates cut them down, and five candidates are cleared for the bench. The results return to the model.
01 · The breakthrough
Bypassing the compute bottleneck.
The run above is a compute problem before it is a physics problem. Solving it took more than faster hardware. It took a theory rebuilt around what machine learning is actually good at.
The Modulated Electron Lattice (MEL) framework is that theory: AI-first by design, not physics code ported to a GPU. It isolates the few variables that decide superconductivity, like local electron crowding, and hands the model exactly those, at a fraction of the compute.
And the model keeps getting better, because every bench result comes back and trains it. That is what lets our machine learning algorithms screen at unprecedented speeds without sacrificing theoretical fidelity: sweeping the whole space instead of modelling one material.
Legacy brute-force compute
Full physics, one candidate at a time.
AI-native MEL pipeline
The critical variables, screened in parallel.
A schematic comparison. On the left, a grid of cells is evaluated one cell at a time, and after six seconds the scan has covered about a seventh of the first two rows. On the right, a stream of candidates converges on a narrow gate over the same six seconds; most deflect away and fade, and the few that pass fill an output column completely. A progress rail under each panel shows how far each got: the right one full, the left one a sliver.
02 · The market
You already live in a world running on superconductors.
The physics is settled, but the engineering is stuck. Breakthroughs in major industries are currently waiting on a conductor that works warmer, carries more, or costs less to run. Every one of them is a brief.
19 fields · 17 of them already run on one
Medicine4 fields
Every scanner is a superconducting magnet. More than 30,000 are installed.
Niobium-titanium and affordable MRI · Physics in PerspectiveA superconducting gantry brought a heavy-ion treatment room from 600 tonnes down to under 300.
A compact superconducting rotating gantry · NIRSThe magnet that resolves a protein structure runs at a few kelvin.
The first 1.2 GHz protein NMR data · BrukerSQUIDs read the magnetic field of brain activity through the skull.
Superconducting magnetometers for brain investigations · PMCEnergy and the grid4 fields
In the ground under Shanghai, Essen and Long Island. Rare because of the cold.
A kilometre of superconducting cable under Essen · KITGoes normal in a millisecond and swallows a short circuit.
Fault-current limiters against a real short circuit · EnergiesA 3.6 MW superconducting rotor ran on a Danish turbine for 650 hours.
A superconducting rotor on a 3.6 MW turbine · EcoSwingCurrent parked in a loop and handed back in milliseconds.
Compute and sensing4 fields
Most of the largest processors are superconducting circuits, held near absolute zero.
Who is leading superconducting quantum computing · EPJ Quantum TechnologySuperconducting nanowires count single photons for quantum links and lidar.
Nanowire detectors at 98 percent efficiency · NISTSwitching at a fraction of the energy of silicon. Demonstrated for decades, never volume-made.
Superconducting logic and what it costs to switch · PMCSensors that measure one photon by the heat it leaves behind.
The SPT-3G focal plane, 16,000 detectors · arXivTransport3 fields
Japan's superconducting maglev reached 603 km/h on a test track.
The superconducting maglev · JR CentralA megawatt motor light enough to fly needs a cryostat that survives a wing.
The high-efficiency megawatt motor · NASAA 36.5 MW superconducting motor was built and tested for naval use.
A 36.5 MW propulsion motor at full power · AMSCScience and heavy industry4 fields
1,232 superconducting dipoles bend the beam around 27 km at CERN.
The LHC's superconducting dipoles · CERNTwenty tesla, and the reason a compact tokamak is credible at all.
A 20 tesla magnet for a compact tokamak · MITHigh-gradient magnets pull iron out of clay at industrial rates.
Superconducting separation of kaolin · Clay MineralsHeats a metal billet far more efficiently than a copper coil.
Induction heaters with high-Tc magnets · SN Applied SciencesThe sweep for any of these is ours to run. The making and the measuring are not, on purpose.
03 · The brief
Most briefs are one stuck step.
Almost nobody asks us for a new superconductor. They ask whether the material they already run survives the next step, and that step is usually where it meets something else.
The answer is usually a known material in a stack nobody had put together that way. A bounded question runs the same way a full search does, and a sweep takes about a week.
Where two materials meet is where devices fail.
The model treats the pair as one system: superconductor on substrate, superconductor against superconductor, and the junction between them.
The bond
A superconducting film that has to survive a semiconductor process without losing what makes it work.
The wire
Leads that carry signal into a cryostat without carrying heat in with them.
The stack
A multilayer that works everywhere except where two of its layers meet.
The window
The deposition and anneal range a fab can actually hold, found before the runs are spent.
The recipe
A known material that needs one property moved: warmer, stronger in field, or cheaper per metre.
Schematic. In each, the heavy line is the part the brief is about.
What comes back
Every answer comes back with the numbers an engineer can build on: transition temperature, critical current in field, the field it survives, a route to make it, and the cost at the operating point.
How teams run it
Run the platform yourself, send the brief and take the answer back, or put us beside your team until the specification is met. Partner labs build and measure whatever survives, so a brief can end with a sample in hand rather than a prediction.
Bring the step you are stuck on. What survives it is built and measured by the benches below.
04 · The network
The wet lab as an API.
We operate a closed-loop empirical pipeline. We do not own physical instrumentation because synthesis requires career-long expertise.
Our predictions are routed to established partner labs run by researchers who have dedicated their lives to mastering these specific machines. Their flawless empirical data feeds directly back into our models, making the engine continuously smarter with every run.
A crystal cleaved open inside the microscope head. The face it exposes has never touched air.
05 · The validation
You do not have to take our word for any of this.
Independent validation
A team at Stanford and SLAC independently measured the cooperation between charge order and superconductivity, and published it in Physical Review Letters. Independent validation in the exact sense of the words: we had no part in the work and no stake in its answer.
DOI 10.1103/g41t-8456
Cooperative phase coherence of charge order and superconductivity in cuprates
Lee et al. · Phys. Rev. Lett., 2026
Exclusive license
Hyunsung TNC
In development at Hyunsung TNC since 2006, licensed exclusively to SuperMatics.
20+ patents issued and pending
Underway
Manuscript · in preparation
The lab results are finalized, and the paper is about what they mean for the framework.
Theory
MEL framework
Measurement
Georgia Tech · STM / STS
Tell us the constraint your device has to hit.
One constraint is enough to start. Let us run the search.
Get in touch