Simulate first.
Build once.
SynapseX is building physics world models for fusion, semiconductors, aerospace and quantum hardware: predictions that come with an uncertainty band, a reference check and a sealed record. The panels on this page are not recordings. They solve the equations in your browser.
The physics outgrew the loop.
Design, simulate, wait, repeat.
Simulation is too slow to explore.
A high-fidelity run of a plasma, a process reactor or an entry flow takes hours to weeks. A design space holds thousands of candidates. Most of them are never simulated.
Experiments are too expensive to guess.
A shot, a wafer lot or a tunnel entry costs real money and real schedule. Trial and error stops scaling long before the design is good.
Results are hard to trust and harder to reuse.
A figure without units, a model without an error bar and a run nobody can reproduce all end the same way: the work is done again.
Four verbs.
One record of what happened.
Simulate
Solve the governing equations, with the method and the reference check stated next to the result.
Quantum circuits and wave packets, error-correction Monte Carlo, your own differential equations, teaching-scale sandboxes.
Finite-element, three-dimensional flow, magnetohydrodynamic and atomistic solvers on managed GPU clusters.
Predict
A world model trained on solver and experimental data returns the whole field, with an uncertainty band, fast enough to interact with.
Gaussian-process surrogates inside Bayesian optimisation.
Neural-operator and physics-informed surrogates for each sector.
Explore
See how every output moves across the design space, instead of sampling a handful of points and hoping.
Device-topology comparison, threshold curves across noise models, your own sweeps as short Python runs.
Interactive design-space maps over the sector world models.
Optimize
Let the search propose the next design, and send only the candidates that earn it to the solver or the experiment.
Bayesian optimisation with three acquisition rules and batch proposals.
Multi-objective, constrained search over world models, closed with cluster submission.
Extreme physics,
one instrument at a time.
Each sector below opens with what we are building and ends with a panel you can run now. The panels compute real, reduced-order physics in your browser: move a dial and the answer is solved again.
Plasma that
answers back.
Fusion design lives on magnetohydrodynamics: equilibrium, stability and transport, across magnetic, inertial and magneto-inertial confinement. We are building world models that return those answers while the design is still on the screen.
- Equilibrium and stability surrogates for magnetic confinement
- Implosion and burn surrogates for inertial confinement
- Compression dynamics for magneto-inertial concepts
- Heat loads on coils, walls and divertors
- Verification against reference codes on HPC
The equilibrium is solved again on every change. In this family of solutions the pressure follows from the current.
Limits checked: q at 95 % flux above 2, normalised beta below 3.5. Raise the current and watch the design leave them.
Running above, in your browser. Plasma equilibrium. Change the machine and read its shape, beta and safety factor from an exact solution of the Grad–Shafranov equation. A reduced-order model, not the sector world model.
The reactor,
before the wafer run.
Process development spends wafers to learn what a reactor does. We are building models of reactor transport and surface chemistry in space and time, so a recipe can be explored before a lot is committed.
- Reactor flow, heat and species transport in three dimensions
- Plasma and surface chemistry for etch and deposition
- Profile evolution at feature scale
- Recipe search under uniformity and throughput constraints
- Run-to-run drift models from tool data
Hotter is faster and less uniform: the reaction outruns the gas supply and the centre of the wafer starves.
Non-uniformity is half the spread between the fastest and slowest points, over the mean.
Running above, in your browser. Film deposition. Set temperature, pressure, showerhead gap and gas flow, and read growth rate and uniformity across a 300 mm wafer. A reduced-order model, not the sector world model.
De-risk it
before it flies.
Mission-critical hardware gets one flight. We are building aerothermal and structural models that screen a design across its whole envelope, so the expensive tests go to the cases that matter.
- Hypersonic aerothermal surrogates with chemistry
- Thermal protection sizing along a trajectory
- Thermal and structural response of hot structures
- Propulsion flow paths and nozzles
- Uncertainty maps over the flight envelope
Halve the nose radius and the peak heating rises by √2. That trade is why entry vehicles are blunt.
Wall temperature is the radiative-equilibrium value at the nose: the temperature at which the surface radiates away what it receives.
Running above, in your browser. Entry heating. Set velocity, altitude and body shape, and read heat flux, wall temperature and shock stand-off from engineering correlations. A reduced-order model, not the sector world model.
The sector
we already ship.
Quantum devices are extreme physics too, and this is where SynapseX simulations are live today. Circuits, wave-packet dynamics, error correction and device topologies run in Chat, on simulators that cost 0 credits.
- Exact circuit simulation with OpenQASM export
- Wave packets checked against a transfer-matrix reference
- Error-correction memory and threshold curves
- Device-topology comparison
- Noisy circuit simulation
- Schrödinger dynamics in three dimensions
- Device world models: pulse response, crosstalk and drift
Units with ħ = m = 1. Move a slider and the run starts again from the same packet.
The reference is the textbook transmission of a single energy. The packet carries a spread of energies, so the simulated value settles near it, not on it.
Running above, in your browser. Quantum dynamics. Send a wave packet at a barrier and watch it split. The scheme is unitary, so the norm staying at 1 is a check on the solver. A reduced-order model, not the sector world model.
Spend the solver
where it counts.
Every sector ends in the same loop: propose a design, evaluate it, learn, propose again. The search engine is shared, and it is already live in Chat for your own objectives.
- Bayesian optimisation with three acquisition rules and batch proposals
- Run planning for HPC: resources, scaling and cost
- Multi-objective search with constraints
- Search over the sector world models
- Submission to GPU clusters from Chat
Each evaluation stands for one expensive run: a simulation, a job on hardware, an experiment.
The optimum is known here because the objective is a test function scanned on 201 points. On a real problem that scan is the cost the search avoids.
Running above, in your browser. Design search. A Gaussian process learns a toy objective and expected improvement picks the next design. Compare its evaluations with the exhaustive scan. A reduced-order model, not the sector world model.
A fast answer
is not yet a right one.
A surrogate is only useful if you know when to believe it. These are the rules a sector world model has to meet before its state on this page changes from in development to live.
An uncertainty band on every prediction
A number without an error bar is not shipped. The band widens where the training data is thin.
A check against a reference
Each model is compared with a named solver or a measurement on named cases, and the comparison is published with it.
Units and provenance on every figure
The quantity, its units and range, the source and the reduction applied are on the figure itself.
A way back to the solver
Outside the range it was trained on, the model says so and hands the case to the full solver.
A sealed record of the run
Inputs, model version and outputs are frozen into a content-hashed record that someone else can re-run.
What runs today.
What we are building.
Two states and nothing else. Live today means you can run it now in SynapseX Chat. In development means it is on the build list and not available yet.
Quantum
- Exact circuit simulationStatevector, Bloch spheres, OpenQASM export.Live today
- Wave-packet dynamics in one dimensionSeven potentials, checked against a transfer-matrix reference.Live today
- Error-correction Monte CarloRepetition and surface codes, three noise models, memory and threshold curves.Live today
- Device-topology comparisonRouting overhead reported as an upper bound.Live today
- Noisy circuit simulationIn development
- Schrödinger dynamics in three dimensionsIn development
- Device world modelsPulse response, crosstalk and drift.In development
Classical dynamics
- Your own systems of differential equationsFourth-order Runge–Kutta, plotted as it integrates.Live today
- Teaching-scale sandboxesGravitating bodies, waves, the Ising model, two-dimensional fluids.Live today
- Short Python runs in a remote sandboxLive today
- Flow in three dimensionsNavier–Stokes and lattice-Boltzmann solvers.In development
- Finite elementsStructures, heat and coupled fields.In development
- Gravitating bodies in three dimensionsIn development
Sector world models
- FusionMagnetohydrodynamic equilibrium, stability and transport across magnetic, inertial and magneto-inertial confinement.In development
- SemiconductorsReactor transport and surface chemistry in space and time.In development
- AerospaceAerothermal and structural response across the flight envelope.In development
- Neural-operator and physics-informed surrogatesWith an uncertainty band on every prediction.In development
Chemistry and materials
- Molecules, crystals and proteins in 3DRendered from structure text.Live today
- Electronic structureDensity-functional and quantum-chemistry calculations.In development
- Atomistic molecular dynamicsIn development
- Structures from line notationSMILES and InChI, with conformers.In development
Search and scale
- Bayesian optimisationGaussian process, three acquisition rules, batch proposals.Live today
- Run planning for HPCResources, scaling and cost for a job. Nothing is submitted.Live today
- Search over sector world modelsMulti-objective, with constraints.In development
- Submission to GPU clusters from ChatIn development
Simulations,
answered plainly.
Bring us
your hardest case.
We are choosing design partners in fusion, semiconductors and aerospace. Their cases decide what is built first.
The panels on this page are reduced-order models. Sector world models are in development.
The panels on this page compute reduced-order physics in your browser for illustration. They are not design tools and their outputs are not qualified engineering results. Capabilities marked in development are not available yet and carry no delivery date. No speed-up or accuracy figure is quoted on this page because none has been measured against a named reference.