Custom electrolyte R&D and matched formation protocols for cell projects already in motion.

Lithiox solves the gap between catalogues, templates, and generic formation protocols with de novo inverse electrolyte design and joint formation protocol optimization.

Custom Electrolyte R&D

For battery manufacturers and R&D teams.

The electrolyte is the single most composition-sensitive layer in a cell. By the time the electrolyte is chosen the active materials are already committed, so the electrolyte is where the remaining performance lives.

Lithiox provides species and loadings discovered from first principles, composition-space search over the simplex, compatibility screening, and feasibility certification.

Formation Protocol Co-Design

For cell development teams with a cathode and anode system.

Between built and cycling sits the formation step, where the cell is charged and discharged under a controlled schedule while the SEI nucleates and grows.

Lithiox provides a current, voltage, and temperature schedule optimized jointly with the composition. The best protocol for one electrolyte is the wrong protocol for another.

Manufacturing and Delivery Support

For operations, procurement, plant management, and finance.

Imported liquid electrolytes create shelf-life loss, lead-time risk, and inventory friction. The electrolyte producer is the last mile and has to manage moisture, metals, and existing HF before delivery.

Lithiox provides high-purity formulations, localized supply chain support, standards for quality and freshness, and just-in-time delivery and inventory support.

Who needs this

Our ideal partners are the battery manufacturers, their R&D and manufacturing teams, and the engineers who own the outcomes of the cell electrolyte and the formation development. They come to us after the cathode, anode, separator, current collectors, and cell format are already picked with the downstream application in mind.

How does it work?

We only need to know what cell configuration each cell producer picked for each battery project, and the target performance outcomes. The prescribed performance envelope gives the optimizer something to point to, negotiate tradeoffs through the coupled models, and find free lunches with non-obvious, non-linear, non-additive interaction effects.

In return, users get

Fast inverse-designed de novo electrolyte composition: species and loadings that best satisfy the envelope for that specific cell, discovered from first principles rather than matched against a catalogue or template.
A matched formation protocol i.e. a schedule of current, voltage, and temperature optimized jointly with the composition against the same envelope. The best protocol for one electrolyte is the wrong protocol for another.
Per-KPI sensitivities decomposed through the physics. A predicted N80 is attributed through SEI thickness, LiF fraction, dead-lithium inventory, LAM, and crack density, so the customer can understand which mechanisms carry the result and which one carries the risk.

What exists now?

Forward simulators are not electrolyte design tools.

Electrolyte design at most manufacturers still runs on multi-month design-of-experiments loops. The large cell makers have proprietary libraries and programs while smaller manufacturers deprioritize electrolyte development altogether and ask for catalogues of formulations.

PyBaMM and COMSOL are Newman-style forward simulators of a cell whose electrolyte parameters you need to already know. The structural gap is the same in both: the electrolyte enters as a hand-filled parameter bag, and formation is exogenous.

	This project	PyBaMM	COMSOL Battery Module
Compositionality	Species identities and loadings are the variables. Speciation, transport, interphase composition, and aging all read from the same recipe.	The electrolyte is a tabulated sigma(c,T), t+0(c), D_e(c,T), dln f / dln c. Swap FEC for VC and the table has to be refit against fresh EIS/GITT cells before the model has anything to say.	Same tabulated electrolyte as PyBaMM, typed in by hand per cell in the GUI.
Inverse Design	Column-generation search over the simplex driven by reverse-mode gradients. Hand it a cell and an envelope, it gives back species, loadings, and a protocol.	Forward simulator. The user picks the electrolyte, the model scores it.	Forward simulator. The user picks the electrolyte, the model scores it.
Sensitivity Analysis	One reverse-mode pass returns dKPI/dcomposition and dKPI/dprotocol for every KPI at once, attributed through every mechanism in the chain.	CasADi forward sensitivities live in fitted-coefficient space. A composition gradient means finite-differencing a slow PDE solve and living with the noise.	Forward-mode Sensitivity Analysis through LiveLink-to-MATLAB. Composition is not on the parameter list.
Speciation & Solvation	Activity coefficients, ion pairing, contact and solvent-separated populations, and Onsager couplings, all derived from molecular structure inside the same differentiable graph.	Folded into the transport fit. There is no speciation layer.	Folded into the transport fit. There is no speciation layer.
Transport Model	OSM ionic transport with Onsager couplings, XGBoost conductivity with finite-difference JVPs through the ensemble, viscosity from the same composition graph.	Concentrated-solution Newman closure consuming the user’s table.	Concentrated-solution Newman closure consuming the user’s table.
Formation Model	Voltage-resolved reduction cascade, bilayer SEI emerging from independent parallel Butler-Volmer kinetics, species-split gas, dead-Li accounting, differentiable stage transitions. The protocol is an optimization variable.	Cycling starts from a prescribed SEI thickness, porosity, and Li inventory. Formation is somebody else’s problem.	Cycling starts from a prescribed SEI thickness, porosity, and Li inventory. Formation is somebody else’s problem.
Aging Model	Composition-dependent SEI growth, three-channel LAM split, dead-Li inventory, HF cascade, mechanical crack coupling, and a temperature-coupled thermal ODE. Each one is a state variable the optimizer moves on.	Arrhenius k_SEI, E_a, a plating overpotential threshold, and a fitted LAM stress coefficient. LiF fraction, donor number, and additive identity never enter the rate laws.	Same phenomenological aging laws as PyBaMM, run on a 3D field at 3D-DFN cost per cycle.
Novel molecules	Functional-group product predictor handles species with zero training data.	Outside the parameter table, the molecule does not exist.	Outside the Material Browser entry, the molecule does not exist.
Speed	Sub-second forward pass on an M4 CPU. Thousands of composition and protocol candidates per hour on a single laptop.	SPMe is seconds per 1000-cycle run, DFN is tens of seconds to minutes. A real composition sweep stretches into hours or days.	3D DFN is minutes to hours per cycle. Aging sweeps become infeasible as soon as the geometry leaves trivial.
Hardware Floor	A consumer M4 laptop. No GPU, no cluster, no licensed solver.	Workstation Python. Laptop SPMe is fine, laptop DFN hurts.	Workstation with a licensed solver, cluster for 3D aging.
Licensing & Deployment	Ships as an API and as a report, can run on a consumer laptop, sits inside an optimizer loop with hundreds of parallel forward evaluations.	Open-source (BSD). Deploys anywhere Python does.	Five figures per seat per year for the Battery Module and LiveLink. Every optimizer-loop instance needs its own seat.
Ease of use	Cell config and performance envelope in, decomposed PDF report out. An R&D engineer runs it from the same desk they use for cycler data.	Python API. A trained electrochemist with a parameter set in hand can drive it.	GUI with mesh convergence, boundary conditions, physics couplings, and reaction source terms. Every new cell is a multi-week setup exercise for a specialist.
Failure reporting	Envelope conflicts surface as a named conflict report listing which axes cannot be jointly satisfied.	Failure shows up as solver non-convergence and gets handed back to the user to interpret.	Failure shows up as solver diagnostics inside the GUI.
Spatial resolution	Lumped ODE. Tab edges, 3D thermal hot spots, and current-collector gradients are outside the model.	1D DFN through the electrode thickness.	Full 3D DFN with thermal and mechanical field coupling. The right pick when the failure mode is genuinely 3D.