AI-Powered Battery Material R&D

From simulation
to validated
material — fast.

We apply deep learning molecular dynamics to find, optimize, and validate next-generation battery materials. Silicon anodes. Solid-state electrolytes. At the speed your production roadmap demands.

3,730×
Faster than DFT
>99%
Prediction accuracy
20k+
Atoms per simulation
dlmd — AI Battery Material Design Platform
$ dlmd simulate --material Li2ZrCl6 --phase disordered-alpha
Loading deep learning potential (DLP)... ✓ 99% accuracy
System size: 12,960 atoms
Simulation time: 5.0 ns
Temperatures: 750 K → 1000 K (6 pts)
Runtime vs AIMD: 3,730× faster
── Results ──────────────────────────
Ionic conductivity (300K): 1.32 × 10⁻³ S/cm
Activation energy: 0.30 eV
Haven's ratio (HR): 0.13 (correlated)
Dominant diffusion: interlayer (z-axis, 70%)
vs. experiment: Δ < 22%
$
Partners
Lawrence Berkeley National Lab
Argonne National Lab
Univ. of Illinois Chicago
DOE BMR Program
Southern Methodist Univ.
Our Approach

Quantum accuracy.
Industrial speed.

Our DLMD pipeline bridges the gap between first-principles science and the throughput your R&D program needs — validated by peer-reviewed research at Argonne and Berkeley Lab.

STEP 01

Ab-Initio MD

DFT-based AIMD via VASP generates gold-standard atomic trajectory data at multiple temperatures — 5,000+ physically consistent frames per material.

STEP 02
🧠

Deep Learning Potential

A deep neural network (DeePMD-kit) is trained on AIMD data, learning the potential energy surface with >99% force and energy accuracy.

STEP 03

Large-Scale DLMD

Classical MD via LAMMPS + DeepMD runs 9,000–20,000 atom systems over 3–5 ns at 6–8 temperatures — 3,730× faster than AIMD.

STEP 04
🔬

Lab Validation

Top computational candidates proceed to physical validation at Berkeley Lab or Argonne. AI predictions confirmed. Results fed back into the next iteration.

Services

Two high-stakes
material challenges.

We focus where the industry bottleneck is tightest — silicon anodes now entering mass production, and solid-state electrolytes racing toward the 2027–2028 production window.

Service Line 01

Silicon Anode Materials

Silicon anodes offer 10× the theoretical capacity of graphite — but volume expansion up to 400% during cycling destroys conventional binders and structures. We use AI to design solutions production lines can actually use.

  • Conductive polymer binder design. AI optimization of multifunctional HOS-PFM polymer binders that conduct both electrons and Li-ions simultaneously — eliminating the trade-off between binding strength and conductivity. Developed by Dr. Gao Liu at Berkeley Lab (Nature Energy 2023).
  • Si secondary particle engineering. AI-designed micron-scale composite particles combining Si nanoparticles + CNT + conductive polymer — bridging nano-lab materials to production-ready powders.
  • Volume expansion modeling. DLMD simulation of lithiation/delithiation cycles — predicts mechanical failure modes before a single physical test is run, saving months of cycle testing.
  • Non-aqueous binder reformulation. For sulfide solid-state battery applications, we model and optimize binder performance in low-polarity, ester-based solvents compatible with moisture-sensitive sulfide electrolytes.
  • Cycle life prediction. DLMD-based capacity retention forecasting over hundreds of simulated cycles — validated against physical battery lab results at Argonne.
HOS-PFM PolymerSi+CNT CompositesPVDF Alternativesπ–π StackingNon-aqueous Systems
Service Line 02

Solid-State Electrolytes

Solid-state batteries require electrolytes that are chemically stable, mechanically compatible, and ionically fast. The composition space is vast. DLMD lets us screen it at a speed physical experiments never could.

  • Halide electrolyte screening. DLMD simulation of Li₂ZrCl₆, Li₃TiCl₆, Li₃YCl₆ and related halide systems — predicting ionic conductivity, activation energy, and Haven's ratio at room temperature.
  • Phase & disorder optimization. We identify which crystallographic phase and structural disorder level maximizes conductivity — directly guiding synthesis conditions and milling parameters.
  • Doping concentration mapping. For sulfide electrolytes, DLMD screens dopant species and concentrations to minimize activation energy barriers — without exhaustive wet-lab synthesis runs.
  • MIEC material development. Amorphous mixed ionic-electronic conductors (Li-Ti-P-S type) enable both ionic and electronic transport — unlocking high-capacity Li-S batteries.
  • Transport mechanism mapping. We reveal whether your material conducts via inter-layer hopping, intra-layer diffusion, or free-volume mechanism — guiding the next synthesis iteration.
Halide SSESulfide SSEMIECAmorphous ModelingLi-S Batteries
AI in Action

Real results from
our DLMD simulations.

The following are actual computational findings from peer-reviewed research by our CSO Dr. Anh T. Ngo, published in ACS Applied Energy Materials, Journal of the Electrochemical Society, and Nature Materials.

DLMD // Li₂ZrCl₆ phase comparisonPUBLISHED · ACS 2025

Challenge: Li₂ZrCl₆ exists in multiple crystal phases. Which phase should your synthesis target?

ordered α-LZC
σ (300K)6.57×10⁻⁴ S/cm
Ea0.33 eV
Haven's HR0.40
OPTIMAL
disordered α-LZC
σ (300K)1.32×10⁻³ S/cm
Ea0.30 eV
Haven's HR0.13
β-LZC
σ (300K)7.66×10⁻⁶ S/cm
Ea0.45 eV
Haven's HR0.97
→ AI identifies disordered α-LZC as 172× more conductive than β-LZC. Root cause: ZrCl₆²⁻ octahedral softening from single-layer Zr arrangement. Synthesis guidance: maximize ball-milling disorder.
DLMD // Li-Ti-P-S doping optimizationPUBLISHED · arXiv 2025

Challenge: Ti doping improves LPS conductivity — but at what concentration?

0% Ti
0.38 eVHigher barrier
10% Ti ★
0.30 eVOPTIMAL
20% Ti
0.32 eVNear-optimal
30% Ti
0.37 eVOver-doped
→ 10% Ti doping: highest configurational entropy, most stable Li-S polyhedra, lowest activation energy. Confirmed vs. experimental DOE data.
DLMD // Li₃TiCl₆ — transport mechanism discovery + speed benchmarkPUBLISHED · J. ELECTROCHEM. SOC. 2024

Challenge: Li₃TiCl₆ transport mechanism was unknown. Physical NMR experiments could not resolve it.

IONIC PROPERTIES
Conductivity (25°C)1.06 mS/cm
Activation energy0.29 eV
vs. experimentΔ < 10%
System size20,000 atoms
TRANSPORT MECHANISM
Mechanism typeInterstitial hopping
Motion typeAnti-correlated
Haven's ratio>1 (rare)
Dself / DTotal14× overestimate
SPEED BENCHMARK
CONVENTIONAL AIMD
150,000 s/atom
AIDEA DLMD
40 s/atom
→ Key discovery: Li₃TiCl₆ shows anti-correlated Li-ion motion (Haven's ratio >1). Conventional analysis overestimates conductivity by 14×. AI reveals the true mechanism — pointing directly to how voids at Ti sites can be engineered to optimize inter-layer hopping.
DLMD // SSE screening — directional conductivityFROM Li₂ZrCl₆ STUDY

AI decomposes ionic conductivity into directional components — revealing where to engineer ion channels.

dis. α-LZC σz
0.844 S/cm
dis. α-LZC σx
0.203 S/cm
ord. α-LZC σz
0.570 S/cm
β-LZC σtotal
0.099 S/cm
→ Disordered α-LZC's z-axis conductivity accounts for 70% of total transport. Engineering synthesis to maximize the single-layer Zr arrangement directly targets the most impactful structural feature.
DLMD // MIEC LPS:Ti — Li-S battery enablementNature Materials 2025

Amorphous Li-Ti-P-S is a mixed ionic-electronic conductor (MIEC) — enabling high-capacity Li-S batteries.

DISCHARGE CAPACITY
>1,450
mAh/g — experimentally validated
CYCLE LIFE
>1,000
cycles with MIEC interlayer
LPS (0% Ti)
10⁻⁴ S/cm
LPS:Ti10% ★
10⁻³ S/cm
→ DLMD identifies free-volume diffusion as the transport mechanism — guiding synthesis to maximize structural disorder at 10–20% Ti, correlated with >1,450 mAh/g capacity.
Technical Depth

What DLMD delivers
that nothing else can.

Our platform is a battery-materials-specific DLMD framework, trained, validated, and refined over years of Argonne and Berkeley Lab research.

01

Multi-scale coverage: quantum to cell

DLMD starts at the quantum level (DFT-AIMD) and scales to simulation cells of 20,000+ atoms — capturing complex dynamics of amorphous and disordered materials that small-scale DFT cannot represent.

02

Anti-correlation and correlated diffusion

Standard tools use Nernst-Einstein with self-diffusion coefficient — which can overestimate conductivity by 14× in materials with anti-correlated ion motion. Our DLMD computes Haven's Ratio for accurate prediction.

03

Amorphous and MIEC material modeling

Amorphous solid electrolytes and mixed ionic-electronic conductors cannot be modeled with crystalline DFT. Our DLMD handles disordered Li-Ti-P-S, amorphous LPS, and other non-crystalline systems.

04

Experimental validation at national labs

Every computational study is validated against physical experimental data — ionic conductivity measurements, X-ray diffraction, and electrochemical cycling at Argonne and Berkeley Lab.

05

Mechanism-level insight, not just numbers

We tell you not just which material is faster — but why. Transport mechanism analysis gives your R&D team the structural insights to improve the next synthesis iteration intelligently.

Core Technologies
DeePMD-kitLAMMPSVASP (AIMD)TensorFlow DNNDeepPot-SEMean Square Displacementvan Hove CorrelationHaven's RatioPotential of Mean ForceRadial DistributionConfigurational EntropyCOHP / DOS AnalysisNEB Migration BarriersArgonne Bebop HPC
>99%
Force and energy prediction accuracy
vs. ab-initio AIMD benchmark — J. Electrochem. Soc. 2024, ACS Appl. Energy Mater. 2025
The Team

Science and industry,
in one founding team.

Three co-founders covering every dimension of battery material R&D: world-class computational science, Berkeley Lab materials invention, and commercial scale-up experience.

TS
Tatsu Suzuki
CEO · Serial Entrepreneur

Former CEO of Zeptor Corporation — an Intel spinout developing silicon anode Li-ion batteries. Brings the industry network and commercial relationships that turn research into paying contracts.

Intel AlumniZeptor Corp.
GL
Dr. Gao Liu
CTO · Battery Materials Scientist

Senior Scientist at Lawrence Berkeley National Laboratory. Inventor of HOS-PFM multifunctional conductive polymer adhesives — enabling simultaneous electron and Li-ion transport in silicon anodes. Published in Nature Energy and Energy & Environmental Science.

Berkeley LabNature EnergySi Anode IP
AN
Dr. Anh T. Ngo
CSO · DLMD Platform Developer

Professor at University of Illinois Chicago and Physicist at Argonne National Laboratory. Developer of the DLMD simulation platform. Published three peer-reviewed papers validating DLMD for halide (Li₂ZrCl₆, Li₃TiCl₆) and sulfide (Li-Ti-P-S) solid electrolytes.

Argonne National LabUIC ChicagoACS · J.Electrochem.
Get in Touch

Ready to accelerate
your material R&D?

Tell us your target material, your current challenge, and your timeline. We will show you what DLMD can deliver — with a scope and timeline tailored to your program.