New Data-driven materials design platform

D3Square

Design of experiments, machine learning, and optimization wired into one loop — reach target properties with fewer experiments.

  • Automated experiment design with DOE and Bayesian optimization
  • Unified management of experimental and simulation data, with version tracking
  • Model training and next-experiment recommendation in the same workspace
  • On-premises and private-cloud deployment options
Request a PoC Request the deck

Fewer experiments, higher confidence

~60%
Fewer experiments
3x
Faster candidate exploration
100%
Data traceability
On-prem
Deployment support
Features

Core capabilities

Everything a materials or process team needs to move from bench-scale experiments to data-driven development.

Automated DOE

Define target properties and constraints — D3Square proposes optimal experiment points. Factorial, LHS, and D-optimal supported.

ML training & evaluation

Train predictive models directly from experimental data, with built-in model comparison and uncertainty estimation.

Bayesian optimization

An active-learning loop that recommends the next experiment. Balances exploration vs exploitation, supports multi-objective targets.

Experiment database

Structured storage of compositions, process variables, and measurements — with per-researcher and per-team permissions and history.

Simulation integration

Connect to Materials Square, in-house codes, or external CAE to build hybrid experimental–simulation datasets.

Enterprise deployment

Docker and Kubernetes-based on-prem and VPC deployments — appropriate for R&D organizations where data sovereignty matters.

How it works

Project flow

From organizing past experiments to recommending the next one — all in one workspace.

  1. 01

    Define targets & constraints

    Set target properties, cost and process constraints, and the exploration ranges for each variable.

  2. 02

    Collect and clean data

    Upload existing experimental and simulation data; normalize missing values and units into a training-ready dataset.

  3. 03

    Train and validate models

    Compare candidate models automatically, review validation performance and uncertainty, and pick the one to adopt.

  4. 04

    Recommend and iterate

    Run the experiments Bayesian optimization recommends, then feed the results back into the loop.

Use cases

Use cases

Secondary-battery composition optimization

Multi-variable optimization across cathode, anode, and electrolyte compositions; exploring lifetime vs capacity trade-offs.

Catalyst screening & combinatorics

Quickly surface promising combinations from a large candidate pool and prioritize which experiments to run first.

Process parameter tuning

Optimize temperature, dwell time, current density and other process variables to improve yield and quality.

Proprietary property prediction models

Build in-house-only predictive models from internal databases that cannot be shared externally.

The first step to turn your data into an asset

Tell us what data you have, your targets, and your constraints — we will sketch out a 30-minute PoC scenario with you.

Request a PoC Request the deck