MIX^™ Platform

It all starts with data. Data is the core of any AI-based process. Without enough high-quality data, it’s garbage-in, garbage-out.

We train our machine learning models on experimental data from thousands of commercial-format prototype cells to maximize the signal-to-noise ratio and make sure the results are commercially-relevant.

Our industry-leading deep learning models enable efficient learning from our large datasets. Models with similar architecture have recently achieved breakthrough results in other machine learning tasks. We have customized these models to our specific use case: predicting battery materials performance.

Pairing the right model with the right dataset is critical. Small datasets typically found in R&D labs do not benefit from deep models.

Battery R&D on autopilot requires autonomous experiment design.

Predictions from our materials design model are used to automatically design batches of dozens of new experiments each day, from highly exploratory to highly optimized.

This automation enables us to work on multiple battery chemistries optimized for a variety of applications simultaneously.

Cycle life evaluation is a unique challenge of battery R&D. Historically, progress has been slow because feedback on a new design takes months.

Our cycle life forecasting model uses the history of a battery’s cycling behavior to forecast its future remaining useful life. This shortens the feedback loop from months to days.

Batteries need to deliver according to a variety of metrics – cycle life, heat generation, rate capability, safety, etc. – under a huge variety of possible use cases.

There is no substitute for comprehensive cell performance testing of commercially-relevant form factors.

We collect all of our own data to ensure confidence in the performance our cells deliver.