The Who

A battery materials team developing a high-voltage LiFSI carbonate electrolyte for NMC811/graphite cells charged to 4.5 V.

The goal: Identify a formulation-process window that improves oxidative stability, reduces impedance growth, maintains conductivity above 8 mS/cm, and remains practical for scale-up.

The Problem

Electrolyte development is slow because the search space compounds quickly.

A single program can span salt concentration, solvent ratio, additive package, moisture limits, mixing conditions, wetting time, formation protocol, and cycling performance. A conventional factorial DOE can quickly grow into hundreds of process conditions before replicates or scale-up validation.

The harder problem is that formulation and process are coupled. A recipe that passes under controlled moisture can fail when environmental conditions drift. Without structured process data, teams risk confusing process failures with chemistry failures.

What Scalarity Needed to Solve

The team needed a faster way to move from broad screening to a validated process window. Scalarity needed to support:

  • AI-driven DOE instead of exhaustive factorial screening
  • Early-signal prediction from conductivity, viscosity, LSV, EIS, formation, and early-cycle data
  • Structured capture of recipe, process, characterization, decision, and next-best action
  • Process-window discovery, not just formulation ranking
  • A benchmark for reducing process-development cycles

Why Existing Approaches Fell Short

Conventional DOE

A representative factorial DOE could test 3 salt concentrations × 4 additive packages × 3 wetting times × 3 moisture levels × 2 formation protocols = 216 process conditions. This gives coverage, but it is slow and does not adapt as results arrive.

Literature-led screening

Literature helps teams choose a starting point, but it does not predict which formulation-process combination will work in a specific customer environment.

Spreadsheets and internal tools

Spreadsheets can track results, but they rarely connect formulation, process, characterization, prediction, and next-best action in one closed loop.

The Scalarity Workflow

Scalarity turns every process run into a learning asset.

Each experiment connects: Hypothesis → Recipe → Process Run → Characterization → Decision → Next-Best Action

Instead of waiting for a full DOE grid to complete, Scalarity uses early experimental signals to predict downstream success and recommend the next most valuable run.

What Scalarity Found

The mock campaign began with a failing LiPF6 baseline. A direct LiFSI analog improved conductivity but still missed high-voltage stability and impedance targets.

Scalarity then moved the campaign toward interphase-stabilizing additive packages. The first passing candidate used LiFSI with FEC and LiDFOB. A follow-on formulation with TMSP delivered the best-balanced result.

This candidate met the core screening targets while maintaining manufacturability headroom.

The campaign also revealed a critical process constraint: the same formulation failed under elevated moisture. That shifted the output from a recipe recommendation to a process-window recommendation.

1.2M LiFSI in EC:EMC:FEC 20:70:10 with 1 wt% LiDFOB and 0.5 wt% TMSP

Results

In the representative campaign, Scalarity reached:

  • First passing candidate by the third LiFSI-focused process run
  • Best-balanced candidate by the next optimization run
  • Initial process-window insight through a moisture robustness check
  • Scale-up recommendation based on viscosity, wetting, and moisture sensitivity

The winning recommendation was not simply a formulation. It was a formulation-process window:

Scalarity recommendation

Use 1.2M LiFSI in EC:EMC:FEC with LiDFOB and TMSP, processed under controlled low-moisture conditions, then advance to pouch-cell validation, low-temperature conductivity testing, and wetting/filling sensitivity studies.

Compared with a representative 216-condition factorial DOE, the mock Scalarity workflow demonstrates how AI-driven DOE can reduce experimental cycles by prioritizing the next best run instead of exhaustively testing the grid.

Pilot benchmark target: reduce electrolyte process-development time by 70–80% using AI-driven DOE and early-signal prediction.

Takeaway

Battery development is not just about finding a promising formulation. It is about finding the formulation-process window that can be reproduced, validated, and scaled.

Scalarity helps teams get there faster by turning each experiment into a prediction for what to do next.