Slide 1: From Molecules to Market: Multiscale Machine Learning & Informatics for Solid-State Electrolyte Design
aterials nformatics & odeling From Molecules to Market: Multiscale Machine Learning & Informatics for Solid-State Electrolyte Design, Integration and Scale-Up 03/25/2026 International Battery Seminar: Al for Energy Storage Solid
Slide 2: Disclaimer
Disclaimer Cautionary Note Regarding Forward-Looking Statements All statements other than statements of present or historical fact contained herein are “forward-looking statements” within the meaning of Section 27A of the Securities Act of 1933, as amended, and Section 21E of the Securities Exchange
Slide 3: Agenda
Agenda Multiscale Machine Learning & Informatics for Solid-State Electrolyte Manufacturing * Overview: Role of Artificial Intelligence and Materials Informatics at Solid Power ¢ Use Case 1: Ultra-Early Cell Performance Prediction and Failure Mode Detection: 80%-90% speedup in testing and integration
Slide 4: Introduction
Introduction Strong coupling across mechanics, interfaces, and electrochemistry What are the key challenges? Anode curfent collector Chemo-Mechanical Coupling + Intercalation strain alters interfacial contacts and electrochemical potentials. * Stack pressure dependence DAS PO Compatibility + All sol
Slide 5: Overview of Materials Informatics & Modeling at Solid Power
Overview of Materials Informatics & Modeling at Solid Power Multiscale Machine Learning & Informatics for Solid-State Electrolyte Manufacturing (1) Progress Multipliers (acceleration), (2) Value Enhancers (optimization), & (3) Breakthrough Discovery & Innovation The Materials Informatics & Modeling
Slide 6: Role of Machine Learning & Informatics
Role of Machine Learning & Informatics Where are the ML models integrated at Solid Power? Jodo} So Inverse Forward Atoms & Structure-Process- === Performance ate ue Commercialization Molecules <4 Property mums §=6Safety & Durability — Design Engineering a Cell & Process A IP R&D fi 7 == | Production
Slide 7: Impact of Materials Informatics & Modeling at Solid Power
Impact of Materials Informatics & Modeling at Solid Power What these tools changed in practice over the last 18 months? Infrastructure: (1) ATLAS: centralized data platform for rapid correlation (~50x faster data evaluation) (2) Custom pipelines: Al for materials discovery & tuning (3) SPARK-iT: AI/
Slide 8: Impact of Materials Informatics & Modeling at Solid Power
Impact of Materials Informatics & Modeling at Solid Power What these tools changed in practice over the last 18 months? Less noise. More signals. Faster decisions. AI/ML Prediction telligent Toolkit Platform Infrastructure: (1) ATLAS: centralized data platform for rapid correlation (~50x faster data
Slide 9: Use Case 1: Ultra-Early and Failure Mode Detection
Use Case 1: Ultra-Early Cell Performance Prediction and Failure Mode Detection
Slide 10: Ultra-Early & Failure Mode Detection
Ultra-Early Cell Performance Prediction & Failure Mode Detection SPARK-IT — An early-warning and decision-support platform for cell testing Typical Battery Testing Protocol Vv <€Days>! < Months of Experimental Data Collection Cell Properties Formation Extended Cycle Life Testing to understand how ma
Slide 11: Ultra-Early & Failure Mode Detection
U Itra-Early Cell Performance Prediction & Failure Mode Detection SPARK-IT — An early-warning and decision-support platform for cell testing Typical Battery Testing Protocol n r 2 |€ Days 2 Months of Experimental Data Collection oO 1 2 ; & Months 8 Formation Extended Cycle Life Testing How Testing O
Slide 12: SPARK-iT: cell performance prediction to materials manufacturing
Cell Performance Prediction and Failure Mode Detection SPARK-iT: cell performance prediction to materials manufacturing A faster workflow for cell analytics and electrolyte manufacturing : and Signal Capacity Analysis Lifetime Chargeability 4 Observe ! Predict ML Predictions | Material Informatics &
Slide 13: Interpretable #1 CE Prediction Model
Interpretable Cell Performance Prediction #1 CE Prediction Model SPARK-iT: cell performance prediction to materials manufacturing >) Part 1: Data Preparation comatned wou) SUN OIKS 1) Extract Features a) Time-series Features: data for every timestep within every with Physics cycle which are: current
Slide 14: Interpretable #1 CE Prediction Model
Interpretable Cell Performance Prediction #1 CE Prediction Model SPARK-iT: cell performance prediction to materials manufacturing ~ Combined Neural Networks with Physics CE model goals: (1) X(time series data) => Y (CE Profile) (2) X(time series data) > Y (EOL Cycle) Historical Data y Data Cleaning
Slide 15: Interpretable #1 CE Prediction Model
Interpretable Cell Performance Prediction #1 CE Prediction Model SPARK-iT: cell performance prediction to materials manufacturing i Y Part 2: Algorithm Training Combined Neural Networks with Physics Neural Network Algorithm CE model goals: Time Series A . . (1) X(time series data) > Y (CE Profile) I
Slide 16: Interpretable #1 CE Prediction Model
Interpretable Cell Performance Prediction CE Prediction Model SPARK-iT: cell performance prediction to materials manufacturing Combined Neural Networks with Physics CE model goals: (1) X(time series data) => Y (CE Profile) (2) X(time series data) > Y (EOL Cycle) Historical Data T Internal data + ope
Slide 17: Interpretable #1 CE Prediction Model
Interpretable Cell Performance Prediction #1 CE Prediction Model Te Sample 32108 Sample 32 Pes SPARK-{T: cell performance prediction enna : Part 3: Using early-cycle data to predict > later behavior Combined Neural Networks — Prediction —>| with Physics CE model goals: (1) X(time series data) => Y (
Slide 18: Interpretable #1 CE Prediction Model
#1 CE Prediction Model Part 3: Using early-cycle data to predict later behavior Interpretable Cell Performance Predic SPARK-iT: cell performance prediction > \ Combined Neural Networks with Physics = a 3 E. 3 CE model goals: (1) X(time series data) => Y (CE Profile) (2) X(time series data) > Y (EOL
Slide 19: Interpretable CE Prediction Model
Interpretable Cell Performance Prediction CE Prediction Model Part 3: Using early-cycle data to predict later behavior Combined Neural Networks Physics Contributions — Cell 1b x with Physics 1 Fastin condoning cry exgonetl) Hl Bukion vansprtstabizaton i expen) IM Late-stgo contact nxn Sal tcp patrwa
Slide 20: Interpretable Failure Mode Classification
* Goal: identify and rank primary and secondary failure modes with physics-grounded interpretability. Interpretable Cell Performance Prediction SPARK-iT: cell performance prediction to materials manufacturing * ML Integration: (1) An encoder maps experimental observables (e.g., time series data) to
Slide 21: Failure mode confusion matrix
True Failure Mode Interpretable Cell Performance Prediction SPARK-iT: cell performance prediction to materials manufacturing Confusion Matrix (% per true class) Predicted Failure Mode 100 80 60 40 20 A physics-guided model for classifying likely failure modes Key takeaways — elliDx + Mechanism: Void
Slide 22: Ultra-Early & Failure Mode Detection
Ultra-Early Cell Performance Prediction & Failure Mode Detection SPARK-iT: cell performance prediction to materials manufacturing #1 CE prediction model Hua #2 Failure Mechanism Typical Battery Testing Protocol Q T = |€ Days>: < Months of Experimental Data Collection 2 : o H a & Formation ; Extended
Slide 23: Use Case 2: Intelligent Design of Experiments (DOEs) and Feedback
Use Case 2: Intelligent Design of Experiments (DOEs) and Feedback for Knowledge-Informed Electrolyte Manufacturing Sag... Q°;.b0vH
Slide 24: Knowledge-Informed Electrolyte Manufacturing
Knowledge-Informed Electrolyte Manufacturing Al-driven design of experiments platform for control and optimization Targets Q1: How to enable faster, consistent, efficient and systematic exploration of solid- state electrolyte manufacturing? Q2: Can Al-driven tools accelerate and improve this process
Slide 25: Knowledge-Informed Electrolyte Manufacturing
Knowledge-Informed Electrolyte Manufacturing Solution: Translate technical goals into optimization targets P Experimental Data Prior Knowledge {| Cy ents / Historical productic fa] ese, Ate Se Physesmods — Tool Speed Expert knowledge L Sensitivity Analysis Al-Optimization Engine Temperature Prior Uk
Slide 26: Knowledge-Informed Electrolyte Manufacturing
Knowledge-Informed Electrolyte Manufacturing yo oo icp occ secor meses Stage 1: Project Setup Knowledge-driven workflow for DOE Existing», ves data? Yo {Project folder intalzed Stage 2: Design Track A: Variables bj Input Stage Interpretable Featurizat + Define parameters * Set bounds (min, max] 1 Ty
Slide 27: Knowledge-Informed Electrolyte Manufacturing
Knowledge-Informed Electrolyte Manufacturing Closed-loop intelligent experimental design framework via SPARK-iT Example/Tutorial: Bayesian optimization of three process parameters for fast charging* *This is a hypothetical example intended to illustrate the workflow and is not based on actual Solid
Slide 28: Knowledge-Informed Electrolyte Manufacturing
Knowledge-Informed Electrolyte Manufacturing Closed-loop intelligent experimental design framework via SPARK-iT Example/Tutorial: Bayesian optimization of three process parameters for fast charging* *This is a hypothetical example intended to illustrate the workflow and is not based on actual Solid
Slide 29: Conclusion & Takeaways
Conclusion & Takeaways Multiscale Machine Learning & Informatics for Solid-State Electrolyte Manufacturing la ~ Materials Informatics & Modeling Multiplication Value Enhancemen Performance + ability Support External Customers igh Discov Al/ML must be deployed against real, operational technical bott
Slide 30: Thank you!
Thank you! Solid Power is a solid-state battery materials company focused on solid-state electrolyte technology. We leverage/design Al/ML/informatics tools when doing so accelerates materials discovery, process understanding, and data-driven decision making. Our core value is advancing electrolyte m