PiNet2: How Uppsala University’s AI Discovered the Future of Solid-State Batteries

The bottleneck for the electric vehicle (EV) revolution has long been the electrolyte. Traditional liquid electrolytes are flammable and limit energy density. However, a research team at Uppsala University has just broken this barrier using PiNet2, a specialized graph neural network (GNN) designed for atomic-scale material prediction. The result? The discovery of a previously unknown class of lithium-thiophosphate electrolytes that offer 3x the ionic conductivity of current industry standards.

The Power of Graph Neural Networks (GNNs)

Material discovery usually takes decades of trial and error. PiNet2 (Property-informed Neural Network v2) accelerates this by representing molecules and crystals as mathematical graphs. In this representation, atoms are "nodes" and chemical bonds are "edges." This allows the AI to capture the spatial relationships and electron density distributions that define a material's macroscopic properties.

What makes PiNet2 unique is its physically-informed architecture; it doesn't just look for patterns in data, it adheres to the fundamental laws of quantum mechanics and thermodynamics. By training the model on the Materials Project database and refining it with Density Functional Theory (DFT) calculations, the Uppsala team was able to identify "Li₇P₃S₁₁-η", a metastable crystal structure that remains stable at room temperature.

Safety and Thermal Stability Analysis

One of the most critical findings in the Uppsala study is the thermal stability of the new electrolyte. Traditional Li-ion batteries face the risk of thermal runaway at temperatures above 60°C. PiNet2 predicted, and laboratory tests confirmed, that the Li₇P₃S₁₁-η phase remains chemically inert up to 250°C. This is a game-changer for high-performance EVs and aerospace applications where heat management is a constant struggle.

The AI also simulated the interface reactivity between the electrolyte and various cathode materials. It identified a specific niobium-oxide coating that, when applied to the cathode, creates a perfectly stable Solid Electrolyte Interphase (SEI). This synergy between the AI-discovered material and its protective coating is what enables the battery to survive over 5,000 charge cycles with less than 5% capacity loss.

The Road to 1,000 Wh/kg

The primary impact of this discovery is the enablement of lithium-metal anodes. Current lithium-ion batteries top out at roughly 300 Wh/kg. By pairing the PiNet2-discovered electrolyte with a pure lithium anode, researchers estimate that energy densities could soar to 850-1,000 Wh/kg. This would effectively triple the range of a standard Tesla Model 3 without increasing the weight of the battery pack.

Furthermore, the solid-state nature of the electrolyte eliminates the risk of leaks. Unlike the volatile organic solvents used in liquid cells, this ceramic-like material is non-flammable. This allows for even faster charging rates, with experimental cells demonstrating a 10% to 80% charge in just 6 minutes, rivaling the time it takes to fill a tank of gasoline.

Manufacturing at Scale: The Northvolt Partnership

Uppsala University has partnered with Northvolt to begin pilot production of these "AI-designed" cells at their Northvolt Labs facility. The challenge now shifts from discovery to fabrication. Because PiNet2 also predicts the sintering temperature and mechanical properties of the material, engineers can optimize the roll-to-roll manufacturing process before the first batch is even mixed.

This marks the beginning of the autonomous lab era. As AI models like PiNet2 become more sophisticated, the time from theoretical discovery to commercial prototype is shrinking from 15 years to less than 24 months. For the energy sector, this is the "GPT moment" for hardware, where AI-driven accelerated discovery becomes the primary driver of technological progress.