Research

Bridging AI with the physical world

Normal is built by engineers at the frontier of physical and mathematical intelligence: co-creators of core TensorFlow frameworks, co-founders of Meta’s Probability team, architects of Google’s first production-scale AI deployments, pioneers of NISQ at Los Alamos, and co-designers of Haskell and C#.

Our AI/ML researchers develop probabilistic methods for generative models and uncertainty quantification, and build agents that translate hardware specifications into formal models. Our open-source work includes Posteriors and Outlines, with contributions to DSPy.

8.10.2026

CN101 - A Digital Thermodynamic Computer for Generative AI

4.23.2026

From Specifications to Formal Models: Autoformalizing Memory Chips with DRAMBench

Manual translation of chip standards into verification artifacts is a major bottleneck in chip design. DRAMBench, developed by Normal Computing and Fraunhofer IESE, benchmarks AI systems that turn natural language DRAM specifications into timed Petri net models.

1.12.2026

A Complete Decomposition of Stochastic Differential Equations

5.30.2024

thermox: The First Thermodynamic Computing Simulator

Fast and exact Ornstein-Uhlenbeck processes with JAX

11.9.2023

A First Demonstration of Thermodynamic Matrix Inversion

Thermodynamic computing offers a natural approach for fast, energy-efficient computations. We report on the first-ever experiment towards thermodynamic artificial intelligence: solving matrix inversion problems by allowing a system of coupled electrical oscillators to thermally equilibrate with its environment.