Normal Computing tapes-out first thermodynamic chip (2025) (www.datacenterdynamics.com)

Normal Computing has announced the successful tape-out of the CN101, the world's first thermodynamic semiconductor designed to enhance AI and high-performance computing (HPC) workloads. This revolutionary chip, dubbed a “physics-based ASIC,” boasts up to 1,000 times greater energy efficiency by leveraging the intrinsic dynamics of physical systems, such as fluctuations and stochasticity, to improve computational efficiency beyond that of traditional semiconductors. By focusing on essential tasks for AI, like linear algebra matrix operations and stochastic sampling, the CN101 marks a significant step towards commercializing thermodynamic computing, promising more AI performance per watt, rack, and dollar. This innovation arrives at a critical time when existing AI capabilities are nearing a plateau due to current energy constraints and architectural limitations. According to Faris Sbahi, CEO of Normal Computing, thermodynamic computing could redefine the scaling laws for AI over the next decade by implementing algorithms that exploit physical phenomena. The company's success in delivering this cutting-edge technology with a small engineering team underscores the potential for thermodynamic computing to tackle the growing demands of AI training and model complexity, setting the stage for further advancements with the upcoming CN201 and CN301 chips designed for sophisticated diffusion models.