Chip startup Normal Computing has announced the successful tape-out of the world’s first thermodynamic semiconductor, CN101.
Designed to support AI and HPC workloads, Normal Computing describes the CN101 as a “physics-based ASIC” that harnesses the “intrinsic dynamics of physical systems... achieving up to 1,000x energy consumption efficiency.”
According to the company, by using natural dynamics such as fluctuations, dissipation, and stochasticity – otherwise known as a lack of predictability - the chips are able to compute far more efficiently than traditional semiconductors.
CN101 specifically targets two tasks necessary for supporting AI workloads: solving linear algebra and matrix operations, and stochastic sampling with lattice random walk, and represents the first step on Normal Computing’s roadmap towards commercializing thermodynamic computing at scale and enabling significantly more AI performance per watt, rack, and dollar, the company says.
“In recent months, we have seen that AI capabilities are approaching a flattening curve with today’s energy budgets and architecture, even as we plan to scale training runs another 10,000x in the next five years,” Faris Sbahi, CEO of Normal Computing. “Thermodynamic computing has the potential to define the next decade’s scaling laws by exploiting the physical realization of AI algorithms, including post-autoregressive architectures. Achieving first silicon success is a historic moment for this emerging paradigm – executed by a radically small engineering team.”
Founded in 2022 by former Google Brain, Google X, and Palantir employees, Normal Computing says that it is now focused on the development of its CN201 and CN301 chips, which support high-resolution diffusion models and advanced video diffusion models, respectively.