rewrite this title Forget Traditional Chips: Thermodynamic Chips Will Revolutionize Energy Efficiency

Extropic aims to revolutionize Artificial Intelligence (AI) with its chips based on thermodynamic principles. The company’s TSU processors model uncertainty, thereby reducing energy consumption.

A new technology startup aiming to revolutionize AI and scientific research is preparing to challenge traditional chip manufacturers. Extropic has developed an unusual chip that uses probabilistic bits (p-bits) and has not only produced its first working hardware but has also proven that more advanced systems can be used for practical tasks.

⚡ Thousands of Times Greater Energy Efficiency

Adopting a completely different approach from traditional CPU and GPU architectures, Extropic’s chips promise to be thousands of times more energy-efficient when scaled. This means solving one of the biggest challenges facing today’s ever-increasing AI investments. While the claim needs to be proven, the energy consumption of AI data centers is becoming a progressively serious issue.

The company calls these new processors Thermodynamic Sampling Units (TSU). TSUs utilize thermodynamic electron fluctuations via silicon components and shape them to simulate the probabilities of complex systems, such as weather forecasting, or AI models that can generate images, text, and video.

Extropic has shared its first working chips with a limited number of partners. These include advanced AI laboratories, startups working on weather modeling, and some government representatives.

One of those testing the new hardware is Johan Mathe, CEO of the startup Atmo. Atmo provides much higher-resolution weather forecasts than normal for customers like the Ministry of Defense. Mathe emphasizes that Extropic’s chips will make it possible to calculate different weather probabilities much more efficiently. The company also offers a software called TRHML. This software allows the behavior of Extropic chips to be simulated on a GPU. Mathe reports observing that the “p-bits behave as expected” using both the software and the actual chip.

Extropic‘s hardware named XTR-0 consists of an FPGA chip that can be reconfigured for different tasks, and two probabilistic chips called X-0, which contain a few qubits. Instead of traditional bits with values of 1 or 0, p-bits can model uncertainty. Although limited in scale, this chip demonstrates the potential of the company’s approach.

💡 What is the Difference from Traditional Chips?

Traditional CPUs and GPUs base their computations on binary logic. Each bit can only be in one of two states: 1 (on) or 0 (off). This structure is the foundation of digital computers, and operations proceed through logical accuracy. However, because these systems reduce every operation to a definite outcome, they have a natural limitation in modeling probabilistic or uncertainty-laden processes like Artificial Intelligence.

The TSU developed by Extropic is based on a completely different physical and mathematical foundation. The p-bit (probabilistic bit) in these units is not a fixed 1 or 0 like a classical bit. These units oscillate probabilistically between the two states at any moment. That is, a p-bit might be “70% probability of 1, 30% probability of 0” at a specific time. This is a random yet controllable behavior, much like thermodynamic fluctuations in nature. TSUs use these random movements to generate statistical models. Thus, AI models, weather forecasts, or molecular simulations can model uncertainty-based systems much more naturally and energy-efficiently than classical processors.

In a paper published on arXiv, Extropic announced that a larger chip (the Z-1) containing thousands of p-bits will be ready next year, and it could be used to create a new type of diffusion model. The Z-1, which is rumored to have 250,000 p-bits, could be used to create AI models for generating images and videos and guiding the movements of robots.

You Might Also Like;

Copy URL
URL Copied