Insider Brief
- Extropic released a paper detailing its technological approach.
- The company is building technology that is underpinned by the development of parameterized stochastic analog circuits.
- The company’s “accelerators” promise vast improvements in both runtime and energy efficiency for algorithms.
Extropic, a relatively under-the-radar startup, released a paper that shed light on the company’s highly anticipated technology. Extropic is at the forefront of developing what it terms as parameterized stochastic analog circuits, according to a story on The Quantum Insider. These advanced, tunable electronic circuits are adept at handling a plethora of tasks by emulating the inherent randomness found in nature, making them particularly suited for complex AI tasks involving prediction and uncertainty.
The technology, which has attracted a lot of attention from both the AI and quantum computing communities, represents a potential shift away from traditional digital computing. Extropic’s “accelerators” are designed to significantly enhance runtime and energy efficiency for algorithms that require sampling from intricate energy landscapes. Drawing inspiration from Brownian motion, these accelerators rely on programmable randomness.
The timing of this type of innovation is important because the tech industry currently is grappling with an unprecedented demand for computing power. This surge is largely fueled by the rapid progression of AI, pushing the boundaries of existing technology to its limits. The lite paper highlights that looming challenge, referred to as “Moore’s Wall,” where the advance of computing efficiency faces a bottleneck due to the physical constraints of CMOS transistor technology.
As energy requirements for AI escalate, the team papers out some of the more drastic measures being considered like nuclear-powered data centers, Extropic offers a sustainable alternative inspired by the efficiency of biological systems. The company explores Energy-Based Models (EBMs) at the intersection of thermodynamics and probabilistic machine learning, promising a new paradigm for AI acceleration that transcends the limitations of digital logic.
The lite paper further details the technical capabilities of Extropic’s superconducting chips, which operate at low temperatures and exploit the Josephson effect to access non-Gaussian probability distributions. These chips underscore Extropic’s dedication to creating highly energy-efficient solutions suitable for high-value applications in government, banking, and private clouds.
Extropic’s vision extends beyond hardware. The company is developing a software layer that translates abstract EBM specifications into practical hardware controls, aiming to overcome the memory-bound challenges of deep learning. This innovative approach has the potential to unlock a new era of AI capabilities, offering speed and efficiency that vastly outpace current digital processors.
Extropic was founded by Guillaume Verdon and Trevor McCourt.
