Quantum-optical spin glass boosts AI memory
A Stanford team created a quantum-optical spin glass from atoms and photons that functions as an associative memory.
A network of atoms and photons called a quantum-optical spin glass can store and recall memories more efficiently than conventional artificial intelligence systems. The advance, published in Science on September 3, 2026, demonstrates a form of associative memory with a capacity up to seven times greater than a traditional Hopfield network of the same size.
According to the Stanford University study, the system also displays short-term plasticity. This phenomenon resembles how synaptic connections in the brain change during learning. Senior author Benjamin Lev, the Stanford Fortitude Professor and professor of physics and applied physics, stated the team can now make neural networks at the atomic level.
"We can now make neural networks at the atomic level, and they adjust themselves in a way that is somewhat similar to how we believe our brains learn," Lev said.
The physics of frustrated spins
A spin glass is a type of frustrated magnet where atomic spins point in random, disordered directions. In 1982, physicist John Hopfield used the properties of frustrated spins in a mathematical model to show a network could store memory patterns. His work, which earned a share of the 2024 Nobel Prize in Physics, underpins modern AI systems. However, a Hopfield network fails when overloaded with memories, collapsing into a cluttered spin glass state that cannot recall information accurately.
The new research overcame this limitation by constructing a functional associative memory from a spin glass state itself. The key was using quantum-optical effects from atoms and photons.
Building a quantum-optical memory
The researchers used laser tweezers to create an array of atomic gases inside an optical cavity, a trap for light formed by two curved mirrors. These gases were Bose-Einstein condensates, clusters of 10,000 or more atoms behaving as single super atoms.
Photons were sent into the cavity, bouncing between the mirrors thousands of times. This created multiple connections among the atoms, analogous to synapses linking neurons. The photons drove the spin orientation of each super atom, settling them into low-energy arrangements that represent stored memories.
The team tested this network, containing up to 20 spins, as an associative memory. It successfully recalled full memories from partial or corrupted inputs. The quantum-optical spin glass's performance was compared directly to a classic Hopfield network.
| Network Type | Number of Spins | Relative Memory Capacity |
|---|---|---|
| Hopfield Network | Same as quantum-optical system | Baseline (1x) |
| Quantum-optical Spin Glass | Up to 20 | Up to 7x greater |
The researchers also observed the photons could induce plasticity, allowing the connections to move and change. This mimics the brain's ability to rewire its neural networks.
Potential and future directions
The system operates with atoms at extremely cold temperatures in a vacuum chamber. Lev emphasized the research is at an early stage, requiring more work to see if it can be scaled for practical applications. The team is now working to develop systems with more spins, including ones that can be quantum entangled, and to investigate other properties of their spin glass.
Lev noted the potential for less power-hungry AI hardware if the system can be improved and scaled. Since it stores more memories in a small network, AI technology might use fewer resources. Beyond applications, the work explores fundamental physics.
"This teaches us a little bit more about how physical systems can compute, not only with the classical laws of physics, but also with quantum laws," Lev said. The team's previous paper in Science in 2025 first described the experimental realization of this quantum-optical spin glass.
