This Tabletop Computer Cracks Hard Math Problems Using Sound Waves Instead of Lasers

Researchers have built a 2,048-spin Ising machine using acoustic wave packets in solid-state delay lines — 10,000× more thermally stable than optical alternatives and cheap enough to run on a desk.

Moving optimization computing off the laboratory bench and onto a desktop, researchers have demonstrated a new kind of Ising machine that replaces fragile laser systems with propagating sound waves in solid-state crystals.

The machine, described in a paper published July 6 on arXiv, uses two serially connected bulk acoustic wave delay lines operating at 20.5 MHz to support 2,048 spins with all-to-all connectivity. It finds approximate MAX-CUT solutions in 341 milliseconds and also solves number partitioning and Sudoku puzzles — outperforming the simulated bifurcation algorithm, the leading software approach, on the more complex problems.

What makes this different from existing Ising machines is the physics. Optical coherent Ising machines use time-multiplexed laser pulses and have demonstrated impressive connectivity and spin counts — but they suffer from large physical footprints, high power consumption, poor thermal stability, and steep costs. The acoustic approach flips this entirely: solid-state delay lines at microwave frequencies are thermally stable, robust, low-power, and affordable enough to run on a tabletop.

The thermal stability improvement is dramatic — four orders of magnitude better than state-of-the-art coherent Ising machines. That is the difference between a system that needs laboratory-grade temperature control and one that works on a desk.

Each delay line provides 707 microseconds of acoustic propagation time, and the system achieves 15-bit coupling resolution between spins. The authors note the architecture could scale to sub-millisecond solution times by moving to higher frequency delay lines.

Ising machines are special-purpose computers designed for combinatorial optimization — the kind of problems that scale exponentially on conventional processors. Applications range from logistics and circuit design to machine learning and drug discovery. Making them smaller, cheaper, and more stable is a prerequisite for moving them out of research labs and into real-world use.

Sources: arXiv