IBM Research Europe demonstrated prototype phase-change memory storing multiple AI vectors in superposition, increasing effective density with an accuracy tradeoff.

Researchers at IBM Research Europe published work titled “Superposed analog memory for vector storage in neural networks.” Semiconductor Engineering describes an analog phase-change memory architecture that stores multiple AI vectors in superposition.

The demonstrated benefit is an increase in effective storage density, with a tunable tradeoff in accuracy. The source does not quantify the density improvement or specify a commercially available memory capacity.

Tests on prototype phase-change memory (PCM) hardware reportedly achieved inference accuracy close to full-precision software. This is a research-hardware result; the supplied summary does not provide benchmark names, exact accuracy differences, endurance measurements, or production-deployment details.