Drift-free memory material for wide-temperature neuromorphic computing

In the era of artificial intelligence (AI), the demand for data storage and processing continues to grow at an astonishing rate. The von Neumann computing paradigm, which separates memory and computing units, faces a critical bottleneck in data transfer, especially when handling massive datasets. Neuromorphic computing [1] offers a promising alternative by integrating storage and computation into the same unit, thereby significantly improving energy efficiency. Among the emerging materials for this technology, chalcogenide phase-change materials (PCMs) are particularly promising [2,3]. These materials exploit the pronounced electrical contrast between their amorphous and crystalline phases to encode information. Switching between the two solid-state phases is achieved via short electrical pulses: crystallization (SET operation) and melt-quench amorphization (RESET operation). By precisely controlling the ratio of amorphous to crystalline domains, multiple intermediate resistance states can be realized within a single PCM cell [2]. This multi-level programmability is crucial for neuromorphic computing, as it reduces the number of cells required to emulate a single synapse. Furthermore, a PCM-based synaptic array can directly perform matrix-vector multiplication (MVM) by leveraging Ohm’s and Kirchhoff’s laws [4].