Exploring Phase Change Memory Programing Towards Energy-Efficient Synaptic Devices

Phase change memory (PCM) is an established storage-class memory technology with strong potential for use as dedicated hardware in neuromorphic computing. PCM relies on chalcogenide materials, such as Ge2Sb2Te5 (GST), which can reversibly switch between polycrystalline and amorphous phases. These two phases exhibit a high resistivity ratio (>10³), allowing different resistance states to encode memory cell data. State transitions are induced by electrical pulses that cause Joule heating: crystallization results in a low resistivity state (set), and amorphization leads to a high resistivity state (reset).

Our research addresses two critical bottlenecks in this technology: 1) the crystallization (set) speed, crucial for fast programming and low energy consumption, and 2) achieving multiple resistance states through amorphization (reset) for neuromorphic applications.

Our first contribution leverages a nanosecond transient measurement methodology to provide deeper insights into the solidification processes within nanoscale device configuration. During the set process, transient resistance measurements probe the phase, while power measurements indicate temperature, uncovering critical details of phase change dynamics. Our approach reveals essential aspects of PCM crystallization, enabling faster programming and reduced energy consumption.

Our second contribution involves applying sub-nanosecond pulses, shorter than the thermal time constant, to achieve near-linear gradual conductance change during reset, which is crucial for synaptic programming. We achieved sub-pJ training pulse energy and ~20 nW inference power, outperforming previous synaptic PCM studies in energy efficiency. This research holds significant potential to advance PCM technology as an energy-efficient synaptic device for neuromorphic hardware.

M.Sc. student under the supervision of Prof. Eilam Yalon.