PCM Applications—An Historical Review to Look to the Future

A historical review of the PCM technology from its first proof of concept up to the most recent achievements of the research will be provided. A parallel historical glance to the product development is also proposed as background to lead the discussion on the opportunities that PCM can bring in the actual scenario of memory eco-system. Against the dream of PCM as universal memory, this paper offers guidelines for PCM applications in the widespread scenario separating the Short-time memory (like DRAM, short both for latency and retention) and the Long-time memory (like NAND, long both for latency and retention), thus contributing to the heterogeneous memory eco-system. In particular, Author will reserve a more extensive discussion for PCM in the Storage Class Memory arena and in the neuromorphic and in-memory computing applications.<br/>The SCM arena represents the ideal realm for the PCM technology from the side of performances like speed and endurance, but a 3-D structure integration must be considered to reduce the cost of PCM devices. To this aim, a cross-point stacked solution combining a PCM with an Ovonic threshold switching as selector has been proposed. In a longer-term perspective, more compact cell design, resembling the 3D vertical NAND of today, should be considered. The primary benefit of such a design is the cell being defined in a single lithographic step, thus reducing the cost. A technological challenge to enable the foundation of this kind of Vertical 3D architectures lies in the ability to deposit conformal and composition-controlled chalcogenide films via ALD technique.<br/>PCMs are also strongly considered as electronic analogue of biological synapses able to implement the Spike-Time-Dependent-Plasticity (STDP) mimicking the synapse functionality. PCM can emulate the biological behavior of synapses and it offers multiple advantages such as scalability, reliability, endurance, multiple programming resistance levels giving them a suitable candidate for implementing large-scale synaptic systems.<br/>PCM technology can also play a relevant role in overcoming the performance limits of the von Neumann architecture, based on a strict separation between the computation unit and memory, where data and instructions for the processing unit use the same data bus, leading to the so-called von Neumann bottleneck.<br/>It will be shown as dense Non-Volatile Memory (NVM) crossbar arrays to few nanometer dimensions is a promising path to build computing systems able to overcome the von-Neumann bottleneck, performing the computation on the memory elements, as the human brain efficiently does, thus avoiding the redundant load-and-store operations. There are still many open challenges to make it a reality, like the improvement of the cell endurance with programming cycling, but definitively, PCM can contribute to mimic what biology figures out: doing the compute in memory. The astounding progress through the Moore’s scaling law made possible demonstrations of truly remarkable tasks of artificial intelligence (AI), but at the expense of an energy consumption that is orders of magnitude higher than the one of the human brain.