The rapidly growing demand for data storage and processing, driven by artificial intelligence (AI) and other data-intensive applications, is posing a serious challenge for current computing devices based on the von Neumann architecture. For every calculation, data sets need to be shuffled sequentially between the processor, and multiple memory and storage units through bandwidth-limited and energy-inefficient interconnects, typically causing 40% power wastage. Phase-change materials (PCMs) show great promise to break this bottleneck by enabling nonvolatile memory devices that can optimize the complex memory hierarchy, and neuro-inspired computing devices that can unify computing with storage in memory cells.

In this tutorial session, four comprehensive talks are scheduled to highlight recent breakthroughs in the fundamental materials science and industrial development. The instructors are senior researchers working in the area of phase-change materials from Stanford University, University of Rome Tor Vergata, STMicroelectronics and IBM.

Embedded Phase Change Memory: From Material Engineering to Technology

Andrea Redaelli, STMicroelectronics

The tutorial starts by introducing the Phase Change Memory (PCM) concept. The basic PCM operation is then reviewed highlighting the main opportunities of this device in terms of performance and reliability comparing it with actual mainstream memories. The novel physics underneath PCM operation is also discussed with special attention to amorphous conduction, low field resistance drift, phase change description and aging. Concerning the possible applications, two possible paths of interest are identified for PCM. PCM for standalone applications required a strong device and process engineering efforts to provide high density and low-cost memories that ended up with the 3DXpoint memory. On the other hand, PCM for embedded applications required a strong material engineering effort, offering much high reliability suitable for more demanding System on Chip (SoC) markets. The challenges from material engineering side are thus discussed in details with main focus on the material interactions with integration process, especially looking to the fabrication thermal budget effect. Novel metrics are introduced to properly describe statistically the segregation effects in Ge-rich GST compounds and a path for process optimization is described. In general, the role of material segregation in the ePCM operation is identified as a key driver of ePCM performance and reliability.

Fundamental, Thermal, and Energy Limits of PCM and RRAM

Eric Pop, Stanford University

This tutorial will introduce the operation mechanism and fundamental limits of phase-change memory (PCM) and resistive random access memory (RRAM), for data storage and neuromorphic applications. The two memory types will be presented in context, with emphasis placed on their thermal and energy limitations, down to atomic scale dimensions. We will also discuss modern devices, challenges, nanoscale metrology, a few different material types (including superlattices), and simple models to understand their operation. The tutorial will give attendees an overview and background sufficient to allow them to think and actively contribute to the discussion on the ultimate (i.e. sub-5 nm) limits of these memory technologies.

Deep Learning Inference and Training Using Computational Phase-Change Memory

Manuel Le Gallo, IBM Research Europe

The computing systems that run today's AI algorithms are based on the von Neumann architecture which is inefficient at the task of shuttling huge amounts of data back and forth at high speeds. Thus, to build efficient cognitive computers, we need to transition to novel architectures where memory and processing are better collocated. In-memory computing is one such approach where the physical attributes and state dynamics of memory devices are exploited to perform certain computational tasks in place with very high areal and energy efficiency. In this tutorial, I will present our latest efforts in employing such a computational memory architecture for performing inference and training of deep neural networks. First, the phase-change memory technology we use as computational memory will be described. Next, the application of computational memory to neural network inference will be explained, and experimental results will be presented based on a state-of-the-art fully-integrated computational phase-change memory core. Finally, our latest efforts in employing such an architecture also for training neural networks will be elaborated.

Molecular Beam Epitaxial Growth and Characterization of Phase Change Materials

Fabrizio Arciprete, University of Rome Tor Vergata

Phase Change Materials (PCMs) are a class of inorganic compounds, generally based on chalcogenides alloys, exhibiting a variety of physical properties, which make them interesting for many technological applications, such as PCM based solid state memories. Very recently, PCMs have attracted further interest as emerging non-volatile memory technology for neuromorphic applications. In this tutorial we will discuss the material synthesis and fabrication of PCMs with a particular focus on epitaxial growth by Molecular Beam Epitaxy. The ability of growing high-quality epitaxial PCM alloys and heterostructures represents a key-point to tailor the material properties, such as a large programming window, or to investigate fundamental properties like, e.g., the electronic structure by photoemission spectroscopy.