Martin Salinga (left) and I collaborated on the research.
Phase change memory (PCM) is an emerging non-volatile memory technology that could play a key role in future computing systems. In collaboration with RWTH Aachen University, my team and I at IBM Research-Zurich went in the opposite direction of the mainstream PCM research by using only one single chemical element—antimony (Sb)—instead of the typical material cocktail. This approach promises not only to make it far easier to miniaturize PCM devices, but also to increase the data density of memory chips and the power efficiency per operation.
Our work is being featured on the cover of the August issue of the peer-reviewed journal Nature Materials.
PCM at a small scale
PCM works by reversibly and rapidly switching a phase change material from a crystalline state with high electrical conductivity to an amorphous state with low conductivity. An electrical pulse thermally induces the transition between the states. Naturally, smaller amounts of material need less heat, and therefore less electrical energy. In the past, research on phase change materials mainly focused on adjusting their physical properties by adding additional chemical elements into the alloys. However, this resulted in very complex compositions that were difficult to create and maintain in memory devices of only a few nanometers in size. At such a small scale, the local variations in the composition can limit the cyclability or lifespan of a device as the distribution of the relevant atoms in the cells can change as a reaction to operation conditions in strong electric fields and high temperatures.
Off the beaten track
Cover of Nature Materials, August 2018
All of these reasons prompted me and my fellow scientists from IBM Research-Zurich and RWTH Aachen University take a different approach using antimony as a valid alternative to conventional approaches.
Antimony is semi-metallic in its crystalline phase and semiconducting as an amorphous thin film and shows a large contrast in resistivity between these two states. It also crystallizes easily and quickly, making it ideal for a PCM in a highly-confined structure – a structure which usually slows down the crystallization kinetics. Instead of fine-tuning new phase change material compositions, we will focus on the effects of material interfaces and confinement with PCM using only one single element.
The challenges ahead
A key challenge going forward will be to adapt to the amorphous state of antimony, which only remains stable for thousands of seconds at room temperature. The indications are that the retention time can be increased, for instance, by further reducing its film thickness, confining antimony in all three dimensions, and designing better confinement materials. My prediction is that the first applications to benefit from a ‘monatomic PCM’ could be in the areas of memory-type storage class memory and in-memory computing, which are considered central to future computing systems for artificial intelligence. These are applications where the retention time is not so critical as in the case of conventional storage applications.
IBM researchers first used the term storage-class memory in 2008 to describe a group of new memory technologies vying to fill the cost-performance gap between DRAM and HDDs. Storage class memory could accelerate several data-centric workloads such as database analysis. PCM could also serve as elements of a computational memory unit where certain computational tasks are performed in place within the memory, unlike conventional computing systems where the memory and the processing units are separated. PCM is also widely explored as elements of neuromorphic hardware. Recently, we at IBM Research showed several promising demonstrations of PCM-based computing for artificial intelligence and machine learning.
Today, as reported in Nature Machine Intelligence, my colleagues and I have demonstrated a novel approach to deep learning that incorporates biologically-inspired neural dynamics and enables in-memory acceleration, bringing it closer to the way in which the human brain works. The results point towards the broad adoption of more biologically-realistic deep learning for applications in […]
Scientists around the world are inspired by the brain and strive to mimic its abilities in the development of technology. Our research team at IBM Research Europe in Zurich shares this fascination and took inspiration from the cerebral attributes of neuronal circuits like hyperdimensionality to create a novel in-memory hyperdimensional computing system. The most […]
Can analog AI hardware support deep learning inference without compromising accuracy? Our research team at IBM Research Europe in Zurich thought so when we started developing a groundbreaking technique that achieves both energy efficiency and high accuracy on deep neural network computations using phase-change memory devices. We believe this could be a way forward in […]