Inspired by Synapses, Researchers Improve Resistive Switching Memory
The team has claimed to solve a common problem in conventional resistive switching memories by introducing a chemical element: barium.
As compute-intensive applications like machine learning become the norm, researchers are reconsidering the efficacy of conventional computer architectures. Specifically, many are turning to in-memory computing powered by resistive RAM (ReRAM) as a potential solution to power-efficient computing.
This week, researchers from the University of Cambridge published a new paper describing how they tackled a longstanding challenge facing resistive switching memories. In this piece, we’ll look at issues with the traditional von Neumann architecture, the idea of resistive switching memory, and the research from Cambridge that builds on these concepts.
Key to the new research was the introduction of barium to thin hafnium oxide films.
Mounting von Neumann Challenges
Most devices created in the last 60 years work operate based on the von Neumann architecture. The von Neumann architecture describes a computer organization in which the computer utilizes a processing unit, a control unit, memory, external mass storage, and input-output mechanisms. The architecture is based on the principle that the data and the instructions to manipulate the data are stored in the same memory, which allows the computer to change its instructions based on the data it processes, enabling program control flow.
The basic structure of the von Neumann architecture. Image courtesy of GCSE
However, as we move into an era of modern computing characterized by artificial intelligence, machine learning, and massive data processing, the von Neumann architecture is presenting several challenges. The primary issue is known as the von Neumann bottleneck, which refers to throughput limitations caused by the standard bus interconnect between the major components of the system. As the speed of processors has increased, the interconnect can become a bottleneck, slowing down the system because it cannot transfer data to and from the memory as fast as the CPU can process it.
Moreover, the architecture's design, which separates memory and processing units, necessitates constant data transfer between the two. This data shuffling is both time-consuming and energy intensive.
Resistive Switching Memory
One potential solution to the von Neumann bottleneck is to spatially combine memory storage and computation in one unit—a process known as in-memory computing. To achieve this, one technology that offers a promising path forward is resistive switching.
Resistive switching is a phenomenon where the electrical resistance of a material changes as different voltages are applied. This principle forms the basis of resistive switching memory or resistive random access memory (ReRAM), a type of non-volatile memory technology. Unlike traditional memory devices that can only exist in two states (0 or 1), resistive switching memory can accommodate a continuous range of states, offering greater memory density and speed.
Resistive switching memories have resistance dependent on voltage. Image courtesy of Science Direct
While resistive switching memory offers significant advantages, it also presents challenges—one of which is the uniformity problem. This issue arises due to the inherent randomness at the atomic level in the materials used for resistive switchings, such as hafnium oxide. The hafnium and oxygen atoms in these materials are randomly mixed, resulting in a lack of structure that makes it difficult to use for memory applications.
The uniformity problem refers to the difficulty in consistently reproducing the same resistance state in each memory cell. This lack of uniformity can lead to variations in the resistance levels that can affect the reliability and performance of the memory device. Overcoming this uniformity problem is a significant challenge in developing resistive switching memory technologies.
Cambridge Researchers Solve the Uniformity Problem
This week, researchers from the University of Cambridge announced that they have found a potential solution to the uniformity challenge.
The introduction of barium helped create stability in the new composite. Image courtesy of Science Advances
As described in their recently published paper, the team overcame the uniformity issue by introducing barium into thin hafnium oxide films. The result of the barium integration was the emergence of unique, vertical, and highly-structured, barium-rich "bridges" that enable electron passage while the surrounding hafnium oxide remains unstructured. By manipulating the height of the energy barrier at the point where these bridges interact with the device contacts, the researchers achieved control over the electrical resistance of the composite material.
Beyond the uniformity of this new composite, an additional aspect of this material is its resemblance to a brain's synapse. By achieving in-memory processing with the new resistive switching composite, the team enabled information to be stored and processed simultaneously in the same place (like a brain’s synapses). The team believes that its finding could have major implications in the rapidly expanding AI and machine learning domains.