Designing artificial synapses for brain-inspired computing

Researchers from the Adolphe Merkle Institute’s Smart Energy Materials group have investigated hybrid halide perovskites for innovative computing systems, focusing on non-volatile memory and brain-inspired applications. Their results highlight the significant potential of these materials in creating more efficient and reliable resistive switching memory devices and artificial synapses.

Brain-inspired, or neuromorphic, computing suggests a fundamentally different way to process information with energy efficiency and the ability to handle vast amounts of data. Conventional computing systems are considered inefficient because of their heightened power consumption due to the so-called “von Neumann bottleneck,” which refers to the constrained throughput between memory and processing units. On the other hand, the brain performs both “tasks” in a single unit, which consumes only about 20 watts, compared to the megawatts of power currently required by a digital supercomputer simulation of an equivalent artificial neural network. By mimicking the brain’s approach to information processing, Neuromorphic systems can achieve substantial energy savings and handle real-world applications more effectively.

The Smart Energy Materials group, led by Prof. Jovana Milic, in collaboration with Prof. Mahesh Kumar from the Indian Institute of Technology in Jodhpur (IITJ), studied layered Ruddlesden-Popper hybrid perovskites based on benzylammonium cations that included either bromide or iodide in their composition. They were interested in hybrid perovskite materials because of their unique soft yet crystalline structure and mixed electronic-ionic conduction that has potential in artificial synapses and numerous other applications, including for solar cells and LEDs. The focus for this project, though, was the switching behavior of the perovskites for use in neuromorphic computing. The memory devices demonstrated notable resistive switching characteristics associated with ion movements, essential for non-volatile memory applications. Unlike conventional 3D perovskites previously explored, these layered (2D) perovskites displayed higher stability during operation. Moreover, the study revealed that iodide-based devices exhibited gradual set and reset processes with reduced power consumption, while bromide-based devices showed more abrupt switching behavior but with superior on/off ratios. These variations provide options for optimizing device performance based on specific application requirements.

A significant discovery was the transformation from digital (a sharp set and reset process) in the bromide perovskite to analog resistive switching (gradual change in resistance states) in the iodide systems, operating like synapses in the brain. This unique transformation is critical for the advancements of modern computing, allowing for more gradual changes in resistance. As a result, iodide-based devices also showed promising synaptic behavior, including key characteristics such as potentiation (the strengthening of synaptic connections between neurons), depression (a reduction in synaptic strength that occurs in response to activity), and paired-pulse facilitation (a process by which neural activity is increased or made more efficient due to prior activation). These properties are vital for mimicking biological synapses.

The findings from this study, published in the journal Materials Advances, highlight the potential of layered hybrid halide perovskites for future applications in neuromorphic computing. The demonstrated ability to emulate synaptic functions with these materials opens new avenues for developing advanced computing systems that can operate more like the human brain.

“Our research showcases the versatility and potential of layered hybrid perovskites for memory and neuromorphic applications,” says Prof. Milic. “The transition from digital to analog resistive switching in these materials is particularly exciting for developing more efficient and biologically inspired computing technologies in the future."

Reference: M. Ganaie, M.; Bravetti, G.; Sahu, S.; Kumar, M.; V. Milić, J. Resistive Switching in Benzylammonium-Based Ruddlesden–Popper Layered Hybrid Perovskites for Non-Volatile Memory and Neuromorphic Computing. Materials Advances 2024, 5 (5), 1880–1886.