Summary: Researchers from Politecnico di Milano, Empa and ETH Zurich have created a more powerful and energy-efficient memristor inspired by the structure and energy efficiency of the human brain. Built from nanocrystals of halide perovskite, the device combines memory and computation in a single element and demonstrates higher-order dynamics useful for neuromorphic computing. While promising for parallel processing of large datasets, the technology still faces integration challenges with conventional silicon-based chips.
Source: Politecnico di Milano
Researchers aiming to reproduce the brain’s efficiency and compact computation have developed a next-generation memristor that is simpler to fabricate and delivers improved performance compared with earlier devices. The work, carried out by teams at Politecnico di Milano in collaboration with Empa and ETH Zurich, appears in Science Advances.
Traditional computers separate memory and processing, requiring constant data transfer between the two. This von Neumann architecture becomes a bottleneck when handling very large datasets, increasing latency and energy use. The human brain, by contrast, stores and processes information at synapses, enabling highly efficient learning and sensory processing while consuming far less energy than conventional computing systems.
Inspired by that biological model, the research team developed memristive devices—memristors—that integrate data storage and computation in a single element. Their devices are based on halide perovskite nanocrystals, a semiconductor material well known in the field of solar cells. These materials support both ionic and electronic conduction, allowing richer internal dynamics than conventional memristors built from single-type carriers.
The team measured device behavior and used those measurements to simulate a complex learning task resembling processes in the brain’s visual cortex. The simulation tested the system’s ability to infer the orientation of a light bar from signals equivalent to retinal input, demonstrating how the device physics can support sensory learning tasks.
According to Rohit John, a postdoctoral researcher at ETH Zurich and Empa, the dual ionic-electronic conductivity of halide perovskites enables more complex calculations that parallel certain brain processes. These devices exhibit second-order dynamics—behavior that depends on both the state of the system and its history—allowing them to implement richer learning rules than first-order memristors.
The devices described function as memdiodes—memristive diodes—with dynamics that enable biologically inspired plasticity rules such as Bienenstock-Cooper-Munro (BCM) and triplet spike timing–dependent plasticity (STDP). The researchers identify ion migration, back diffusion, and tunable Schottky barriers as key physical mechanisms that can be combined to design higher-order memristive elements. These intrinsic device properties make it possible to reproduce complex features such as binocular orientation selectivity within neural networks without adding elaborate external circuitry.
Despite the promising results, practical adoption faces hurdles. Halide perovskites are not currently compatible with standard silicon manufacturing temperatures (typically 400–500 °C), which complicates direct integration with existing CMOS processes. The team notes that alternative materials with similar ionic-electronic properties may offer better compatibility with silicon-based fabrication and that the memristor concepts developed can be tested with other material systems.
Daniele Ielmini, professor at Politecnico di Milano, emphasizes that the goal is not to replace conventional computer architectures, but to create complementary, brain-inspired hardware that can perform specific tasks—such as massively parallel data processing—more efficiently. Potential applications range from agricultural sensing systems to space exploration, wherever large-scale, energy-efficient pattern recognition and learning are required.
About this neurotech research news
Author: Emanuele Sanzone
Source: Politecnico di Milano
Contact: Emanuele Sanzone – Politecnico di Milano
Image: The image is in the public domain
Original Research: Open access.
“Ionic-electronic halide perovskite memdiodes enabling neuromorphic computing with a second-order complexity” by Rohit John et al., Science Advances
Abstract
Ionic-electronic halide perovskite memdiodes enabling neuromorphic computing with a second-order complexity
Rising computational demands are exposing limits of serial von Neumann architectures built from zeroth-order digital circuits. Brain-inspired approaches using memristive devices offer large parallelism, low energy consumption, and resilience to noise, making them attractive alternatives. However, most prior demonstrations have replicated only simple, lower-order biological behaviors with first-order devices.
This work introduces memdiodes based on halide perovskites that show second-order dynamics, enabling the implementation of learning rules that capture both timing- and rate-based plasticity. A triplet spike timing–dependent plasticity scheme driven by ion migration, back diffusion, and adjustable Schottky barriers provides general design principles for higher-order memristors. Such advanced device physics allows neural networks to perform complex functions—like binocular orientation selectivity—directly from the intrinsic behavior of the devices, reducing the need for elaborate external circuitry.