Crossbar AI: Practical Neuromorphic Memory and VLSI

Crossbar compute‑in‑memory performs vector–matrix multiplication by exploiting Ohm’s and Kirchhoff’s laws in dense arrays of resistive or magnetic elements, effectively collapsing many digital operations into a single physical step. Recent demonstrations using ReRAM, MRAM, PCM, Ferroelectric, and related multi‑state devices show attractive energy–delay and area benefits, but they also reveal a strong dependence on the underlying conduction, switching, and relaxation processes. Variability, non‑linearity, drift over time, and cycle‑to‑cycle fluctuations can often be traced back to filament dynamics, ion migration, domain evolution, or spin‑related phenomena, and these effects directly influence how “synapses” behave in large arrays.​As the field moves from isolated devices to full mixed‑signal Neuromorphic systems, there is a clear need for a physics‑aware body of work that connects materials, compact models, circuit techniques, and learning algorithms under realistic workloads and constraints.This Research Topic aims to bring the underlying physics of crossbar and Neuromorphic memory to the foreground, while keeping a clear line of sight to practical AI hardware. The focus is on how charge, spin, ionic transport, and phase transitions in Non‑Volatile Memories give rise to synaptic weights in large arrays, and how these physical mechanisms set the limits on multi‑bit precision, stability, and learning dynamics. Questions of interest include how conductance modulation, stochasticity, noise, retention, and endurance emerge from material physics and then propagate through circuits and algorithms to shape system‑level energy, latency, and accuracy.​We invite contributions that spell out these physics‑driven trade‑offs between device structures, array architectures, and mixed‑signal interfaces, and that propose models and calibration schemes linking material behavior directly to AI performance. By bringing device physicists, circuit designers, and Neuromorphic system researchers into one collection, we hope to establish experimentally grounded principles and metrics that reduce the gap between small‑scale prototypes and robust, deployable crossbar‑based AI accelerators.This Research Topic invites manuscripts that address the physics of Neuromorphic and crossbar memories and show how that physics shapes the design and behavior of AI hardware. We welcome work on: a) Device physics of resistive, magnetic, phase‑change, Ferroelectric, and related synaptic elements b) Selector and interconnect physics c) Array‑level phenomena such as sneak paths, IR‑drop, and crosstalk d) Stochastic switching, noise, and variability e) Reliability test such data retention and endurance mechanisms f) Mixed‑signal MAC circuits and calibration guided by physical models g) CMOS/BEOL and 3D integrationh) Physics‑based compact models for SNN and DNN accelerators, including benchmarking approaches that expose physical limits under realistic tasks. Topic editors: Dr. Shounak De, Dr. (Prof.) Siddheswar Maikap, Dr. Sourav Roy Source: https://www.frontiersin.org/research-topics/78196/crossbar-ai-practical-neuromorphic-memory-and-vlsi