Programmable memristive circuits for reconfigurable collective pseudomagnetic ordering, transport, and learning

Abstract

Programmable hardware typically uses fixed architectures, whereas complex physical systems obtain diverse behavior by connecting reusable local elements through different interaction rules. Memristive devices combine writable states with nonlinear dynamics, but translating these local properties into reconfigurable collective order and symmetry-governed transport remains challenging. Here we develop modular circuits configured through programmable connections between identical memristive units. Experiments establish binary conductance states as Ising-type pseudospins and show that signed feedback between units selects uniform or staggered pseudospin order. Circuit-model simulations then examine controllable domain-wall motion and pattern recovery. A symmetry encoder combines measured conductance patterns with sublattice labels and a selected symmetry rule to configure a fixed transport layer, implementing circuit analogs of ferromagnetic, antiferromagnetic, and altermagnetic transport symmetries. Although the antiferromagnetic and altermagnetic encodings share a compensated pattern, their distinct voltage spreading illustrates how programmed circuit symmetry affects transport. Device-calibrated simulations show that memristive dynamics generates history-dependent features for reservoir computing. By separating local memory from programmable interactions, symmetry encoding and task-specific readout, this architecture provides a modular route from device dynamics to collective order, symmetry-dependent transport, and learning.