Atomistic Origins of\nConductance Switching in an ε‑Cu<sub>0.9</sub>V<sub>2</sub>O<sub>5</sub> Neuromorphic Single Crystal Oscillator

Building artificial neurons and synapses is key to achieving\nthe\npromise of energy efficiency and acceleration envisioned for brain-inspired\ninformation processing. Emulating the spiking behavior of biological\nneurons in physical materials requires precise programming of conductance\nnonlinearities. Strong correlated solid-state compounds exhibit pronounced\nnonlinearities such as metal–insulator transitions arising\nfrom dynamic electron–electron and electron–lattice\ninteractions. However, a detailed understanding of atomic rearrangements\nand their implications for electronic structure remains obscure. In\nthis work, we unveil discontinuous conductance switching from an antiferromagnetic\ninsulator to a paramagnetic metal in ε-Cu<sub>0.9</sub>V<sub>2</sub>O<sub>5</sub>. Distinctively, fashioning nonlinear dynamical\noscillators from entire millimeter-sized crystals allows us to map\nthe structural transformations underpinning conductance switching\nat an atomistic scale using single-crystal X-ray diffraction. We observe\nsuperlattice ordering of Cu ions between [V<sub>4</sub>O<sub>10</sub>] layers at low temperatures, a direct result of interchain Cu-ion\nmigration and intrachain reorganization. The resulting charge and\nspin ordering along the vanadium oxide framework stabilizes an insulating\nstate. Using X-ray absorption and emission spectroscopies, assigned\nwith the aid of electronic structure calculations and measurements\nof partially and completely decuprated samples, we find that Cu 3<i>d</i> and V 3<i>d</i> orbitals are closely overlapped\nnear the Fermi level. The filling and overlap of these states, specifically\nthe narrowing/broadening of V 3<i>d</i><sub><i>xy</i></sub> states near the Fermi level, mediate conductance switching\nupon Cu-ion rearrangement. Understanding the mechanisms of conductance\nnonlinearities in terms of ion motion along specific trajectories\ncan enable the atomistic design of neuromorphic active elements through\nstrategies such as cointercalation and site-selective modification.