Four-Year Center Goals
Thrust 1 will fuse an inverse design approach based on an analytical framework developed for Mott oxides with a forward design approach linking atomistic and electronic structure to coherent and incoherent electronic transport. Using redox transitions in molecular films of TM complexes of N-heterocyclic ligands and Li-ion insertion in βʹ-CuxV2O5 and LixCoO2 as model systems, leveraging carefully constructed existing libraries with site-selective modification, and with the help of Bayesian machine learning (ML) and multimodal operando measurements, we will connect target neuronal behavior to the necessary molecular/material properties.
To untangle transformation characteristics and achieve REMIND’s vision of massively reconfigurable, ultra-fast, and low-energy switching, Thrust 2 will investigate the mechanistic origins of MITs and design of instabilities in intercalation compounds, where electrochemical ion intercalation precisely modulates electron/hole density and drives a series of displacive or minimally distortive transitions. These can be exploited to reconfigure neuromorphic elements in different regions of the phase diagram with distinct transformation characteristics between internal states.
Thrust 3 builds on a novel mode of neuronal emulation and dynamic reconfiguration based on deterministic, repeatable switching between distinct conductance states demonstrated by Williams for thin films of redox-active TM coordination complexes with N-heterocyclic ligands. This recent work highlights the importance of intermolecular electronic coupling between ligands, suggesting further exploration of the relevance of Marcus electron transfer to carrier transport.
Thrust X will be a cross-cutting edifice, which will provide mechanistic understanding of conductance switching and reconfigurability across our exemplary material classes. TX will develop electrothermal, spectroscopic, and imaging toolsets to interrogate dynamic evolution of local electronic and atomistic structures in response to external stimuli; link atomistic and mesoscopic behavior to predict the response of electrically-stimulated microscopic device elements; and develop ML frameworks (including graph-based representations of material/molecular configurations) to work backward from neuronal and synaptic function to predict and design molecules and crystal structures.
Four-Year Center Goals
- Identify fundamental neuromorphic conductance switching mechanisms that enable the desired 4 to 5 orders of magnitude improvements (see below) in speed and energy efficiency of neuromorphic analog computing over scaled digital CMOS, underpinned by the design of materials emulating the neuronal nonlinear dynamical responses such as thresholding, integration, and amplification.
- Develop experimental tools to interrogate form and function, bridging length, time, and energy scales, and link with atomistic and mesoscopic modeling approaches to predict the cumulative nonlinear electrically-triggered response of (a) microscopic elements and (b) nontrivial ensembles of such elements interfaced within physical networks.
- Create inverse design rules that map neuronal/synaptic function to material/interfacial properties.
- Tailor conductance switching and reconfigurability across two material classes exhibiting low-entropy transformations: (a) intercalation materials and (b) coordination complexes, using materials design, host-guest chemistry, site-selective modification, and lattice strain to establish rules for decoupling conductance differential, threshold voltage, sharpness of the transition, and hysteresis.
- Demonstrate in-situ device reconfiguration by tuning material properties, thereby enabling a small-scale reconfigurable decision network and explore ultimate limits of speed and energy consumption.