Dec 4, 2024
8:00pm - 10:00pm
Hynes, Level 1, Hall A
Jakub Zdzieblowski1,Nicolas Barreau2,Pawel Zabierowski1
Warsaw University of Technology1,Université de Nantes2
Jakub Zdzieblowski1,Nicolas Barreau2,Pawel Zabierowski1
Warsaw University of Technology1,Université de Nantes2
In this work, we demonstrate the first realization of devices based on cadmium-indium sulfide (CdIn<sub>2</sub>S<sub>4</sub>), capable of acting as a binary memory cell as well as exhibiting basic synaptic functions, i.e., potentiation, depression including optical stimulation, and leaky integrate and fire (LIF) behavior. Neuromorphic behaviors are controlled by more than one physical process and rely only on native defects of the CdIn<sub>2</sub>S<sub>4</sub>, thus offering a wide range of technological possibilities.<br/>Cadmium-indium sulfide is a widely studied n-type chalcogenide semiconductor with a spinel structure, primarily researched for its applications in photocatalysis and photovoltaics. Past investigations have revealed its photosensitive nature, along with distinctive memory behaviors. The CdIn<sub>2</sub>S<sub>4</sub> possesses several native defect states, including Cd<sub>In</sub> and In<sub>Cd</sub> antisite defects induced by inversion of the spinel structure. Ab initio simulations show that Cd<sub>In</sub> and sulfur vacancy - another native defect in CdIn<sub>2</sub>S<sub>4</sub> - form metastable complex states.<br/>Polycrystalline CdIn<sub>2</sub>S<sub>4</sub> thin films were grown via physical vapor deposition on SLG/Mo substrates. By controlling the deposition temperature, we controlled the stoichiometry of the layer. Afterwards, the films were coupled with a polycrystalline ZnO:Al layer acting as a transparent electrode, with aluminum electrodes completing the device.<br/>The structure consists of a double rectifying junction: Mo/CdIn<sub>2</sub>S<sub>4</sub> and CdIn<sub>2</sub>S<sub>4</sub>/ZnO. The barrier heights of both junctions can be gradually modulated electrically and optically, which is a core process responsible for its neuromorphic behavior. Applying a train of learning electrical pulses leads to gradual barrier reduction on one of the junctions, reducing the resistance and thus practically realizing electrical synaptic potentiation. Train of negative pulses resets the first barrier and impacts the opposite one. Applying sufficiently high voltage results in abrupt switching between two opposite states, notably transitioning between the state when the barrier on ZnO is completely reduced and the opposite state when the barrier on Mo is reduced, practically realizing a binary switching operation. The same switching procedure can be realized with a sufficient number of training pulses with an amplitude just below the switching voltage. Those processes combined mimic the 'Leaky integrate and fire' synaptic behavior, as the process decay time is relatively fast. Our measurements indicate that the switching mechanism is related to the repining of the Fermi level by defect states, and its intricate details are currently under our investigation. As the In<sub>Cd</sub>&V<sub>S</sub> complex is metastable, persistent photoconductivity phenomena occur. It is then possible to convert the charge state of the metastable defects by illuminating the compound with a particular wavelength, practically altering its conductivity, which can be done even with the CdIn<sub>2</sub>S<sub>4</sub> bulk material alone. In devices, light pulses influence both junctions similarly to electrical pulses. As the absorption coefficient monotonously grows with the photon energy, the learning function can be controlled with incident light wavelength.<br/>All the switching behaviors, including switching voltage values and learning performance, strongly depend on stoichiometry and post-deposition treatment, which indicates the implicit influence of the CdIn<sub>2</sub>S<sub>4</sub> native defects. It opens promising possibilities for utilizing defect engineering, including metastable defects, to create neuromorphic devices. This work marks an initial exploration into utilizing CdIn<sub>2</sub>S<sub>4</sub> in neuromorphic computing, offering promising avenues for future advancement.