SEMINAR

Alkali Ion-Based Memristors for Neuromorphic Computing Applications

 

Hong-Sub Lee

Department of Advanced Materials Engineering for Information and Electronics, Kyung Hee University, Korea

E-mail : h.s.lee@khu.ac.kr

 

The matrix-vector multipliers combining analog memory (memristor) and crossbar array architecture (CAA) is the foremost scheme among the neuromorphic hardware currently considered. For implementing energy-efficient analog in-memory computing, four major memristors with analog memory characteristics, such as redox-based (ion-migration), ferroelectric tunnel junction, phase-change (crystal-amorphous), and magnetic tunnel junction, are competing. Among them, a redox-based memristor is particularly well-suited for neuromorphic and analog computation due to its fast-switching speed and multi-bit capability (large on/off ratio). However, the stochastic nature of defect-induced switching coupled with limited control over intrinsic materials defects have been identified as the primary factors undermining the reliability of memristors in scaled CAAs.

In this talk, I introduce a Na-doped TiO₂ memristor grown by atomic layer deposition (ALD) that uses high mobility sodium cations instead of oxygen anions (oxygen vacancies) as the main agent for resistive switching. Unlike conventional memristors based on oxygen anions, the high mobility of Na ions can be expected to produce memristive behavior regardless of the underlying oxygen vacancy concentration and even under a self-rectifying characteristics.  The resulting Na-doped TiO₂ memristors show electroforming-free and self-rectifying resistive switching behavior that are ideally suited for selectorless CAAs. Effective addressing of selectorless nodes is demonstrated via electrical measurement of individual memristors in a 6×6 crossbar using a read current of less than 1 μA with negligible sneak current at or below the noise level of ~100 pA. Finally, the long-term potentiation and depression synaptic behavior from these Na-doped TiO₂ memristors achieves greater than 99.1% accuracy for image recognition tasks using a convolutional neural network based on the selectorless of crossbar arrays.