2D Heterostructure for High‐Order Spatiotemporal Information Processing. (3rd December 2021)

• Record Type: Journal Article
• Authors: Zhai, Yongbiao; Xie, Peng; Feng, Zihao; Du, Chunyu; Han, Su‐Ting; Zhou, Ye
• Is Part Of: Advanced functional materials. Volume 32:Number 7(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-03
• Subjects: Bienenstock–Cooper–Munro theory -- high‐order spatiotemporal -- memtransistors -- neuromorphic computing -- triplet‐triplet‐spike timing dependent plasticity; Materials -- Periodicals; Chemical vapor deposition -- Periodicals; 620.11
• DOI: 10.1002/adfm.202108440
• Languages: English
• ISSNs: 1616-301X

Abstract

Abstract: The Bienenstock, Cooper, and Munro (BCM) theory of synaptic plasticity is regarded as the most precise model of the synapse, and is more compatible with neuromorphic computing. However, the development in BCM synaptic modification is rather limited since the memristive devices used to emulate the BCM lack tunable forgetting rate. Compared with memristors, memtransistors provide another gate‐tunable freedom degree, which will help to modulate the forgetting rate. In this work, the authors demonstrate a perfect BCM learning rule based on the 2D heterostructure memtransistor through using triplet‐spike timing dependent plasticity model. Two critical characteristics of the BCM rule, sliding frequency threshold and enhanced depression effect, are perfectly presented due to their spontaneous/gate‐assistant forgetting effect. The experimental results are extremely consistent with the BCM learning rule and suggest the potential application of 2D memtransistors in high‐order spatiotemporal recognition. Abstract : A 2D heterostructure‐based memtransistor is designed to emulate the Bienenstock–Cooper–Munro (BCM) theory, since this structure not only induces spontaneous forgetting process but also offers another gate‐tunable forgetting effect. BCM learning rule is perfectly demonstrated on this memtransistor using triplet‐STDP. Furthermore, high‐order spatiotemporal recognition is achieved in a feedforward neuron network based on the memtransistor.