As researchers seek to employ neuromorphic computing to overcome the limitations of conventional von Neumann architecture, mimicking the biological properties of neural systems has become increasingly critical. In tripartite synapses, astrocytes modulate synaptic activity and maintain homeostasis, thereby enabling more robust and adaptive neural systems. Inspired by biological tripartite synapses, we present an artificial tripartite synaptic transistor (synaptor) that integrates both synaptic and homeostatic functionalities within a single, CMOS-compatible device. The tripartite synaptor, with a split-gate silicon-oxide-nitride-oxide-silicon (SONOS) structure, uses two independently addressable gates: the primary gate controls synaptic weight via charge trapping/detrapping for weight updates, while the secondary gate modulates transmission current for homeostasis. The proposed tripartite synaptor demonstrates not only long-term retention and distinct potentiation and depression characteristics using the primary gate, but also dynamic conductance regulation using the secondary gate. The tripartite synaptor achieves higher accuracy than conventional bipartite synapse-based systems when applied to neural networks to classify grayscale handwritten digits and real-world RGB images. This work provides a scalable hardware platform for reliable neuromorphic computing with homeostasis.
Yun et al. (Fri,) studied this question.