Bi-Modal Synapse Based on a Short-Channel Ferroelectric van der Waals Heterostructure

Authors:

Abstract
The growing demand for energy-efficient and -adaptive computing drives research into neuromorphic architectures. Van der Waals (vdW) ferroelectric field-effect transistors offer nonvolatile polarization control and a highly tunable semiconductor channel, enabling multilevel states and making them promising for brain-inspired electronics. Here, we present a reconfigurable bimodal ferroelectric synapse based on the CuInP2S6/hBN/WSe2 vdW heterostructure, extending beyond conventional single-modal synaptic devices by introducing added functionality. Transport measurements and piezoresponse force microscopy reveal precise electrical control over the ferroelectric domain landscape, enabling continuous tuning of WSe2 channel conductance and its threshold voltage. Crucially, the ambipolar nature of WSe2 allows for real-time switching between excitatory and inhibitory synaptic behaviors, mimicking multimodal neurotransmission observed in the human brain. Moreover, the bimodal synapse is demonstrated at channel lengths down to 50 nm, venturing into previously uncharted territory for ferroelectric vdW synapses. Neural network simulations incorporating our device show excellent learning performance for both synaptic modes, highlighting its potential for next-generation neuromorphic computing. This work expands the functional and scaling capabilities of vdW ferroelectric technology, highlighting its potential for next-generation artificial intelligence electronics.
Funding Information
  • Agence Nationale de la Recherche (ANR 10 IDEX 0002, ANR 20 SFRI 0012, ANR-21-CE09-0029, ANR-22-PEXD-0006, ANR-23-CE09-0007, ANR-24-EXSP-0008, QMat-ANR-18-EUR-0016)
  • Ministry of Education, Culture, Sports, Science and Technology (JPMJCR24A5)
  • European Commission (847471)
  • Ministry of Education, Culture, Sports, Science and Technology (21H05233, 23H02052)
  • Institut Universitaire de France
  • National Science Foundation (DMR-2212965)

This publication has 47 references indexed in Scilit: