Chinese Physics B, Volume. 29, Issue 9, (2020)
Recent advances, perspectives, and challenges in ferroelectric synapses
Fig. 1. (a) Typical multiple polarization (
Fig. 2. Schematic representation of the energy potential profile in an FTJ for polarization pointing to the electrode one (Metal1) (a) and for polarization pointing to the electrode two (Metal2) (b). Reproduced from Ref. [
Fig. 3. (a) Evolution of the junction resistance and ferroelectric domain configuration in a 180 nm wide (Ca,Ce)MnO3/BiFeO3/Pt/Co FTJ. (b) Symbols are experimental resistance as a function of percentage of down domains extracted from the PFM phase images and the black curve is a simulation in a parallel resistance model shown in inset. Reproduced from Ref. [
Fig. 4. The FTJ could be treated as a two-terminal synapse where the top and bottom electrodes transmit signals from pre and post neurons. (a) Sketch of pre- and post-neurons connected by a synapse. (b) Sketch of the (Ca,Ce)MnO3/BiFeO3/Pt/Co FTJ. YAO stands for YAlO3. (c) Single-pulse hysteresis loop of the resistance versus voltage amplitude, there are voltage thresholds ±
Fig. 5. (a) STDP measurement in La0.7Sr0.3MnO3/BaTiO3/La0.7Sr0.3MnO3 MTJ with the parallel (p) or antiparallel (AP) states of magnetization alignment in electrodes. (b) Schematic illustration of the magnetoelectrically coupled memristor and the corresponding synapses. Reproduced from Ref. [
Fig. 6. (a) Ferroelectricity in ultrathin thin films of PVDF. Topography (top), PFM phase (middle), and amplitude (bottom) images, after switching the polarisation upward in the central 5 × 5 μm2 area of a 2 L-thick (left) and 1 L-thick (right) PVDF films. Scale bar, 1 mm. (b) Three-dimensional schematic diagram of PVDF FTJs. (c)
Fig. 7. (a) Evolution of the junction resistance as a function of voltage amplitude in an Au/P(VDF-TrFE)/Nb:SrTiO3 FTJ. (c), (d) Programmable conductance, corresponding to LTP/D of synaptic weight, is achieved by modulating either amplitude (b) or number (c) of applied voltage pulses. (d) Measurements of LTP/D. (e) Measurements of Hebbian and anti-Hebbian STDP learning rules. Reproduced from Ref. [
Fig. 8. (a) Sketch of Pt/BaTiO3/Nb:SrTiO3 FTJs. (b) Linear and symmetric conductance update with 200 states and 500 reproducible cycles. (c) Schematic diagram of a three layers neural network (left) and neural core with a crossbar structure to perform the analog matrix operations (right). (d) The simulation results, the training accuracy of BaTiO3 FTJs and SrTiO3 based resistive devices for a small image (left) and a large image (right). Reproduced from Ref. [
Fig. 9. (a) The schematic diagram of high-density 3D vertical HZO-based FTJS array with high resolution transmission electron microscopy image of the 3D TiN/ HZO/Pt devices. (b) Measurements of Hebbian and anti-Hebbian STDP rule in the HZO-based synaptic array. The red lines are the exponential fits to the experimental data. (c) The schematic diagram of multilayer perceptron network (left), and simulated output signal in which “correct” class of the applied letter patterns is larger than the other “false” outputs after learning process (right). Reproduced from Ref. [
Fig. 10. (a) Schematic diagram of charge distribution and energy-band diagrams of a Pt/BiFeO3/Nb:SrTiO3 FD at low resistance state. (b) Schematic diagram of charge distribution and energy-band diagrams of a Pt/BiFeO3/ Nb:SrTiO3 FD at high resistance state. Reproduced from Ref. [
Fig. 11. (a) Schematic diagram of Au/BaTiO3/SmNiO3 FDs. (b) Off-pulse
Fig. 12. (a) The lower half part is out-of-plane phase image and the upper half part is the corresponding current mapping with read voltage of –0.5 V measured on BiFeO3/SrRuO3/ SrTiO3 (substrate). (b) Out-of-plane phase images after heating the sample at 25 °C, 300 °C, and 500 °C for 5 min, respectively. (c)
Fig. 13. (a) Schematic structure of an HZO FeFET. (b) Transfer curve of the HZO FeFET. (c) LTP/D properties of the HZO FeFET with incremental pulse scheme. (d) Schematic illustration of two-layer perceptron neural network for simulation of pattern recognition. Reproduced from Ref. [
Fig. 14. (a) Sketch of a biological synapse. (b) Sketch of an MoS2 FeFET with ferroelectric P(VDF-TrFE) gate insulator. Note that the input port can be either the gate-source/drain port or the drain-source port. For supervised learning in general SNN, the channel usually connects pre and post synaptic neurons while the gate terminal accepts signals to modulate the channel conductance. (c) Channel conductance versus gate electric field curves in a MoS2 FeFET. (d) Gate current versus gate electric field curve of the MoS2 FeFET. (e) Evolution of channel conductance as a function of the different voltage pulse sequences. (f) Measurements of Hebbian STDP learning rule in a MoS2 FeFET. Reproduced from Ref. [
Fig. 15. (a) Schematic view that when the polarization is upward/downward, the graphene channel becomes hole/electron dominated due to different positions of Fermi levels within the graphene energy bands. (b) The channel conductance will be decreased/increased due to reduction/enhancement of hole/electron density caused by the corresponding change of ferroelectric polarization. (c) A remote supervise method (ReSuMe) module composed of graphene FeFET complementary synapses, leaky integrate-and-fire (LIF) neuron, and supervise circuits (left). Time chart of signals (right). (d) Sketch of single-layer perceptron for classification of 3 × 3 binary images of z, v, and n, where classification is represented by the different timings of the output neuron. (e) The simulation results, the evolution of output signals, where output timings for z, v, and n inputs gradually approach the desired timings labeled by dash lines. Reproduced from Ref. [
|
|
Get Citation
Copy Citation Text
Bo-Bo Tian, Ni Zhong, Chun-Gang Duan. Recent advances, perspectives, and challenges in ferroelectric synapses[J]. Chinese Physics B, 2020, 29(9):
Received: May. 11, 2020
Accepted: --
Published Online: Apr. 29, 2021
The Author Email: Tian Bo-Bo (cgduan@clpm.ecnu.edu.cn)