New parallel computing architectures based on neuromorphic computing are needed due to their advantages over conventional computation with regards to real‐time processing of unstructured sensory data such as image, video, or voice. However, developing artificial neuromorphic system remains a challenge due to the lack of electronic synaptic devices, which can mimic all the functions of biological synapses with low energy consumption. Here it is reported that two‐terminal organometal trihalide perovskite (OTP) synaptic devices can mimic the neuromorphic learning and remembering process. Various functions known in biological synapses are demonstrated in OTP synaptic devices including four forms of spike‐timing‐dependent plasticity (STDP), spike‐rate‐dependent plasticity (SRDP), short‐term plasticity (STP) and long‐term potentiation (LTP)), and learning‐experience behavior. The excellent photovoltaic property of the OTP devices also enables photo‐read synaptic functions. The perovskite synapse has the potential of low energy consumption of femto‐Joule/(100 nm)2per event, which is close to the energy consumption of biological synapses. The demonstration of energy‐efficient OTP synaptic devices opens a new plausible application of OTP materials into neuromorphic devices, which offer the high connectivity and high density required for biomimic computing.
For hardware artificial intelligence, the central task is to design and develop artificial synapses with needed characteristics. Here, the design and experimental demonstration of a van der Waals (vdW) photo‐ferroelectric synapse are reported. In the photo‐ferroelectric synapse, the synaptic memory is extracted by reading the photocurrent, and written or edited by electrical pulses. The semiconducting vdW organic‐inorganic halide perovskite ((R)‐(–)‐1‐cyclohexylethylammonium)PbI3(R‐CYHEAPbI3) photo‐ferroelectric serves as the model photo‐ferroelectric channel. Here, the vdW organic layer provides ferroelectric dipole and the PbI6octahedron is responsible for photon absorption and charge transport. The R‐CYHEAPbI3photo‐ferroelectric synapse show a writing/reading dynamics with >200 synaptic states, close to 103on/off ratio, and reasonable endurance and retention characteristics. With the experimentally measured weight dynamics (parallel reading through ferroelectric photovoltaic effect and writing by electrical pulses) of R‐CYHEAPbI3synapses, the feasibility of using a crossbar circuit to implement classic training and inference of hand‐written digits is presented. An image recognition accuracy of up to 90% is obtained. The demonstration of such a vdW photo‐ferroelectric synapse opens a window in the design of advanced devices for artificial intelligence.
more » « less- Award ID(s):
- 1916652
- NSF-PAR ID:
- 10373738
- Publisher / Repository:
- Wiley Blackwell (John Wiley & Sons)
- Date Published:
- Journal Name:
- Advanced Electronic Materials
- Volume:
- 8
- Issue:
- 10
- ISSN:
- 2199-160X
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
-
more » « less
Abstract Optoelectronic synapses combine the functionalities of a non-volatile memory and photodetection in the same device, paving the path for the realization of artificial retina systems which can capture, pre-process, and identify images on the same platform. Graphene/Ta2O5/graphene phototransistor exhibits synapse characteristics when visible electromagnetic radiation of wavelength 405 nm illuminates the device. The photocurrent is retained after light withdrawal when positive gate voltage is applied to the device. The device exhibits distinct conductance states, modulated by different parameters of incident light, such as pulse width and number of pulses. The conductance state can be retained for 104 s, indicating long term potentiation (LTP), similar to biological synapses. By using optical and electrical pulses, the device shows optical potentiation and electrical LTD repeatably, implying their applicability in neural networks for pattern recognition.
-
null (Ed.)Abstract Optical data sensing, processing and visual memory are fundamental requirements for artificial intelligence and robotics with autonomous navigation. Traditionally, imaging has been kept separate from the pattern recognition circuitry. Optoelectronic synapses hold the special potential of integrating these two fields into a single layer, where a single device can record optical data, convert it into a conductance state and store it for learning and pattern recognition, similar to the optic nerve in human eye. In this work, the trapping and de-trapping of photogenerated carriers in the MoS 2 /SiO 2 interface of a n-channel MoS 2 transistor was employed to emulate the optoelectronic synapse characteristics. The monolayer MoS 2 field effect transistor (FET) exhibits photo-induced short-term and long-term potentiation, electrically driven long-term depression, paired pulse facilitation (PPF), spike time dependent plasticity, which are necessary synaptic characteristics. Moreover, the device’s ability to retain its conductance state can be modulated by the gate voltage, making the device behave as a photodetector for positive gate voltages and an optoelectronic synapse at negative gate voltages.more » « less
-
more » « less
Abstract An electrical‐biased or mechanical‐loaded scanning probe written on the ferroelectric surface can generate programmable domain nanopatterns for ultra‐scaled and reconfigurable nanoscale electronics. Fabricating ferroelectric domain patterns by direct‐writing as quickly as possible is highly desirable for high response rate devices. Using monolayer
α ‐In2Se3ferroelectric with ≈1.2 nm thickness and intrinsic out‐of‐plane polarization as an example, a writing‐speed dependent effect on ferroelectric domain switching is discovered. The results indicate that the threshold voltages and threshold forces for domain switching can be increased from −4.2 to −5 V and from 365 to 1216 nN, respectively, as the writing‐speed increases from 2.2 to 10.6 µm s−1. The writing‐speed dependent threshold voltages can be attributed to the nucleations of reoriented ferroelectric domains, in which sufficient time is needed for subsequent domain growth. The writing‐speed dependent threshold forces can be attributed to the flexoelectric effect. Furthermore, the electrical‐mechanical coupling can be employed to decrease the threshold force, achieving as low as ≈189±41 nN, a value smaller than those of perovskite ferroelectric films. Such findings reveal a critical issue of ferroelectric domain pattern engineering that should be carefully addressed for programmable direct‐writing electronics applications. -
more » « less
Neuromorphic computing has the great potential to enable faster and more energy‐efficient computing by overcoming the von Neumann bottleneck. However, most emerging nonvolatile memory (NVM)‐based artificial synapses suffer from insufficient precision, nonlinear synaptic weight update, high write voltage, and high switching latency. Moreover, the spatiotemporal dynamics, an important temporal component for cognitive computing in spiking neural networks (SNNs), are hard to generate with existing complementary metal–oxide–semiconductor (CMOS) devices or emerging NVM. Herein, a three‐terminal, Li
x WO3‐based electrochemical synapse (LiWES) is developed with low programming voltage (0.2 V), fast programming speed (500 ns), and high precision (1024 states) that is ideal for artificial neural networks applications. Time‐dependent synaptic functions such as paired‐pulse facilitation (PPF) and temporal filtering that are critical for SNNs are also demonstrated. In addition, by leveraging the spike‐encoded timing information extracted from the short‐term plasticity (STP) behavior in the LiWES, an SNNs model is built to benchmark the pattern classification performance of the LiWES, and the result indicates a large boost in classification performance (up to 128×), compared with those NO‐STP synapses.
