Organometal Halide Perovskite Nanoparticle Artificial Synapses with Ultrahigh Sensitivity

OHP-NPs were synthesized by solubility-difference-induced crystallization methods at room temperature. We prepared the clear perovskite precursor solution by dissolving 0.3 mmol of CH₃NH₃Br, 0.4 mmol of PbBr₂, and 0.6 μmol of n-hexylamine into 5 mL of dimethylformamidinium (DMF). After dripping 1 mL of precursor solution into 5 mL of toluene solution containing 100 μL of oleic acid under vigorous stirring, the solution was centrifuged at 10,000 rpm for 10 min. The supernatant solution was collected to fabricate the synaptic devices.

p+Si wafer was cleaned by sequential sonication in acetone and isopropanol (15 min for each process), boiled in isopropanol, and then blown dried with N₂. OHP-NP thin film was spin-coated from solution, and then baked at 90°C for 10 min. ^{Finally, 100-nm-thick circular Al metal dots were thermally evaporated through a shadow mask to define the pixel area of 100 μm2.}

Atomic force microscopy (AFM) images were captured using a Dimension 3,100 microscope (Digital Instruments). Scanning transmission electron microscopy (STEM) images were obtained from TEM equipment 2100F with Cs correction. All electrical characteristics of the electronic devices were characterized using a Keithley 4200-SCS semiconductor parameter analyzer in a nitrogen atmosphere.

First, we evaluated the size distribution of OHP-NPs used in our artificial synapse. The OHP-NPs were pre-synthesized using methylammonium lead bromide (MAPbBr₃) and dispersed in p-xylene. The dispersion was then drop-cast onto polycarbon-coated copper grids and dried. STEM images of the OHP-NPs revealed spherical shapes with diameters ranging from 5 to 10 nm (Fig. 2(A)).

Fig. 2. Characteristics of materials and thin films. (A) TEM images of dispersed perovskite nanoparticles, a single OHP-NP and its EDS mapping. (B) High-resolution TEM image of a single OHP-NP, its Fast Fourier transform (FFT) image and lattice spaces. (C) X-ray diffraction patterns of an OHP-NP thin film. (D) FT-IR spectrum of ONP-NPs. (E) Three-dimensional AFM images of OHP-NP thin films.

Download Original Figure

Energy dispersive spectrometry (EDS) mapping and X-ray photoelectron spectroscopy (XPS) further confirmed the successful synthesis of OHP-NPs with Pb and Br elements embedded inside the crystallites (Fig. 2A; SI Figs. 1, 2). The crystal structure within the OHP-NPs was clearly evident from the magnified view of a specific nanocrystallite (Fig. 2B) and was validated using X-ray diffraction (XRD) patterns (Fig. 2C). Fourier-transform infrared spectroscopy (FT-IR) analysis of the perovskite nanoparticles showed characteristic peaks corresponding to C-N stretches (~1,200 cm^–1), C-O bends (~1,250 cm^–1), C-H bends (~1,500 cm^–1), C-H stretches (~2,950 cm^–1), and N-H stretches (~3,300 cm^–1). These peaks provided evidence of well-adhered ligand chains on the NP surface in film states (Fig. 2D).

Furthermore, AFM imaging revealed a uniform thin film surface, with a root-mean-square roughness of 5.20 nm (Fig. 2E). Overall, our characterization of the OHP-NPs demonstrates their suitability for use in the fabrication of artificial synapses.

An artificial synapse utilizing OHP-NP thin films was fabricated, and the cross-sectional view of the devices confirmed a 15-nm-thick OHP-NP layer in a sandwiched structure (Fig. 3A, Fig. 3B). Synaptic plasticity, which plays a fundamental role in signal transmission in synapses, was investigated by measuring the synaptic weight, modulated by sensorimotor impulses and recovering rapidly during PNS signal transitions. Upon the application of an external pulse, a sudden current shoot is triggered, and after the pulse is removed, it decays within several tens of milliseconds (Fig. 3C), closely resembling the EPSC observed in biological synapses. This phenomenon arises from the redistribution of ions in response to external stimuli, followed by their spontaneous drift back to their original positions. Consequently, this process emulates the arrival of a presynaptic spike on the presynaptic membrane to cause neurotransmitter releasement and the generation of EPSC through the postsynaptic membrane.

Fig. 3. Short-term plasticity and long-term plasticity. (A) Cross sectional view of the top electrode/OHP-NP/conductive polymer/bottom electrode structure. (B) Geometry of the artificial synapse. (C) EPSC triggered by a single presynaptic spike and EPSC triggered by two successive pulses. (D) PPF index as a function of time interval of the two successive presynaptic spikes. (E) Maximal EPSC (Cyan) and EPSC (10 s after stimuli, Magenta) as a function of spike number (Reading bias 0.1 mV, pulse amplitude 0.1 mV). (F) EPSC triggered by 60 successive pulsesfrom an OHP-NP artificial synapse.

Download Original Figure

Upon subjecting the artificial synapse to two successive pulses, we observed two EPSC peaks, with the second peak being more intense than the first (Fig. 3C). This behavior closely resembles PPF in biological synapses, a phenomenon underlying various critical brain functions, including information processing and sound localization. The PPF index, representing the incremental percentage of synaptic weight, decreases as the time interval between the two presynaptic spikes increases (Fig. 3D). Remarkably, the PPF index drops to below 10% when the time interval is extended to approximately 7 seconds.

By applying additional consecutive pulses, we observed a gradual step-by-step increase in EPSC amplitude (Fig. 3E, Fig. 3F). This result suggests that ion migration during these spiking periods occurs faster than their recovery. In MAPbBr₃ perovskites, Br⁻ exhibits a lower activation energy for ion migration (≈0.2 eV) compared to MA⁺ and Pb²⁺ (both≈0.8 eV). Therefore, we consider that Br⁻ ions predominantly contribute to the ion migration and the modulation of synaptic strength in our system. We have seen that both maximal EPSC (Cyan) and EPSC (10 s after stimuli, Magenta) gradually increase with increasing spike number (Fig. 3E, Fig. 3F).

In order to demonstrate the diverse synaptic functions of our artificial synapses based on OHP-NPs, we conducted measurements of EPSC peaks while varying the intensity of spike duration (spike-duration-dependent plasticity (SDDP), Fig. 4A, Fig. 4B), the spike amplitude (spike-voltage-dependent plasticity (SVDP), Fig. 4C), and the spike rate (SRDP, Fig. 4D). The artificial synapses utilizing OHP-NPs exhibited gradual increases in EPSC peaks with increasing spike duration, spike amplitude, and spike rate, demonstrating the versatility and adaptability of the artificial synapse in response to different spiking patterns and durations. Particularly, the spike rate and frequency demonstrated a linear relationship with EPSC peaks (Fig. 4E).

Fig. 4. Emulation of different types of synaptic plasticity (reading bias 0.1 mV). (A) Schematic demonstration that multiple synaptic functions can be realized using the single electronic device. (B) PSC vs. spike duration (pulse amplitude=0.6 V). (C) PSC vs. spike amplitude (pulse width=380 ms). (D) PSC triggered by 10 pulses at different spike rates. E) Maximum PSC vs. spiking rate (0.1 Hz–1.2 Hz).

Download Original Figure

Next, we investigated STDP, a crucial form of the Hebbian learning rule, which posits that “neurons that fire together, wire together” [42]. STDP describes the modulation of synaptic weight when the relative timing between presynaptic and postsynaptic spikes is changed. To evaluate the applicability of STDP in our artificial synapses, we applied a series of pulse pairs to the presynaptic and postsynaptic terminals and recorded the changes in postsynaptic current. We observed an anti-symmetric plot of STDP (SI Fig. 3A, Fig. 3B), consistent with the biological phenomenon where synapses typically strengthen when presynaptic spikes precede postsynaptic spikes, while the reverse sequence leads to synaptic depression.

To study the sensitivity and energy consumption of the synaptic devices, presynaptic spikes with various amplitudes were applied and concomitant EPSC was recorded (Fig. 4C). Remarkably, the device exhibited exceptionally high sensitivity, successfully realizing all the aforementioned functions with stimuli applied at a voltage level as low as 100 mV. Furthermore, the electronic device demonstrated the ability to respond to even smaller presynaptic spikes, as low as 10 mV. This sensitivity level represents one of the highest among state-of-the-art two-terminal synaptic devices, surpassing that of a biological synapse, where action potentials of approximately 70 mV are typically applied. Such high sensitivity is crucial for the development of bio-integrating electronics, enabling the detection of weak signals at the bio/electronics interface, and represents a significant step towards constructing ultrasensitive neuromorphic electronics for brain-inspired applications.

The human brain remarkably functions at a power level comparable to that of a household lightbulb, despite its billions of neurons and trillions of synapses. This energy efficiency is primarily due to the low power consumption of synapses, typically in the range of 1–10 fJ per SE. Consequently, for the construction of a truly energy-efficient artificial system, low-energy consumption by artificial synapses is a critical requirement. Our calculations based on the pulse amplitude (A), the triggered current (I), and the pulse width (t) of the EPSC peak, triggered by a 10–mV pulse, revealed an extremely low energy consumption of approximately 5 fJ per SE. This energy consumption level represents one of the lowest observed among two-terminal synapse-emulating electronic devices to date and is comparable to that of a biological synapse [6,20]. Furthermor7, these values are comparable with those of other two- or three-terminal synaptic devices (~10 fJ/synaptic event) in previously reported literature [43]. These findings demonstrate the significance of the high sensitivity and remarkably low energy consumption for the development of an exceptionally sensitive and energy-efficient artificial neural system. With such capabilities, our artificial synapses based on OHP-NPs hold great promise for advancing neuromorphic computing and brain-inspired technologies.