The miniaturization of electronic synapses to the molecular scale is crucial for increasing integration density and reducing energy consumption in neuromorphic computing. Current CMOS technology faces limitations as devices shrink, and molecular electronics offers a potential solution. While molecular switches, rectifiers, transistors, memory, and logic gates have been demonstrated, the development of molecular synapses that mimic the brain's information processing capabilities remains a significant challenge. Neuromorphic computing, inspired by the brain's architecture, promises improved computation and energy efficiency by operating under a Compute-in-Memory paradigm, thus circumventing the von Neumann bottleneck. This requires artificial synapses capable of emulating the change in connection strength between neurons. Memristors, with their electric-field-induced memory phenomena, have shown promise. However, creating molecular junctions that exhibit continuous conductance modulation, rather than discrete switching, for emulating synaptic functions, remains a significant hurdle. This paper focuses on overcoming this hurdle by demonstrating a molecular electronic synapse that enables continuous weight modulation.

The field of molecular electronics has advanced significantly in recent years due to breakthroughs in wiring techniques and a deeper understanding of charge transport mechanisms. Researchers have successfully fabricated various molecular-scale electronic components, including switches, rectifiers, transistors, memory devices, and logic gates. These advancements have spurred interest in replacing silicon-based electronics with molecular counterparts. However, the von Neumann architecture's separation of computation and memory presents a significant challenge, especially with the explosive growth of data. Neuromorphic computing offers a potential solution by mimicking the brain's parallel processing capabilities. The development of artificial synapses, capable of modulating conductance in response to electrical stimuli, is central to this approach. Previous research has explored the use of molecular junctions for this purpose, leveraging phenomena such as redox state changes and conformational changes of molecules or electrode-molecule couplings to achieve resistance or conductance switching. However, achieving the continuous conductance modulation required for emulating synaptic plasticity has remained a significant challenge.

The researchers fabricated a molecular synapse device by sandwiching a self-assembled monolayer (SAM) of a specific peptide (CAAAAKAAAAK) between an active Ag/AgOx electrode and a liquid eutectic Gallium-Indium (EGaIn) electrode. The Ag/AgOx electrode, prepared by partial oxidation of a silver film, serves as a reservoir for Ag+ ions. The peptide molecules contain functional groups that interact with the Ag+ ions, facilitating their migration. The electrical characteristics of the molecular junction were investigated using consecutive triangular pulses. The conductance modulation was analyzed by applying positive and negative voltage pulses. To investigate the mechanism, the researchers conducted experiments with different peptides, including those with long alkyl chains, and varied the electrode materials. The role of Ag+ in the conductance modulation was studied. The researchers developed a continuum charge transport model, combining the Nernst-Planck diffusion and Poisson's equations, to simulate the transport of both Ag+ ions and electrons within the peptide junction. Atomic force microscopy (AFM), angle-resolved X-ray photoelectron spectroscopy (AR-XPS), and ellipsometry were used to characterize the peptide monolayer, its thickness, and surface coverage. Synaptic plasticity was evaluated by measuring the weight change (Δwt) in response to various pulse parameters, including paired pulses and spike-timing-dependent plasticity (STDP) protocols. Finally, the device's reproducibility and durability were tested through multiple write/read cycles, and its performance in waveform recognition was evaluated using a reservoir computing approach.

The study demonstrated that the conductance of the molecular junction could be dynamically modulated by applying electrical pulses. The application of negative pulses led to a significant increase in current (potentiation), while positive pulses resulted in a decrease (depression). This behavior is reversible and can be cycled multiple times. The researchers demonstrated both short-term plasticity, characterized by paired-pulse facilitation (PPF), and long-term plasticity, characterized by spike-timing-dependent plasticity (STDP). PPF exhibited a positive weight change (Δwt) that decreased with increasing pulse amplitude and increased inter-pulse intervals. STDP showed that the synapse could be potentiated (Δt < 0) or depressed (Δt > 0) depending on the timing of pre- and post-synaptic spikes. A continuum charge transport model accurately captured the experimental observations, suggesting that Ag+ injection into the peptide monolayer, combined with chemical gating and coordination effects, is the primary mechanism governing the conductance modulation. Importantly, the molecular synapse showed high reproducibility and stability, maintaining consistent performance over multiple write/read cycles. Finally, the synapse was successfully used in a reservoir computing system for waveform recognition, achieving 100% accuracy at a small mask length. Detailed analysis showed that the time constants of both short-term and long-term plasticity correlated with specific mechanisms of Ag+ interaction within the peptide molecules.

The findings address the significant challenge of creating molecular-scale synapses with continuous conductance modulation, essential for emulating biological synaptic plasticity. The successful demonstration of both short-term (PPF) and long-term (STDP) plasticity in a single molecular device is a major advance. The developed continuum charge transport model provides a strong theoretical framework for understanding the underlying mechanisms. The high accuracy achieved in waveform recognition underscores the potential of these molecular synapses for practical applications in neuromorphic computing. The reversible and stable nature of the conductance modulation suggests that such devices are viable building blocks for larger-scale neuromorphic circuits. This work bridges the gap between fundamental research in molecular electronics and the practical implementation of neuromorphic computing hardware.

This research successfully demonstrated a molecular-scale artificial synapse capable of continuous conductance modulation, exhibiting both short-term and long-term synaptic plasticity. The device's performance in waveform recognition highlights its potential for neuromorphic computing. Future research could focus on exploring different peptide sequences, investigating the integration of these molecular synapses into larger-scale circuits, and exploring applications beyond waveform recognition.

While the study demonstrates excellent performance in a controlled laboratory environment, further research is needed to assess the long-term reliability and stability under varied operating conditions. The scalability of the fabrication process to larger arrays of molecular synapses also requires investigation. The current model does not account for all possible complexities of charge transport and interaction within the molecular junction, and future work should seek to refine this model. The current waveform recognition was performed on simple waveforms; more complex signal processing needs to be explored.