An artificial synapse based on molecular junctions

Abstract

Shrinking the size of the electronic synapse to molecular length-scale, for example, an artificial synapse directly fabricated by using individual or monolayer molecules, is important for maximizing the integration density, reducing the energy consumption, and enabling functionalities not easily achieved by other synaptic materials. Here, we show that the conductance of the self-assembled peptide molecule monolayer could be dynamically modulated by placing electrical biases, enabling us to implement basic synaptic functions. Both short-term plasticity (e.g., paired-pulse facilitation) and long-term plasticity (e.g., spike-timing-dependent plasticity) are demonstrated in a single molecular synapse. The dynamic current response is due to a combination of both chemical gating and coordination effects between Ag⁺ and hosting groups within peptides which adjusts the electron hopping rate through the molecular junction. In the end, based on the nonlinearity and short-term synaptic characteristics, the molecular synapses are utilized as reservoirs for waveform recognition with 100% accuracy at a small mask length.

Fig. 1. Molecular synapse architecture and dynamic current–voltage characteristics.

a Scheme of the molecular synapse device containing a bottom Ag/AgO_x electrode, peptide SAM, and a liquid GaO_x/EGaIn top electrode, Ag/AgO_x//CAAAAKAAAAK//GaO_x/EGaIn. The Ag/AgO_x electrode is prepared by thermal annealing electron-beam evaporated Ag layer (200 nm) at 150 °C for 40 min in air. Current signals (black, red, and blue curves) recorded by consecutively placing 75 positive b, negative c, and positive d, triangle pulses (light blue curves, amplitude, 0.45 V, width, 9 s). The bottom Ag/AgO_x is grounded. e Device endurance test by placing continuous potentiation (amplitude, −0.45 V, red markers, left y-axis) and depression (amplitude, 0.45 V, blue markers, right y-axis) pulses for 65 cycles. 30 square pulses (width, 0.5 s) are placed for each potentiation or depression.

Fig. 2. Operation mechanism of the molecular synapse.

a Schematic showing the release and distribution of Ag⁺ in Ag/AgO_x//CAAAAKAAAAK//GaO_x/EGaIn device under negative voltage. The coordination between Ag⁺ and –C=O, –NH groups enables the long-distance transport of silver cations. b Ag⁺ couldn’t be activated in Ag//CAAAAKAAAAK//GaO_x/EGaIn device. Much higher energy is required. c Ag⁺ couldn’t migrate across the long alkyl junction in Ag/AgO_x///HS-(CH₂)₁₂H//GaO_x/EGaIn device due to the lack of cation bridges. d Current decay curves of Ag/AgO_x//CAAAAKAAAAK//GaO_x/EGaIn molecular synapse after potentiated by 20, 60, and 100 square pulses (amplitude, −1 V). The colored solid lines are fitted curves with a double exponential function I = a ∗ exp (−t/τ₁) + b ∗ exp (−t/τ₂) + c in which the time constant τ₁ and τ₂ correspond to the fast and slow decaying process. e Dependence of the fitting time constants τ₁ and τ₂ on the number of potentiation pulses. The amplitude is −1 V.

Fig. 3. Modeling of the molecular synapse.

a The schematic picture of the peptide with two Ag⁺ cations. b The inversed hopping rates of the peptides with (red curves) and without (blue curves) binding Ag⁺ were calculated based on the Marcus hopping theory. The inset shows the log(reduced 1/k) along with the inverted temperature (1000/T). c Simulated potentiation (red square, left y-axis) and depression (blue diamond, right y-axis) of Ag/AgO_x//CAAAAKAAAAK//GaO_x/EGaIn synapse. The Ag⁺ is injected by placing −1.3 V for 0.5 s while the depression process is monitored by placing a reversed 0.1 V for 1 s. The same voltages are placed on a control device (w/o Ag⁺), and no potentiation/depression characteristic is found (black circle, right y-axis). d Simulated potentiation (solid squares, left y-axis) and depression (hollow diamonds, right y-axis) of Ag/AgO_x//CAAAAKAAAA //GaO_x/EGaIn synapse under voltages from −0.6 to −1.5 V. e Simulated current decay curves after potentiated by −0.7, −1.1, and −1.5 V (hollow circles). The fitted solid lines are obtained by using a double exponential function $I=a*\exp \left(-t/{\tau }_1\right)+b*\exp \left(-t/{\tau }_2\right)+c$. f Dependence of the fitting time constants τ₁ and τ₂ on the potentiation voltage.

Fig. 4. Synaptic plasticity characterization.

a Synaptic weight change (∆wt) as a function of the number of potentiation pulses under different pulse amplitudes. The pulse width and time interval are 200 ms. Paired pulse behavior of molecular synapse by changing the pulse amplitude b (time interval, 200 ms), and time interval c (amplitude, −1 V). The width of the pulse is 200 ms. The red line in (b) is a guideline of the dependent tendency while the red line in (c) shows exponential decay (time constant, 192 ms) with the time interval of the paired-pulse. d Spike-timing-dependent plasticity of the molecular synapse. The solid lines are exponential fitting with the time constants of 560 ms (blue) and 308 ms (red). The insets in (b)–(d) are the placed pulses. The error bars in (b) and (c) are based on ten independent devices.

Fig. 5. Signal processing based on the molecular synapse.

a The current response (red markers) by placing 120 periodic pulses (blue line). The amplitude and width of the write (potentiation) pulse are −1.0 V and 1 s while the amplitude and width of the read (depression) pulse are −0.25 V and 10 s. The inset shows the magnified current decay of the last read pulse, which is fitted by using a double exponential function. b, c Statistical results of time constant τ₁ (read pulse) and synaptic weight change ∆wt (write pulse) in (a). The solid red lines are fitting to the Gaussian distribution. The performance of reservoir computing system on training d, and testing dataset e (ground truth, black line, output, red line). Training and testing data consist of 250 sine and square waveforms. The recognition result is dependent on the output 1 (sine) or 0 (square). f The NRMSE (red curve) and accuracy (blue curve) as a function of mask length. The error bars are based on ten devices.