Tunable Optoelectronic Neuromorphic Synaptic Link Based on Nanoscale Resonant Tunnelling Diode-Photodetector Spiking Neurons

We report an optoelectronic neuromorphic synaptic link built with high-speed nanoscale resonant tunnelling diode-photodetectors (RTD-PDs), which reproduces temporal-coded spike generation and transmissions in biological neurons and synapses. The artificial optoelectronic neuron is composed of a nanoscale RTD-PD built with a double barrier quantum well (DBQW) structure for nonlinearity and excitability, and an optical-sensitive layer to enable light-triggered spike firing. Due to the resonant tunnelling effect, RTDs exhibit a typical N-shaped current-voltage (I-V) curve with a nonlinear negative differential conductance (NDC) region bringing about a number of distinct functionalities, such as high-speed self-oscillations and controllable excitability [1]. For optical excitability, the nanoscale RTD-PD consist of an InAlGaAs light-sensitive spacer layer to receive infrared optical stimuli [2]. Due to the photodetection and carrier accumulation in the device, the optical stimuli can shift the I-V curve and trigger an optically-induced spiking event (provided the RTD-PD is biased in the vicinity of its NDC region) [1]. We have demonstrated that RTD-PDs are capable of implementing neuromorphic excitability with well-defined thresholding and refractoriness [3–5]. We build an optoelectronic synaptic link based on two nanoscale RTD-PD (pre-synaptic and post-synaptic) neurons to reproduce the generation, weighting and propagation of spiking signals. Fig.1 illustrates the schematic diagram of the experimental optical synaptic link. Optical pulses are injected into the pre-synaptic nanoscale RTD-PD neuron which in response elicits electrical spikes in an all-or-nothing manner. Information is coded in sparse temporal intervals. The electrical spikes are converted into the optical domain via direct modulation of a 1550nm vertical cavity surface emitting laser (VCSEL) and then the amplitude (weight) of the optical spikes is tuned by bespoke intensity modulation waveforms (created with an arbitrary waveform generator, AWG) via a Mach-Zehnder modulator (MZM). Subsequently, the weighted optical spikes are injected into the post-synaptic nano RTD-PD neuron. Due to the thresholding spike-firing property of RTDs, only weighted spikes with sufficient intensity can be regenerated by the post-synaptic neuron. This first report of nanoscale RTD-PD neurons and optoelectronic synaptic link demonstrates a promising solution for the core component of optical neuromorphic networks for high-speed, energy-efficient AI computation.

We report an optoelectronic neuromorphic synaptic link built with high-speed nanoscale resonant tunnelling diode-photodetectors (RTD-PDs), which reproduces temporal-coded spike generation and transmissions in biological neurons and synapses. The artificial optoelectronic neuron is composed of a nanoscale RTD-PD built with a double barrier quantum well (DBQW) structure for nonlinearity and excitability, and an optical-sensitive layer to enable light-triggered spike firing. Due to the resonant tunnelling effect, RTDs exhibit a typical N-shaped current-voltage (I-V) curve with a nonlinear negative differential conductance (NDC) region bringing about a number of distinct functionalities, such as high-speed self-oscillations and controllable excitability [1]. For optical excitability, the nanoscale RTD-PD consist of an InAlGaAs light-sensitive spacer layer to receive infrared optical stimuli [2]. Due to the photodetection and carrier accumulation in the device, the optical stimuli can shift the I-V curve and trigger an optically-induced spiking event (provided the RTD-PD is biased in the vicinity of its NDC region) [1]. We have demonstrated that RTD-PDs are capable of implementing neuromorphic excitability with well-defined thresholding and refractoriness [3–5]. We build an optoelectronic synaptic link based on two nanoscale RTD-PD (pre-synaptic and post-synaptic) neurons to reproduce the generation, weighting and propagation of spiking signals. Fig.1 illustrates the schematic diagram of the experimental optical synaptic link. Optical pulses are injected into the pre-synaptic nanoscale RTD-PD neuron which in response elicits electrical spikes in an all-or-nothing manner. Information is coded in sparse temporal intervals. The electrical spikes are converted into the optical domain via direct modulation of a 1550nm vertical cavity surface emitting laser (VCSEL) and then the amplitude (weight) of the optical spikes is tuned by bespoke intensity modulation waveforms (created with an arbitrary waveform generator, AWG) via a Mach-Zehnder modulator (MZM). Subsequently, the weighted optical spikes are injected into the post-synaptic nano RTD-PD neuron. Due to the thresholding spike-firing property of RTDs, only weighted spikes with sufficient intensity can be regenerated by the post-synaptic neuron. This first report of nanoscale RTD-PD neurons and optoelectronic synaptic link demonstrates a promising solution for the core component of optical neuromorphic networks for high-speed, energy-efficient AI computation. Fig. 1

Schematic setup of the optical synaptic link and measured time-traces showing the input stimuli and the pre-synaptic, weighted and post-synaptic spikes in the link. (a) and (b) present the I-V characteristics of two RTD-PD devices, along with the biasing operating conditions respectively.