The need for innovation in the realm of computer hardware is imminent. Transistor densities will soon reach their physical boundaries and Dennard scaling will limit further power density improvements. Emulating the adaptive signal processing performed by neural synapses in the human brain is one way to combat this stagnation, as they are computationally dense and highly power efficient. Brain-inspired—*neuromorphic*—circuitry is also likely to enable new types of autonomous and intelligent systems with embedded signal processing, computing, and learning. Yet, current neuromorphic hardware, including solid-state memory resistors (memristors) operate via poorly understood, non-biological switching mechanisms and require high voltages.

In contrast, we are pursuing a biomolecular approach using voltage-activated membranes that captures key synaptic behaviors including voltage-dependent conductance changes and activity-dependent plasticity. Specifically, we are studying biomimetic membranes formed by joining lipid-coated aqueous droplets in a hydrophobic phase that show synapse-like short term memory and operate through ionic current. We have shown that droplet interface bilayers (DIBs) doped with pore forming molecules can exhibit memory resistance, while lipid-only DIBs exhibit memory capacitance, as indicated by pinched hysteresis in either current or charge in the applied voltage plane, respectively. These effects arise from a dynamic interaction of ion channel formation and changes in membrane area via electrowetting, both of which depend on the history of applied voltage. Synapse connection strength is similarly dependent on a history of received signals, and the ion channels found in neurons are known to be memristors.

               We demonstrate that biomolecular synaptic mimics can be interfaced with solid-state spiking neuron circuits, whereby dynamically changing synapse conductance affects the frequency of neuron spiking. Two leaky integrate-and-fire neurons that generate constant amplitude voltage pulses are connected to separate droplets of the DIB (hereafter, “synapse”) such that the output voltage from the pre-synaptic neuron affects the conductance of the synapse, which in turn modulates current supplied to the post-synaptic neuron and thus its firing rate. In our test platform, the firing of the pre-neuron is controlled by the output current of a photoresistor (PR). This combination makes it a simple model of a sensory neuron.  Six individually controllable LEDs on a separate circuit are positioned directly above the PR. Activating any number of LEDs rapidly decreases the PR resistance, supplying the pre-neuron with more current and increasing its firing rate and, thus, the root mean square voltage sensed by the synapse.

 It is our goal to investigate how voltage-dependent changes in synapse conductance, controlled by membrane composition, can be used to provide adaptive filtering of neural signaling. The activities of two voltage-gated pore-forming species are compared: alamethicin and monazomycin. The responses of the synapses and neurons to bright light, low light, and their transition between these states will be characterized to elucidate how neuron- and synapse-inspired devices may be integrated into existing hardware. A tunable, modular, and self-adapting synapse holds great potential in advancing neuromorphic hardware for increased computing efficiency and adaptive filtering for embedded intelligence.


Biomolecular Synapses for Adaptive Filtering in Physical Neuromorphic Networks

Division / Sponsors: 
Aerospace Division

Paper Proceedings: https://asmedigitalcollection.asme.org/SMASIS/SMASIS2020/volume/84027

SMASIS2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.