Plastic-based artificial synapse beats all energy-efficiency benchmarks

Plastic-based artificial synapse beats all energy-efficiency benchmarks
Technology News |
Led by associate professor Alberto Salleo from Stanford University, an international team of researchers has devised a low-cost, compliant and very energy-efficient artificial synapse mostly made out of polymers.
By Julien Happich

Share:

The so-called ENODe (electrochemical neuromorphic organic device) can be switched into over 500 distinct, non-volatile conductance states within a 1V range and beats all previous energy-efficiency benchmarks relating to artificial synapses.

According to the paper “A non-volatile organic electrochemical device as a low-voltage artificial synapse for neuromorphic computing” published in Nature Materials Letters, the fairly large (103µm2) ENODes drew under 10pJ per synaptic event and were proven to achieve high classification accuracy (between 93% and 97%) when implemented in neural network simulation. The researchers projected that with the switching energy being proportional to the electrode area, a submicron 0.3×0.3µm ENODe fabricated by photolithography would only require 35 aJ (10-18 Joules) per switching event, several orders of magnitude less than biological synapses as they operate in the human brain.

Operating with a fundamentally different mechanism from existing memristors based on filament-forming metal oxides (FFMOs) or phase change memory (PCM) materials, the ENODe consists of a stack of two similar polymers films, one of which is partially reduced, separated by an electrolyte layer transporting ions/protons. By applying a voltage bias across the two films (forming the pre- and post-synaptic layers), one can reversibly control the conductivity of this organic mixed ionic/electronic device, setting it in different non-volatile states.

a) Sketch of the device structure. Pre- and postsynaptic layers are separated by an electrolyte layer transporting ions/protons (red spheres). b) A positive Vpre drives protons into the postsynaptic electrode, which results in the compensation of some PSS by the protonated PEI. This reaction causes the reduction of PEDOT in the same electrode due to charge neutrality, which eliminates a polaron (in red) and decreases the polymer conductivity. The reaction is reversed upon applying a negative Vpre.

The researchers describe the ENODe as a type of non-volatile redox cell (NVRC) in which the state of charge determines the electronic conductivity and where the barrier for state retention is decoupled from the barrier for changing states. This, they explain, allows for extremely low switching voltages (down to 10mV pulses) while maintaining non-volatility.


The conductance states are monitored using a postsynaptic potential Vpost and the conductance of the interface layer represents the synaptic weight of the connection between two neurons. The postsynaptic state is programmed by varying the amplitude or the duration of the presynaptic pulse.

The ENODe was found to properly emulate short-term to long-term potentiation found in nature and its scalability (size and geometry dictate operating speed and switching energy levels) makes it a promising candidate for designing ultra-low power neuromorphic computers.

Schematic of the flexible all solid-state neuromorphic
device.

With the ENODe, full plastic neural electrode arrays could be implemented in large-area systems for implantable prosthetics, where they could fold to form three-dimensional densely connected neuromorphic devices.

The researchers even envisage such biocompatible devices to enter advanced neural prostheses with integrated brain-machine interfaces that combine neural sensing with training.

 

Related articles:

IBM emulates neurons with phase-change materials

Closely mimicking synapses: diffusive memristors

BrainChip provides details of neural network architecture

Synaptic transistor learns as it switches

Linked Articles
eeNews Europe
10s