Organic transistor mimics synapse, suitable for ML/AI

February 06, 2019 //By Peter Clarke
Organic transistor mimics synapse, suitable for ML/AI
Researchers from Linköping University have reported the development of an organic electrochemical transistor (OECT) that shows short- and long-term memory effects and demonstrates the behaviour of a biological synapse.

The OECT is formed by electropolymerizing a self‐doped conjugated monomer, sodium 4‐(2‐(2,5‐bis(2,3‐dihydrothieno[3,4‐b][1,4]dioxin‐5‐yl)thiophen‐3‐yl)ethoxy)butane‐1‐sulfonate (ETE‐S), as the transistor channel and this channel can be formed, grown, shrunk or obliterated in situ and under operation.

This transistor can be trained to react to a certain input signal so that the channel becomes more conductive and the output signal larger. For the research the PETE‐S OECT channel was fabricated on a silicon oxide substrate patterned with source and drain electrodes to give a channel length of 30 microns and a channel width of 1 micron. The voltage scheme for action is at around 2V and below.

The authors report that the long‐term modulation of the channel conductance persists for months while short‐term synaptic plasticity occurs on a time‐scale of seconds.

"It is the first time that real-time formation of new electronic components is shown in neuromorphic devices," said Simone Fabiano, principal investigator in organic nanoelectronics at the Laboratory of Organic Electronics, Campus Norrköping, in a statement. The claim is that this is the first synaptic device that can generate new synapses within its working environment in a similar way to how biological synapses establish, evolve, and operate.

Although machine learning and artificial intelligence are supported quite directly in hardware in some commercial systems these are nearly always digital systems and would typically use multiple transistors to mimic synaptic activity and therefore have higher power consumption.

The monomer ETE-S has a number of properties that make suitable for synaptic transistor operation: it forms sufficiently long polymer chains, is water-soluble while the polymer form is not, and it produces polymers with an intermediate level of doping.

The authors conclude that they are confident the PETE-S OECT can be implemented in a cross-bar array and is singularly suited for neuromorphic applications.

Related links and articles:

www.liu.se

An Evolvable Organic Electrochemical Transistor for Neuromorphic Applications

News articles:

Organic electronic breakthrough doubles efficiency

Memristor synapse supports unsupervised learning

Synaptic transistor learns as it switches

SmartKem raises funds for organic semiconductor materials


Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.