Researchers at Linköping University have designed a new transistor based on organic materials. It has the ability to learn and is fitted with both long-term and short-term memory. This effort is a huge step toward creating technology that imitates the human brain.
Thus far, brains have been exclusive in being able to form connections where none existed. In a scientific article in Advanced Science, scientists from Linköping University describe a transistor that can develop a new connection between an input and an output. They have integrated the transistor into an electronic circuit that learns how to connect a certain stimulus with an output signal, in the identical way that a dog absorbs that the sound of a food bowl being prepared means that dinner will soon be served.
A standard transistor acts as a valve that dampens or amplifies the output signal, subject on the properties of the input signal. In the organic electrochemical transistor that the scientists have created, the channel in the transistor comprises of an electropolymerized conducting polymer. The channel can be formed, shrunk, or grown, or totally eliminated during operation. It can also be trained to react to a specific stimulus, a definite input signal, such that the transistor channel turns out to be more conductive and the output signal superior.
“It is the first time that real-time formation of new electronic components is shown in neuromorphic devices”, says Simone Fabiano, principal investigator in organic nanoelectronics at the Laboratory of Organic Electronics, Campus Norrköping.
The channel is grown by expanding the material’s degree of polymerization in the transistor channel, thus increasing the number of polymer chains that transmit the signal. Alternatively, the material can be overoxidized by applying a high voltage rendering the channel inactive. Temporary variations of the conductivity can also be attained by doping or dedoping the material.
“We have shown that we can induce both short-term and permanent changes to how the transistor processes information, which is vital if one wants to mimic the ways that brain cells communicate with each other”, says Jennifer Gerasimov, postdoc in organic nanoelectronics and one of the authors of the article.
By altering the input signal, the strength of the transistor response can be moderated across a broad range, and connections can be formed where none formerly existed. This provides the transistor with a performance that is comparable with that of the synapse, or the communication interface between two brain cells.
Hardware for machine learning
It is also a huge step towards machine learning with the help of organic electronics. Software-based artificial neural networks are presently used in machine learning to accomplish what is known as “deep learning”. Software requires that the signals are conveyed between a vast number of nodes to mimic a single synapse, which takes substantial computing power and thus devours sizeable energy.
“We have developed hardware that does the same thing, using a single electronic component”, says Jennifer Gerasimov.
“Our organic electrochemical transistor can, therefore, carry out the work of thousands of normal transistors with an energy consumption that approaches the energy consumed when a human brain transmits signals between two cells”, confirms Simone Fabiano.
Newly developed monomer
The transistor channel has not been built using the most basic polymer used in organic electronics, PEDOT, but rather using a polymer of a newly-created monomer, ETE-S, designed by Roger Gabrielsson, who also works at the Laboratory of Organic Electronics and is one of the authors of the paper. ETE-S has numerous unique properties that make it ideally suited for this application. The properties include formation of adequately long polymer chains, water solubility while the polymer form is not, and production of polymers with an intermediate level of doping. The polymer PETE-S is created in its doped form with an intrinsic negative charge to balance the positive charge carriers (it is p-doped).
The study has been supported by, among other sources, the Knut and Alice Wallenberg Foundation, Vinnova, the Swedish Research Council and the Swedish Foundation for Strategic Research.