Osaka, Japan-Reservoir Computing (RC) solves complex problems by mimicking the way information is processed in the animal brain. It relies on a randomly connected network as a repository of information, which ultimately leads to more efficient output. In order to implement RC directly in matter (rather than simulation in a digital computer), many reservoir materials have been studied so far. Now, a team including researchers from Osaka University has designed a sulfonated polyaniline network for RC.
Neural networks in the brain use electrochemical signals carried by ions. Therefore, when choosing a material system for RC, the electrochemical method is a logical choice. Organic electrochemical field effect transistors (OECFETs) are very popular in bioelectronics; however, they have not been widely used in RC.
The key to the reservoir material is that it has a rich (time-dependent) behavior and is disordered, which makes polymer materials a good choice because they form a random network on their own.
Polyaniline is a promising polymer for RC applications because it is easy to polymerize, has good stability in the atmosphere, and has reversible doping/dedoping behavior, which means its conductivity can be changed.
Researchers studied sulfonated polyaniline (SPAN), which, in addition to the advantages of polyaniline, also has high water solubility and self-doping behavior. These make SPAN easier to use and doping more uniformly.
“The protons in the atmosphere are directly injected into the polymer chains of SPAN to make it conductive,” explains Yuki Usami, the lead author of the study. "Then this conduction can be controlled by adjusting the humidity."
The researchers used a simple drop casting method to assemble SPAN on gold electrodes to provide an organic electrochemical network device (OEND).
RC test was performed on SPAN OEND by examining the waveform and evaluating its performance in short-term memory tasks. The test result of testing speech recognition ability reached 70% accuracy. This capability of SPAN OEND is comparable to RC's software simulation.
"We have proven that our SPAN OEND system can be applied to RC," said research correspondence author Takuya Matsumoto. "Future steps to build a humidity-independent system will provide more practical options; however, the success of our SPAN-based system is a positive step in material-based reservoir calculations and is expected to have a significant impact on the next generation of artificial intelligence devices."
-This press release was originally published on the Osaka University website
Tags: Artificial Intelligence Industry News Material Science Product Resources: Industry News
© 2021 Laboratory Manager. all rights reserved.