An Energy-Efficient Solid-State Organic Device Array for Neuromorphic Computing

The slowing-down of Moore’s law is shifting the computing paradigm towards solutions based on quantum and neuromorphic computing elements. Unlike conventional digital computing, neuromorphic computing is based on analog devices. In this work, we propose a three-terminal neuromorphic organic device (NODe) capable of providing both analog computing and memory in a single device by tuning its conductance. The availability of three-terminal devices enables the independent tuning of the NODes, preventing write sneak path issues typical of the two-terminal memristor crossbar array. The NODe conductance relaxes exponentially with a measured time constant of 2.9 h, furthermore, it can be operated at 51 MHz, corresponding to an estimated energy efficiency of 0.1 pJ per multiply-accumulate (MAC) operation. To demonstrate the NODe’s computing capabilities, a
$3\times3$
crossbar array has been successfully used to perform edge detection and blurring on an image with
$128\times64$
pixels.