Neuromorphic-based Boolean and reversible logic circuits from organic electrochemical transistors
- PDF / 731,312 Bytes
- 6 Pages / 585 x 783 pts Page_size
- 39 Downloads / 274 Views
oduction An important goal in the advancement of organic neuromorphic circuit (ONC) concepts is to implement organic electrochemical transistors (OECTs)1,2 and related devices in ways that allow for clear design strategies toward generalized computation, while taking advantage of the intrinsic properties of mixedmode conduction in organic electrochemical materials.3–9 A unique aspect of OECTs is their ability to function in a multigate geometry (Figure 1) that mimics neuronal behavior by summing input signals from multiple neighboring devices to determine the conductivity of the active channel,10–12 commonly made from poly(3,4-ethylenedioxythiophene) poly(styrene sulfonate) (PEDOT:PSS). This, coupled with the nonlinear dependence of the channel conductivity on the sum total gate voltage that approximates neuronal activation, makes OECTs a natural choice to mimic neuronal behavior. The question becomes: How can OECT-based neuromorphic circuits be designed in systematic, scalable, and energy- efficient ways given the complexity of connecting devices with more than three terminals? We have proposed a design strategy that takes advantage of the basic neuromorphic functionality of OECTs to build small-scale circuits that carry out Boolean and reversible logic functions, which can then be scaled to perform more complex operations through conventional digital or reversible
logic design.13 It represents an extension of previous, experimental work in which multiple organic field-effect transistors (OFETs) and write-once-read-many (WORM) devices were integrated into a monolithic design in an early demonstration of a polymer-based neuromorphic circuit. This earlier design, which used the OFETs to simulate neural activation, was shown to perform a continuous value classification task aimed at soft robotics.14,15 The strategy proposed here is instead based on OECTs as the fundamental neuronal unit in the circuit, taking advantage of their unique multigate behavior. The strategy involves mimicking a conventional neural network structure with input, hidden, and output layers built with OECT neurons. However, the circuits operate anomalously compared to standard artificial neural networks, in that the connection weights, which are the weighting factors that determine the signal amplitudes between neurons in a neural network, are not fixed quantities, instead, they vary depending on the input values. We demonstrated that such in operando weighting can be effectively used if the bias voltages on the source and drain electrodes are carefully controlled by the circuit design. A key outcome of the approach is implementation of the XOR Boolean function that then easily allows for more complicated circuits and reversible logic functions to be built. The approach also provides a potential pathway to energy-efficient operation, with energy per computation (a critical figure of merit for the
Jake C. Perez, University of Colorado Boulder, USA; [email protected] Sean E. Shaheen, University of Colorado Boulder, USA; sean.shaheen@colorad
Data Loading...
