Implementation of the Revised Pavlov’s Associative Learning Circuit to Avoid Parasitic Training Effects

Abstract

To enable organisms to learn behaviors from past experiences, biological systems rely on associative learning, a fundamental mechanism that forms connections between simultaneous events. Developing an electrical counterpart to model this associative learning phenomenon could be highly beneficial for edge AI applications. This study focuses on Pavlovian learning, an associative process in dogs where the sound of a bell becomes linked to food, leading to drooling as a response based on memory. Existing circuit models that attempt to replicate this phenomenon face limitations, especially in distinguishing between different stimuli types and often exhibiting a false learning effect without genuine learning. Consequently, these circuits only demonstrate associative learning under specific signal sequences, failing to do so with other configurations. To address this, we propose a novel circuit architecture using non-volatile memory components, designed to overcome sequencing limitations and enable true associative learning regardless of stimulus order. Simulation results obtained through PSPICE using 0.18 µm CMOS technology validate the functionality of the proposed circuits.