A Retinomorphic Graphene Transistor Implementing Direction- and Velocity-Selective Computation for In-Sensor Image...

Area of Science:

• Neuroscience
• Materials Science
• Electrical Engineering

Background:

• Biological neural circuits in the retina, specifically direction-selective ganglion cells (DSGCs) and starburst amacrine cells (SACs), perform complex motion detection.
• Replicating these sophisticated neural computations in solid-state hardware remains a significant challenge.

Purpose of the Study:

• To demonstrate a single multi-gate ion-gel graphene transistor (MIG-GFET) capable of emulating the essential functions of the DSGC-SAC microcircuit.
• To achieve intrinsic motion direction and velocity selectivity in hardware without relying on algorithmic post-processing.

Main Methods:

• Fabrication of a MIG-GFET with spatially distributed gates to create artificial dendritic compartments.
• Utilizing ionic dynamics within these compartments for nonlinear summation of sequential inputs.
• Implementing an oppositely biased inhibitory gate to replicate SAC-like null-side inhibition and achieve velocity tuning.

Main Results:

• The MIG-GFET exhibited intrinsic direction selectivity due to nonlinear summation of sequential inputs.
• The device demonstrated a band-pass velocity response with tunable preferred speed, mimicking biological DSGCs.
• The transistor functioned as an analog motion filter, enabling on-sensor image stabilization by suppressing specific frequency jitter with low latency and power.

Conclusions:

• Dendritic-level computation can be effectively embedded within a graphene transistor.
• This approach paves the way for vision hardware that co-localizes sensing and early neural processing at the device level.
• The compact, low-power MIG-GFET is suitable for focal-plane array integration and real-time motion analysis applications.