Realization and training of an inverter-based printed neuromorphic computing system

Area of Science:

• Materials Science
• Computer Engineering
• Neuroscience

Background:

• Rigid silicon electronics face limitations in cost and conformity for emerging applications like soft robotics and wearables.
• Conventional digital logic in printed electronics is constrained by performance, power, and integration density.

Purpose of the Study:

• To propose and demonstrate an inkjet-printed neuromorphic computing architecture as a viable alternative to rigid silicon.
• To overcome the challenge of implementing non-linear elements in printed neuromorphic systems.

Main Methods:

• Development of printed hardware building blocks including inverter-based weight representation and resistive crossbars.
• Integration of printed transistor-based activation functions.
• Creation of a custom learning algorithm tailored for the printed neuromorphic computing architecture.

Main Results:

• Successful demonstration of printed hardware components for neuromorphic computing.
• Implementation of non-linear activation functions using printed transistors.
• Validation of a learning algorithm for training the printed neuromorphic computing system.

Conclusions:

• Inkjet-printed neuromorphic computing offers a promising solution for flexible electronics, bridging the gap between soft materials and high-performance computing.
• The developed printed building blocks and learning algorithm enable the creation of advanced, adaptable systems for next-generation devices.