Four levels of in-sensor computing in bionic olfaction: from discrete components to multi-modal integrations.

Area of Science:

• Materials Science
• Computer Engineering
• Biotechnology

Background:

• Sensing and computing enable digital interaction with the analog world via analog-to-digital converters (ADCs) and data buses.
• Increasing sensor nodes and deep neural networks amplify energy and time consumption, limiting data throughput.
• In-sensor computing offers a paradigm shift by integrating sensing, storage, and processing within a single device, bypassing ADCs and data transfer.

Purpose of the Study:

• To review and categorize four levels of in-sensor computing integration for artificial olfactory applications.
• To explore advancements in in-sensor computing for enhanced artificial olfaction.
• To provide an outlook on utilizing metal nanoparticle devices for bionic olfaction.

Main Methods:

• Categorization of in-sensor computing based on integration degree in artificial olfaction.
• Review of discrete component functions, in-memory computing architectures, single-device integration, and multi-modal approaches.
• Exploration of metal nanoparticle devices for future bionic olfaction.

Main Results:

• Four distinct levels of in-sensor computing integration were identified, progressing from discrete components to multi-modal systems.
• In-memory computing architectures exempt data conversion and transfer, while single-device integration streamlines functionality.
• Multi-modal in-sensor computing enhances classification accuracy and reliability in artificial olfaction.

Conclusions:

• In-sensor computing significantly reduces energy and time consumption in artificial olfaction systems.
• The progression of integration levels demonstrates increasing efficiency and capability.
• Metal nanoparticle devices show promise for realizing advanced in-sensor computing in bionic olfaction.