Neuromorphic object localization using resistive memories and ultrasonic transducers

Area of Science:

• Neuromorphic Engineering
• Bio-inspired Computing
• Sensory Processing

Background:

• Real-world applications demand compact, low-latency, low-power computing.
• Hybrid memristive-Complementary Metal-Oxide Semiconductor (CMOS) neuromorphic architectures excel at in-memory, event-driven computing.
• These architectures are ideal for efficient sensory processing.

Purpose of the Study:

• To demonstrate a full-potential, end-to-end sensory processing solution.
• To develop a bio-inspired, event-driven object localization system.
• To leverage neuromorphic resistive memories for computational mapping.

Main Methods:

• Coupled piezoelectric micromachined ultrasound transducer sensors with a neuromorphic resistive memories-based computational map.
• Fabricated a system including resistive memories-based coincidence detectors and delay line circuits.
• Utilized experimental results to calibrate system-level simulations for performance estimation.

Main Results:

• Presented measurement results from the fabricated end-to-end object localization system.
• Calibrated simulations using experimental data to estimate angular resolution and energy efficiency.
• Demonstrated orders of magnitude greater energy efficiency compared to microcontroller-based systems.

Conclusions:

• The proposed bio-inspired, event-driven neuromorphic system shows significant potential for real-world object localization.
• Hybrid memristive-CMOS architectures offer a highly energy-efficient hardware substrate for sensory processing.
• This approach paves the way for more efficient edge computing solutions.