An analog silicon retina with multichip configuration

Area of Science:

• Neuroscience
• Computer Engineering
• Integrated Circuit Design

Background:

• Neuromorphic silicon retinas emulate biological retinal circuits using analog very large scale integrated circuits.
• Previous designs suffered from low spatial resolution and computational variability, limiting applications.
• Sample/hold circuits were previously integrated to suppress output fluctuations.

Purpose of the Study:

• To develop a multichip silicon retina with enhanced spatial resolution and reduced computational variability.
• To demonstrate real-time image spatial filtering capabilities.
• To propose a novel architecture for analog neuromorphic multichip systems.

Main Methods:

• Fabricated a multichip silicon retina separating photoreceptor (P chip) and horizontal cell (H chip) networks.
• Implemented analog voltage transfer between chips via a line-parallel transfer bus.
• Integrated sample/hold circuits in both P and H chips to compensate for pattern noise.
• Utilized an off-chip differential amplifier for real-time spatial filtering.

Main Results:

• Achieved real-time spatial filtering with orientation-selective receptive fields.
• Demonstrated successful compensation for pattern noise across analog communication pathways.
• The multichip architecture effectively addressed limitations of previous single-chip designs.

Conclusions:

• The multichip silicon retina offers improved performance for neuromorphic vision systems.
• The analog data transfer method is suitable for hierarchical neuromorphic architectures.
• This approach facilitates the design of systems that mimic the visual system's structure and function.