A multichip aVLSI system emulating orientation selectivity of primary visual cortical cells

Area of Science:

• Neuroscience
• Computer Engineering
• Artificial Intelligence

Background:

• The primary visual cortex contains simple cells that exhibit orientation selectivity.
• Mimicking these biological neural processes is crucial for developing advanced artificial vision systems.

Purpose of the Study:

• To design and fabricate a multichip neuromorphic analog very large scale integrated (aVLSI) system.
• To emulate the orientation selective response of simple cells in the primary visual cortex.

Main Methods:

• A system comprising a silicon retina and an orientation chip was designed and fabricated.
• The silicon retina filters images using a concentric center-surround receptive field.
• Analog signals transfer image data to the orientation chip, which aggregates pixels mimicking the Hubel and Wiesel feedforward model.

Main Results:

• The orientation chip provides orientation-selective (OS) outputs tuned to 0, 60, and 120 degrees.
• Feed-forward aggregation effectively reduces fixed pattern noise caused by transistor mismatch.
• Spatial properties of the orientation selective response were analyzed based on adjustable chip parameters.

Conclusions:

• The developed multichip aVLSI architecture successfully emulates simple cell orientation selectivity.
• This architecture can be extended to implement higher-order visual cortex cells, such as complex cells.
• The system offers a robust platform for advancing neuromorphic engineering and artificial vision research.