Nanoscale photonic artificial neuron with biological signal processing

Area of Science:

• Optoelectronics
• Artificial Intelligence
• Neuroscience

Background:

• Artificial intelligence (AI) infrastructure faces unsustainable energy demands.
• Photonic approaches offer speed and low energy but suffer from large footprints and lack biological functions like inhibition.
• Scaling neuromorphic hardware requires reduced footprints and enhanced functionality.

Purpose of the Study:

• To report a novel nano-optoelectronic artificial neuron.
• To address limitations of existing photonic neuromorphic solutions, specifically footprint size and biological function integration.
• To enable efficient photonic neuromorphic computing and adaptive optical sensing.

Main Methods:

• Development of a nano-optoelectronic artificial neuron device.
• Integration of excitatory and inhibitory inputs with deterministic control.
• Implementation of neural weighting via tunable input gains.
• Utilizing commercial silicon technology and multi-wavelength operation.

Main Results:

• Achieved at least 100-fold reduction in circuit footprint compared to existing solutions.
• Demonstrated picowatt-level operating power consumption.
• Device performs nonlinear transfer operations and exhibits biologically relevant temporal dynamics.
• Tunable input gains enable controlled summation and thresholding for neural weighting.

Conclusions:

• The developed artificial neuron offers a significant reduction in footprint and power consumption for neuromorphic hardware.
• The device integrates key biological functions, including inhibition, crucial for advanced AI.
• This technology paves the way for compact, low-power photonic neuromorphic computing and versatile adaptive optical sensing applications.