A new wide range Euclidean distance circuit for neural network hardware implementations

Area of Science:

• Neuromorphic Engineering
• Analog Circuit Design
• VLSI Systems

Background:

• Synapse circuits are fundamental components in artificial neural networks.
• Efficient Euclidean distance computation is essential for synapse functionality.
• Existing circuits often face limitations in size, power consumption, or operational range.

Purpose of the Study:

• To design and implement a compact analog very large-scale integration (VLSI) circuit for wide-range Euclidean distance computation.
• To optimize the circuit for high synapse density, wide operational range, and low power consumption.
• To integrate this circuit into a self-organizing map (SOM) for a bio-inspired vision system.

Main Methods:

• Analog VLSI implementation of a wide-range absolute value circuit.
• Fabrication using 1.5-/spl mu/m technology via MOSIS.
• Simulation and experimental validation of circuit performance.
• Integration into a hardware implementation of a self-organizing map (SOM).

Main Results:

• A compact circuit (80 x 76 /spl mu/m) was successfully fabricated.
• The circuit operates over a wide range (0.5 to 4.5 V) with low power consumption (< 200 /spl mu/W).
• Simulated and experimental results demonstrate the circuit's effectiveness.
• The circuit is suitable for high-density synapse implementations.

Conclusions:

• The developed analog VLSI circuit provides an efficient solution for Euclidean distance computation in synapse circuits.
• This technology enables high-density hardware implementations of self-organizing maps (SOMs).
• The SOM, utilizing this circuit, is a key component for advanced bio-inspired vision systems.