NeuroPipe-Chip: A digital neuro-processor for spiking neural networks

Area of Science:

• * Computer Engineering
• * Artificial Intelligence
• * Neuroscience

Background:

• * Conventional hardware struggles with real-time computation of Spiking Artificial Neural Networks (SANNs).
• * Existing platforms do not meet the demands for complex SANNs in real-world applications.
• * There is a need for specialized hardware to accelerate SANN processing.

Purpose of the Study:

• * To introduce novel on-chip concepts for accelerating SANN computation.
• * To present the design and simulation of a prototype neuro-processor, NeuroPipe-Chip.
• * To evaluate the performance of NeuroPipe-Chip for SANN acceleration.

Main Methods:

• * Development of a neuro-processor (NeuroPipe-Chip) with two new on-chip acceleration concepts.
• * Implementation of a prototype NeuroPipe-Chip.
• * System simulation using a hardware description language (HDL) for performance evaluation.
• * Fabrication of the NeuroPipe-Chip using Alcatel 0.35-μm digital CMOS technology.

Main Results:

• * NeuroPipe-Chip operating at 100 MHz demonstrated over a two-order-of-magnitude improvement in SANN computation speed compared to an Alpha 500 MHz workstation.
• * The prototype approaches real-time processing for SANNs involving approximately 10^6 neurons.
• * The hardware structure was successfully simulated, validating the design.

Conclusions:

• * NeuroPipe-Chip effectively accelerates the computation of complex Spiking Artificial Neural Networks.
• * The developed neuro-processor enables SANNs to meet real-time requirements for demanding tasks.
• * This accelerator facilitates real-world applications, including real-time image processing, using complex SANNs.