Pulse stream VLSI circuits and systems: the EPSILON neural network chipset

Area of Science:

• Neuromorphic Engineering
• Integrated Circuit Design
• Artificial Intelligence Hardware

Background:

• Development of specialized hardware for artificial intelligence is crucial for advancing computational capabilities.
• Analogue Very-Large-Scale Integration (VLSI) circuits offer potential for power-efficient neural network implementation.
• Pulse-stream signalling is an emerging method for representing neural information in hardware.

Purpose of the Study:

• To design and fabricate an analogue CMOS VLSI neural processing chip.
• To evaluate the chip's performance in terms of synaptic connection speed and real-world problem-solving.
• To inform the design of a next-generation chip with enhanced features.

Main Methods:

• Design and fabrication of an analogue CMOS VLSI chip.
• Implementation of pulse-stream neural state signalling.
• Characterization of chip performance, including synaptic computation rate.
• Demonstration of the chip's efficacy in solving practical problems.

Main Results:

• Successful design and fabrication of an analogue CMOS VLSI neural processing chip.
• The chip achieves a computation rate of approximately 360 million synaptic connections per second.
• Demonstrated performance in solving real-world problems.
• Experience gained led to the design of an improved second-generation chip.

Conclusions:

• The developed analogue CMOS VLSI chip represents a significant advancement in neural processing hardware.
• The pulse-stream signalling approach enables high-speed computation.
• The insights gained are driving the development of more advanced systems for applications like robotic control.