Unforgetting AI Chip Genesis Developed

Researchers at the MATRIX AI Consortium of the University of Texas at San Antonio have developed a chip named Genesis, which enables artificial intelligence systems to learn new information without losing old data.

Artificial Intelligence and the Problem of Forgetting

One of the most persistent problems in artificial intelligence systems is the ability to accumulate new information without losing what was previously learned.

This situation poses an obstacle that results in systems continuously writing new data over old data.

Working Principle of the Genesis Chip

The team led by Dhireesha Kudithipudi designed special hardware that mimics the lifelong learning capability of the human brain structure.

Based on the principle of metaplasticity, this hardware prevents the deletion of old information while ensuring that new information is directed to flexible connections.

Energy Efficiency and Areas of Use

The system offers energy efficiency by using synaptic neural networks that process information in pulses similar to biological neurons.

Consuming thirty to one hundred times less energy than traditional hardware, this chip operates at millisecond levels, making it suitable for edge computing devices.

Supporting Institutions and Production

This research work is supported by a five-year grant from the Air Force Research Laboratory.

The chips were manufactured using IBM's sixty-five nanometer technology in partnership with SUNY Albany.