UT San Antonio team designs Genesis chip to curb AI forgetting

Researchers at UT San Antonio’s MATRIX AI Consortium have developed Genesis, a spiking neuromorphic accelerator chip designed to address catastrophic forgetting, a problem in which AI systems lose previously learned abilities when trained on new tasks. The chip is intended to support on-device continual learning across its operational lifetime, enabling autonomous systems to keep improving without depending on cloud infrastructure.

Genesis draws on metaplasticity, a brain-inspired principle that regulates how readily connections change. Each processing element tracks its use, contribution and firing history, allowing important connections to resist overwriting while new learning is directed toward more flexible parts of the system. The team is also using spiking neural networks, which process information in pulses and remain idle when no task is active.

The design targets energy-constrained uses such as implantable medical devices, field-deployed drones and wearable sensors. Testing suggests the chip could consume 30 to 100 times less energy than traditional hardware. The current Genesis architecture followed earlier prototypes and is supported by related learning algorithms, a scaling framework and MetaplasticNet. The chips are fabricated through a partnership with SUNY Albany using IBM’s 65nm technology.