Researchers at The University of Texas at San Antonio have developed Genesis, a spiking neuromorphic accelerator chip designed to enable on-device continual learning across an AI system's operational lifetime. The chip targets one of artificial intelligence's most persistent technical problems: catastrophic forgetting, the phenomenon in which a neural network loses previously acquired knowledge the moment it learns new information.

Consider a security drone trained over months to detect wildfire smoke in a dense forest. Reassign that drone to a coastal flood-monitoring mission, and the instant it learns to interpret the new imagery it may lose its ability to detect forest fires entirely. Genesis addresses this failure mode in hardware.

Dhireesha Kudithipudi, PhD, leads the project. She is the founding director of UT San Antonio's MATRIX AI Consortium and the Neuromorphic Artificial Intelligence Laboratory (NUAI Lab) in the College of AI, Cyber and Computing. Her team designs specialized silicon that mimics the brain's capacity to learn continually without overwriting prior knowledge. What distinguishes the effort is a close collaboration with theoretical and computational neuroscientists, whose understanding of the biological mechanisms of learning and memory informs the hardware design.

Metaplasticity in silicon

The design principle at the core of Genesis is metaplasticity — the brain's regulation of how readily its connections change. Not all synapses are treated equally. Junctions between neurons that fire frequently become resistant to modification, while less-used connections remain available for new learning.

Genesis replicates variants of this behavior on silicon. Each processing element on the chip tracks not only its current use but also its history: how much it has contributed and how often it has fired. High-importance connections resist overwriting. The system routes new learning toward connections that remain flexible. The result is a device that accumulates knowledge rather than replacing it.

Energy consumption

The team also employs spiking neural networks (SNNs), which process information in pulses similar to biological neurons and rest when no tasks are pending. This approach consumes far less energy than the standard artificial neural networks used in modern large language models.

Genesis saves additional power through a custom data movement strategy. In a conventional chip, moving data between memory and processors creates a bottleneck that wastes power. Genesis uses a shortcut that stores and accesses all information required for learning in one place. The chip is still in the testing phase, but the researchers project it could consume 30 to 100 times less energy than traditional hardware.

Built for the edge

Genesis targets deployments without data centers or reliable cloud connectivity. It runs on milliwatts — a power envelope suited to implantable medical devices, field-deployed drones, and wearable sensors that must keep learning for years without a recharge or reset. The architecture is intended for cumulative workloads where system performance improves the longer the device operates.

The chip did not arrive in a single iteration. Doctoral students and postdoctoral fellows Vedant Karia, Fatima Tuz Zohora, Abdullah M. Zyarah and Nicholas Soures worked through two earlier prototypes over a five-year period before arriving at the current Genesis architecture. Alongside the hardware, the team developed the learning algorithms, the framework for understanding and scaling the AI agent, and MetaplasticNet, a brain-inspired neural network architecture. All components were engineered together to support the chip's energy efficiency by keeping power usage low.

"We have spent years building the theoretical and hardware foundations for how systems can learn continuously the way the brain does without erasing what came before," said Kudithipudi. "Genesis is proof that this isn't just a concept anymore. UT San Antonio is setting the precedent for students solving these problems by building the actual chips, and I expect more doctoral and post-doctoral scientists to pursue this specialization on our campus."

Fabrication and funding

The chips are fabricated through a partnership with SUNY Albany using IBM's 65nm technology. The research effort is supported by a multi-million-dollar, five-year grant from the Air Force Research Laboratory (AFRL), underlining UT San Antonio's position in AI hardware innovation, semiconductor design and energy-efficient intelligent systems.

The work connects to a broader national push as AI permeates daily life. From medical devices to autonomous vehicles, the Genesis architecture points toward systems that build on accumulated knowledge over many years at a fraction of the energy cost of other AI learning models.

For Kudithipudi, Genesis is not the endpoint. The next phase of the team's work focuses on scaling the metaplasticity mechanisms, preparing them for real-world deployment, and integrating them with new hardware and software to increase efficiency and capability.