Forty Tested Neuromorphic Capabilities on BrainChip Akida Hardware
Year: 2026
Type: preprint
Abstract: Forty brain-inspired capabilities were built and tested on BrainChip Akida AKD1500 neuromorphic processors, working alone and in combination with Tang Primer 20K FPGAs and host computers. The capabilities fall into four families. Memory includes fast weights, associative recall, sparse distributed memory, hippocampal indexing, multi-timescale synapses and a working memory held in the wiring of the network itself. Plasticity and learning include structural plasticity, metaplasticity, three-factor learning, predictive coding, dendritic segments, sleep, neurogenesis, neuroevolution, learning to learn, one-shot place fields and exact perturbation learning. Dynamics and computation include neuromodulation, neural fields, two-timescale control, oscillatory gating, event-driven dendrites, hyperdimensional binding, Boltzmann sampling, path integration, a dendritic XOR and an apical switch. A final family covers self-reference across chips: self-repair, a rewindable network, reflexive plasticity, introspection, exact unlearning, unihemispheric sleep, a split brain, dreaming by recombination, neural Darwinism and a causal self-map. Each capability was built and run on silicon, and every chip output used was checked against an exact integer reference of the same computation, with zero mismatches. All forty work, several within stated bounds. Each result is paired with a potential application, from verifiable data deletion and self-monitoring confidence signals to dead reckoning without GPS and fault-tolerant edge hardware.
Author: Johnson, Kevin D.
Institution: Arizona State University
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Keywords: neuromorphic hardware, neuromorphic computing, associative memory, sparse distributed memory, structural plasticity, metaplasticity, predictive coding, neuroevolution, meta-learning, place fields, neuromodulation, implicit neural representations, Boltzmann sampling, path integration, self-reference, self-reflection, unlearning, unihemispheric sleep, split-brain, neural Darwinism, self-monitoring, dead reckoning
Open Access status: green
OpenAlex ID: W7234697377
