Neuromorphic Computing for AGI: A Systematic Review Beyond GPU Limitations

Solorzano B.M.; Rivasplata V.V.; Portugal E.S. · Proceedings of the LACCEI international Multi-conference for Engineering, Education and Technology (2026) · DOI: 10.18687/laccei2026.1.1.1050

Resumen

This paper presents a Systematic Literature Review (SLR) analyzing the role of neuromorphic computing in advancing Artificial General Intelligence (AGI) beyond the limitations of GPU-based systems. Following the PICOC framework and PRISMA 2020 guidelines, 71 records were retrieved from Scopus (2024-2025), of which 36 full-text articles were selected after rigorous application of inclusion and exclusion criteria. The review is structured around four research questions examining chip architecture evolution, hardware comparisons with conventional accelerators, AGI-oriented applications, and reported performance outcomes. Key findings include energy improvements of 13×-49× in segmentation tasks, 50× device-level energy reduction via spin-neuron scaling, competitive accuracies of 92-97% for SNN/ANN/MPNN models, and strong adaptive behavior in reinforcement-learning-based cognitive control. The paper concludes that neuromorphic computing represents a promising and biologically plausible pathway toward sustainable AGI, while acknowledging speed-accuracy trade-offs, hardware variability, and tooling fragmentation as current limitations.

Instituciones: Universidad Tecnológica del Perú