|
Memristive technologies in spatial artificial intelligence architectures: A comprehensive review of advantages, bottlenecks, and mitigation strategies
|
Authors: Haoyue Deng, Huiqin Ma, Tuo Shi, Xiaoxin Xu
Status: Final
Date of publication: 30 September 2026
Published in: ITU Journal on Future and Evolving Technologies, Volume 7 (2026), Issue 3, Pages 464-476
Article DOI : https://doi.org/10.52953/DSXX8218
|
Abstract:
Space-based Artificial Intelligence (AI) and orbital edge computing are constrained by the physical limitations of traditional memory devices. Conventional architectures face the dual challenges of the von Neumann "memory wall" and strict operational requirements of space environments regarding energy, thermal dissipation, and radiation tolerance. Memristive In-Memory Computing (IMC) and In-Sensor-Memory Computing (ISMC) address these challenges by co-locating data storage and executing analog Matrix Vector Multiplication (MVM) within crossbar arrays. This review evaluates memristive technologies tailored for spatial AI architectures. We compare emerging Non-Volatile Memory (NVM) technologies under space environments based on their radiation tolerance and power consumption. Based on these trade-offs, Resistive Random Access Memory (RRAM) is highlighted as a primary option for orbital edge computing hardware. We discuss its structural radiation-tolerance mechanisms and low-power characteristics (exceeding 10 TOPS/W) across critical aerospace applications, including on-board hyperspectral image processing, autonomous deep-space asteroid navigation, and satellite cognitive radio orchestration. Despite these characteristics, implementing RRAM in flight-ready hardware is hindered by physical and architectural non-idealities. We trace accuracy degradation pathways from device-level variations to array-level parasitic and thermal constraints. To address these constraints, we present hardware-software co-design strategies that combine One-Selector-One-Memristor (1S1R) topologies and vertical thermal conduits (TSVs) with algorithmic compensations, including closed-loop write-verify schemes, bit-slicing, ensemble averaging, and dynamic Analog-to-Digital Converter (ADC) reference scaling. Based on these developments, we also provide an outlook on deploying neuromorphic processors in future orbital and deep-space missions.
|
Keywords: Edge computing, in-memory computing, RRAM, space environments, spatial AI
Rights: © International Telecommunication Union, available under the CC BY-NC-ND 3.0 IGO license.
|
|
| ITEM DETAIL | ARTICLE | PRICE | |
|---|
ENGLISH 
Full article (PDF) |
|
| 0
| Free of charge | DOWNLOAD |
|
|
| |