Neuromorphic Devices for Extreme Environments: State‐Variable Reliability and Stabilization Strategies

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Neuromorphic devices are increasingly considered for edge‐intelligence systems operating in spaceborne, aerospace, nuclear, cryogenic, and other harsh environments, where local decision‐making is constrained by limited power, communication bandwidth, and latency. Reliable operation in these environments requires more than the preservation of basic electrical functionality because neuromorphic computation is physically encoded in material and interfacial state variables. Environmental stress can reshape these variables and alter analog modulation, plasticity, memory retention, threshold dynamics, and readout stability. This review discusses neuromorphic devices for extreme environments from the perspective of state‐variable reliability and stabilization. Radiation, temperature extremes, and vacuum or low‐pressure conditions are examined as major stressors that perturb defect generation, ionic and electronic transport, interfacial chemistry, switching pathways, and signal readout. Emerging space‐relevant factors, including microgravity‐assisted fabrication and mechanical loading, are also considered with respect to device formation, structural integrity, and long‐term reliability. Stabilization strategies are organized across material‐level design, environmental boundary control, and programming or readout compensation. By connecting device‐level degradation to neuromorphic metrics, array operation, soft‐error response, and inference accuracy, this review provides a framework for evaluating and designing reliable neuromorphic hardware under realistic environmental constraints.
Funding Information
  • National Research Foundation of Korea (RS‐2024‐00438999)
  • National Research Foundation of Korea (RS‐2026‐25493157)
  • National Research Foundation of Korea (RS‐2026‐25540048)
  • National Research Foundation of Korea (RS‐2026‐25493690)

This publication has 108 references indexed in Scilit: