a-t-0
commented
1 year ago Neuron death: THE EFFECTS OF RADIATION ON MEMRISTOR-BASED ELECTRONIC SPIKING NEURAL NETWORKS Radiation in such cases can lead to neuron death due to circuit failure (CMOS threshold shift, oxide breakdown, gate rupture, displacement damage [176], [177]) in the SNN.
E.g. for Mars P168 of source:
Sources
https://www.semanticscholar.org/paper/Evaluation-of-Radiation-Effects-in-RRAM-Based-for-Ye-Liu/028ce71eebaf83055d9eb120b7d286dc0b8bfe31 Says: With ions randomly striking the network, the weights (conductance) of RRAM devices change according to the ion LET. By using the SPICE simulation method presented in Section II, the conductance tuning of RRAM devices is observed so that the updated weight patterns are obtained at specific irradiation conditions. Then, we rerun the MLP network simulation with the updated weight patterns.
https://www.semanticscholar.org/paper/Neutron-Induced%2C-Single-Event-Effects-on-Vision-A-Roff%C3%A9-Akolkar/1be9e6677d0b7f5bf45c6da730013beb734dc28c https://www.semanticscholar.org/paper/Neutron-Induced%2C-Single-Event-Effects-on-Vision-A-Roff%C3%A9-Akolkar/b48d2639036b0f5ec6d95cca2f76804df145cf6f https://www.semanticscholar.org/paper/Radiation-Effects-on-LiNbO-%24_2%24-Memristors-for-Greenlee-Shank/4a7a288a6a71ce28ba75e0cdee0420e5630e9ced https://www.semanticscholar.org/paper/Radiation-Effect-on-Learning-Behavior-in-Circuit-Dahl-Ivans/6a350cb81d31dee8071f78f9f3c083858a166d7d Says: It is found that although radiation changes the synaptic weights, the network is able to recover and relearn the pattern after radiation ceases. Recovery time is proportional to flux, intensity, and duration of the radiation. If the network is exposed to radiation events of low intensity and flux even for an indefinite period (as observed in space), it may be able to retain its pattern recognition capabilities.
And: Literature examines the effect of different types and intensity of radiations on memristive devices with different active material and structure [16]–[20].
16] E. Deionno, M. D. Looper, J. V. Osborn, H. J. Barnaby, and W. M. Tong, “Radiation effects studies on thin film TiO2 memristor devices,” IEEE Aerosp. Conf. Proc., pp. 1–8, 2013. [17] W. M. Tong et al., “Radiation hardness of TiO2 memristive junctions,” IEEE Trans. Nucl. Sci., vol. 57, no. 3 PART 3, pp. 1640–1643, 2010. [18] M. J. Marinella et al., “Initial Assessment of the Effects of Radiation on the Electrical Characteristics of Memristive Memories,” Nucl. Sci. IEEE Trans., vol. 59, no. 6, pp. 2987–2994, 2012. [19] Y. Gonzalez-Velo, H. J. Barnaby, and M. N. Kozicki, “Review of radiation effects on ReRAM devices and technology,” Semicond. Sci. Technol., vol. 32, no. 8, 2017. [20] E. Deionno, M. D. Looper, J. V. Osborn, and J. W. Palko, “Displacement damage in Tio2 Memristor devices,” IEEE Trans. Nucl. Sci., vol. 60, no. 2, pp. 1379–1383, 2013.
And:
The network and its components were designed and simulated in Cadence Virtuoso Spectre using Verilog-A. The literature presents a few of the experimental radiation studies performed on memristive devices. The results are inconclusive due to the variables like radiation source energy and exposure trajectory, the thickness of active and other layers, and device shielding and packaging techniques used. However, TiO2 devices consistently show an increase in the synaptic state of the device in the presence of x-ray, alpha or proton irradiation [16]–[18]. This state-altering radiation behavior was modeled directly in the non-linear memristor model using radiation current as Irad_sc, more model details can be found in [22].
[22] S. G. Dahl, R. Ivans, and K. D. Cantley, “Modeling Memristor Radiation Interaction Events and the Effect on Neuromorphic Learning Circuits,” in Proceedings of the International Conference on Neuromorphic Systems - ICONS ’18, 2018, pp. 1–8.
And:
B. Quantifying Radiation The radiation model presented in [22] and used in this study is agnostic to the type of memristor material, fabrication process, process variation, and the radiation source. For example, 10, 20, 25, and 30 current pulses of 50 μA (magnitude) induced 30%, 77%, 90%, and 95% change in resistance (from off state) of the memristor in the model, respectively. Similar changes in resistance are experimentally presented in [20] , [16], [23] and [18] using a total fluence of 7.7x1015 350-keV proton/cm2, 1.4x1011 1-MeV alpha/cm2, 4.9x1012 14.1-MeV neutrons/cm2, and 7.75x1016 10-keV x-rays/cm2 respectively. Thus, if the effect of radiation on the desired memristive device is known, it can be simulated using the given model by modifying a few parameters. Increase in mean current magnitude and interval (frequency) of radiation pulses increases the change in weight of the memristive device [22]. Flux calculations in the study are based on 100 nm x 100 nm interaction size memristive devices. https://www.semanticscholar.org/paper/Modeling-Memristor-Radiation-Interaction-Events-and-Dahl-Ivans/1785b4af0ebfe1bb177aab2a02d72c325e49dd19
https://elicit.org/ https://www.humata.ai/