REPROGRAMMABLE HARDWARE FOR DATA PROCESSING AT THE EDGE : A NEW COMPUTING PARADIGM BASED ON NEUROMORPHIC SYSTEMS
The brain’s ability to perform efficient and fault-tolerant data processing is strongly related with its peculiar interconnected adaptive architecture, based on redundant neural circuits interacting at different scales. By emulating the brain’s processing and learning mechanisms, computing technologies strive to achieve higher levels of energy efficiency and computational performance. [1] Although efforts to address neuromorphic solutions through hardware based on top-down CMOS-based technologies have obtained interesting results in terms of energetic efficiency improvement, the replication of brain’s self-assembled and redundant architectures is not considered in the roadmaps of data processing electronics.
In materia computing has been proposed as competitive and alternative strategy, exploiting the complexity and collective phenomena originating from various classes of physical substrates, to perform data processing.[2] In particular, the employment of random-assembled memristive materials is a strategic solution for the development of energy efficient and neuromorphic computing devices, well beyond von Neuman bottleneck.[3] In this contest, films obtained by the assembling of metallic nanoparticles and, in particular, Au cluster-assembled films have shown interesting non-linear electrical properties and complex resistive switching phenomena.[4,5]
Here the implementation of data processing devices based on nanostructured thin films is explored. Reversible electronic switches and reprogrammable threshold logic gates (TLGs) have been implemented on both hard[6] and soft substrates.[7] An experimental strategy, based on micro-thermography, has been exploited for the study of the spatial and temporal dynamic of the resistive switching activity of the nanostructured gold network. The adaptive reorganization of the network, at different scales, has been described and a network correlation coefficient of the local activities has been here proposed.
We also demonstrated the potentiality of the use of these resistive switching devices to classify with high accuracy and in real-time neuronal traces corresponding to physiological and evoked spiking activity recorded from the barrel cortex of a rat. The classification was carried out with a linear classifier, requiring limited datasets for training and limited memory storage, and characterized by higher interpretability and accuracy with respect to artificial neural networks.
The hardware implementation of neuromorphic material, characterized by nonlinear dynamic and memory mechanisms, as also complex and redundant morphological network, is proposed for the development of novel electronic architectures, and energy efficient reconfigurable data processing devices. These results constitute a promising milestone for a fruitful combination of physical and computing intelligence directly integrated on edge systems to efficiently interact with the environment.
References
[1] F. Borghi, T. R. Nieus, D. E. Galli, P. Milani, Front. Neurosci. 2024, 18, DOI 10.3389/fnins.2024.1465789.
[2] H. Jaeger, Neuromorph. Comput. Eng. 2021, 1, 012002.
[3] A. Vahl, G. Milano, Z. Kuncic, S. A. Brown, P. Milani, J. Phys. D : Appl. Phys. 2024, 57, 503001.
[4] M. Mirigliano, F. Borghi, A. Podestà, A. Antidormi, L. Colombo, P. Milani, Nanoscale Advances 2019, 1, 3119.
[5] F. Borghi, M. Mirigliano, D. Dellasega, P. Milani, Applied Surface Science 2022, 582, 152485.
[6] G. Martini, M. Mirigliano, B. Paroli, P. Milani, Jpn. J. Appl. Phys. 2022, 61, SM0801.
[7] G. Nadalini, A. Dallinger, D. Sottocorno, F. Greco, F. Borghi, P. Milani, Advanced Electronic Materials n.d., n/a, 2400717.
