CogniGron Seminar: Matteo Mirigliano "The RECEPTRON: a Boolean classifier based on multi-electrode cluster-assembled Au films with non-local and non-linear electrical conduction properties"

High-complexity tasks like pattern recognition are a very difficult challenge for digital computing. The use of electronic devices with a deterministic behaviour and the need of rigidly specified instructions to implement the desired tasks represent major obstacles. This essentially limits the speed capability and the amount of processed information due to high number of instructions and the high integration of electronic components in a single chip needed to achieve the desired computing performances. A possible approach to overcome the limitations of digital technologies is represented by the so-called unconventional computing technologies (UCTs). UCTs are based on the exploitation of the physical properties of a broad class of complex nanostructured materials that mimic the complex network topology of neurons in the brain in terms of nanoscale junctions with non-linear electrical properties. The challenge consists in finding reliable fabrication process and the way to perform information processing exploiting resilient and self-organized materials avoiding pre-determined constraints on the elaboration and precisely designed devices. Recently we reported that continuous cluster-assembled Au films produced by supersonic cluster beam deposition, show anomalous electric and resistive switching behaviour with non-local and correlation properties [1-3]. The results suggest the possibility of fabricating neuromorphic devices based on cluster-assembled gold films. Here, I will discuss the emergence of this anomalous conduction properties from the complex structure achieved at the nanoscale, resulting from the random stacking of differently shaped crystalline clusters directly connected by junctions of different cross sections with an extremely high number of defects and grain boundaries. I will present one possible approach to exploit the electrical conduction properties of cluster-assembled Au films to implement a Boolean classifier. The device processes multiple input electrical signals thanks to non-local and non-linear electrical conduction properties avoiding the use of training protocols. The advantages of the device and the perspectives opened by the presented approach will be discussed [4].

[1] M. Mirigliano, D. Decastri, A. Pullia, D. Dellasega, A. Casu, A. Falqui, and P. Milani, Nanotechnology 31, 234001 (2020).
[2] M. Mirigliano, F. Borghi, A. Podestà, A. Antidormi, L. Colombo, and P. Milani, Nanoscale Adv. 1, 3119 (2019).
[3] M. Mirigliano, S. Radice, A. Falqui, A. Casu, F. Cavaliere, and P. Milani, Sci. Rep. 10, 1 (2020).
[4] M. Mirigliano, B. Paroli, G. Martini, M. Fedrizzi, P. Milani, “Classification of Boolean Functions with a Reconfigurable Dense Network of Metallic Nanojunctions”, submitted