Neuromorphic computing

Main researcher: Javier del Valle

Neuromorphic computing is an emerging field that takes inspiration from biological brains and seeks to create new electronic devices that mimic the functionalities of neurons and synapses [1]. One of the most promising approaches is based on a phenomenon known as resistive switching, in which the resistance of a material can be changed by applying a voltage [2-4]. We study the two basic types of resistive switching: volatile and non-volatile.

Volatile switching is observed in correlated oxides that feature an insulator-to-metal transition [5], and can be used to reproduce neuron spiking [6-8]. Non-volatile switching is ideal for mimicking synapses and it has even been used to perform pattern classification [9,10]. It is observed in most oxides and takes place when oxygen vacancies drift under strong electric fields [11,12].

In Geneva, we fabricate resistive switching devices to study the underlying mechanisms that govern this phenomenology and, using tools such as ion irradiation, try to find new ways of manipulating it.

Voltage spiking in VO2
With proper device design, the metal-insulator transition in VO2 can be engineered to produce a spiking pattern in the MHz regime.

[1]. G. Indiveri and S.-C. Liu, Proc. IEEE. 103, (2015)

[2]. J. J. Yang et al, Nat. Nanotechnol. 8, (2013)

[3]. J. del Valle et al, J. Appl. Phys. 124, (2018)

[4]. Y. Zhou and S. Ramanathan, Proc. IEEE. 103, (2015)

[5]. J. del Valle et al, Nature. 569 (2019)

[6]. J. del Valle et al, Sci. Rep. 10, (2020)

[7]. M. D. Pickett et al, Nat. Mater. 12, (2013)

[8]. M. Ignatov et al, Front. Neurosci. 9, (2015)

[9]. M. Prezioso et al, Nature. 521, (2015)

[10]. I. Boybat et al, Nat. Commun. 9, (2018)

[11]. R. Waser and M. Aono, Nat. Mater. 6, (2007)

[12]. A. Beck et al, Appl. Phys. Lett. 77, (2000)