Integrating polycrystalline VO2 films for neuromorphic computing applications
Siegfried Karg a, Nele Harnack a, Olivier Maher a, Bernd Gotsmann a
a IBM Reseach Europe - Zurich
Proceedings of Materials, devices and systems for neuromorphic computing 2022 (MatNeC22)
Groningen, Netherlands, 2022 March 28th - 29th
Organizers: Jasper van der Velde, Elisabetta Chicca, Yoeri van de Burgt and Beatriz Noheda
Poster, Siegfried Karg, 019
Publication date: 23rd February 2022

The structural phase transition of vanadium dioxide (VO2) and its associated pronounced insulator-to-metal transition (IMT) has gained considerable attention in the scientific community. Due to the significant changes of the electrical, thermal, and optical properties at a transition temperature close to room temperature, VO2 is an appealing material for electronics, photonic devices as well as many other exciting applications [1]. Very compact and energy-efficient electronically triggered relaxation oscillators can be fabricated exploiting the crystal-phase transition of VO2 [2]. These are perfectly suited to build oscillating neural networks for neuromorphic computing [3,4].

Research efforts have shown that epitaxial VO2 layers can be grown on crystalline substrates such as sapphire. However, the phase transition suffers on technologically relevant CMOS-compatible substrates, including silicon-oxide (glass) and silicon. The limitation in substrate choice has so far restricted VO2’s potential implementations despite its unique and versatile characteristics. A deeper understanding of the interplay between substrate, stoichiometry, nanostructure, and nature of the phase transition is required to successfully integrate this material on a CMOS platform and compete in the field of neuromorphic computing.

Our team at IBM has fabricated electrical oscillator-based arrays relying on the IMT of VO2 on a silicon platform and demonstrated neuromorphic computing capabilities through phase relations between restively coupled oscillators in a neural network [3, 4]. These results motivate the desire to push further the exploration of CMOS technology cointegration. An atomic-layer deposition (ALD) technique is used to form a smooth amorphous vanadium oxide film sitting on a SiO2-coated silicon wafer. A post-deposition annealing step carried out under oxidized atmosphere conditions converts the ALD-deposited films into a defined oxidation state of the VO2. We tested a series of conditions adapting the pressure, the temperature along with several other parameters, in both a low-oxygen pressure heat chamber with slow heating and cooling ramps and a millisecond flash-lamp annealer. VO2 stoichiometry was confirmed using micro-Raman analysis. The morphology of VO2 layers can be varied in a large range spanning from isolated crystallites over polycrystalline, granular films to almost amorphous layers. Correspondingly, large variations of the IMT properties can be observed.

Thanks to our in-house developed scanning thermal-microscopy (SThM) technique [5], one is able to visualize and quantify the thermal properties and conductance paths of operating devices. The scanning force microscopy-based approach provides topology and temperature at nanoscale resolution. This allows for the correlation of the microscopic nature of granular VO2 films with the electrical and thermal properties of the IMT measured. Filamentary current paths are observed while scanning IV characteristics, revealing the electrothermal dynamics within the device during operation. This helps to provide a deeper understanding of the switching behavior of this highly promising material.

VO2 oscillators have been fabricated in planar and crossbar configurations. The strong dependence on geometry, fabrication process, and film morphology are brought to light by the electrical characteristics of devices and circuits. Oscillating neural networks of resistively coupled VO2 oscillators were investigated and an application for neuromorphic computing was demonstrated.

We thank operations team of the Binnig and Rohrer Nanotechnology Center(BRNC) at IBM Research Europe - Zurich for their help and support. In particular, Ronald Grundbacher for ALD and Steffen Reidt for annealing support. We thank Pengyan Wen, Elisabetta Corti and Federico Balduini  for experimental support in SThM

This project has received funding from the EU’s Horizon 2020 program under the Marie Skłodowska-Curie grant agreement No 861153 (MANIC) and No 871501 (NeurONN).

© FUNDACIO DE LA COMUNITAT VALENCIANA SCITO
We use our own and third party cookies for analysing and measuring usage of our website to improve our services. If you continue browsing, we consider accepting its use. You can check our Cookies Policy in which you will also find how to configure your web browser for the use of cookies. More info