Improving VO2 Granular Films’ CMOS Compatibility to Perform Oscillation-Based Computing
Abstract
The growing interest in polycrystalline phase-change materials to develop brain-inspired learning and novel computing approaches makes $ VO_2 $ a first-choice material, with its ability to switch from a high to a low resistive state around room temperature (68$^{\circ}$C). Natural oscillations triggered by running a current through this material are offering the means to fabricate electronic oscillators that once coupled, create an oscillating neural network (ONN), best suited to compute pattern recognition tasks and solve optimization problems [1, 2, 3]. This type of network shows the promise of energy-efficient computing by processing data locally while avoiding issues related to the different frequency operations from the memory access to the processor’s calculation times [1, 3, 4]. Improving CMOS compatibility with pristine material quality and a highly reliable fabrication process to produce uniform devices is paramount for industrial purposes [5, 6]. The integration of $ VO_2 $ layers on a Si-based platform typically results in polycrystalline films presenting granular structures with a rugged surface topology [5]. This introduces undesired variability among our electrical oscillators that needs to be reduced to a minimum before their integration in a compound circuit, such as our ONN [6]. To minimize device-to-device variations, $ VO_2 $ -based oscillators in planar and crossbar configurations have been fabricated. The devices excel in high-frequency operation and power efficiency. A 50 nm thick Vanadium oxide film is deposited by a water-based ALD reaction on a $ Si/SiO_2/HfO_2 $ substrate for CMOS compatibility. A post-annealing step is required to turn the amorphous film into a crystalline VO2 layer. We use two different annealing techniques which vary in steepness of temperature ramps and oxygen pressure to obtain reproducible behavior in our devices. The impact of the different annealing methods on the grain size, the surface roughness, the electric behavior, and the quality of the resulting VO2 film are studied through Raman Spectroscopy, Atomic Force Microscopy (AFM), and by measuring the resistance of the film as a function of temperature. The devices are then placed in a circuit to measure the resulting natural oscillations before being coupled with one another. The power efficiency and the oscillation frequency could be significantly improved by device scaling and optimizations. Our network of $ VO_2 $ -based oscillators shows the promise of an attractive and scalable computing unit for hardware accelerators [3, 4], thanks to its high-performance switching properties and CMOS compatibility. This project has received funding from the EU’s Horizon 2020 program under projects No 871501 (NeurONN) and No 861153 (MANIC). Bibliography [1] E. Corti and S. Karg, "Coupled VO2 oscillators circuit as analog first layer filter in convolutional neural networks," Frontiers in Neuroscience, 11 February 2021. [2] S. Dutta, A. Khana, H. Paik, D. Schlom, A. Raychowdhury, Z. Toroczkai and S. Datta, "An Ising Hamiltonian Solver using Stochastic Phase-Transition Nano-Oscillators," 2021. [Online]. Available: https://www.researchsquare.com/article/rs-93438/v1. [3] G. Csaba and W. Porod, "Coupled oscillators for computing: A review and perspective," Applied Physics Review, 3 January 2020. [4] G. Indiveri and S.-C. Liu, "Memory and Information Processing in Neuromorphic Systems," Proceedings of the IEEE, vol. 103, no. 8, pp. 1379 - 1397, 2015. [5] H. Guo, Y. Wang, H. Fu, A. Jain and F. Chen, "Influence of dopant valence on the thermochromic properties of VO2 nanoparticles," Elsevier Ceramics International, 2021. [6] E. Corti, B. Gotsmann, K. Moselund, A. M. Ionescu, J. Robertson and S. Karg, "Scaled resistively-coupled VO2 oscillators for neuromorphic computing," Solid-State Electronics, vol. 168, p. 107729, 2020.