Abstract Body

The Adiabatic Quantum-Flux-Parametron (AQFP) logic [1] emerges as a promising technology for the next generation of energy-efficient, high-performance information processing systems. This is largely due to its remarkably low power dissipation, achieved through the adiabatic switching of Josephson junctions. In this paper, we present the design and implementation of a neuron circuit, which incorporates analog accumulation and activation via magnetic coupling confluence and an AQFP current comparator, respectively. This is specifically tailored for binarized neural networks [2, 3] that have a crossbar structure with pre-stored weights. To demonstrate feasibility, we successfully designed and implemented neuron circuits with 4, 8, and 16 inputs using the AIST 10kA/cm² 4-layer niobium process and verified their functionality experimentally at 100 kHz.

References

[1] N. Takeuchi, D. Ozawa, Y. Yamanashi, and N. Yoshikawa, “An adiabatic quantum flux parametron as an ultra-low-power logic device, Superconductor Science and Technology, vol. 26, no. 3, p. 035010, 2013.
[2] M. Rastegari, V. Ordonez, J. Redmon, and A. Farhadi, Xnor-net: Imagenet classification using binary convolutional neural networks, in Computer Vision - ECCV 2016 - 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part IV, ser. Lecture Notes in Computer Science, B. Leibe, J. Matas, N. Sebe, and M. Welling, Eds. , vol. 9908. Springer, 2016, pp. 525–542. [Online]. Available: https://doi. org/10. 1007/978-3-319-46493-0\_32.
[3] M. Courbariaux, I. Hubara, D. Soudry, R. El-Yaniv, and Y. Bengio, “Binarized neural networks: Training deep neural networks with weights and activations constrained to +1 or -1, 02 2016.

Acknowledgment

This work was supported by JST FOREST Program (Grant Number JPMJFR226W, Japan) and JSPS KAKENHI Grant Number JP22H0220.