top of page
An SFQ-Based Compact and High-Precision Deterministic Adder for Stochastic Computing

Zeyu Han (Yokohama National University); Yuya Mandai (Yokohama National University); Nobuyuki Yoshikawa (Yokohama National University); Yuki Yamanashi (Yokohama National University)

Stochastic computing [1] represents data as random bit sequences, enabling complex arithmetic with simple logic. However, it suffers from high latency and limited precision. Deterministic stochastic computing [2] uses shorter deterministic bit sequences, reducing latency and improving accuracy. Moreover, single-flux quantum (SFQ) circuits, known for their ultra-high speed, are promising candidates for implementing stochastic computing hardware [3]. We proposed a superconducting deterministic adder based on SFQ circuits. Compared with the conventional SFQ stochastic adder that relies on a multiplexer (MUX) controlled by random bit sequences [4], our design uses only an AND gate, an XOR gate, and a toggle flip-flop, significantly lowering area overhead. Additionally, the deterministic adder requires no random control signal, reducing the number of random number generators and avoiding precision loss due to probability deviations of the control signal. We fabricated this adder using a 10 kA/cm² Nb four-layer superconducting process and verified its high-speed operation up to 35.4 GHz. For a 1024-bit sequence, our adder achieves a 32× reduction in root-mean-square error, a 49.0% reduction in circuit area, and a 28.1% decrease in Josephson junction count compared to the conventional MUX-based design.

Acknowledgement

This work was supported by JSPS KAKENHI Grant Number JP24KJ1148 and JP25K01284. This work was also supported by Support Center for Advanced Telecommunications Technology Research (SCAT). The circuits were fabricated in the Superconducting Quantum Circuit Fabrication Facility (Qufab) in National Institute of Advanced Industrial Science and Technology (AIST).

References
[1] B. R. Gains, "Stochastic computing," in Proc. Spring Joint Computer Conference, pp.149-156, 1967.
[2] D. Jenson and M. Riedel, "A deterministic approach to stochastic computation," in Proc. IEEE/ACM Int. Conf. Comput.-Aided Design (ICCAD),  pp. 1-8, Nov. 2016
[3] K. Asaka et al., "Large fan-out single-flux-quantum stochastic number splitter utilizing frequency-synchronized superconductor random number generators," Supercond. Sci. Technol., vol. 38, no. 4, Art. no. 045009, Mar. 2025.
[4] Y. Yamanashi et al., "Stochastic Matrix Multiplier Using Superconductor Single-Flux-Quantum Circuit," Supercond. Sci. Technol., vol. 37, no. 11, Art. no. 115024, Oct. 2024.

Poster

|

Device and Circuit

|

October 27, 13:30 → 15:00

bottom of page