Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

IEEE Transactions on Applied Superconductivity

Superconducting Hyperdimensional Associative Memory Circuit for Scalable Machine Learning

Kylie Huch, Patricia Gonzalez-Guerrero, Darren Lyles, George Michelogiannakis

Published May 1, 2023• 3 citations
Publisher

Abstract

We propose a generalized architecture for the first rapid-single-flux-quantum (RSFQ) associative memory circuit. The circuit employs hyperdimensional computing (HDC), a machine learning (ML) paradigm utilizing vectors with dimensionality in the thousands to represent information. HDC designs have small memory footprints, simple computations, and simple training algorithms compared to superconducting neural network accelerators (SNNAs), making them a better option for scalable SFQ machine learning (ML) solutions. The proposed superconducting HDC (SHDC) circuit uses entirely on-chip RSFQ memory which is tightly integrated with logic, operates at 33.3 GHz, is applicable to general ML tasks, and is manufacturable at practically useful scales given current SFQ fabrication limits. Tailored to a language recognition task, SHDC consists of$\sim$2–20 M Josephson junctions (JJs) and consumes up to three times less power than an analogous CMOS HDC circuit while achieving 78–84% higher throughput. SHDC is capable of outperforming the state of the art RSFQ SNNA, SuperNPU, by 48-99% for all benchmark NN architectures tested while occupying up to 90% less area and consuming up to nine times less power. To the best of the authors' knowledge, SHDC is currently the only superconducting ML approach feasible at practically useful scales for real-world ML tasks and capable of online learning.

DOI
10.1109/tasc.2023.3271951

Related papers

Toward Practical Superconducting Accelerators for Machine Learning Using U-SFQ

Patricia Gonzalez-Guerrero, Kylie Huch, Nirmalendu Patra +2 more

Most popular superconducting circuits operate on information carried by ps-wide, μV-tall, single flux quantum (SFQ) pulses. These circuits can operate at frequencies of hundreds of GHz with orders of magnitude lower switching energy than complementary-metal-oxide-semiconductors (CMOS). However, under the stringent area constraints of modern superconductor technologies, fully-fledged, CMOS-inspired superconducting architectures cannot be fabricated at large scales. Unary SFQ (U-SFQ) is an alternative computing paradigm that can address these area constraints. In U-SFQ, information is mapped to a combination of streams of SFQ pulses and in the temporal domain. In this work, we extend U-SFQ to introduce novel building blocks such as a multiplier and an accumulator. These blocks reduce area and power consumption by 2 \(\times\) and 4 \(\times\) compared with previously proposed U-SFQ building blocks and yield at least 97% area savings compared with binary approaches. Using these multiplier and adder, we propose a U-SFQ Convolutional Neural Network (CNN) hardware accelerator capable of comparable peak performance with state-of-the-art superconducting binary approach (B-SFQ) in 32 \(\times\) less area. CNNs can operate with 5–8 bits of resolution with no significant degradation in classification accuracy. For 5 bits of resolution, our proposed accelerator yields 5 \(\times\) to 63 \(\times\) better performance than CMOS and 15 \(\times\) to 173 \(\times\) better area efficiency than B-SFQ.

ACM Journal on Emerging Technologies in Computing Systems•Apr 9, 2024•5 citations

Design and Implementation of a Single Flux Quantum Logic-Based Memory Controller for Josephson-CMOS Hybrid Memory Systems

Seda Demirhan, Ali Bozbey

Single flux quantum (SFQ) digital circuits have shown the potential for high speed/low power computation applications. Unfortunately, because of the low integration density and low driving capability of SFQ circuits, realizing large-scale memories by using only SFQ circuits is still a challenge. Josephson-CMOS hybrid memory, hybridizing high-speed, low power SFQ circuits, and high-density CMOS memories is already proposed as a solution to the large-scale memory problem in RSFQ digital systems. In this article, an SFQ-based memory controller which works up to 10 GHz clock frequencies is designed for Josephson-CMOS hybrid memory systems. The memory controller acts as a SFQ/CMOS interface between the SFQ circuit and SRAM module and generates the required waveforms for SRAM read and write operations. The circuit which has 4-b data and 2-b address lines is fabricated with AIST 2.5 kA/cm2STP2 process and operations of the circuits are verified. To demonstrate the scalability of the circuit, memory controller was scaled to 8-b data and 13-b address to control of the 64kb SRAM. Operations of the scaled memory controller are verified with analog simulations. Scaled circuit consumes 0.76 mW power in an area of 1.64 mm × 1.60 mm.

IEEE Transactions on Applied Superconductivity•Jul 28, 2020•5 citations

Quantum-StateHD: Leveraging Random Quantum States for Hyperdimensional Learning

Yangjie Xu, Hui Huang, Radu State

Hyperdimensional Computing (HDC) is a brain-inspired computational paradigm that utilizes high-dimensional vector spaces to represent, bind, and process information efficiently. Its robustness, scalability, and adaptability make HDC particularly effective for machine learning tasks such as classification and pattern recognition. In this work, we introduce Quantum-StateHD (QSHD), a novel approach that leverages random quantum states as encoding mechanisms for high-dimensional spaces within the HDC framework. Quantum states, with their inherent properties of high-dimensionality, near-orthogonality, and probabilistic distributions, provide a natural substrate for constructing robust representations. Quantum-StateHD encodes classical data into the coefficients of random quantum states and seamlessly integrates these representations into the HDC pipeline. Experimental results on several benchmark classification datasets demonstrate that Quantum-StateHD achieves competitive or superior performance compared to traditional state-of-the-art HDC methods, showcasing its potential for robust and efficient high-dimensional information processing. This study bridges the gap between quantum information science and HDC, offering a promising direction for hybrid quantum-classical machine learning architectures.

International Joint Conference on Neural Networks (IJCNN)•Jun 30, 2025•0 citations

An Area Efficient Superconducting Unary CNN Accelerator

Patricia Gonzalez-Guerrero, Kylie Huch, Nirmalendu Patra +2 more

In superconducting circuits, information is carried by ps-wide, µV-tall, Single Flux Quanta (SFQ) pulses. These circuits can operate at frequencies of hundreds of GHz with orders of magnitude lower switching energy than complementary-metal-oxide-semiconductors (CMOS). However, under the stringent area constraints of modern superconductor technologies, fully-fledged, CMOS-inspired superconducting architectures cannot be fabricated at large scales. Unary SFQ (U-SFQ) is an alternative computing paradigm that addresses these area constraints. In U-SFQ, information is mapped to a combination of streams of SFQ pulses and in the temporal domain. In this work, we propose a U-SFQ Convolutional Neural Network (CNN) hardware accelerator capable of comparable peak performance with state-of-the-art superconducting binary (B-SFQ) approaches in 32× less area. CNNs can operate with 5 to 8 bits of resolution with no significant degradation in classification accuracy. The proposed CNN accelerator effortlessly supports this variable resolution and, for less than 7 bits, yields 5×-63× better performance than CMOS and 15×-173× better area efficiency than B-SFQ.

International Symposium on Quality Electronic Design (ISQED)•Apr 5, 2023•3 citations

Superconductive single-flux quantum technology

Oleg A. Mukhanov

Key digital circuits for signal processing with tens of gigahertz speed and a few microwatts power dissipation have been recently demonstrated using a simple, 2 /spl mu/m linewidth process. These are the fastest digital circuits reported to date. These circuits use novel rapid single flux quantum (RSFQ) technology that takes advantage of the fundamental properties of superconductivity. In this technology, data and clock are picosecond-long quantized voltage pulses corresponding to the transfer of single magnetic flux quanta, /spl Phisub 0/=h/2e=2 mVps. These SFQ pulses are regenerated at each gate and are processed by circuits comprised of overdamped Josephson junctions interconnected via microstrip inductances. The inherent internal memory of all RSFQ gates (primarily flip-flops), combined with high switching speed, makes serial, bit-pipelined circuit design advantageous. This paper presents experiments in RSFQ digital circuits, including an N/spl times/8b serial multiplier, and on-chip test systems based on RSFQ shift registers and on SFQ samplers. They enable test and study of RSFQ circuits at full GHz speed while communicating with semiconductor electronics at low speed.>

Digest of technical papers/Digest of technical papers - IEEE International Solid-State Circuits Conference•Dec 17, 2002•8 citations

Low power single flux quantum qubit control circuit without high-frequency input

Bicong Weng, Wei Peng, Jie Ren

Abstract The use of high-frequency input signals from room-temperature microwave sources makes it difficult to scale up the number of quantum bits in universal quantum computers. To address this issue, superconducting single flux quantum (SFQ) integrated circuits are being explored as suitable candidates for qubit manipulation in universal quantum computers. This paper deals with a scalable SFQ qubit control circuit (SQCC) structure that requires only low-frequency input. The circuit mainly consists of a pulse generator and a counter, that output the SFQ pulse train with adjustable frequency and a controllable number of pulses, which is applicable to control single-qubit Clifford operations. The design of low-voltage rapid single flux quantum (LV-RSFQ) and energy-efficient rapid single flux quantum (ERSFQ) for the SQCC achieves low power consumption and provides a basis for scaling up SQCC to control more qubits. The proposed circuits are fabricated under the SIMIT-Nb03 process and successfully pass test verification. The achieved test results reveal that the adjustable output frequency ranges of the SQCC based on the LV-RSFQ and ERSFQ designs in order are [2.40, 8.11] GHz and [4.81, 5.14] GHz. In the operating frequency range, the circuit is able to generate the correct number of SFQ pulses under control. The controllable number range is from 1 to 128. When the circuits operate at 5 GHz, the total power consumptions of the above circuits in order are 23.88 μW and 6.2 μW. All input signals are low-frequency signals, which frees the control of large-scale qubits from limitations caused by high-frequency inputs produced by room-temperature microwave sources.

Superconductor Science and Technology•Jul 11, 2023•4 citations
We propose a generalized architecture for the first rapid-single-flux-quantum (RSFQ) associative memory circuit. The circuit employs hyperdimensional computing (HDC), a machine learning (ML) paradigm utilizing vectors with dimensionality in the thousands to represent information. HDC designs have small memory footprints, simple computations, and simple training algorithms compared to superconducting neural network accelerators (SNNAs), making them a better option for scalable SFQ machine learning (ML) solutions. The proposed superconducting HDC (SHDC) circuit uses entirely on-chip RSFQ memory which is tightly integrated with logic, operates at 33.3 GHz, is applicable to general ML tasks, and is manufacturable at practically useful scales given current SFQ fabrication limits. Tailored to a language recognition task, SHDC consists of∼\sim∼2–20 M Josephson junctions (JJs) and consumes up to three times less power than an analogous CMOS HDC circuit while achieving 78–84% higher throughput. SHDC is capable of outperforming the state of the art RSFQ SNNA, SuperNPU, by 48-99% for all benchmark NN architectures tested while occupying up to 90% less area and consuming up to nine times less power. To the best of the authors' knowledge, SHDC is currently the only superconducting ML approach feasible at practically useful scales for real-world ML tasks and capable of online learning.
Most popular superconducting circuits operate on information carried by ps-wide, μV-tall, single flux quantum (SFQ) pulses. These circuits can operate at frequencies of hundreds of GHz with orders of magnitude lower switching energy than complementary-metal-oxide-semiconductors (CMOS). However, under the stringent area constraints of modern superconductor technologies, fully-fledged, CMOS-inspired superconducting architectures cannot be fabricated at large scales. Unary SFQ (U-SFQ) is an alternative computing paradigm that can address these area constraints. In U-SFQ, information is mapped to a combination of streams of SFQ pulses and in the temporal domain. In this work, we extend U-SFQ to introduce novel building blocks such as a multiplier and an accumulator. These blocks reduce area and power consumption by 2 ×\times× and 4 ×\times× compared with previously proposed U-SFQ building blocks and yield at least 97% area savings compared with binary approaches. Using these multiplier and adder, we propose a U-SFQ Convolutional Neural Network (CNN) hardware accelerator capable of comparable peak performance with state-of-the-art superconducting binary approach (B-SFQ) in 32 ×\times× less area. CNNs can operate with 5–8 bits of resolution with no significant degradation in classification accuracy. For 5 bits of resolution, our proposed accelerator yields 5 ×\times× to 63 ×\times× better performance than CMOS and 15 ×\times× to 173 ×\times× better area efficiency than B-SFQ.
Single flux quantum (SFQ) digital circuits have shown the potential for high speed/low power computation applications. Unfortunately, because of the low integration density and low driving capability of SFQ circuits, realizing large-scale memories by using only SFQ circuits is still a challenge. Josephson-CMOS hybrid memory, hybridizing high-speed, low power SFQ circuits, and high-density CMOS memories is already proposed as a solution to the large-scale memory problem in RSFQ digital systems. In this article, an SFQ-based memory controller which works up to 10 GHz clock frequencies is designed for Josephson-CMOS hybrid memory systems. The memory controller acts as a SFQ/CMOS interface between the SFQ circuit and SRAM module and generates the required waveforms for SRAM read and write operations. The circuit which has 4-b data and 2-b address lines is fabricated with AIST 2.5 kA/cm2STP2 process and operations of the circuits are verified. To demonstrate the scalability of the circuit, memory controller was scaled to 8-b data and 13-b address to control of the 64kb SRAM. Operations of the scaled memory controller are verified with analog simulations. Scaled circuit consumes 0.76 mW power in an area of 1.64 mm × 1.60 mm.
Hyperdimensional Computing (HDC) is a brain-inspired computational paradigm that utilizes high-dimensional vector spaces to represent, bind, and process information efficiently. Its robustness, scalability, and adaptability make HDC particularly effective for machine learning tasks such as classification and pattern recognition. In this work, we introduce Quantum-StateHD (QSHD), a novel approach that leverages random quantum states as encoding mechanisms for high-dimensional spaces within the HDC framework. Quantum states, with their inherent properties of high-dimensionality, near-orthogonality, and probabilistic distributions, provide a natural substrate for constructing robust representations. Quantum-StateHD encodes classical data into the coefficients of random quantum states and seamlessly integrates these representations into the HDC pipeline. Experimental results on several benchmark classification datasets demonstrate that Quantum-StateHD achieves competitive or superior performance compared to traditional state-of-the-art HDC methods, showcasing its potential for robust and efficient high-dimensional information processing. This study bridges the gap between quantum information science and HDC, offering a promising direction for hybrid quantum-classical machine learning architectures.
In superconducting circuits, information is carried by ps-wide, µV-tall, Single Flux Quanta (SFQ) pulses. These circuits can operate at frequencies of hundreds of GHz with orders of magnitude lower switching energy than complementary-metal-oxide-semiconductors (CMOS). However, under the stringent area constraints of modern superconductor technologies, fully-fledged, CMOS-inspired superconducting architectures cannot be fabricated at large scales. Unary SFQ (U-SFQ) is an alternative computing paradigm that addresses these area constraints. In U-SFQ, information is mapped to a combination of streams of SFQ pulses and in the temporal domain. In this work, we propose a U-SFQ Convolutional Neural Network (CNN) hardware accelerator capable of comparable peak performance with state-of-the-art superconducting binary (B-SFQ) approaches in 32× less area. CNNs can operate with 5 to 8 bits of resolution with no significant degradation in classification accuracy. The proposed CNN accelerator effortlessly supports this variable resolution and, for less than 7 bits, yields 5×-63× better performance than CMOS and 15×-173× better area efficiency than B-SFQ.
Key digital circuits for signal processing with tens of gigahertz speed and a few microwatts power dissipation have been recently demonstrated using a simple, 2 /spl mu/m linewidth process. These are the fastest digital circuits reported to date. These circuits use novel rapid single flux quantum (RSFQ) technology that takes advantage of the fundamental properties of superconductivity. In this technology, data and clock are picosecond-long quantized voltage pulses corresponding to the transfer of single magnetic flux quanta, /spl Phisub 0/=h/2e=2 mVps. These SFQ pulses are regenerated at each gate and are processed by circuits comprised of overdamped Josephson junctions interconnected via microstrip inductances. The inherent internal memory of all RSFQ gates (primarily flip-flops), combined with high switching speed, makes serial, bit-pipelined circuit design advantageous. This paper presents experiments in RSFQ digital circuits, including an N/spl times/8b serial multiplier, and on-chip test systems based on RSFQ shift registers and on SFQ samplers. They enable test and study of RSFQ circuits at full GHz speed while communicating with semiconductor electronics at low speed.>
Abstract The use of high-frequency input signals from room-temperature microwave sources makes it difficult to scale up the number of quantum bits in universal quantum computers. To address this issue, superconducting single flux quantum (SFQ) integrated circuits are being explored as suitable candidates for qubit manipulation in universal quantum computers. This paper deals with a scalable SFQ qubit control circuit (SQCC) structure that requires only low-frequency input. The circuit mainly consists of a pulse generator and a counter, that output the SFQ pulse train with adjustable frequency and a controllable number of pulses, which is applicable to control single-qubit Clifford operations. The design of low-voltage rapid single flux quantum (LV-RSFQ) and energy-efficient rapid single flux quantum (ERSFQ) for the SQCC achieves low power consumption and provides a basis for scaling up SQCC to control more qubits. The proposed circuits are fabricated under the SIMIT-Nb03 process and successfully pass test verification. The achieved test results reveal that the adjustable output frequency ranges of the SQCC based on the LV-RSFQ and ERSFQ designs in order are [2.40, 8.11] GHz and [4.81, 5.14] GHz. In the operating frequency range, the circuit is able to generate the correct number of SFQ pulses under control. The controllable number range is from 1 to 128. When the circuits operate at 5 GHz, the total power consumptions of the above circuits in order are 23.88 μW and 6.2 μW. All input signals are low-frequency signals, which frees the control of large-scale qubits from limitations caused by high-frequency inputs produced by room-temperature microwave sources.