Superconductor bistable vortex memory for data storage and in-memory computing

Mustafa Altay Karamuftuoglu, Beyza Zeynep Ucpinar, Sasan Razmkhah, Massoud Pedram

Introduction

Superconductor electronics (SCE) offer a path toward ultra-fast, energy-efficient computing, yet they are currently limited by a significant memory bottleneck. Traditional SCE memory designs often force a compromise between density, speed, and nonvolatility. For instance, Non-Destructive Readout (NDRO) RAM typically relies on bulky magnetic transformers to couple signals, which severely restricts integration density and limits capacity to a few kilobits.

The Bistable Vortex Memory (BVM) Cell

To address these challenges, this research introduces the Bistable Vortex Memory (BVM) cell. The BVM design achieves high density by eliminating transformers in favor of direct electrical coupling and shared inductors, significantly reducing the physical footprint of each cell.

The BVM cell utilizes a dual-loop SQUID architecture for transformer-free data storage and readout.

The cell features a dual-loop SQUID (Superconducting Quantum Interference Device) architecture:

  • S-Loop ($\Phi_S$): Stores the data bit as a persistent circulating current, or magnetic vortex. Because this current flows in a zero-resistance superconducting loop, the memory is nonvolatile and consumes zero static power.
  • R-Loop ($\Phi_R$): Handles the readout process. This design ensures non-destructive readout, meaning the stored data is not erased or altered during a read cycle.

Scalable Architecture and In-Memory Computing

The BVM cells are organized into a scalable crossbar architecture. This physical structure naturally supports analog in-memory computing, specifically Multiply-Accumulate (MAC) operations, which are the backbone of modern neural networks. By activating multiple rows simultaneously, the output current on a shared sense line represents the analog sum of the stored values.

The BVM array architecture (a) and detailed cell schematic (b) showing the integration into a crossbar for high-speed operation and MAC functionality.

Performance and Evaluation

Circuit simulations using the JoSIM framework based on the MIT Lincoln Laboratory SFQ5ee fabrication process demonstrate significant performance gains over traditional logic:

  • High-Speed Operation: While individual cells can reach 50 GHz, a 32×32 memory array operates reliably at 20 GHz, maintaining signal integrity across long word and bit lines.
  • Density and Efficiency: A BVM-based 4-bit multiplier was found to be approximately 17 times smaller (0.236 mm² vs. 4.14 mm²) and 13 times more power-efficient in terms of static power than a comparable design using standard RSFQ (Rapid Single Flux Quantum) logic.
  • Zero Static Power: Data is maintained without power consumption, with energy only required during active read/write cycles.

Conclusion and Future Directions

The BVM cell represents a significant step toward practical, high-density superconductor memory. While the requirement for cryogenic cooling (4K) limits its use to specialized environments like supercomputers or space applications, its potential for high-performance computing and AI acceleration is substantial. Future work will focus on scaling the sense line and optimizing peripheral driver circuits to fully leverage the cell's ultra-fast switching capabilities.

Summary updated

Abstract

Abstract Despite superconductor electronics (SCE) advantages, the realization of SCE logic faces a significant challenge due to the absence of dense and scalable nonvolatile memory. While various nonvolatile memory technologies, including Non-destructive readout, vortex transitional memory, and magnetic memory, have been explored, designing a dense crossbar array and achieving a superconductor random-access memory remains challenging. This work introduces a novel, nonvolatile, high-density, and scalable vortex-based memory design for SCE logic called bistable vortex memory. Our proposed design addresses scaling issues with an estimated area of 10 × 10 um 2 while boasting zero static power with the dynamic energy consumption of 12 aJ for single-bit read and write operations. The current summation capability enables analog operations for in-memory or near-memory computational tasks. We demonstrate the efficacy of our approach with a 32 × 32 superconductor memory array operating at 20 GHz. Additionally, we showcase the accumulation property of the memory through analog simulations conducted on an 8 × 8 superconductor crossbar array.
BibTeX
@inproceedings{karamuftuoglu2024superconductor,
  author = {Mustafa Altay Karamuftuoglu and Beyza Zeynep Ucpinar and Sasan Razmkhah and Massoud Pedram},
  title = {{Superconductor bistable vortex memory for data storage and in-memory computing}},
  booktitle = {Superconductor},
  year = {2024},
  doi = {10.1088/1361-6668/ad9863},
}
Figure