Abstract
Digital single-flux quantum (SFQ) technology promises to meet the demands of ultra low power and high speed computing needed for future exascale supercomputing platforms. However, clocking SFQ logic circuits remains a challenge due to the presence of a large number of on-chip clock sinks, and hence, decomposing large designs into multiple independent clock domains similar to CMOS, have been proposed. However, such clock domains demand efficient synchronizing First-in-first-out (FIFO) buffers and robust interfaces to safely transfer data from one clock domain to another. In this brief, we propose such a FIFO synchronizer and clock-domain crossing interface for both uni and bi-directional data transfer without any significant degradation of the clock frequency. Our proposal scales to complex gate-level pipelined SFQ logic cores while demonstrating extremely low Bit Error Rate (BER) and is unaffected by noise, given current SFQ lithography feature sizes.
Introduction
Digital Single-Flux Quantum (SFQ) technology is a leading candidate for future exascale supercomputing due to its ability to operate at hundreds of Gigahertz with orders of magnitude less power than traditional CMOS. However, SFQ logic faces a significant "clocking nightmare": almost every logic gate is pipelined and requires its own clock pulse. Distributing a perfectly synchronized global clock at 100+ GHz is nearly impossible due to process variations and signal skew.
To address this, designers partition chips into independent clock domains. The challenge then shifts to Clock Domain Crossing (CDC). When data travels between asynchronous domains, timing mismatches can lead to metastability—where a circuit cannot resolve a signal to a stable '1' or '0' state—resulting in high Bit Error Rates (BER). Previous SFQ synchronizers often lacked the flexibility for bi-directional communication or real-world data bursts.
Proposed Synchronizer and Interface Architecture
The researchers introduce a novel FIFO synchronizer and a custom bi-directional interface designed specifically for the unique constraints of SFQ logic. The design incorporates several key innovations:
- Enhanced Synchronizer Design: Unlike previous models, this design adds an
enablesignal to facilitate a "handshake" protocol, ensuring data is only sent when the receiving side is ready. - Metastability Mitigation: A "two-flop" synchronization strategy is employed. By placing an OR gate followed by a D-Flip-Flop (DFF) in the control path, the system provides extra time for metastable signals to resolve, drastically lowering the BER.
- Bi-Directional Interface using NDRO Cells: The interface utilizes Non-Destructive Read-Out (NDRO) cells to create a control loop. This allows the system to effectively "park" a data pulse until the next domain is ready, bypassing the need for complex and often unreliable high-speed clock gating.
Experimental Results and Evaluation
The designs were verified using JSIM simulations with parameters from the MIT Lincoln Laboratory (MIT LL) SFQ5ee fabrication process. A 32-bit wide, 8-stage circular shift register served as a complex testbench to prove the bi-directional interface's functionality in a realistic, high-speed data loop.
Key performance metrics include:
- Ultra-Low Error Rates: At 30 GHz, the write-side BER was measured at approximately $0.128 \times 10^{-15}$. At 20 GHz, the error rate was effectively zero in simulation.
- High Throughput: The system supports reliable operation up to ~33-40 GHz.
- Area Efficiency: By integrating synchronization logic directly into the FIFO control path, the design achieved robustness with an area increase of only ~1.39% for a 5-stage FIFO.
- Noise Resilience: Simulations confirmed that errors are dominated by timing metastability rather than thermal noise, which has negligible effect on the two-flop synchronizer at current feature sizes.
Conclusion
This paper presents a significant advancement in SFQ logic by providing a robust, high-speed solution for clock domain crossing. While current throughput is limited to 50% (one new data point every two cycles) due to the control loop delay, the design's extreme reliability and low area overhead make it a viable building block for future superconducting processors. Future work will focus on increasing throughput and physical validation in cryogenic environments.
Summary updated