高级检索

    多通道感知的RSFQ电路长度匹配布线算法

    A Multi-Channel Aware Routing Algorithm with Length-Matching for Rapid Single-Flux-Quantum Circuits

    • 摘要: 得益于超导材料的进步以及快速单通量量子(rapid single-flux-quantum, RSFQ)逻辑在高性能计算的巨大潜力, RSFQ 电路在未来的计算技术中有着广阔的应用前景. 然而, RSFQ电路因其超高的工作频率, 在物理设计中面临严峻的时序挑战, 需要在布线阶段引入长度匹配约束. 同时, 随着工艺节点的进步, 布线层已支持跨逻辑门列进行布线. 为此, 该文提出一种多通道感知的RSFQ电路长度匹配布线算法. 该算法通过基于交叉图的无源传输线段分配方法与布线区域最小化策略, 结合通孔生成及单调河流布线构建宽度最小化的初始路径; 引入基于扩散布线的区域调整方法, 重构布线资源以减少布线资源的浪费; 利用基于布线路径方向的线长扩展方法提高整个电路布线资源利用率, 从而实现电路面积优化. 实验结果表明, 在16位RSFQ-Sklansky加法器中, 所提算法相较于现有算法, 布线所需的区域面积减少了55%、44%和12%; 在随机生成连接的测试用例中, 布线所需的区域面积减少了74%、73%、36%、15%和7%, 取得了高质量的布线结果.

       

      Abstract: Benefiting from advances in superconducting materials and the great potential of rapid single-flux-quantum (RSFQ) logic for high-performance computing, RSFQ circuits are expected to have broad application prospects in future computing technologies. However, due to their ultra-high operating frequencies, RSFQ circuits encounter severe timing challenges during physical design, necessitating the introduction of length-matching constraints at the routing stage. Simultaneously, with advancements in process nodes, routing layers now support routing across logic gate columns. To address these issues, this paper proposes a multi-channel aware routing algorithm with length-matching for RSFQ circuits. The algorithm constructs width-minimized initial routing paths by employing a cross-graph-based passive transmission line (PTL) allocation method and a routing area minimization strategy, combined with via generation and monotonic river routing. Furthermore, it introduces a diffusion-routing-based area adjustment method to reconfigure routing resources and effectively reduce resource waste. Finally, a direction-based wire length extension method is utilized to enhance global resource utilization, thereby optimizing the total circuit area. Experimental results demonstrate that, for a 16-bit RSFQ-Sklansky adder, the proposed algorithm reduces the routing area by 55%, 44%, and 12% compared to existing algorithms. Additionally, in randomly generated connection test cases, the routing area is reduced by 74%, 73%, 36%, 15%, and 7%, respectively, indicating the achievement of high-quality routing results.

       

    /

    返回文章
    返回