qBSA: Logic Design of a 32-bit Block-Skewed RSFQ Arithmetic Logic Unit

Souvik Kundu, Gourav Datta, Peter A. Beerel, Massoud PedramOpenReview 2019 · DOI · OpenReview

Abstract

Single flux quantum (SFQ) circuits are an attractive beyond-CMOS technology because they promise two orders of magnitude lower power at clock frequencies exceeding 25 GHz. However, every SFQ gate is clocked, creating very deep gate-level pipelines that are difficult to keep full, particularly for sequences that include data-dependent operations. This paper proposes to increase the throughput of SFQ pipelines by re-designing the datapath to accept and operate on least-significant bits (LSBs) clock cycles earlier than more significant bits. This skewed datapath approach reduces the latency of the LSB side, which can be fed back earlier for use in subsequent data-dependent operations. We propose a block-skewed MIPS-compatible 32-bit ALU (qBSA) using 4-bit blocks. Our design improves the throughput of 32-bit data-dependent operations by 2x and 1.5x compared to previously proposed 4-bit bit-slice and 32-bit Ladner-Fischer ALUs, respectively.

Introduction

Traditional CMOS technology is approaching fundamental limits in power consumption and scaling. Rapid Single Flux Quantum (RSFQ) technology offers a promising alternative with ultra-high speeds (>25 GHz) and extremely low power. However, RSFQ architectures face a unique challenge: gate-level pipelining. Since every logic gate is clocked, even simple operations result in deep pipelines. For data-dependent instructions, where one operation depends on the result of a previous one, the processor must wait for the entire pipeline to clear, leading to high "Initiation Intervals" (II) and significantly reduced Instructions Per Cycle (IPC).

The qBSA Architecture

To address the latency bottleneck, the authors introduce the qBSA (quantum Block-Skewed ALU). The core idea is a block-skewed datapath. Instead of processing a 32-bit word all at once, the ALU is divided into eight 4-bit blocks.

The qBSA architecture (a) and the internal logic of the 4-bit Sklansky ALU block (b).

As shown in Figure 1, the processing is temporally skewed: the Least Significant Bits (LSBs) enter the pipeline first, followed by more significant blocks in subsequent clock cycles. This allows the LSB results of one instruction to be fed back to the start of the pipeline to begin the next dependent instruction's LSB calculation before the current instruction has fully finished its Most Significant Bit (MSB) calculation. Each 4-bit block utilizes a Sklansky prefix-tree adder topology to minimize internal latency and provides an "early carry-out" signal to speed up the ripple between blocks.

Performance Evaluation

The qBSA design was evaluated against existing RSFQ ALU architectures using RISC-V benchmarks like Median, qSort, and Dhrystone. The primary advantage of the block-skewed approach is the reduction of the initiation interval for data-dependent operations from 15 cycles to just 8 cycles.

IPC improvement of the qBSA-based processor over 4-bit bit-slice (4BSP) and 32-bit Ladner-Fischer (32LFP) designs across various benchmarks.

As illustrated in Figure 2, the qBSA-based processor (qBSP) demonstrates significant IPC improvements. Compared to a 4-bit bit-slice ALU (4BSP), the IPC gains range from approximately 3x to 4.7x depending on the benchmark and system latency. Against a more advanced 32-bit Ladner-Fischer ALU (32LFP), it still achieves a consistent improvement of around 1.3x to 1.5x. Overall, the design doubles the throughput for data-dependent operations compared to bit-sliced alternatives.

Conclusion

The qBSA architecture successfully mitigates the performance impact of deep gate-level pipelines in RSFQ technology. By leveraging temporal skewing and early feedback at the block level, it achieves substantial throughput and IPC gains without excessive hardware overhead. This work provides a critical building block for future high-performance superconductive processors.

Summary updated

Abstract

Single flux quantum (SFQ) circuits are an attractive beyond-CMOS technology because they promise two orders of magnitude lower power at clock frequencies exceeding 25 GHz.However, every SFQ gate is clocked creating very deep gate-level pipelines that are difficult to keep full, particularly for sequences that include data-dependent operations. This paper proposes to increase the throughput of SFQ pipelines by re-designing the datapath to accept and operate on least-significant bits (LSBs) clock cycles earlier than more significant bits. This skewed datapath approach reduces the latency of the LSB side which can be feedback earlier for use in subsequent data-dependent operations increasing their throughput. In particular,we propose to group the bits into 4-bit blocks that are operatedon concurrently and create block-skewed datapath units for 32-bit operation. This skewed approach allows a subsequent data-dependent operation to start evaluating as soon as the first 4-bit block completes. Using this general approach, we developa block-skewed MIPS-compatible 32-bit ALU. Our gate-level Verilog design improves the throughput of 32-bit data dependent operations by 2x and 1.5x compared to previously proposed 4-bit bit-slice and 32-bit Ladner-Fischer ALUs respectively.
BibTeX
@misc{kundu2019qbsa,
  author = {Souvik Kundu and Gourav Datta and Peter A. Beerel and Massoud Pedram},
  title = {{qBSA: Logic Design of a 32-bit Block-Skewed RSFQ Arithmetic Logic Unit}},
  year = {2019},
  eprint = {2001.10715},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2001.10715},
  howpublished = {OpenReview},
}
Figure