Approximate Scan Flip-flop to Reduce Functional Path Delay and Power Consumption

Lakshmi Bhanuprakash Reddy Konduru, Vijaya Lakshmi, Jaynarayan T. TuduarXiv 2022 · DOI · arXiv

TL;DR

This paper introduces two novel scan flip-flop designs that aim to reduce both the functional path delay and power consumption during testing. By cleverly modifying the scan flip-flop architecture, the authors achieve significant improvements in performance and energy efficiency, particularly for high-performance digital designs.

What Problem Does This Paper Address?

Modern integrated circuits (ICs) are becoming increasingly complex, making testing a significant challenge. Design-for-Test (DfT) mechanisms, like scan chains, are essential for ensuring the quality of manufactured chips. However, the fundamental building block of these scan chains, the scan flip-flop, introduces its own set of problems. Specifically, the multiplexer used to switch between functional and test modes adds extra delay to the critical functional path, which can limit the overall performance of the chip, especially in advanced, smaller technology nodes. Furthermore, the high switching activity during test mode, particularly when shifting test patterns through the scan chain, leads to excessive power dissipation, potentially causing chip overheating. This paper tackles these two critical issues: the delay penalty introduced by the scan flip-flop's multiplexer and the high power consumption during test mode.

Methods

The authors propose two new scan flip-flop designs to address the identified problems.

  1. Approximate Scan Flip-flop: This design aims to eliminate the delay introduced by the multiplexer in the functional path. It achieves this by directly connecting the functional data input (DI) to the flip-flop's input. To manage potential issues where the DI might override the scan input (SI) during test mode, the aspect ratios of transistors in the scan path's transmission gate are increased. This design focuses on reducing the functional path delay.
  2. GDI-based Scan Flip-flop: This design utilizes the Gate-Diffusion Input (GDI) technique to implement the multiplexer part of a scan flip-flop. The GDI method is known for its low complexity, reduced power-delay product, and smaller area. This approach aims to improve power efficiency and reduce area overhead compared to traditional multiplexer-based designs.

Both proposed designs are based on a negative-edge triggered master-slave D flip-flop structure. The functionality of these designs was verified using post-layout simulations in a 45nm CMOS technology node.

Datasets

Not discussed in the paper. The evaluation is based on simulations using Cadence Spectre and Virtuoso physical layout tools.

Key Insights and Ideas

  • Eliminating the Multiplexer for Delay Reduction: The most significant insight is that removing the multiplexer from the functional path of a scan flip-flop can directly reduce the critical path delay. This is a straightforward yet effective approach to improving chip performance.
  • Trade-offs in Scan Flip-flop Design: The paper highlights the inherent trade-offs in scan flip-flop design. For instance, while the approximate design reduces delay, careful transistor sizing is needed to ensure correct functionality in test mode. Similarly, GDI offers power and area benefits but can introduce slightly higher delay compared to a pure transmission gate multiplexer.
  • Test Mode Power is a Major Concern: The paper reiterates that test mode power consumption can be significantly higher than functional mode power due to increased switching activity. This underscores the importance of low-power test solutions.
  • Approximate Design's Efficiency: The "Approximate" scan flip-flop emerges as particularly efficient, offering a good balance of reduced delay and power consumption.

Key Results

The experimental results, obtained through post-layout simulations at 1GHz clock frequency, demonstrate the effectiveness of the proposed designs:

  • Approximate Scan Flip-flop:
    • Achieved a time gain of 0.02ns in functional mode and 0.025ns in test mode compared to the multiplexer-based design.
  • Showed a power gain of 85.9% in functional mode and 85.3% in test mode compared to the multiplexer-based design.
  • Required a smaller area (14 units) compared to existing designs.
  • GDI-based Scan Flip-flop:
    • Offered a power gain of 70.7% in functional mode and 64% in test mode compared to the multiplexer-based design.
  • Required a smaller area (12 units) compared to existing designs.
  • Overall: The proposed approximate scan flip-flop design was found to be the most efficient in terms of both delay and power consumption among the three proposed designs. All proposed designs consume less power and require less area than existing scan flip-flop designs.

Limitations and Future Work

  • Limitations:
    • The "Approximate" design's functionality relies on careful transistor sizing to prevent the functional input from overriding the scan input. This might require specific tuning for different technology nodes or design contexts.
  • The paper focuses on a specific technology node (45nm) for simulations, and the performance gains might vary with different technology scaling.
  • Future Work:
    • The authors suggest that the proposed approximate scan flip-flop can significantly reduce test power and time when used in scan chains of large lengths, implying potential for further exploration in large-scale designs.
  • Further investigation into optimizing the GDI-based design for even better performance or exploring hybrid approaches could be future directions.

Summary updated

BibTeX
@misc{konduru2022approximate,
  author = {Lakshmi Bhanuprakash Reddy Konduru and Vijaya Lakshmi and Jaynarayan T. Tudu},
  title = {{Approximate Scan Flip-flop to Reduce Functional Path Delay and Power Consumption}},
  year = {2022},
  eprint = {2212.12360},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2212.12360},
}
Figure