A new mechanism for SOC scan test scheduling

Ning Huang, En ZhuIEICE Electron. Express 2012 · OpenReview

Introduction

In semiconductor manufacturing, the efficiency of System-on-Chip (SOC) testing is a critical factor in overall production costs. Testing complex SOCs requires shifting vast amounts of data through limited Test Access Mechanism (TAM) channels. Traditionally, scan test scheduling has treated the entire test set of a core as an indivisible block. However, a standard scan test pattern consists of two distinct phases: the Load-Unload phase, which heavily utilizes TAM channels, and the Capture phase, during which the actual logic test occurs and the TAM channels remain idle.

This paper, originally published in IEICE Electronics Express, identifies this idle time as a significant inefficiency. The authors propose a granular scheduling mechanism that operates at the individual scan pattern level rather than the core level, allowing for the overlap of capture phases with the loading of subsequent patterns.

The Pattern-Level Scheduling Model

The core innovation of this research is the shift in the scheduling "object." By breaking down test sequences into individual patterns, the TAM can finish loading one core and immediately begin loading another while the first core is still performing its internal capture.

To optimize this process, the authors model the scheduling task as a 2D Bin Packing problem under color constraints:

  • The Bin: Represents the total available TAM width and the total test time.
  • The Rectangles: Each scan pattern is represented as a rectangle, where the width corresponds to the channel requirement and the height corresponds to the load-unload time.
  • The Colors: Each core is assigned a unique color to distinguish its patterns.
  • The Constraint: To prevent data corruption and respect hardware limitations, only one pattern from a specific core can be active at a time. Mathematically, this requires a vertical gap between rectangles of the same color, equivalent to the capture time ($TC$).
The 2D bin packing model with color constraints, illustrating the required capture time (TC) gap between patterns of the same core.

Optimization Strategy

Since pattern-level scheduling is an NP-hard optimization problem, the authors employ a combination of Sequence Pairs and Simulated Annealing (SA). Sequence pairs provide a symbolic representation to map the relative positions of rectangles efficiently. The SA algorithm then iteratively perturbs the schedule—swapping patterns to find a global minimum for test time—while occasionally accepting suboptimal changes to avoid local optima.

Experimental Results

The proposed mechanism was evaluated using three large-scale industrial SOC designs: T273 (6 cores, 2009 patterns), T5527 (15 cores, 5176 patterns), and T9017 (30 cores, 8965 patterns). These designs represent high-complexity chips where test scheduling is a major bottleneck.

Comparing the pattern-level mechanism against a state-of-the-art Ant Colony Optimization (ACO) approach across different TAM widths (16 and 32), the authors achieved:

  • An average reduction in total test application time of 3.7% to 4.15%.
  • Consistent performance improvements across all industrial designs, regardless of core count or channel width.

While a 4% reduction may seem modest, in high-volume manufacturing, it translates to significant savings in tester overhead and increased factory throughput.

Conclusion and Future Work

The study demonstrates that increasing the granularity of SOC test scheduling can effectively reclaim wasted idle time on test channels. However, the authors note that overlapping multiple operations across different cores could lead to power spikes. Future research will likely focus on integrating power-dissipation constraints into this pattern-level scheduling model to ensure that optimized schedules remain physically safe for the hardware.

Summary updated

BibTeX
@inproceedings{huang2012new,
  author = {Ning Huang and En Zhu},
  title = {{A new mechanism for SOC scan test scheduling}},
  booktitle = {IEICE Electron. Express},
  year = {2012},
  url = {https://openreview.net/forum?id=CdlEISx57f2},
}

Loading related content...

Figure