Time Borrowing and Cycle Stealing in Sequential Circuits

This direction is merged into another direction: process variation. Its last version stays readable and citable here.

Version 1 of 117 papers35 conceptsfrozen 2026-10-04

In traditional digital design, timing is often viewed as a "bucket brigade." Each flip-flop acts as a rigid wall: a signal must arrive before the clock edge (setup) and stay stable shortly after (hold). If a signal is late, the entire circuit fails. This rigid partitioning is simple to analyze but leaves significant performance on the table.

A more fluid approach, championed by researchers like David G. Chinnery and Bing Li, treats timing as a shared resource. By using level-sensitive latches or tunable clock buffers, designers can implement time borrowing (or cycle stealing), allowing a slow logic path to "borrow" time from a subsequent faster path.

The Fluidity of Time Borrowing

In a standard edge-triggered system, the clock period is dictated by the single slowest path in the entire design. If one stage takes 110% of the clock cycle and the next takes only 50%, the circuit fails.

As David G. Chinnery explains in Closing the Gap Between ASIC and Custom, replacing flip-flops with level-sensitive latches makes the boundaries between clock cycles transparent. Because a latch is "open" for the duration of a clock phase rather than a single instant, a signal that arrives late can still pass through. This effectively averages the delay across multiple stages, allowing the 110% path to "steal" the 10% it needs from the 50% path.

The Challenge of Circular Dependencies

If time borrowing is so effective, why isn't it the default? The answer lies in the complexity of analysis. Traditional Static Timing Analysis (STA) assumes that each stage is independent. Once you allow time borrowing, the arrival time at one gate depends on the arrival time at the previous gate, which depends on the one before that.

This creates what Pinhong Chen and Desmond A. Kirkpatrick identify as a circular dependency in Static Crosstalk-Noise Analysis. In deep sub-micron designs, these dependencies are exacerbated by crosstalk: the timing of a signal depends on the noise from its neighbors, but that noise itself depends on when those neighbors transition.

Beyond Constant Delay Models

To make time borrowing predictable, we must move beyond the "constant delay" simplification. In physical hardware, the time a latch takes to output a signal (clock-to-Q) isn't a fixed number; it actually increases as the input signal gets closer to the timing deadline.

Grace Li Zhang, Bing Li, and Ulf Schlichtmann addressed this in PieceTimer. They modeled the interdependency between setup time, hold time, and clock-to-output delay as a three-dimensional piecewise linear surface. By recognizing that these parameters are coupled, they showed that designers can squeeze more performance out of a circuit without risking the stability of the sequential loops created by cycle stealing.

Sequential Criticality and Post-Silicon Tuning

The rise of manufacturing variations means that a path that is fast on one chip might be slow on another. To combat this, designers use Post-Silicon Tunable (PST) clock buffers—essentially "knobs" that can delay or advance the clock edge for specific registers after the chip is manufactured.

This introduces a new concept: Sequential Criticality. In a traditional design, a "critical path" is simply the longest one. But in a design with PST buffers, Bing Li and Ulf Schlichtmann argue in Statistical Timing Analysis and Criticality Computation that a gate is only truly critical if its delay cannot be mitigated by borrowing time from its neighbors.

Using graph-transformation methods, they developed algorithms to identify these true bottlenecks. This allows designers to focus their optimization efforts—like gate sizing—only on the components that actually limit the final yield, rather than wasting area and power on paths that can be "fixed" via clock tuning.

The Balancing Act

The ultimate goal of this research direction is to move timing analysis from a local, pessimistic check to a global, statistical optimization. As Grace Li Zhang notes in EffiTest, this is always a balancing act: adjusting a clock buffer to fix a setup violation (a signal arriving too late) might inadvertently cause a hold violation (a signal arriving so early it overwrites the previous data).

By treating the entire sequential path as a single, interconnected system—rather than a series of isolated boxes—this research enables chips that are faster, more resilient to manufacturing defects, and closer to the theoretical limits of their silicon.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_time_borrowing_and_cycle_stealing_in_seq_368e12d5,
  title = {Time Borrowing and Cycle Stealing in Sequential Circuits},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/time-borrowing-and-cycle-stealing-in-sequential-circuits/rdr_368e12d5bd8e4357ba55be4ec511ceeb},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_time_borrowing_and_cycle_stealing_in_seq_368e12d5_v1,
  title = {Time Borrowing and Cycle Stealing in Sequential Circuits},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/time-borrowing-and-cycle-stealing-in-sequential-circuits/rdr_368e12d5bd8e4357ba55be4ec511ceeb/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}