Access Official Technical PDF

Share Summary:LinkedInEmail

In modular ASIC design, the ability to assemble reusable hardware blocks depends on more than their logic and interfaces. Clock distribution, power delivery and energy estimation must also behave predictably when blocks are placed together. The SiLago framework addresses these physical-design challenges by raising the design abstraction from standard cells toward microarchitectural components, with the goal of supporting composition by abutment: placing predesigned blocks side by side without redesigning their internal structure.

Three areas are central to that goal: ASIC power characterization, hierarchical clock tree synthesis and power grid dimensioning. Together, they show why physical-design planning must account for interactions between blocks as well as the behavior of each block on its own.

Why modular physical design needs accurate models

Traditional standard-cell design flows require repeated refinement because high-level design stages lack details about placement, routing and physical cell arrangements. That missing information makes it harder to predict area, timing and power early. A higher-level methodology can reduce the search space for synthesis, but its predictions are useful only if the constituent blocks and their connections are characterized accurately.

SiLago uses coarse-grain, microarchitectural elements, including dynamically reconfigurable resource arrays and distributed memory. Its design approach seeks to preserve regularity as these elements are assembled into larger systems. Irregular physical changes—such as different clock-buffer counts from block to block—can undermine that regularity and make system-level behavior harder to predict.

ASIC power characterization must include coupling

Power estimation is especially sensitive to physical context. It depends not only on the operation being performed, but also on signal activity, the distance signals travel and interactions between nearby wires. When a switching signal, known as an aggressor, is capacitively coupled to a neighboring victim net, it can induce a voltage disturbance. Depending on the circuit and timing, this crosstalk may contribute to functional errors or setup and hold violations.

For a modular fabric, characterizing each block in isolation may therefore miss effects that arise when blocks are placed close together. The SiLago work extended power characterization to a flat design so that interactions between neighboring data-path units could be considered alongside the energy behavior of individual cells.

The characterization process used a sequence of design and analysis steps:

  1. Generate test cases that exercise different operations and signal connections.
  2. Translate those cases into testbenches and corresponding assembly code.
  3. Simulate the fabric and produce value change dump files that record signal activity.
  4. Use the activity data to generate power reports from the physical-design flow.

Intra-cell characterization describes the energy associated with an individual cell’s modes and connections. Inter-cell characterization captures coupling between data-path units. Taken together, these models can help designers estimate energy for workloads without rerunning every experiment for every new design instance. The work examined patterns across operations such as addition, multiplication and multiply-accumulate, and used CNN, two-dimensional DCT and face-recognition workloads to illustrate power distribution.

The practical lesson is that activity alone does not tell the whole story. Signal routes and neighboring circuitry can change the observed power behavior. Characterization intended for modular designs should therefore reflect the physical relationships that will exist when blocks are assembled.

Hierarchical clock tree synthesis for predictable blocks

Clock tree synthesis (CTS) distributes the clock from its source to sequential elements across a chip. Its implementation must manage insertion delay, skew and signal transition constraints. Automated CTS tools typically insert and size buffers to meet timing objectives, but the resulting buffer arrangement can vary with placement and other design conditions.

That variability creates a challenge for composition by abutment: if nominally identical blocks receive different clock structures, their physical and timing characteristics may no longer be uniform. The SiLago approach therefore considered a hierarchical clock tree, separating local clock distribution within a block from regional distribution across blocks and global clock generation.

At the regional level, the intended scheme uses structured clock distribution and programmable delay buffering. The aim is to make clock behavior and its associated cost more predictable while preserving the ability to assemble blocks into larger fabrics. In the described implementation, a multiplexer-register-buffer structure provides selectable buffering, while the local clock tree remains part of the predesigned block.

This is not simply a matter of minimizing skew at any cost. The design must also maintain acceptable slew and preserve the regular structure required by the methodology. Static timing analysis and comparisons with functionally equivalent clock structures are relevant checks when evaluating whether a proposed tree meets its timing objectives.

Power grid dimensioning across levels

Clock planning is only one part of physical integration. Every standard cell must receive power through a distribution network that can support the assembled design. A hierarchical power-grid concept connects global supply nets to regional structures, then to block-level rings and ultimately to local power rails.

For a modular fabric, the geometry and interfaces of these power structures matter as much as their electrical purpose. A grid that cannot align across abutted blocks may require additional engineering after assembly, weakening the benefits of reusable physical components. The proposed SiLago scheme emphasizes regular distribution and block compatibility, with global nets feeding local structures at planned intervals.

Power-grid dimensioning must ultimately account for the electrical requirements of the design and the physical arrangement of its blocks. The framework describes the hierarchy and the need for compatibility; it does not remove the need to verify the resulting implementation for supply integrity.

Practical implications for design teams

For engineers and system integrators evaluating modular physical-design flows, the key takeaway is to treat power, clocking and block composition as connected design problems. Useful review questions include:

  • Does power characterization include both individual-cell behavior and interactions between physically close blocks?
  • Can the clock structure remain regular when blocks are assembled, and are delay and skew checked at the appropriate hierarchy levels?
  • Do global, regional and local power structures connect consistently across block boundaries?
  • Are the assumptions behind early energy and timing estimates aligned with the physical implementation?

These checks help expose gaps between a logically reusable block and a physically composable one. They also clarify what must be characterized or verified before a design methodology can provide dependable system-level predictions.

Conclusion

Modular ASIC design requires physical infrastructure that scales with the logic. The SiLago work highlights how power characterization that considers coupling, hierarchical clock tree synthesis and structured power-grid planning can support more predictable block-based integration. For design teams, the practical priority is to validate these elements together: regular blocks are valuable only when their clock, power and interconnect behavior remain compatible in the assembled design.

Access Official Technical PDF

About The Author