Explore / Methods & Algorithms / Clock Tree Synthesis / Clock Mesh

Clock Mesh

Clock Tree Synthesis industry 1990s onward industry practice; analysis literature variation-proof power-expensive

The idea

Trees are efficient and fragile: each sink hangs off one path, so any variation along that path (device, wire, voltage, temperature) lands directly in its arrival time. A mesh inverts the trade. Lay a grid of horizontal and vertical clock wire over the region; drive the grid at many points from a balanced pre-tree; connect each flip-flop to the nearest grid segment. Now every sink is electrically connected to every other through many parallel paths: local variation is averaged away by sheer copper, and nearby sinks, the pairs that matter most for short-path timing, see nearly identical arrivals no matter what the drivers do. Measured and simulated local skews on meshes sit far below comparable trees, and tolerance to voltage droop and variation is the objective the structure serves.

The bill: enormous wire capacitance switching every cycle (clock power multiplies), many strong drivers, and analysis pain, since a multi-driven shorted grid is one giant coupled circuit requiring careful extraction and simulation rather than simple tree traversal.

Where you meet it today

High-performance microprocessors and anywhere timing margins are worth more than milliwatts: server cores, high-speed datapaths. Design questions become mesh pitch, driver placement, and stub lengths, an optimization literature of its own.

References and further reading