Explore / Methods & Algorithms / Clock Tree Synthesis / H-Tree and X-Tree

H-Tree and X-Tree

Clock Tree Synthesis foundational 1980s folklore and many authors zero skew by symmetry assumes uniform sinks

The idea

Perfect symmetry needs no analysis. Draw an H; drive it from the exact center; the four tips are equidistant from the driver by construction. Put a smaller H at each tip and recurse: after k levels, 4^k endpoints all see identical wire length, identical topology, identical delay. Skew, the difference in clock arrival between sinks, is zero not because anything was computed but because nothing distinguishes one leaf from another. The X-tree variant uses diagonal branches for shorter paths where 45-degree routing is allowed.

Where symmetry breaks

The construction assumes what real chips never provide: sinks arranged uniformly, identical loads at every tip, homogeneous wire layers, and no blockages forcing detours. Any asymmetry in load or route converts directly into skew, and pure H-trees also spend a lot of wire reaching regions that may contain few flip-flops. So in practice the H-tree survives as the top of the network: a symmetric spine distributing the clock to a grid of regional tap points, below which load-aware algorithms (the next three pages) take over, or a mesh shorts the taps together.

Where you meet it today

Top-level clock distribution of high-performance processors, standard figures in every textbook, and as the mental model for why 'balanced' and 'symmetric' are the first words anyone says about clock networks.

References and further reading