Zero skew clock routing with minimum wirelength
read more
Citations
Clock distribution networks in synchronous digital integrated circuits
Performance optimization of VLSI interconnect layout
Modeling and analysis of nonuniform substrate temperature effects on global ULSI interconnects
A Clustering-Based Optimization Algorithm in Zero-Skew Routings
Zero-skew clock routing trees with minimum wirelength
References
The Transient Response of Damped Linear Networks with Particular Regard to Wideband Amplifiers
The Rectilinear Steiner Tree Problem is $NP$-Complete
Signal Delay in RC Tree Networks
Clock skew optimization
Approximation of wiring delay in MOSFET LSI
Related Papers (5)
Frequently Asked Questions (21)
Q2. Why is the load capacitance of each node equal to the amount of wire in its?
Because unit resistance and capacitance both equal one, and because loading capacitances at the leaves are zero, the tree capacitance of each node equals the amount of wire in its subtree.
Q3. What is the motivation behind the work?
The primary motivation behind their work is to minimize the total wirelength of clock routing trees while maintaining exact zero skew with respect to the appropriate delay model.
Q4. What is the linear delay model for clock nets?
With smaller device dimensions and higher ASIC system speeds, a distributed RC tree model for signal delay in clock nets is often required to derive accurate timing information.
Q5. What is the effect of minimization of total wire length?
minimization of total wirelength will lead to reduction of wiring area, with the added e ect of less blockage for subsequent routing phases of layout.
Q6. How can DME be used for a given problem?
The DME algorithm can also be used for problems with allowed skew [1] [13] [25], where the signal must arrive at each sink within a prescribed segment of time.
Q7. What is the way to reduce clock skew?
The H-tree structure can signi cantly reduce clock skew [10] [26], but is applicable only when all of the sinks have identical loading capacitances and are placed in a symmetric array.
Q8. What is the definition of a Manhattan arc?
A Manhattan arc is de ned to be a line segment, possibly of zero length, with slope +1 or -1; in other words, a Manhattan arc is a line segment tilted at 45 degrees from the wiring directions.
Q9. What is the cost of merging two trees?
The merging cost depends on the distance between the two roots of the ZSTs, the delay of each ZST, and the tree capacitance of each ZST.
Q10. What was the surprising result of their experiments?
the authors used the circuit simulator SPICE2G.6 [20] to evaluate13A surprising outcome of their experiments was the strong performance of topologies generated by the KCR algorithm.
Q11. How long does it take to build a tree of segments?
By Lemma 1, procedure Build Tree of Segments requires constant time to compute each new mergingsegment, and time linear in the size of S to construct the entire tree of merging segments.
Q12. What is the topology of the merging segment?
The merging segment of a node depends on the merging segments of its two children, so the connection topology must be processed in a bottom-up order.
Q13. What is the linear delay between a source and a sink?
Normalized by an appropriate constant factor, the linear delay between any two nodes u and w in a source-sink path istLD(u;w) = Xev2path(u;w)jevj:While less accurate than the distributed RC tree delay formulas of Rubinstein et al [22], the linear delay model has been e ectively used in clock tree synthesis [18] [21].
Q14. How does the DME algorithm achieve the optimal clock tree length?
In practice, the DME algorithm begins with an initial clock tree computed by any previous method, then maintains exact zero clock skew while reducing the wirelength.
Q15. What is the method for achieving the optimal zero skew clock tree?
In regimes where the linear delay model applies, their method produces the optimal (i.e., minimum wirelength) zero skew clock tree with respect to the prescribed topology, and this tree will also enjoy optimal source-sink delay.
Q16. How do the authors minimize the merging cost?
to minimize the merging cost the authors should therefore choose topologies such that merged subtrees have minimum distance between their roots, along with similar capacitances and delays, so as to avoid the extra cost 0 .
Q17. What is the average improvement of BB+DME over Tsay's algorithm?
It should be noted that DME alone resulted in an average improvement of only 2% over Tsay's algorithm, which can be attributed to the fact that Tsay's embedding algorithm allows deferral of the choice of placements for one level in the tree (the two endpoints of each merging segment are selected and carried to the next level, where the actual embedding is chosen to be the point which allows the minimum connection cost).
Q18. What is the robust method of clock tree construction for cellbased layouts?
A more robust clock tree construction for cellbased layouts is due to Jackson, Srinivasan and Kuh [17]: their \\method of means and medians" (MMM) algorithm generates a topology by recursively partitioning the set of sinks into two equal-sized subsets, then connecting the center of mass of the entire set to the centers of mass of the two subsets.
Q19. What is the NP-completeness of the linear delay problem?
[18] [9] showed that a closely related problem (in the linear delay model), the \\bounded-skew pathlength-balanced tree problem", is trivially NP-complete since it reduces the minimum rectilinear Steiner tree problemwhen the allowed pathlength skew is in nite.
Q20. How is the tree of segments constructed?
Once the tree of segments has been constructed, the exact embeddings of internal nodes in the ZST are chosen in a top-down manner.
Q21. What is the way to calculate the edge lengths needed to merge two trees of merging?
To calculate the edge lengths needed to merge two trees of merging segments TSa and TSb with minimum merging cost in the Elmore model, the authors use the analysis of Tsay [25].