300 likes | 561 Views
Zhe-Wei Jiang, Meng-Kai Hsu, Yao-Wen Chang and Kai-Yuan Chao From DAC2009. Spare-Cell-Aware Multilevel Analytical Placement. Outline. Background Problem formulation Analytical placement framework Integrate spare-cell consideration to AP Multilevel spare-cell-aware insertion
E N D
Zhe-Wei Jiang, Meng-Kai Hsu, Yao-Wen Chang and Kai-Yuan Chao From DAC2009 Spare-Cell-Aware Multilevel Analytical Placement
Outline • Background • Problem formulation • Analytical placement framework • Integrate spare-cell consideration to AP • Multilevel spare-cell-aware insertion • Experimental result
Introduction • There are two common ways to performing post-silicon debugging. • Metal fix : reuse the transistor masks and change only the metal layer masks. • FIB(Focus Ion Beam): cut unwanted electrical connections, or to deposit conductive material to make a connection after chip is manufactured Both of them require spare cells insertion
Outline • Background • Problem formulation • Analytical placement framework • Integrate spare-cell consideration to AP • Multilevel spare-cell-aware insertion • Experimental result
Problem formulation • Given a hypergraph H = (V,E) • Let V= be blocks E= be nets We intend to determine the optimal positions of movable blocks such that the wirelength is minimized, and there is no overlap among blocks.
Previous Work • PostSpare[26][22] : inserting spare cells after design placement. • UniSpare[10]: place spare cells among the chip uniformly before placement.
Outline • Background • Problem formulation • Analytical placement framework • Integrate spare-cell consideration to AP • Multilevel spare-cell-aware insertion • Experimental result
Flow chart Upper-level Uncoarsening stage Lower-level uncoarsening stage
Log-sum-exp wirelength Smooth the function When is small, log-sum-exp wirelength is close to the HPWL.
Potential function Bell-shaped smooth function
Solve Unconstraint programming Solve by conjugate gradient Apply quadratic penalty method. We solve a sequence of unconstrained minimization problem with increasing
Outline • Background • Problem formulation • Analytical placement framework • Integrate spare-cell consideration to AP • Multilevel spare-cell-aware insertion • Experimental result
Cluster Expansion • In most cases, the spare cell requirement is determined by designer’s experience. Designer need to set this parameter
Outline • Background • Problem formulation • Analytical placement framework • Integrate spare-cell consideration to AP • Multilevel spare-cell-aware insertion • Experimental result
Multilevel Spare Cell Insertion Top-down partition & spare cell allocation Partition according to block distribution 35:15 15:20 10:5 bottom-up spare cell position computation
Determination of Optimal Spare Cell Positions • The optimal positions that minimize the avg distance form blocks to spare cells.
Determination of Optimal Spare Cell Positions The sub-region usually contain less than 5 spare cell.
Outline • Background • Problem formulation • Analytical placement framework • Integrate spare-cell consideration to AP • Multilevel spare-cell-aware insertion • Experimental result
Experimental Result • Quality Comparison of Spare Cell Insertion IWLS 2005 benchmark
Experimental Result • HPWL Comparison after Spare Cell Insertion
Conclusions • The paper proposed the first spare-cell-aware analytical placement framework. • Experimental result have shown that this algorithm can achieve a much better solution quality of spare cell insertion than existing algorithms with only slight quality overhead.