This work proposes a high-quality analytical placement algorithm considering wirelength, preplaced blocks, and density based on the log-sum-exp wirelength model proposed by Naylor and the multilevel framework and uses the conjugate gradient method to find better macro positions.
Abstract:
In addition to wirelength, modern placers need to consider various constraints such as preplaced blocks and density. We propose a high-quality analytical placement algorithm considering wirelength, preplaced blocks, and density based on the log-sum-exp wirelength model proposed by Naylor and the multilevel framework. To handle preplaced blocks, we use a two-stage smoothing technique, i.e., Gaussian smoothing followed by level smoothing, to facilitate block spreading during global placement (GP). The density is controlled by white-space reallocation using partitioning and cut-line shifting during GP and cell sliding during detailed placement. We further use the conjugate gradient method with dynamic step-size control to speed up the GP and macro shifting to find better macro positions. Experimental results show that our placer obtains very high-quality results.
TL;DR: EDA/VLSI practitioners and researchers in need of fluency in an "adjacent" field will find this an invaluable reference to the basic EDA concepts, principles, data structures, algorithms, and architectures for the design, verification, and test of VLSI circuits.
TL;DR: SimPL is a self-contained, flat, force-directed algorithm for global placement that is simpler than existing placers and easier to integrate into timing-closure flows.
TL;DR: This work presents a learning-based approach to chip placement, and shows that, in under 6 hours, this method can generate placements that are superhuman or comparable on modern accelerator netlists, whereas existing baselines require human experts in the loop and take several weeks.
TL;DR: In this article, the authors presented a deep reinforcement learning approach to chip floorplanning, which can automatically generate chip floorplans that are superior or comparable to those produced by humans in all key metrics, including power consumption, performance and chip area.
TL;DR: A shortest augmenting path algorithm for the linear assignment problem that contains new initialization routines and a special implementation of Dijkstra's shortest path method is developed.
TL;DR: The authors present a placement method for cell-based layout styles that is composed of alternating and interacting global optimization and partitioning steps that are followed by an optimization of the area utilization.
TL;DR: The algorithm is capable of addressing the problems of global placement, floorplanning, timing minimization and interaction to logic synthesis, and its iterative nature assures that timing requirements are precisely met.
TL;DR: A generalized force-directed algorithm embedded in mPL2's multilevel framework is presented, which produces the shortest wirelength among all published placers with very competitive runtime on the IBM circuits used in [29].
Q1. What are the contributions in "Ntuplace3: an analytical placer for large-scale mixed-size designs with preplaced blocks and density constraints" ?
The authors propose a high-quality analytical placement algorithm considering wirelength, preplaced blocks, and density based on the log-sum-exp wirelength model proposed by Naylor et al. and the multilevel framework. To handle preplaced blocks, the authors use a two-stage smoothing technique, i. e., Gaussian smoothing followed by level smoothing, to facilitate block spreading during global placement ( GP ). The authors further use the conjugate gradient method with dynamic step-size control to speed up the GP and macro shifting to find better macro positions.
Q2. What is the effect of smoothing potential levels on the placement area?
Smoothing potential levels reduce “mountain” (highpotential regions) heights so that movable blocks can smoothly spread to the whole placement area.
Q3. What is the white space of the root?
The white space of the root is 3, and it should always be greater than or equal to 0, or the blocks can never fit into the placement region.
Q4. How does the algorithm reduce the densities of overflowed bins?
The authors divide the placement region into uniform bins, and then their algorithm iteratively reduces the densities of overflowed bins by sliding the cells from denser bins to sparser ones while the cell order is preserved.
Q5. How does the WDP algorithm solve the bipartite matching problem?
The WDP algorithm finds a group of exchangeable cells inside a given window and formulates a bipartite matching problem by matching the cells to all empty slots in the window.
Q6. How can the authors obtain a smoother base potential?
Applying convolution to the Gaussian function G with the basepotential P asP ′(x, y) = G(x, y) ∗ P (x, y) (15)we can obtain a smoother base potential P ′.
Q7. What is the way to solve the bipartite matching problem?
Though the bipartite matching problem can optimally be solved in polynomial time, the optimal assignment cannot guarantee the optimal HPWL result because the HPWL cost of a cell connected to each empty slot depends on the positions of other connected cells.
Q8. How do the authors allocate white space to the two children?
2) If the two children both have white spaces greater than or equal to 0, the authors allocate the white space proportional to their original white space amount.
Q9. What is the CG search with dynamic step size control?
For the analytical search, the CG search with dynamic step-size control is adopted to speed up the search for a desirable solution.
Q10. What is the average HPWL of the placer?
On average, their resulting HPWL is smaller than that of APlace 2.0 by 5% and similar to mPL6’s, and their placer is 10.32× and 2.56× faster than APlace 2.0 and mPL6, respectively.