VLSI placement optimization using self-supervised graph clustering
A VLSI placement optimization framework receives a cell connectivity representation and cell characteristics and uses self-supervised graph clustering to optimize cell cluster assignments for power, performance, and area (PPA). The framework provides cell clustering constraints as placement guidance to commercial placers. Specifically, graph learning techniques are used to formulate the PPA metrics as machine learning loss functions that can be minimized directly through gradient descent. The framework improves the PPA metrics at the placement stage and the improvements endure to the post-route stage.
1. A computer-implemented method, comprising:
receiving a connectivity representation for cell instances of an integrated circuit;
processing, according to parameters, the connectivity representation and characteristics for each cell instance to generate clustering guidance for each cell instance;
updating the parameters to optimize metrics using the clustering guidance;
repeating the processing using the updated parameters to update the clustering guidance; and
producing cell cluster assignments for the cell instances based on the updated clustering guidance.
2. The computer-implemented method of claim 1 , further comprising initializing the parameters by:
producing learned embeddings for each cell instance from the connectivity representation and the characteristics; and
computing a first portion of the parameters to minimize a similarity loss for the learned embeddings.
3. The computer-implemented method of claim 2 , further comprising computing a second portion of the parameters from the learned embeddings.
4. The computer-implemented method of claim 1 , wherein the connectivity representation comprises a netlist graph and each cell instance corresponds to a node in the netlist graph.
5. The computer-implemented method of claim 4 , further comprising, before processing the connectivity representation, transforming the netlist graph by inserting an edge corresponding to a timing path between a start point and an end point of the timing path, the edge bypassing at least one node between the start point and the end point.
6. The computer-implemented method of claim 1 , wherein the characteristics for each cell instance comprise estimated timing, power consumption, and congestion for an initial placement.
7. The computer-implemented method of claim 1 , wherein the clustering guidance comprises clustering probabilities for each cell instance.
8. The computer-implemented method of claim 1 , wherein the processing comprises:
producing learned embeddings for each cell instance by applying a first portion of the parameters to the connectivity representation and the characteristics; and
applying a second portion of the parameters to the learned embeddings to compute the clustering guidance for each cell instance.
9. The computer-implemented method of claim 1 , wherein updating the parameters comprises minimizing a loss function computed using the clustering guidance for the metrics that include at least one of a timing loss, a congestion loss, and a power loss.
10. The computer-implemented method of claim 9 , wherein the loss function includes a clustering loss that is computed using the clustering guidance.
11. The computer-implemented method of claim 1 , wherein producing the cell cluster assignments comprises, for each cell instance, identifying a highest probability value defined in the updated clustering guidance.
12. The computer-implemented method of claim 1 , wherein at least one of the steps of receiving, processing, updating, and producing are performed on a server or in a data center to produce the cell cluster assignments, and the cell cluster assignments are streamed to a user device.
13. The computer-implemented method of claim 1 , wherein at least one of the steps of receiving, processing, updating, and producing are performed within a cloud computing environment.
14. The computer-implemented method of claim 1 , wherein the integrated circuit is employed in a machine, robot, or autonomous vehicle.
15. The computer-implemented method of claim 1 , wherein at least one of the steps of receiving, processing, updating, and producing is performed on a virtual machine comprising a portion of a graphics processing unit.
16. The computer-implemented method of claim 1 , wherein updating the parameters comprises:
computing a congestion score for each cell instance; and
equalizing the congestion scores across the cell clusters to minimize a congestion loss function.
17. The computer-implemented method of claim 1 , wherein updating the parameters comprises:
computing switching activity for each cell instance; and
aggregating the cell instances with high switching activity into a subset of the cell clusters to minimize a power loss function.
18. The computer-implemented method of claim 1 , wherein updating the parameters comprises:
computing cut sizes of timing critical paths for each cell instance; and
minimizing a timing loss function using the cut sizes.
19. A system, comprising:
a memory that stores a connectivity representation for cell instances of an integrated circuit; and
a processor that is connected to the memory, wherein the processor is configured to produce cell cluster assignments for the cell instances by:
process, according to parameters, the connectivity representation and characteristics for each cell instance to generate clustering guidance for each cell instance;
update the parameters to optimize metrics using the clustering guidance;
repeat the processing using the updated parameters to update the clustering guidance; and
produce the cell cluster assignments based on the updated clustering guidance.
20. The system of claim 19 , wherein the characteristics for each cell instance comprise estimated timing, power consumption, and congestion for an initial placement.
21. The system of claim 19 , wherein updating the parameters comprises minimizing a loss function computed using the clustering guidance for the metrics that include at least one of a timing loss, a congestion loss, and a power loss.
22. A non-transitory computer-readable media storing computer instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of:
receiving a connectivity representation for cell instances of an integrated circuit;
processing, according to parameters, the connectivity representation and characteristics for each cell instance to generate clustering guidance for each cell instance;
updating the parameters to optimize metrics using the clustering guidance;
repeating the processing using the updated parameters to update the clustering guidance; and
producing cell cluster assignments for the cell instances based on the updated clustering guidance.
23. The non-transitory computer-readable media of claim 22 , wherein the characteristics for each cell instance comprise estimated timing, power consumption, and congestion for an initial placement.