RL-Based CBL Placement for Flexible Digital Circuit Floorplanning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning-based methods for circuit floorplanning have limited practical applicability, requiring significant runtime and inflexible block placement sequences, and struggle to generalize to new circuit designs, leading to increased design time and sub-optimal floorplans.
Innovation Solution
Utilize reinforcement learning (RL) with a corner block list (CBL) representation and a hypergraph graph convolutional network (GCN) to encode netlist graphs, enabling efficient block placement and optimization of floorplans without predefined sequences, using a RL agent to generate CBL actions that minimize wirelength and adapt to circuit modifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If machine learning-based methods are used for floorplanning, then design time is reduced, but the methods have limited practical applicability and require significant runtime for training
Solution Approach 1:
The system performs preliminary action by pre-training the reinforcement learning agent on a dataset of circuit floorplans before actual floorplanning tasks. This pre-training phase prepares the model in advance, allowing it to quickly adapt to new circuits with minimal retraining, thus reducing both design time and training runtime.
Solution Approach 2:
The system uses dynamic adaptation by fine-tuning the pre-trained RL agent on specific circuit types or designs. The model can dynamically adjust its policy based on the characteristics of the target circuit, optimizing performance for different circuit complexities while maintaining efficient runtime.
2Extent of automation
If current machine learning-based floorplanning methods are used, then automation is improved, but the methods struggle to generalize to new circuit designs
Solution Approach 1:
The system exposes internal representations (such as state embeddings or feature vectors) from the pre-trained model during the floorplanning process. These exposed representations can be analyzed and used to adapt the model to new circuit types, improving generalization while maintaining high automation levels.
Solution Approach 2:
The system changes parameters by adjusting the pre-trained model's policy or embedding space to suit different circuit types. By modifying model parameters based on circuit characteristics, the system achieves better generalization across diverse designs without sacrificing automation.
3Adaptability or versatility
If reinforcement learning is used with traditional representations, then learning efficiency is reduced, but flexibility in block placement is improved
Solution Approach 1:
The system transitions from traditional discrete block placement representations to continuous embedding space representations. This dimensional change allows the RL agent to operate in a more expressive continuous space, improving learning efficiency while maintaining the flexibility to generate diverse block placement configurations.
Data Source
AI summary
Methods and systems for generating a floorplan for a circuit are disclosed. A netlist graph of the circuit and block features associated with blocks of the circuit are obtained. A reinforcement learning (RL) agent is used to generate a sequence of corner block list (CBL) actions. Each CBL action is generated by: generating a current state embedding representing a current state of the floorplan; and inputting the current state embedding to a policy network of the RL agent to generate a predicted output vector, which is used to generate the CBL action. After each CBL action is generated, the current CBL representation of the floorplan and the block features are updated to reflect the state of the floorplan after applying the CBL action. The CBL representation is outputted as a final floorplan after all blocks have been placed.