Abstract:Applying deep learning (DL) techniques in the electronic design automation (EDA) field has become a trending topic. Most solutions apply well-developed DL models to solve specific EDA problems. While demonstrating promising results, they require careful model tuning for every problem. The fundamental question on "How to obtain a general and effective neural representation of circuits?" has not been answered yet. In this work, we take the first step towards solving this problem. We propose DeepGate, a novel representation learning solution that effectively embeds both logic function and structural information of a circuit as vectors on each gate. Specifically, we propose transforming circuits into unified and-inverter graph format for learning and using signal probabilities as the supervision task in DeepGate. We then introduce a novel graph neural network that uses strong inductive biases in practical circuits as learning priors for signal probability prediction. Our experimental results show the efficacy and generalization capability of DeepGate.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to obtain a general and effective circuit representation method in the field of electronic design automation (EDA). Although existing deep - learning techniques have shown good results in solving specific EDA problems, these methods usually require careful model tuning for each problem. In addition, there is currently no method to solve the fundamental question of "how to obtain a general and effective neural circuit representation?". For this reason, the authors propose DeepGate, a new representation - learning solution that can effectively embed the logical functions and structural information of a circuit into the vector of each gate. Specifically, the main contributions of the paper include: 1. **Proposing a new design flow**: This includes converting the circuit into a unified And - Inverter Graph (AIG) form, using the probability of logical simulation as a supervised task, and using a specially - designed Graph Neural Network (GNN) model for representation learning of circuit diagrams. 2. **Designing a new GNN model**: This model takes advantage of the unique properties of the circuit, including an attention mechanism that mimics the logical calculation process and a back - propagation layer that takes into account the logical influence effect. 3. **Handling the reconvergent structures in multi - level logic networks**: These structures are the main challenges in logical analysis. The paper treats them as first - class citizens in model design and introduces new solutions. Through the above methods, DeepGate can not only effectively predict the signal probability of circuit nodes, but also show good generalization ability on large - scale circuits. Experimental results show that DeepGate is superior to existing GNN models in terms of prediction accuracy and generalization ability.

DeepGate: Learning Neural Representations of Logic Gates

DeepGate3: Towards Scalable Circuit Representation Learning

DeepSeq: Deep Sequential Circuit Learning

Logic Synthesis with Generative Deep Neural Networks

Deep Differentiable Logic Gate Networks

GoodFloorplan: Graph Convolutional Network and Reinforcement Learning-Based Floorplanning

Versatile Multi-stage Graph Neural Network for Circuit Representation

EDA-ML: Graph Representation Learning Framework for Digital IC Design Automation

Understanding Graphs in EDA: From Shallow to Deep Learning

Circuit Transformer: End-to-end Circuit Design by Predicting the Next Gate

DE-HNN: An effective neural model for Circuit Netlist representation

DeepSeq2: Enhanced Sequential Circuit Learning with Disentangled Representations

Convolutional Differentiable Logic Gate Networks

BoolGebra: Attributed Graph-learning for Boolean Algebraic Manipulation

Graph Theory Based Machine Learning for Analog Circuit Design

Rethinking Arithmetic for Deep Neural Networks

The prediction of the quality of results in Logic Synthesis using Transformer and Graph Neural Networks

DeepLogic: Joint Learning of Neural Perception and Logical Reasoning.

Circuit as Set of Points

Circuit design completion using graph neural networks

Neural Combinatorial Logic Circuit Synthesis from Input-Output Examples