Abstract:Despite the great success of high-level synthesis (HLS) tools, we observe several unresolved challenges: 1) the high-level abstraction of HLS programming styles sometimes conceals optimization opportunities; 2) the actual quality of resulting RTL designs is hard to predict; and 3) existing HLS tools do not provide flexible tradeoff (Pareto) solutions among different objectives and constraints. To this end, we propose an end-to-end framework, namely, IronMan-Pro. The primary goal is to enable a flexible and automated design space exploration (DSE), to provide either optimized solutions under user-specified constraints or Pareto tradeoffs among different objectives (such as resource types, area, and latency). IronMan-Pro consists of three components: 1) GPP, a highly accurate graph-neural-network-based performance and resource predictor; 2) RLMD, a reinforcement-learning-based multiobjective design exploration engine for optimal resource allocation strategies, aiming to provide Pareto solutions among different objectives; and 3) CT, a code transformer to assist RLMD and GPP, which extracts the data flow graphs from original HLS C/C++ and automatically generates synthesizable code with optimized HLS directives. Experimental results show that, 1) GPP achieves high prediction accuracy, reducing the prediction errors of HLS tools by in resource utilization and in critical path (CP) timing; 2) compared with meta-heuristic-based techniques, IronMan-Pro generates superior solutions improving resource utilization by and CP timing by 7.6%–16.5%; and - ) under user-specified constraints, IronMan-Pro can find satisfying solutions over 96% of the cases, more than twice as many as that of meta-heuristic-based techniques and with a speedup of up to . This work demonstrates the great potential of applying machine learning algorithms in the electronic design automation domain, especially for the hard-to-solve problems, such as timing estimation and optimization. IronMan-Pro is available at https://github.com/lydiawunan/IronMan.

Graph Representation Learning for Microarchitecture Design Space Exploration.

Prior-Boosted GRL: Microarchitecture Design Space Exploration Via Graph Representation Learning

Effective and efficient microprocessor design space exploration using unlabeled design configurations

GoodFloorplan: Graph Convolutional Network and Reinforcement Learning-Based Floorplanning

Microarchitecture Design Space Exploration Via Pareto-Driven Active Learning.

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

Automated Design Space Exploration of Coarse-Grained Reconfigurable Architecture Via Bayesian Optimization

Parallel Multi-Objective Bayesian Optimization Framework for CGRA Microarchitecture

Graph Neural Networks for High-Level Synthesis Design Space Exploration

AUGER: A Multi-Objective Design Space Exploration Framework for CGRAs

BSSE: Design Space Exploration on the BOOM With Semi-Supervised Learning

A Benchmark on Directed Graph Representation Learning in Hardware Designs

IronMan-Pro: Multiobjective Design Space Exploration in HLS via Reinforcement Learning and Graph Neural Network-Based Modeling

A graph placement methodology for fast chip design

GRAEBO: FPGA General Routing Architecture Exploration Via Bayesian Optimization.

CDE: A Novel CGRA Development Environment with Fast Design Space Exploration Framework

ROI-HIT: Region of Interest-Driven High-Dimensional Microarchitecture Design Space Exploration

HierCGRA: A Novel Framework for Large-Scale CGRA with Hierarchical Modeling and Automated Design Space Exploration

HGBO-DSE: Hierarchical GNN and Bayesian Optimization Based HLS Design Space Exploration

Automated Design Space Exploration of CGRA Processing Element Architectures using Frequent Subgraph Analysis

BOOM-Explorer: RISC-V BOOM Microarchitecture Design Space Exploration