ArtNet: Hierarchical Clustering-Based Artificial Netlist Generator for ML and DTCO Application

TL;DR

ArtNet addresses the critical bottleneck of limited real-world data for ML-guided PPA optimization and DTCO by generating artificial netlists that faithfully reproduce topology, interconnect complexity via Rent’s rule, and timing characteristics. It uses a hierarchical clustering–based approach with PI/PO balancing and timing-aware net generation to produce scalable, macro-capable netlists, which are validated against real designs through Rent’s analysis, P&R QoR, and design locality metrics. The framework yields tangible benefits: CNN routability predictions improved with ArtNet augmentation (F1 up by 0.16), and DTCO mini-brains achieve PPA-matching losses as low as 2.06% while slashing P&R runtime by up to 95%. These results demonstrate ArtNet’s value as a fast, scalable proxy for real circuits, accelerating ML data generation and DTCO exploration while maintaining structural and timing fidelity; open-source integration with OpenROAD further broadens its impact for EDA research and practice.

Abstract

In advanced nodes, optimization of power, performance and area (PPA) has become highly complex and challenging. Machine learning (ML) and design-technology co-optimization (DTCO) provide promising mitigations, but face limitations due to a lack of diverse training data as well as long design flow turnaround times (TAT). We propose ArtNet, a novel artificial netlist generator designed to tackle these issues. Unlike previous methods, ArtNet replicates key topological characteristics, enhancing ML model generalization and supporting broader design space exploration for DTCO. By producing realistic artificial datasets that moreclosely match given target parameters, ArtNet enables more efficient PPAoptimization and exploration of flows and design enablements. In the context of CNN-based DRV prediction, ArtNet's data augmentationimproves F1 score by 0.16 compared to using only the original (real) dataset. In the DTCO context, ArtNet-generated mini-brains achieve a PPA match up to 97.94%, demonstrating close alignment with design metrics of targeted full-scale block designs.

Figures (13)

• Figure 1: Use cases of artificial netlist generation. ArtNet enables netlist generation from user-specified input parameters, and from parameters of a given target design. The output netlist can be used for design benchmarking, ML data augmentation, DTCO and other applications.
• Figure 2: Overall ArtNet framework.
• Figure 3: Hierarchical clustering from a given priority queue.
• Figure 4: Net generation flow: (a) calculate In/Out terminals to be created; (b) determine additional FFs from $Q_{seq}$; and (c) make connections.
• Figure 5: PI/PO matching methods: (a) add PI; (b) delete PI; (c) add PO; and (d) delete PO.