The field of automated design and optimization is moving towards increased efficiency and accuracy in various areas, including power modeling, transistor-level simulation, and synthesis. Researchers are exploring new methods and techniques to improve the performance of existing tools and algorithms, such as automated power modeling, self-supervised learning, and equivalence of RC long-chain structures. Notable papers in this area include AutoPower, which achieves a low mean absolute percentage error of 4.36% in architecture-level power modeling, and ATLAS, which predicts time-based layout power with a mean absolute percentage error of only 0.58%. Other noteworthy papers are e-boost, which demonstrates a 558x runtime speedup over traditional exact approaches in e-graph extraction, and Cristal, which constructs higher-quality choice networks for synthesis and technology mapping. These advancements have the potential to significantly impact the field of automated design and optimization, enabling faster and more accurate design and optimization of complex systems.
Advances in Automated Design and Optimization
Sources
Automating the Derivation of Unification Algorithms: A Case Study in Deductive Program Synthesis
AutoPower: Automated Few-Shot Architecture-Level Power Modeling by Power Group Decoupling
ATLAS: A Self-Supervised and Cross-Stage Netlist Power Model for Fine-Grained Time-Based Layout Power Analysis
e-boost: Boosted E-Graph Extraction with Adaptive Heuristics and Exact Solving
Accelerating Transistor-Level Simulation of Integrated Circuits via Equivalence of RC Long-Chain Structures
Bisimilarity and Simulatability of Processes Parameterized by Join Interactions
Revisit Choice Network for Synthesis and Technology Mapping
Close is Good Enough: Component-Based Synthesis Modulo Logical Similarity