2025-04-18 | | Total: 2
The independence of logic optimization and technology mapping poses a significant challenge in achieving high-quality synthesis results. Recent studies have improved optimization outcomes through collaborative optimization of multiple logic representations and have improved structural bias through structural choices. However, these methods still rely on technology-independent optimization and fail to truly resolve structural bias issues. This paper proposes a scalable and efficient framework based on Mixed Structural Choices (MCH). This is a novel heterogeneous mapping method that combines multiple logic representations with technology-aware optimization. MCH flexibly integrates different logic representations and stores candidates for various optimization strategies. By comprehensively evaluating the technology costs of these candidates, it enhances technology mapping and addresses structural bias issues in logic synthesis. Notably, the MCH-based lookup table (LUT) mapping algorithm set new records in the EPFL Best Results Challenge by combining the structural strengths of both And-Inverter Graph (AIG) and XOR-Majority Graph (XMG) logic representations. Additionally, MCH-based ASIC technology mapping achieves a 3.73% area and 8.94% delay reduction (balanced), 20.35% delay reduction (delay-oriented), and 21.02% area reduction (area-oriented), outperforming traditional structural choice methods. Furthermore, MCH-based logic optimization utilizes diverse structures to surpass local optima and achieve better results.
A delayed feedback reservoir (DFR) is a reservoir computing system well-suited for hardware implementations. However, achieving high accuracy in DFRs depends heavily on selecting appropriate hyperparameters. Conventionally, due to the presence of a non-linear circuit block in the DFR, the grid search has only been the preferred method, which is computationally intensive and time-consuming and thus performed offline. This paper presents a fast and accurate parameter optimization method for DFRs. To this end, we leverage the well-known backpropagation and gradient descent framework with the state-of-the-art DFR model for the first time to facilitate parameter optimization. We further propose a truncated backpropagation strategy applicable to the recursive dot-product reservoir representation to achieve the highest accuracy with reduced memory usage. With the proposed lightweight implementation, the computation time has been significantly reduced by up to 1/700 of the grid search.