Paprica: A Pre-Synthesis Area Predictor for Baseband Circuit Auto-Generators

Auto-generation and auto-optimization for baseband circuits have significantly enhanced design productivity. Nevertheless, existing auto-optimizers primarily focus on optimizing RTL code for algorithm performance, neglecting critical hardware constraints. To overcome this limitation, this paper proposes Paprica, a pre-synthesis area predictor for baseband circuit auto-generators. Paprica bridges the gap between hardware parameters and area, providing auto-optimizers with accurate area predictions in milliseconds. To validate Paprica, we estimate the area of three baseband circuits (polar encoder, MIMO detector, and polar decoder). Experimental results demonstrate remarkable accuracy and efficiency in area prediction, achieving a mean absolute percentage error (MAPE) of 0.21%, 1.49%, and 1.20% with average end-to-end runtimes of 0.9 ms, 8.0 ms, and 6.6 ms, respectively. Paprica outperforms the state-of-the-art approach in area prediction, reducing MAPE by 98.0%, 91.8%, 78.4%, respectively, and decreasing average end-to-end runtime by over 99% across the three circuits.