This study proposes a framework that integrates Physics-Informed Neural Networks (PINNs) into Conditional Generative Adversarial Networks (cGANs) to enhance model performance in data scarcity [1-2]. While developing generative models for new domains typically requires extensive datasets, real-world engineering data is often scarce and non-uniformly distributed [3]. In this research, we quantify data sparsity by analyzing the distances between image data points and develop a generator framework that adaptively incorporates these sparsity characteristics. This approach aims to build a robust model capable of capturing the unique features of sparse data while maintaining physical consistency. To validate the proposed method, we applied it to the design of tire tread patterns, which are critical to vehicle performance [4-6]. Specifically, when generating winter tread patterns where data is limited, the model augments sparse datasets with snow traction physics as auxiliary information. This synergy between data-driven distribution learning and PINN-based physical constraints significantly improves overall performance. Numerical results demonstrate that our approach outperforms conventional generative methods, successfully synthesizing high-quality winter patterns even under data scarcity.