The rapid growth of Internet of Things (IoT) applications, such as autonomous vehicles and extended reality (XR), has led to an increased reliance on artificial intelligence (AI). To meet latency requirements and data privacy, training should be done on IoT devices. While federated learning (FL) can satisfy these data privacy requirements, concerns regarding the energy-intensive training pipeline still exist. Such large energy cost further imposes challenges on the sustainability of FL algorithms. Hence, in this chapter, we provide an overview of the emerging area of green FL that is primarily concerned with the design of energy-efficient and sustainable FL systems. We first discuss what challenges lie in the area of green FL. In particular, designing green FL requires maintaining energy efficiency and sustainability during the life cycle of FL algorithms in both the training and inference phases. Further, the scalability of green FL is also of importance for a real-world implementation over a large number of devices. These challenges are coupled with developing efficient learning algorithms under strict resource constraints in both devices and networks. Hence, green FL requires an end-to-end approach rather than focusing on solely improving the communication system or training phase. We then show how such a green approach requires the codesign of computing, learning algorithm, and communication systems and an understanding of the tradeoff between them. Subsequently, important research problems are presented to tackle those challenges. As an example of green FL research, the problem of optimizing the tradeoff between quantization level, energy, and learning performance is investigated.