The article reviews recent advances in integrated photonic neuromorphic systems, discussing current and future challenges, and outlining the scientific and technological advancements needed to address these challenges. Photonic computing has seen significant growth due to the integration of optoelectronic components on photonic platforms, enabling ultrafast artificial neural networks. These networks have the potential to meet the growing demand for machine learning and artificial intelligence in various fields such as medical diagnosis, telecommunications, and high-performance computing. Neuromorphic photonics offers sub-nanosecond latencies, providing a complementary solution to conventional computing architectures, which are inefficient for distributed, massively parallel, and adaptive computational models like those used in neural networks. The review highlights the advantages of using optics for matrix multiplication and interconnects, which are crucial for implementing neural networks. It discusses various hardware implementations, including weighted interconnects (synapses) and nonlinearities (neurons), and the challenges and solutions in each area. Key technologies needed for scaling neuromorphic photonic hardware to practical systems are also covered, such as active on-chip electronics and light sources. The article emphasizes the importance of integrating electronics and photonics, and explores emerging ideas and technologies, such as non-volatile memory, photonic digital-to-analog converters (DACs), and frequency comb sources, to enhance the performance and efficiency of neuromorphic photonics.The article reviews recent advances in integrated photonic neuromorphic systems, discussing current and future challenges, and outlining the scientific and technological advancements needed to address these challenges. Photonic computing has seen significant growth due to the integration of optoelectronic components on photonic platforms, enabling ultrafast artificial neural networks. These networks have the potential to meet the growing demand for machine learning and artificial intelligence in various fields such as medical diagnosis, telecommunications, and high-performance computing. Neuromorphic photonics offers sub-nanosecond latencies, providing a complementary solution to conventional computing architectures, which are inefficient for distributed, massively parallel, and adaptive computational models like those used in neural networks. The review highlights the advantages of using optics for matrix multiplication and interconnects, which are crucial for implementing neural networks. It discusses various hardware implementations, including weighted interconnects (synapses) and nonlinearities (neurons), and the challenges and solutions in each area. Key technologies needed for scaling neuromorphic photonic hardware to practical systems are also covered, such as active on-chip electronics and light sources. The article emphasizes the importance of integrating electronics and photonics, and explores emerging ideas and technologies, such as non-volatile memory, photonic digital-to-analog converters (DACs), and frequency comb sources, to enhance the performance and efficiency of neuromorphic photonics.