Microcombs have sparked a surge of applications over the past decade, ranging from optical communications to metrology^1–4. Despite their diverse deployment, most microcomb-based systems rely on a large amount of bulky elements and equipment to fulfil their desired functions, which is complicated, expensive and power consuming. By contrast, foundry-based silicon photonics (SiPh) has had remarkable success in providing versatile functionality in a scalable and low-cost manner^5–7, but its available chip-based light sources lack the capacity for parallelization, which limits the scope of SiPh applications. Here we combine these two technologies by using a power-efficient and operationally simple aluminium-gallium-arsenide-on-insulator microcomb source to drive complementary metal–oxide–semiconductor SiPh engines. We present two important chip-scale photonic systems for optical data transmission and microwave photonics, respectively. A microcomb-based integrated photonic data link is demonstrated, based on a pulse-amplitude four-level modulation scheme with a two-terabit-per-second aggregate rate, and a highly reconfigurable microwave photonic filter with a high level of integration is constructed using a time-stretch approach. Such synergy of a microcomb and SiPh integrated components is an essential step towards the next generation of fully integrated photonic systems.

The emergence of parallel convolution-operation technology has substantially powered the complexity and functionality of optical neural networks (ONN) by harnessing the dimension of optical wavelength. However, this advanced architecture faces remarkable challenges in high-level integration and on-chip operation. In this work, convolution based on time-wavelength plane stretching approach is implemented on a microcomb-driven chip-based photonic processing unit (PPU). To support the operation of this processing unit, we develop a dedicated control and operation protocol, leading to a record high weight precision of 9 bits. Moreover, the compact architecture and high data loading speed enable a preeminent photonic-core compute density of over 1 trillion of operations per second per square millimeter (TOPS mm⁻²). Two proof-of-concept experiments are demonstrated, including image edge detection and handwritten digit recognition, showing comparable processing capability compared to that of a digital computer. Due to the advanced performance and the great scalability, this parallel photonic processing unit can potentially revolutionize sophisticated artificial intelligence tasks including autonomous driving, video action recognition and image reconstruction.

The rising demand for high scanning accuracy and resolution in sensors for self-driving vehicles has led to the rapid development of parallelization in light detection and ranging (LiDAR) technologies. However, for the two major existing LiDAR categories—time-of-flight and frequency-modulated continuous wave—the light sources and measurement principles currently used for parallel detection face severe limitations from time- and frequency-domain congestion, leading to degraded measurement performance and increased system complexity. In this work we introduce a light source—the chaotic microcomb—to overcome this problem. This physical entropy light source exhibits naturally orthogonalized light channels that are immune to any congestion problem. Based on this microcomb state, we demonstrate a new type of LiDAR—parallel chaotic LiDAR—that is interference-free and has a greatly simplified system architecture. Our approach also enables the state-of-the-art ranging performance among parallel LiDARs: millimetre-level ranging accuracy and millimetre-per-second-level velocity resolution. Combining all of these desirable properties, this technology has the potential to reshape the entire LiDAR ecosystem.

Silicon modulators are key components to support the dense integration of electro-optic functional elements for various applications. Despite numerous advances in promoting the modulation speed, a bandwidth ceiling emerges in practices and becomes an obstacle toward Tbps-level throughput on a single chip. Here, we demonstrate a compact pure silicon modulator that shatters present bandwidth ceiling to 110 gigahertz. The proposed modulator is built on a cascade corrugated waveguide architecture, which gives rise to a slow-light effect. By comprehensively balancing a series of merits, the modulators can benefit from the slow light for better efficiency and compact size while remaining sufficiently high bandwidth. Consequently, we realize a 110-gigahertz modulator with 124-micrometer length, enabling 112 gigabits per second on-off keying operation. Our work proves that silicon modulators with 110 gigahertz are feasible, thus shedding light on its potentials in ultrahigh bandwidth applications such as optical interconnection and photonic machine learning.

Optical implementations of neural networks (ONNs) herald the next-generation high-speed and energy-efficient deep learning computing by harnessing the technical advantages of large bandwidth and high parallelism of optics. However, due to the problems of the incomplete numerical domain, limited hardware scale, or inadequate numerical accuracy, the majority of existing ONNs were studied for basic classification tasks. Given that regression is a fundamental form of deep learning and accounts for a large part of current artificial intelligence applications, it is necessary to master deep learning regression for further development and deployment of ONNs. Here, we demonstrate a silicon-based optical coherent dot-product chip (OCDC) capable of completing deep learning regression tasks. The OCDC adopts optical fields to carry out operations in the complete real-value domain instead of in only the positive domain. Via reusing, a single chip conducts matrix multiplications and convolutions in neural networks of any complexity. Also, hardware deviations are compensated via in-situ backpropagation control provided the simplicity of chip architecture. Therefore, the OCDC meets the requirements for sophisticated regression tasks and we successfully demonstrate a representative neural network, the AUTOMAP (a cutting-edge neural network model for image reconstruction). The quality of reconstructed images by the OCDC and a 32-bit digital computer is comparable. To the best of our knowledge, there is no precedent of performing such state-of-the-art regression tasks on ONN chips. It is anticipated that the OCDC can promote the novel accomplishment of ONNs in modern AI applications including autonomous driving, natural language processing, and scientific study.

Optical chaos is vital for various applications such as private communication, encryption, anti-interference sensing, and reinforcement learning. Chaotic microcombs have emerged as promising sources for generating massive optical chaos. However, their inter-channel correlation behavior remains elusive, limiting their potential for on-chip parallel chaotic systems with high throughput. In this study, we present massively parallel chaos based on chaotic microcombs and high-nonlinearity AlGaAsOI platforms. We demonstrate the feasibility of generating parallel chaotic signals with inter-channel correlation <0.04 and a high random number generation rate of 3.84 Tbps. We further show the application of our approach by demonstrating a 15-channel integrated random bit generator with a 20 Gbps channel rate using silicon photonic chips. Additionally, we achieved a scalable decision-making accelerator for up to 256-armed bandit problems. Our work opens new possibilities for chaos-based information processing systems using integrated photonics, and potentially can revolutionize the current architecture of communication, sensing and computations.