Research Overview
As modern computing systems continue to scale, the central bottleneck is no longer only the speed of individual processors, but the efficiency of data movement across the system. Artificial intelligence, high-performance computing, cloud infrastructure, and scientific simulation all depend on massive data exchange among processors, memories, accelerators, and network nodes. In these platforms, communication increasingly determines latency, energy consumption, and system complexity. Conventional electrical interconnects have supported computing for decades, but they now face growing limits in bandwidth density, signal loss, power efficiency, and scalability as data rates and communication distances continue to rise.
Photonic interconnects address this bottleneck by using light as the information carrier. Optical links offer high bandwidth, low propagation loss, and native support for wavelength-division multiplexing, allowing multiple data channels to share the same physical path. Their value is therefore not simply to act as faster wires, but to enable communication fabrics that are denser, more parallel, and more energy efficient than purely electrical approaches. Our research focuses on the design of optical engines that integrate the key functions needed for practical photonic communication, including optical transmission and reception, wavelength management, switching, coupling, and chip-scale routing. This direction is closely connected to co-packaged optics, where optical engines are placed near switching or computing ASICs to reduce electrical reach and energy cost, and to optical I/O, where high-bandwidth optical interfaces bring communication closer to processors, memory, and accelerators. The goal is to understand how photonic links should be co-designed with electronic systems rather than added as isolated components.
Beyond communication, photonics also creates opportunities for computation itself. Many important workloads, such as matrix-vector multiplication, convolution, signal transformation, and other linear operations, can be naturally mapped onto optical interference, modulation, diffraction, and propagation. Because light can carry and process information in parallel, photonic computing has the potential to provide high throughput and low latency for selected tasks in machine learning, sensing, optimization, and scientific computing. The key question, however, is not whether light can compute, but where optical computation provides a meaningful advantage over mature electronic processors. We therefore view photonic computing as a specialized accelerator rather than a universal replacement for digital electronics, with its greatest promise in computational primitives where optical physics can directly perform useful operations with lower energy or shorter delay.
A major theme of our work is the convergence of photonic interconnect and photonic computing. These two areas are often discussed separately, but future computing systems will increasingly require communication and computation to be considered together. If data must be repeatedly converted, moved, stored, and then processed in separate domains, much of the potential efficiency of photonics can be lost. A more compelling direction is to transmit, route, and partially process information within the optical domain, enabling new forms of in-network computing, near-memory optical acceleration, optical tensor processing, and reconfigurable photonic fabrics. Realizing this vision requires progress not only in devices, but also in system architecture, packaging, control, calibration, and software-hardware co-design.
Our future research will develop photonic interconnect and computing technologies that are physically efficient, architecturally meaningful, and relevant to real workloads. This means designing optical engines around concrete system bottlenecks, evaluating photonic computing blocks in end-to-end applications, and exploring how optical links and optical processors can reshape the balance among communication, memory, and computation. By combining integrated photonics, electronic-photonic co-design, and application-driven architecture, we aim to contribute to next-generation intelligent computing infrastructure. In this direction, photonics is not merely a replacement for electrical links or processors; it is a way to rethink how information flows through computing systems and how part of that information processing can be embedded directly into the physical medium of communication.
