NVIDIA Explores Optics on the Silicon Interposer
At SEMICON West 2026 in San Francisco, NVIDIA researcher Nikola Nedovic introduced Optics-on-Interposer (OOI), a research approach that moves optical engines from the package substrate onto the silicon interposer shared with compute chips. The design shortens high-speed electrical paths with denser interconnects; NVIDIA says it can reduce power use and increase interconnect bandwidth density, but has not disclosed measured results or a commercialization timeline. NVIDIA also presented dense wavelength division multiplexing research using eight wavelengths over one fiber to achieve 256Gb/s of total bandwidth.
OOI would bring optical engines closer to compute chips, linking optical-interconnect development more closely to advanced packaging as AI-cluster bandwidth demand grows. However, tighter integration raises thermal, testing, yield, alignment, repair and maintenance challenges, and the report does not establish whether the added integration costs can be offset by lower power consumption.
近日，英伟达展示了下一代 CPO（共封装光学）光电集成方案，准备将光引擎进一步向计算芯片靠近。
上个月，在旧金山举办的 SEMICON West 2026 大会上，英伟达研究员 Nikola Nedovic 介绍了这项名为 Optics-on-Interposer（OOI，中介层集成光学）的新技术。相比现有 CPO 方案，它将负责光电转换的光引擎从封装基板移到计算芯片所在的硅中介层上，进一步缩短高速电信号的传输距离，以降低功耗、提高互连带宽密度。
英伟达同期还展示了一项密集波分复用（DWDM）研究，利用 8 个不同波长在同一根光纤中传输数据，实现 256Gb/s 的总带宽。两项研究分别着眼于缩短芯片与光引擎之间的电连接，以及提升单根光纤的传输容量。
#NVIDIA#Optics-on-Interposer#Spectrum-X Ethernet Photonics#SN6800#Vera Rubin#TSMC#ASE Technology Holding#Lumentum#Coherent#Foxconn#TrendForce