Huawei is pushing standardisation of NPO (near-packaged optics) for AI networks.

It is promoting it as a technology to replace existing detachable optical modules, instead of CPO (co-packaged optics), which has a heavy cost and maintenance burden.

It is positioning it as a technology that can replace existing detachable optical modules while also addressing cost and maintenance burden issues surrounding CPO (co-packaged optics).

Huawei unveiled a 7.2-terabit-per-second (Tbps) NPO module at the China International Optoelectronic Exposition (CIOE) held in Shenzhen from Sept. 9 to 11. The module will be installed in next-generation SuperPOD AI systems.

NPO places the optical engine as close as possible to the switch application-specific integrated circuit (ASIC), but plugs it into the mainboard as a separate package. Detachable optical modules include a digital signal processor (DSP) that conditions signals, and as speeds rise, power consumption and heat generation also increase. As AI infrastructure pushes connection speeds to 800 gigabits per second (Gbps) and further to 1.6 Tbps, these problems could become bigger, it said.

CPO places the optical engine on the same substrate as the switch ASIC to further reduce latency and power consumption. But Huawei points out that this packaging method is expensive and makes it difficult to replace a failed optical engine in a data centre. The company said, "NPO is more technologically mature than CPO, making mass production and maintenance easier," and added, "For now, it is the most practical choice."