Co-Packaged Optics Breaks Data Bottlenecks to Boost Chip Performance Amid AI Boom

The wave of artificial intelligence continues to sweep across the globe, driving graphics processing units and various computing chips to pursue ever-greater performance. However, behind the rapid escalation of chip computing power, the industry faces an imminent and severe challenge: severe data transmission congestion. As AI models grow increasingly large and computational tasks become extremely complex, a single processor can no longer bear the load alone. It must rely on thousands of processors operating simultaneously, interconnected via high-speed networks to exchange massive amounts of data. If data fails to arrive in time during transmission, computing units are forced to idle, preventing high-cost hardware investments from achieving their full potential efficiency.

To address this transmission bottleneck that plagues major tech companies, academia and industry have recently turned their attention to optical communications, with silicon photonics technology and optical interconnect solutions becoming the absolute focal point. Traditional data transmission primarily relies on copper wires to carry electronic signals. However, as bandwidth demands surge exponentially, copper wires suffer from severe signal attenuation during long-distance transmission, accompanied by extremely high heat generation and power consumption. In contrast, optical transmission uses photons as the carrier, offering high bandwidth, low latency, and superior energy efficiency. It can break through the physical limits of traditional metal wires, serving as the critical key to extending Moore’s Law and the growth of AI computing power.

Among the various optical transmission technologies, co-packaged optics (CPO), which is gradually entering mass production, is regarded by the industry as the most breakthrough ultimate solution. In the past, optical transceiver modules were typically placed at the edge of the motherboard, a certain distance away from the processor. Consequently, signals still lost considerable energy and time during the opto-electrical conversion and transmission process. Co-packaged optics, however, integrates the optical engine and the processor directly on the same substrate, bringing optical components extremely close to the computing chip. This significantly shortens the transmission path for electrical signals, not only markedly reducing transmission latency but also effectively lowering overall power consumption. For AI data centers pursuing ultimate performance, this technology offers immense appeal.

The breakthroughs and mass production progress of this technology have successfully ignited a competitive race among global tech giants and the semiconductor supply chain. From chip design majors and foundry leaders to packaging and testing firms, all are actively investing resources in the research, development, and deployment of co-packaged optics-related technologies, aiming to seize a dominant position in the next generation of high-speed transmission standards. This is not merely a simple technological upgrade; it also drives the restructuring of the future global AI hardware ecosystem. Whoever first solves the challenges of mass production yield and cost will take the lead in capturing the vast market opportunities.

Looking ahead, as AI applications continue to penetrate various industries, data transmission demands will only increase. The widespread adoption of co-packaged optics will become a critical driver determining the upper limits of AI development. From overcoming communication barriers between chips to reducing the massive electricity costs of data centers, this technology bears the heavy responsibility of propelling the next-generation digital revolution. As the technology in the related industrial chain matures and moves toward commercialization, we will witness another leap in AI computing performance.

(Source: CNA)