Keep abreast of the latest corporate news and focus on industry hot topics.
"Optical speed computing power" leads the future
In the summer of 2025, the global computing power race will reach a critical turning point - when traditional electronic chips encounter the "physical high wall" of quantum tunneling effect in processes below 7 nanometers, the contradiction between heat loss and computing power growth becomes increasingly acute. Photonic chips are emerging as the core direction to break through the shackles of Moore's Law in the form of "optical speed computing power". From the innovative breakthroughs in Chinese laboratories to the strategic layout of the US Department of Defense, a competition over the next-generation computing power infrastructure has been fully launched.
Release time:
2025-09-23
In the summer of 2025, the global computing power race will reach a critical turning point - when traditional electronic chips encounter the "physical high wall" of quantum tunneling effect in processes below 7 nanometers, the contradiction between heat loss and computing power growth becomes increasingly acute. Photonic chips are emerging as the core direction to break through the shackles of Moore's Law in the form of "optical speed computing power". From the innovative breakthroughs in Chinese laboratories to the strategic layout of the US Department of Defense, a competition over the next-generation computing power infrastructure has been fully launched.
Nowadays, the training of large AI models is caught in a dilemma of "scale and energy consumption" : training a model with hundreds of billions of parameters consumes millions of kilowatt-hours of electricity and takes several weeks. Even the latest GPU clusters from NVIDIA find it difficult to balance computing power density and energy efficiency. What is more serious is that the transmission bandwidth of electronic chips is limited by the resistance effect of copper wires. The energy consumption of the cooling system in data centers accounts for 40%, and the energy consumption of data transfer accounts for 30% to 40% of the energy consumption of IT equipment. This means that the traditional electronic architecture is approaching the physical limit where "no matter how much optimization is done, it is difficult to break the deadlock".
The emergence of photonic chips has precisely hit this pain point. Photons, as information carriers, have the natural advantages of bosons. There is no need to worry about mutual interference of charges. Their transmission speed can reach the order of light speed, and their parallel processing capacity far exceeds that of electrons. More importantly, when completing the same computing task, the energy consumption of photonic chips is only one-tenth of that of electronic chips, and the latency can be reduced to the nanosecond level. This dual feature of "high energy efficiency + high speed" makes photonic technology the core engine of the computing power revolution in the post-Moore's Law era.
At present, photonic chips have formed dual technical routes of "silicon-based" and "thin-film lithium niobate" : Silicon-based photonics achieve low-cost integration relying on mature CMOS processes (such as Intel HorseRidge III chips), which is suitable for large-scale scenarios such as data center interconnection. Thin-film lithium niobate dominates the high-performance market with its superlinear electro-optic effect (such as the ray tracing chip jointly developed by Huawei), and is more suitable for high-precision requirements such as AI inference and quantum communication. The coordinated development of the two routes has further broadened the application boundaries of photonic technology.
➡
Latest Blog
Contact Us
Address: No.4 Tangtou Road, Cuibei Industrial Park , Henglin Town, Economic Development Zone, Changzhou City, Jiangsu Province
微信扫码我们
