Imagine if a hair were cut longitudinally into 500 pieces, each piece would roughly match the size of the protagonist we are discussing today—the “world’s smallest laser.” This incredibly tiny light source is not science fiction, but a real breakthrough unfolding in the semiconductor field. Why is it so important? Because it directly addresses an urgent bottleneck in the development of artificial intelligence (AI), a decisive factor concerning the future computational speed and energy efficiency of AI.
In the era of large AI models, the computational power of chips has skyrocketed, but a huge challenge has arisen: data transmission. The volume of data exchanged internally and between AI chips is growing exponentially, while traditional copper wire interconnection technology, like narrow country roads, can no longer keep up with the pace of data flow.
The limitations of copper wire interconnections are apparent: first, the bandwidth bottleneck; the more data there is, the more the transmission speed is restricted; second, high energy consumption; when electric signals pass through copper wires, a large amount of heat is generated, wasting energy and causing overheating, which affects performance and stability; finally, signal attenuation; the longer the transmission distance, the more severe the signal loss.
Confronted with these challenges, the advantages of optical interconnections become evident. Optical signals transmit at the speed of light, featuring ultra-high bandwidth, ultra-low energy consumption, and nearly no attenuation. Using light to transmit data is like upgrading narrow country roads to a super-fast fiber optic network, enabling data to shuttle freely between chips and even data centers at unprecedented speed and efficiency. This is the inevitable path for the future development of AI chips.
It’s against this backdrop that a French deep tech startup, NcodiN, in collaboration with the French Atomic Energy Commission’s Electronics and Information Technology Laboratory (CEA-Leti), has made an exciting breakthrough. They are developing the “NConnect” integrated optical interconnection platform, centered around a nanoscale laser praised as the “smallest laser on silicon.”
Its astonishing feature lies in its size: it is 500 times smaller than the devices currently standard in the industry! Imagine what this means? It enables ultra-high-density integration—over 5000 such nanoscale lasers can fit into every square millimeter of area. This unprecedented integration density provides a physical foundation for the mass data transmission within chips.
Even more commendable is its ultra-low energy consumption. This nanoscale laser requires only about 0.1 picojoules (pJ) of energy per bit of data transmitted. What does this concept translate to? Compared to the energy consumption of traditional electrical interconnection, this is a revolutionary reduction. It means that when processing the same amount of data, the energy consumption of AI chips will significantly decrease, effectively alleviating the pressure from chip heating and power supply.
Any disruptive technology must ultimately move toward industrialization to truly exert its influence. The collaboration between NcodiN and CEA-Leti is a crucial step in pushing this cutting-edge technology from the laboratory to mass production.
The core of this cooperation is to integrate NcodiN’s nanoscale laser technology into a 300mm wafer-level integrated photonic process. The 300mm wafer is currently the mainstream size in semiconductor manufacturing, and its significance lies in its high compatibility with existing semiconductor manufacturing processes, allowing for direct use of mature wafer fabrication equipment and processes, greatly lowering the barriers and costs of industrialization.
Sébastien Dauvé, the CEO of CEA-Leti, clearly pointed out that transitioning photonic technology to a 300mm CMOS-compatible process is a “turning point” for optical interconnection, indicating that this technology can finally be produced at the scale, cost, and reliability required by the AI industry. This not only accelerates the industrialization process of NcodiN’s concept validation work but also paves the way for in-package and long-distance optical links in future AI chips and high-end computing systems, achieving a significant increase in cost effectiveness.
This nanoscale laser technology is far more than just a patch on existing chips; it heralds a revolutionary impact on AI chip architecture.
Imagine when optical signals replace electrical signals as the primary “information highway” within chips; AI chip design will no longer be constrained by the physical and energy constraints of copper wire interconnections. We can construct more complex, more efficient, and more energy-saving chip architectures, achieving a true form of “optical computing.” This will greatly enhance the training speed and reasoning efficiency of AI models, making larger and smarter AI models possible.
The potential application scenarios are extremely broad. Beyond AI chips, it can also empower high-performance computing (HPC), data centers, autonomous driving, the Internet of Things, and other fields with high demands on data transmission bandwidth and energy efficiency. From data exchange between servers within data centers to real-time data processing in edge computing devices, nanoscale lasers will play a key role.
Although large-scale commercialization is still some time away, NcodiN secured €16 million in seed funding last November and is accelerating its concept validation work. Its collaboration with CEA-Leti marks an important step toward industrial-grade 300mm processes. We have reason to believe that within the next 5 to 10 years, this technology will gradually mature and spark a profound transformation in the AI field.
From electrical signals to optical signals, from copper wires to nanoscale lasers, this is not just a change in technological routes, but a profound evolution in computation paradigms. This nanoscale laser, tiny to the point of being invisible to the naked eye, is redefining our understanding of computation and the future of AI with its astonishing potential. It illuminates an avenue toward higher performance and lower energy consumption for AI chips with the power of “light.” Perhaps, we are standing at the dawn of a new computational era.
