The AI business opportunity continues to ferment, and silicon photonics is set to become a key technology in the future semiconductor industry.
The AI business opportunity continues to ferment, and high-speed transmission has also become a focus under the trend of accelerated computing. TSMC Chairman Mark Liu has pointed out that silicon photonics will become a key technology in the semiconductor industry. Not only that, but ASE Group CEO Jason Wu has even stated outright: Silicon photonics will be the breakthrough point for the semiconductor industry in the next 5 to 10 years.
In fact, major companies such as IBM and Intel have already begun researching silicon photonics technology, and TSMC and ASE have also been investing in R&D for many years. Wu believes that Taiwan currently has a manufacturing advantage, and if silicon photonics can also take root, it will help to consolidate all aspects of Taiwan's current manufacturing production chain.
TSMC further pointed out that it is currently developing Compact Universal Photonic Engine (COUPE) technology. COUPE uses SoIC-X chip stacking technology, stacking electronic bare dies on top of photonic bare dies. Compared with traditional stacking methods, it can provide the lowest resistance and higher energy efficiency for die-to-die interfaces.
Advertisement
TSMC expects to complete the COUPE verification supporting small pluggable connectors in 2025, and in 2026, integrate CoWoS packaging into Common Photonic Optical Components (CPO), directly introducing optical connections into the packaging.
How to change electrons into photons?
On a general circuit board, there are many circuits made of copper wire processes, and the electrons in them will follow these copper circuits to transmit signals. Silicon photonics replaces most of the electron signals with photons for transmission. Since photons are faster than electrons and less likely to generate heat, they can achieve better transmission effects.
How to change electrons into photons? In fact, the back end of the current server casing can be inserted with an optoelectronic signal converter module, which can be called an optical transceiver. When electrons reach the end of the server and enter this module, the signal will be converted from electrons to photons. Subsequently, photons will carry the signal along optical waveguide materials such as optical fibers.This technology has been around for many years, and what TSMC (Taiwan Semiconductor Manufacturing Company) aims to do now is to integrate the module within the transceiver responsible for converting optical signals into electrical signals with the chip using advanced packaging, creating silicon photonics. This technology of packaging the optoelectronic integrated module and the chip together is called CPO (Co-Packaged Optics).
In the past, electrons would start from the chip and be converted into light only after traveling to the optical transceiver at the end of the server via copper wires. After becoming silicon photonics, electrons will enter the signal conversion point and become photons as soon as they start, without waiting until the end of the server to be converted. This reduces the distance electrons travel through copper wires, and the latter part is entirely transmitted by photons, allowing the chip's performance in both efficiency and power consumption to be further upgraded.
What are the challenges in the development of silicon photonics?
However, the ideal is very rich, but the reality is very thin. The current technological development can only achieve moving the optoelectronic integrated module inside the server, fixed on the motherboard. In the future, TSMC and ASE (Advanced Semiconductor Engineering) hope to package the chip and the optoelectronic integrated module together through 2.5D advanced packaging to create real silicon photonics. The next step will be to stack the module on the chip, and through 3D packaging, the copper wires will be shortened to the extreme.
MIC senior analyst Zheng Kai'an pointed out that it will take several years for silicon photonics to become popular, and there are still three difficulties to overcome in the future. First, it is how to successfully miniaturize the optoelectronic integrated module; second, the efficiency problem of signal conversion still needs to be overcome, as conversion will cause signal loss.
Third, Zheng Kai'an pointed out that how to build silicon photonics servers in the future is a major challenge. Now, the optical transceiver module is plugged in behind the server, which is easy to install; in the future, when it comes to silicon photonics, if the optical fiber is broken, the circuit board must be disassembled. He believes that portability is a major bottleneck in the development of silicon photonics in the future, and it is very likely that future server suppliers will need a team dedicated to maintenance.
As for future application scenarios, Zheng Kai'an named data centers, cars, drones, and ultra-large display screens as fields that will take the lead in introducing silicon photonics.Wu Tianyu: Taiwan, China, Should Seize Advantages and Develop the Silicon Photonics Industry Chain
In fact, it will take several years for silicon photonics to gradually increase in volume. Sun Moon Lake has been investing in silicon photonics R&D for 13 years. Wu Tianyu pointed out that the boundary between future packaging and testing factories and electronic contract manufacturers will become blurred, and the business model also needs continuous observation.
Wu Tianyu said that currently, silicon photonics is mostly still in the research stage, with no actual products appearing. The ones that have truly taken root in communication are the Americans.
Developing silicon photonics helps to extend the life of Moore's Law and also consolidate Taiwan's position in the global semiconductor industry. Wu Tianyu calls on Taiwan, which has very good R&D conditions, that the semiconductor industry must join hands in investment. Everyone should come together to discuss and gradually shape this (silicon photonics industry chain), so as not to be taken away in the end.