Silicon is sparking again: a new article reveals its potential

Photonics is a field that promises to radically transform high technology by using light (infrared radiation or visible spectrum) as the signal carrier. However, the dimensions of photonic waveguides and logic circuits are usually several orders of magnitude larger than those of comparable silicon electronic components. This creates serious difficulties when considering the use of photonics for heavy computing tasks such as training large language models (LLMs). Deep neural networks require hundreds of billions or even trillions of matrix multiplication operations; if each of these multiplications were performed on a separate physical node of a photonic processor, the required equipment would exceed any reasonable limits.

Moreover, fabricating large-scale photonic circuits will require a complete technology cycle: silicon is a material that has already been nearly perfected in microelectronics, but it is not suitable for ordering and processing infrared radiation due to its non‑polar nature. Therefore even the most promising prototypes of photonic computers remain expensive, bulky, and difficult to manufacture.

Why does silicon photonics still generate optimism?

1. Indirect bandgap of silicon

When an electron transitions between a free and valence state with photon emission, additional energy and time losses occur, making silicon lasers extremely inefficient.

2. Hybrid solutions

To create quantum‑optical integrated circuits (QOICs, PICs) hybrid technology is used: waveguides and logic circuits are fabricated on silicon‑on‑insulator (SOI) wafers, while mini‑ and nanolasers are made from more suitable direct‑bandgap materials. This results in QOICs being not only larger than conventional electronic integrated circuits but also significantly more expensive to produce.

3. Economic sensitivity of LLMs

Modern large language models heavily depend on the cost of the “hardware” they run on. Hybrid circuits generally lose out to monolithic integrated solutions in terms of unit cost.

4. Scaling problem for direct‑bandgap materials

Using direct‑bandgap semiconductors for all components (waveguides, circuits, and lasers) would require a decade‑long investment cycle in an entirely new microprocessor industry—practically impossible under current macroeconomic conditions.

Why silicon photonics is still considered promising

Silicon is the second most abundant element on Earth, and humanity has been working with it for more than half a century. This makes it attractive for developing new technologies:

- Existing infrastructure – millions of factories, specialists, and component suppliers.
- Integration potential – the ability to combine photonics with existing silicon processors.

In April 2026 researchers from the University of California (name not specified) presented a new approach that could significantly accelerate the development of silicon photonics and make it more competitive compared to hybrid solutions.

Conclusion:

Photonics promises a revolution in high technology, but currently faces serious technical and economic barriers. Silicon photonics remains one of the most realistic paths for its development thanks to existing infrastructure and experience with silicon. New research in 2026 could shift the balance between hybrid and monolithic solutions, opening new opportunities for large‑scale photonic computing.