Silicon integrated circuitry can transfer information over tremendous distances at the cost of an ever more economical manufacturing process. Silicon photonics will revolutionize computing, sensing, and data centers—worthy goals in an era dominated by information. Scientists are focusing on this technology for a different reason, though: its high-volume production through CMOS-like fabrication.
Silicon might be promising, but it has shortcomings. Its basic form isn’t ideal for creating photodetectors or optical modulators, but the academic world is already making progress in these areas.
The Future of Silicon Photonics
Optical computing is quickly moving into the mainstream. IBM and Intel have commercialized silicon photonics already. The future of light seems bright despite the failures of the last 30 years. Hyperscale data centers are more necessary than ever before, and experts predict that silicon photonics will help the world to achieve cooler, more efficient data transfer within the next few years. In 2015, their use in data centers had barely hit the $50 million mark, but by 2020, the market is projected to be worth $4 billion.
Silicon photonics removes the need for external laser packaging and exploits nonlinear multiwavelength sources. This has made it possible for Princeton University to build a neuromorphic chip, with each node functioning at a specific light wavelength. Futuristic as light-based neural networks may seem, the project has shown the world how impressive photonic networks can be and what their speed might mean for future technologies.
Next generation IC designs have been an adequate alternative to photonics until now, but Moore’s law evidently has its limits, so their scaling designs have become more difficult to achieve. Today’s chips can and should be capable of achieving 25.6Tbts/s, and silicon photonics is making those speeds possible.
For a thorough discussion, check out “Light in a Package” from Ed Sperling on Semiconductor Engineering.