Researchers are developing computing systems that use light (photons) instead of electricity (electrons) to perform calculations. This breakthrough, known as optical or photonic computing, has shown impressive results in AI tasks and may overcome the limitations of traditional silicon-based systems.
Why Optical Computing?
Conventional electronic processors generate heat, consume significant energy, and face speed limits due to electrical resistance. Optical computing offers:
- Ultra-fast data transmission – photons travel at light speed
- Higher bandwidth – more data processed at once
- Energy efficiency – minimal heat generation
- Greater parallelism – simultaneous data processing
Breakthrough Experiments Using Optical Fibres
Two independent research teams demonstrated how thin optical fibres with nonlinear properties can carry out machine learning tasks:
- Used intense light pulses in optical fibres to exploit nonlinear effects
- Implemented an Extreme Learning Machine (ELM) model
- Used optical signals to perform complex computations physically, rather than digitally
Impressive AI Accuracy with Light
- The optical computing setup achieved over 91% accuracy in handwritten digit recognition using anomalous dispersion in fibres.
- Accuracy crossed 93% in normal dispersion conditions.
- These results match traditional machine learning benchmarks—without electronic processors.
How Light Performs Computation
- Images are encoded onto light pulses by modifying their phase and amplitude.
- These encoded light pulses pass through optical fibres where nonlinear interactions transform them.
- The output light pattern forms a unique spectral fingerprint—acting as the hidden layer of a neural network.
- Only the final weights are computed digitally, making this a hybrid AI system.
Extreme Learning Machine (ELM): Ideal for Photonic AI
- ELM is a type of single-layer neural network known for fast training.
- Only output weights are trained, reducing computation time.
- Its simplicity makes it suitable for early stages of hardware-based AI like optical computing.
Challenges and Future Scope
Despite promise, light-based computing faces hurdles:
- Current models do not account for polarisation changes in light
- Integration with conventional computer systems is still evolving
- Hardware miniaturisation is crucial for real-world applications
Future enhancements could come from:
- Optical neural networks
- Photonic integrated circuits
- Fully light-based AI chips
Conclusion
Photonic AI computing could dramatically change how AI systems are built—making them faster, scalable, and energy-efficient. With continued research, light-based processors may one day replace GPUs and silicon chips for AI tasks.