A team from the University of Florida has engineered an innovative AI chip that blends optical and electrical techniques, embedding photonic components directly onto a silicon chip. Utilising lasers and microscopic Fresnel lenses, it performs convolution operations with 10× to 100× greater power efficiency than conventional chips. The prototype’s handwritten digit recognition accuracy in tests was about 98%, which is comparable to current electronic chips.

Using common semiconductor techniques, ultrathin Fresnel lenses – narrower than a human hair – are etched into the fabrication. The computation sequence goes: data → laser light → Fresnel lenses (computation) → digital signal. Notably, the chip supports wavelength multiplexing, enabling parallel processing by using different wavelengths of light simultaneously.

This breakthrough could significantly reduce the energy demands of AI systems, paving the way for more sustainable, high-performance AI technologies in the future!

#OptoelectronicChip #EfficientAI #EnergyEfficiency #AIChip #WavelengthMultiplexing #GreenTech #UFlorida #FresnelLens

Source: https://techxplore.com/news/2025-09-based-chip-boost-power-efficiency.html