Photonic Processors Can Help Accelerate AI Growth

Researchers are looking to light to address the mounting need for improved data processing in terms of speed and volume, according to a study published in Nature journal. Photonic processors will be used to improve efficiency.

This new development will “harness the distinctive properties of light” to “inspire a renaissance of optical computing.” With the particular use of photon, the study seeks to use AI, specifically artificial neural networks, for operations.

According to the study, “These networks perform complex mathematical operations using many layers of interconnected artificial neurons. The fundamental operation that uses most of the computational resources is called matrix-vector multiplication.”

Photonic Processors Accelerate AI Growth

The use of photonic processors is an effort to create a system that will speed up the artificial neural network using optical frequency combs. These offer advancements for integrated photonic processors.

The Nature journal study said, “Optical frequency combs are sets of light sources with emission spectra that consist of thousands or millions of sharp spectral lines that are uniformly and closely spaced in frequency.”

Such technology has been successfully used for various applications such as optical-clock metrology, telecommunication, and spectroscopy.

For data processing, these combs can be combined within a computer chip for power efficiency that will optimize optical computing.

The researchers referred to a study by Xu and colleagues that has achieved data parallelization by wavelength multiplexing which can produce a versatile integrated photonic processor. The device was able to perform a convolution for image-processing applications.

The paper said, “They first used chromatic dispersion – whereby the speed of transmitted light depends on its wavelength – to produce different time delays for wavelength-multiplexed optical signals.”

The combination of these signals allowed for the full exploitation of the various photon wavelengths.

The Nature article also referred to Feldmann and colleagues’ study which created an integrated photonic processor on their own. This device was also able to perform a convolution that uses optical signals that cover two dimensions.

Moreover, the device was able to completely parallelize the input data through “wavelength multiplexing and conducted analogue matrix-vector multiplication using an array of integrated cells of the phase-change material.“

However, the high-computing power shown by Feldmann and colleagues’ device is hampered by its lack of energy efficiency.

The Nature article noted that the photonic processor approach in artificial intelligence can also be researched through the development of advanced nonlinear integrated photonic computing architectures instead of a one- or two-dimensional linear approach.