Research Note

Abstract

Optical Biosensors enabled by integrated optics and AI

A Proposal from LightCounting

What if AI can create a new kind of optical sensor on a chip? Something less intuitive for engineers, yet more practical.

Current efforts in developing optical sensors on a chip rely on replicating deigns based on bulk optical elements. Spectrometers, OCT instruments and even flow cytometers on a chip are good examples. None of these chip-scale replicas performs as well as their non-chip scale originals and probably never will.

Should the industry consider a different approach? What if we design a chip-scale optical sensor that does not have a bulk-optics analog? What if this sensor was designed with and for AI capabilities in mind?

One of the AI miracles is extracting signal from noise. Speech recognition is a simple example. Image and video recognitions are more complex, but exceedingly powerful. These are also good examples of noisy data produced by optical sensors (cameras) and sorted out by AI.

A new type of camera can be designed using pixels which do not just detect light, but provide information about chemicals on their surface. Applications of micro-ring resonators (MRRs) in bio-sensing have already been demonstrated. What if we scale them to 1000x1000 arrays of pixels, similar to digital cameras, and let AI extract signals from all the noise? Think of the difference between a single element detector, a 10,000-pixel array and a 10,000,000-pixel camera.

Humans rely heavily on their visual perception, but animals use a wider variety of sensors. Pulitzer prize winning author Ed Yong presented a fascinating story on how animal senses reveal the hidden realms around us in a beautifully written book “An Immense World”. Even insects have developed a huge variety of sensors connected to their tiny brains – a combination perfected by evolution. Insects rely heavily on listening to surface waves. Fish’s skin incorporates 2D arrays of sensors detecting static electric fields.

Can we add to this variety by designing new on-chip sensors and letting AI sort all the noise produced by them? Can we replicate the evolutionary approach by co-designing sensors and AI models supporting them? The answer is YES! Many of such projects have already started, but few results have yet been disclosed.

“When AI is discussed in the context of biosensing or other scientific fields, people tend to jump immediately to data analysis. While it is true that leveraging AI for data processing and analysis is powerful, that represents only a fraction of what AI is truly capable of. In fact, AI approaches can be applied at each stage of the biosensor development process, assisting in the selection of analytes, development of recognition elements, enhancement of signal transduction, and analysis and interpretation of data.”

The quote above is taken from a recently published review article: “Artificial Intelligence in Point-of-Care Biosensing: Challenges and Opportunities”.

The latest industry events discussing opportunities for optical sensors highlighted continuing progress in development of chip-scale devices, but none of them are taking advantage of AI capabilities in design or data analysis. This field seems to be open for companies to explore.

Full text of this research note is available to subscribers at: https://www.lightcounting.com/login