Cognifiber Introduces Glass-based Photonic Chip for Edge Computing

Cognifiber, a deep technology company specializing in photonic computing hardware, recently announced a new glass-based photonic chip that could significantly boost computing capabilities while enabling smaller edge devices.

Cognifiber was founded in 2018 by Dr. Eyal Cohen, who has a background in electrical engineering, biology, and neuroscience, and Professor Zeev Zalevsky, an entrepreneur and dean of the faculty of engineering at Bar-Ilan University.

 

Dr. Eyal Cohen, co-founder and CEO of Cognifiber. Image courtesy of Cognifiber

 

“Upon finishing my Ph.D., I turned to Professor Zalevsky and proposed that we research deep learning systems in photonics,” Dr. Cohen explained. “Together with Professor Mickey London (Brain Research, Hebrew University), we invented and implemented a breakthrough technology to revolutionize the world of computing. Two years later, we wrote our first patent and published a peer-reviewed paper in Nature’s Scientific Reports (2016). Zeev and I founded Cognifiber in May 2018, completed our first funding round in January 2019, and established the company HQ in Rosh HaAyin, Israel, 20 minutes east of Tel Aviv.”

 

A “Revolutionary” Glass-based Chip

The chip developed by Dr. Cohen, Professor Zalevsky, and Professor London differs greatly from previously proposed photonic chips. 

“First, we perform pure photonic start-to-finish computing within the photonic system, which allows us to avoid any memory read and write during computation, accelerating the computation significantly,” Dr. Cohen said. “Next, we don’t rely on coherent light sources, which allows us to avoid phase noise that can deter computational quality in silicon photonic chips.”

The chip produced by Cognifiber relies on off-the-shelf fiber optics and optical communication devices. Cognifiber says this significantly increases its reliability and extends its life span, reducing the mean time between failure (MTBF).

 

The glass-based chip developed by Cognifiber. Image courtesy of Cognifiber

 

“Our approach combines in-fiber capabilities with fiber optic communication devices, overcoming interface issues and delivering 100x greater speed while consuming a fraction of the energy used by today’s leading silicon-based solutions,” Dr. Cohen explained.

The fiber optic components contained in Cognifiber’s glass-based chip are said to be easily integrated with existing optical systems and work with standard amplitude encoding sources of varying wavelengths, which could greatly facilitate their widespread adoption. The team’s technology can incorporate several features on a simple chip, promoting the miniaturization of existing devices.

“By using a neuromorphic approach and a distributed architecture, our glass chips do not play a central role, as is seen in traditional system architectures,” Dr. Cohen said. “This distributed approach simplifies the design and manufacturing of our chips and greatly increases their yield and reliability.”

 

The Advantages of Glass Over Silicon

The Cognifiber representatives explained the numerous advantages photonic chips have over existing chips based on silicon. First, glass is more readily available and more affordable than silicon. It's also less prone to degradation and overheating.

 

Comparison of glass substrates over silicon and organic laminate. Image used courtesy of Samtec

 

“The difficult truth is that for silicon photonics, the yield is very low due to the size of the chip and the complexity of manufacturing,” Dr. Cohen said. “Our chips don’t face these issues. They can be created without reliance on traditional fabs, allowing for locally distributed manufacturing without the high upfront infrastructure costs. Using glass also has significant business implications, such as reduced cost of manufacturing, lower power consumption per chip, and improved cost of operations.”

The Cognifiber team estimated that their glass-based chips could result in a 100-fold boost in computing capabilities and an 80% reduction in the costs required to train artificial intelligence (AI) and machine learning (ML) algorithms.

 
The Potential of Glass for Edge Computing

While the new glass-based chips may enhance numerous areas of computing, they could be particularly promising for edge computing applications, the Cognifiber team said. 

“Downsizing will make it easier to bring a boost in computing power to the edge with lower power consumption,” Dr. Cohen said. “On the system level, miniaturization of systems and significantly reduced power consumption supports the scaling of major systems. For example, two racks with our glass and fiber technology can handle more than what twenty racks do today.”

The Cognifiber team is focused on production scaling to facilitate the up-and-coming growth and sales phase.

“We are excited to be a part of a global effort toward the automation of the post-production inspection process. Typical critical dimensions are far larger than electronics, making inspection far easier,” Dr. Cohen said. “In the coming weeks, we will be evaluating new chip designs that will optimize our performance. We have many directions that we could explore in the future, but for now we are focused on bringing our product to the market.”