Photonic Computing
The Revolution of Information Technology
Photonic Computing - The Revolution of Information Technology

With every passing year, it becomes clearer that AI is the technology to dominate the coming decades. AI is projected to be the foundation for a $100B industry, but for that to happen, it must overcome the challenges inherent to its nature.

The demand for speed, processing power and efficiency is ever increasing – faster than it can be supplied by conventional means.

CogniFiber will change that.

With our DeepLight™.

The AI hardware technology that will change the world of computing forever.

What makes DeepLight™ so powerful?

By applying cutting edge research and technology from the fields of photonics, AI and advanced electronics, we can deliver devices that are 10.000x faster, consume only 1% of the power and need no buffering at all.

This symbolizes the next step in the evolution of information technology.

DeepLight will solve speed, power consumption and buffering problems.

This makes scalable photonic computing the emerging technology of the coming decades – allowing AI to be cheaper, increasingly mobile and more accessible in the process.

Aside from their performance superiority over existing technology, DeepLight devices have been carefully engineered to come with multiple added benefits:

  • Scalability
  • Easy to Manufacture
  • Fault Resistance
  • Connectable to any existing Communication Platform

This and more make DeepLight the perfect solution for the growing need of the global information economy.

Current Status

After years of researching and our successful early seeding phase, we are moving from prototyping to early production and plan on rolling out our first products in early 2023 to our strategic partners for field testing and optimization.

We will begin with an IoT oriented system the size of a 45 Unit server rack, a cut down version of the optical data centers we plan on rolling out eventually. This DeepLight system will be ideal for the implementation of neural networks and already uses inference at the speed of light.

This modestly sized server will be replacing a huge part of an entire data center with a minimal power consumption of 10kW/h. It is this incredible increase in efficiency that makes us so confident that DeepLight Technology will be the next step in the evolution of information technology.


The concept behind DeepLight Technology is converting a flaw into a feature. When using fiber optics that transmit data by means of light, precautions need to be made to prevent light from crossing between fibers and producing errors and noise.
DeepLight lets this cross-talk between fibers happen in a controlled way. This way cross-talk is being transformed from noise to computation. This is possible through our unique design and the use of smart algorithms that allow us to program, how the fibers interact and influence one another. What this does is essentially convert strands of fiber into programmable nodes that can be used as neurons in a neural network, enabling the training r of sophisticated AI systems and their light speed running after training.
CogniFiber pioneers this one of a kind technology of the future. Nobody else has yet managed to make any progress on this bleeding edge of research. Not only will DeepLight revolutionize the world of computing but it’ll also make AI cheaper, more mobile and more accessible to the broader markets. Its hidden power lies in the scalability, that’s simply not possible with other technologies.

Technology in Detail

Thanks to years of research and ceaseless striving for perfection, our engineers have reached new heights. We are proud to present to you some of the specs our DeepLight Technology comes with.

  • In clocking benchmark tests in June, DeepLight operated at up to 1 Giga (billion) frames per second.
  • Layers:
    • Input/ Output Layer: Up to 128 (each) fiber connections using standard 86 fiber connectors. Each fiber carries up to 100 channels at 100Gb/s x 4 Bits/ symbol (16QAM)
    • Electronics Layer (control, synchronization, management, and neural network topology programmability)
    • Programmable Light Amplification Layer (for training and implementing the in-fiber inference system).
      • Inference-only option: use a pre-fabricated, un-reprogrammable layer that reduces power consumption and eliminates external interference for high-security environments.
    • Input Optical Neural Network Layer: Up to 128×100 inputs and 128×100 outputs using a combination of variable-sized optical devices.
    • Hidden Layers: Any number of hidden layers with up to 128×100  inputs and 128×100  outputs using a combination of variable-sized optical devices for each layer.
    • Output Optical Neural Network Layer: Up to 128×100 inputs and 128×100 outputs using a combination of variable-sized optical devices.

Current models are trained within the server. Future deployments will include automated conversions from other neural network architectures such as TensorFlow and ONNX etc.

Our next-generation product is a larger scale deployment which we call an optical data center.

About Us
After 7 years of intense research, we incorporated CogniFiber in February of 2019 and have been growing and improving ever since. Thanks to the uniqueness and potential that DeepLight™ Technology presents, we managed to secure strategic partnerships that will help greatly in realizing the bright and promising future ahead for CogniFiber. For example, we successfully graduated from the Intel’s Accelerator Program and have officially become a part of Intel’s Ingenuity Partner Program (IPP), setting the company up for invaluable connections and guidance from industry experts.
Our Founders
Our team of Co-Founders bring plenty of experience, not only in the intellectual and scientific arena, but in the world of business as well. Together they can boast more than 30 years of start-up experience (software, hardware, product and business) in more than 14 companies, 5 of which were co-founded, over 40 years of research and more than 50 patents. As the whole venture moves from the development and research of the DeepLight™ Technology towards the commercialization and implementation, these are the very same researchers and entrepreneurs that started it all. They are still on board and making sure their vision becomes a reality:
Co-Founder, CEO

Dr. Cohen has been envisioning and implementing photonics-based solutions for various uses since 2012. He is an experienced hardware and photonics architect, a machine learning aficionado, and a neurobiology lecturer. He completed his B.Sc. In Electronics engineering and Biology (Technion), M.Sc. and Ph.D. in neurobiology (Weizmann Institute) and his post-doc in optical neuromorphics (Bar Ilan and Hebrew Universities). Eyal has previously held positions of algorithm team leader (M.S. Tech) and hardware engineer (Mellanox Semiconductor, Saifun Semiconductor).

Co-Founder and Chief Technology Officer

Together with Dr. Cohen in 2012, Prof. Zavlevsky began the promising research project that ended up showing that neural networks can be implemented in-fiber and thus enable light speed inferencing. Prof. Zalevsky is a professor and Dean of the Faculty of Engineering at Bar Ilan University. His research expertise involves biomedical optics, super-resolution, and electro-optical devices. He has published more than 450 peer-reviewed papers, 10 books, 30 book chapters, and around 100 patents. For his work and great service to science, he received many national and international scientific prizes. He also already is a co-founder and technological leader in many successful startup companies that originated from technologies commercialized from his lab.

Additionally to our founders, CogniFiber relies on an extraordinary team of experts across a wide variety of fields. Optics, AI and advanced electronics are just the tip of the iceberg. Developing and creating DeepLight™ Technology is a multidisciplinary effort, that wouldn’t be possible without the great and diverse team at CogniFiber.


NN FPGA subcontractor engineer

  • Vast knowledge in Neural-Network
  • Vast Experience in implementation of Neural-Network (include learning stage) on FPGA – Must
  • Design experience in Xilinx Ultrascale+ FPGA family
    • HDL code (VHDL/Verilog)
    • RT-C programming
    • System Linux (Device tree, HW interfaces…)
    • Experience in high-speed FPGA design
    • Design experience in RFSoC as an advantage
    • Design experience in PyTorch, Matlab as an advantage

Contact us for more information and send relevant CV.



Israeli startup CogniFiber strives to reinvent the chip
The company, which has won through to the 4YFN startup competition final at the Mobile World Congress in Barcleona, is planning mega-fast processors.

COGNIFIBER HAS BEEN SELECTED TO TAKE PART IN INTEL’S INGENUITY PARTNER PROGRAM (INTEL IPP) As part of the program, CogniFiber will be mentored by experts from Intel’s business organizations, will be exposed to state-of-the-art technology, along with an educational infrastructure and local and international events.

Cognifiber: Make AI cheaper, more mobile and more accessible with In-fiber photonic computing devices.
4YFN began as a complimentary event to Mobile World Congress with the aim of connecting startups with corporations and investors. In it’s 5th edition, and with more than 20,000 attendees, we can now say it is no longer a side event.

4YFN Awards to Identify Best International Startup
Cognifiber (Israel): a world-leading company in the field of scalable photonic computing.

    HQ - Rosh HaAin, Israel