Israeli startup develops photonic processors 1000 times faster than Nvidia

Israeli startup CogniFiber's DeepLight processors promise a groundbreaking advancement in data processing, particularly for data centers and supercomputers. Currently, supercomputers rely on traditional processors, including those from Nvidia. However, the increasing demand for computational power, driven by the rise of AI applications, necessitates new solutions.

AI applications require highly efficient processors, leading to escalating energy consumption. Against this backdrop, CogniFiber's developments are highly promising. The company is creating processors that use photonic connections instead of electrical ones, reducing energy consumption and enhancing performance. These are neuromorphic processors, differing from the traditional von Neumann architecture where memory and the processor are separate, necessitating data exchange between them. This data exchange speed acts as a "bottleneck" limiting computational efficiency.

CogniFiber is developing a neuromorphic processor without the bottleneck, as both memory and computations occur on a single chip. According to Calcalist, the latest version of the DeepLight processor surpasses Nvidia's latest processors by 1000 times in data processing speed and reduces energy consumption by 100 times compared to Nvidia processors.

Investors have shown strong support for CogniFiber, with the company raising $5 million in its latest funding round, bringing the total investment to $14 million, including funding from the Israel Innovation Authority. The company plans to complete the development of the first full photonic processor version by 2027 and aims to build a photonic supercomputer based on it, marking a significant leap in the microprocessor industry.