New Hardware Paradigms Needed to Run AI


Regardless of the accuracy of the dollar amounts, there is an urgent need for new hardware paradigms to run AI (Generative AI, self driving cars, …), which is why there is a new wave of well-funded chip startups focused on optimizing AI both in data centers and at the edge. This includes:


l  New architectures mainly designed to accelerate inference and reduce power consumption (Graphcore, Samba Systems, Rebellions, Cerebras, Neureality, and Hailo).

l  Analog computing and in-memory computation (Mythic, Analog Inference):

l  Silicon Photonic chip for faster computation with lower power consumption (Lightmatter, Luminous, Celestial, Lightelligence, and SalienceLabs, including enablers such as Xscape Photonics or iPronics)