“The amount of computation we need is quite incredible. We need to build one giant chip. And we need it cause in the future AI is gonna talk to itself, is gonna be thinking, reflecting, processing!”
I was fascinated by Jensen Huang’s keynote at #CES in Las Vegas. I can still remember, back in 2015, how incredible it was for me to think about transistors getting below 10nm, while studying them at the University. 10 years later, the focus is completely different: the world needs larger and larger systems to maximize performance and power efficiency. And in the long term they will look like spartan shields. This “mega-chip” will be capable of executing more than 1 exaFLOPS (slightly more than today’s most powerful supercomputer in the world) and will have a data rate of almost 10 Pbps!! Sure, for the moment it is just a prop, many challenges are still to be faced to make it possible. But students must be as excited as I used to be 10 years ago!
This is what always amazed me about microelectronics: years go by, chip requirements evolve and technology is always capable to adapt, improve and offer novel solutions.
There is a critical element though with those future systems: transferring data among all the entities as fast and efficiently as possible, to make them work in parallel and keep them busy. And copper is hitting a wall. There’s nothing worse from a cost efficiency perspective than having computing entities to wait to receive data with very long idle times. It basically means that most of the time, while training AI models, you have extremely advanced and expensive GPUs which are not operating!
I am convinced that #photonics is the key to go beyond this limitation. And I am convinced that in the next few years, we will see it replacing copper more and more in data centers, as 15 years ago the optical fiber replaced the coaxial cable to bring the internet connection at everybody’s home.
This is why with Bruno Garbin and Fabrice Raineri we created NcodiN: our mission is to make what Jensen is holding in that picture possible. Our technology unlocks unlimited scaling and enables chipmakers’ roadmap to conceive those ultra-powerful processors.
I still remember ourselves few years ago, brainstorming in front of a beer and dreaming about those gigantic chips integrating the power of an entire supercomputer. Well, watching NVIDIA’s CEO showing it today on the main stage is extremely moving. And motivating.
The best is yet to come.
Ad maiora!
#startup #innovation #AI
