Hardware technology in Europe

We have unprecedented challenges in Hardware design, just to support machine learning and AI, and I think we really need a broad collaboration. We need collaboration among countries because of manufacturing and design capabilities. We need people, we need more engineers to come and share their experience.

There's lots of talent, there are lots of engineers in the Far East, in Asia and they are very welcome to come to Europe, as they are welcome to go to the US, because they bring not only their skills but also their enthusiasm. They like to work, they like to see a product, and when I talk to my European students in Switzerland I try to tell them it's not that you do a job just for the money: you do it because you want to achieve something that stays.

Silicon Valley is an explosive mix of Americans, Asians and Europeans all together on this kind of common ground, people are able to do the unthinkable. We don't do this enough in Europe?

If you look at future systems, and what we want to achieve within the energy budget that we can afford, then we need to open up. We need to have more diversity in terms of heterogeneity, so we need to be able to find for specific functions the computational fabric that can match the computation. This is not to say that CMOS will disappear - CMOS will stay. It will actually be the material that will connect everything together, but for achieving specific objectives we will need accelerators.

The most striking example is the computation of matrix time-vectors. These are operations that happen all over in machine learning algorithms and you want to split them up. There is a significant advantage if you can speed them up, and you can do that with CMOS, you can do it with specialized structures in FPGA but you get the best bang for your buck if you use specific technology.

One realization at EPFL was by a colleague of mine - Andras Kis - who uses a new material, molybdenum disulfide (MOS2) which is very interesting because it is flat electronics, single layer electronics, and allows very low power realization computational functions.

The next level will be just mixing and matching various types of devices, various types of architectures, to try to achieve the best. Keep in mind that if we look at AI applications nowadays, we are 1,000x off from the level of energy that we want to put in computers.

AI will be great to better the current tools because AI can capitalize on previous experiences, in design. But when you have something new you have to do a step ahead. You need to have ingenuity, certain sensitivity towards the new type of product that you are designing by understanding the array of technologies that you have. This kind of guess, this educated guess is something that a human can do and not an engine.

There is a real need for broad education, letting people know what technology does for them, so that they don't just buy a smartwatch or a telephone as a box that they rely on. That should understand the effort in designing it, so that they can use it better. Also you have to make people know that this domain is interesting and rewarding, which is something that we in Europe sometimes neglect. Technology jobs are great and we have fantastic industries for manufacturing semiconductors, fantastic industries for vehicles, planes, cars but in the computer domain Europe is still lagging behind.