The UPC leads a project to develop faster, lower-power processors for artificial intelligence

+
Download

Picture of Laura Ockel (Unsplash)

+
Download

Akshay Jain, Eduard Alarcón, Robert Guirado, Sergi Abadal, Hamidreza Taghvaee and Albert Cabellos-Aparicio, researchers at the UPC working in the WIPLASH project

The European Commission grants three million euros in funding to the WiPLASH project within the framework of the FET-Open call. It aims to prototype miniaturised wireless graphene antennas in the terahertz band to provide future computing platforms with plasticity and reconfigurability. The project involves seven European universities and companies, under the supervision of Sergi Abadal, a researcher at the UPC’s Department of Computer Architecture.

Oct 27, 2020

Researchers at the Universitat Politècnica de Catalunya · BarcelonaTech (UPC) are leading an international project to develop processors for artificial intelligence and machine learning that are much faster and have lower energy consumption than the current ones. The project, which is called WiPLASH (Wireless Plasticity for Heterogeneous Massive Computer Architectures), involves seven European computing schools and companies and has received three million euros in funding from the EU Horizon 2020 programme, within the prestigious and highly competitive FET Open call.

Specifically, over the next three years, thirty researchers from a wide range of disciplines, led by Sergi Abadal, a researcher at the UPC’s Department of Computer Architecture, will collaborate to prototype miniaturised wireless graphene antennas in the terahertz band for providing future communication platforms with plasticity and reconfigurability.

 “These new computer chips can be used in implants, the Internet of Things, mobile phones and large servers. We want to open the door for artificial intelligence to reach more areas, where size and energy consumption are critical”, explains Eduard Alarcón, an ICREA Acadèmia professor who leads the Energy Processing and Integrated Circuits (EPIC) group at the UPC’s Department of Electronic Engineering and a professor at the Barcelona School of Telecommunications Engineering (ETSETB). Alarcón is also the scientific co-head of the interdisciplinary centre N3CAT-UPC, which focuses on research on wireless communications using nanotechnology. 

So far, there are two types of processors or computer chips: general purpose processors, which can perform any function at a certain speed, and highly specialised processors for specific tasks, which perform a single function very efficiently and quickly. An example of this is real-time facial recognition technology on most new smartphones, a sophisticated authentication method for unlocking the phone or verifying online payments. In this case, a processor is performing hundreds of billions of operations per second to process images.

“Our goal with this project is to try to develop a processor as generic as possible, so that it can perform many different functions, that is also as fast as a specialised processor”, explains Abadal. In order to combine these two opposing properties, “we need communications within the chips and the system that are very fast and able to reconfigure very quickly, which was not possible until now.”

Miniaturised wireless graphene antennas, which are up to 100 times smaller than a metal antenna and capable of operating at terahertz frequencies, are thus essential. “Since they are wireless, we can radiate information throughout the chip and provide the processor with plasticity and reconfigurability”, highlights Alarcón.

The processor prototype that will be developed in the WiPLASH project is designed for artificial intelligence and machine learning, which have experienced exponential growth in recent years. When an algorithm, which is a programming code, is run on a large server, it consumes large amounts of energy. “Power consumption is a bottleneck in artificial intelligence, and WiPLASH will try to solve it. Wireless graphene antennas will also reduce energy consumption”, state Abadal and Alarcón.

The idea of using graphene, a two-dimensional material with very interesting properties, to make antennas emerged 10 years ago in the UPC group led by Albert Cabellos and Eduard Alarcón, in collaboration with researchers from the Massachusetts Institute of Technology (MIT), in the United States. In his doctoral thesis, Abadal took the idea a step further and proposed using them to allow communications within the chip. Now WiPLASH will test whether graphene antennas can create communication networks within a chip for artificial intelligence processors.

 The WiPLASH project has been chosen within the FET Open call, a highly competitive funding programme that aims to catalyse breakthrough innovative technologies and that is open to all sciences and disciplines. More than 400 projects participated in the 2019 call, and only 53 were selected for funding, including WiPLASH.