Prof. Lior Klein & Group Win Grant for Novel AI Hardware Research

A group of researchers from Israel, Portugal, Belgium, and Italy, led by Prof. Lior Klein of the Department of Physics at Bar-Ilan University, has been awarded a research grant of over €3 million from the European Innovation Council (EIC).

The grant is awarded through the Pathfinder Open program, promoting high-risk/high-gain technological development. The team, combining researchers from academia and industry, will develop innovative computer hardware in the field of artificial intelligence (AI) based, among other, on patents from Prof. Klein's laboratory in the field of Spintronics.

The innovative hardware is a device expected to lead to a breakthrough in the speed and energy cost of computations needed for artificial intelligence, particularly impacting applications that require computational capabilities at the Edge, such as in transportation, medicine, robotics, and more.

The increasing use of artificial intelligence will require a significant increase in computational resources. There is an urgent need for faster and more energy-efficient ways for performing artificial intelligence computations. One promising approach is based on a device capable of performing computations analogically rather than digitally. This means the device will receive electrical signals as input, and the output, determined by the electrical properties of its components, will be equivalent to performing a large number of digital operations. This requires a complex device comprising a large number of components whose electrical properties can be altered to suit various computational tasks.

The research team will develop such a device based on Spintronics - a field focused on studying physical phenomena related to the fact that, in addition to its electric charge, an electron also possesses a magnetic moment called spin. The Spintronic component, which will be the heart of the new device, is a magnetic tunneljunction where two magnetic layers are separated by a thin insulating layer. The resistance of the magnetic tunnel junction depends on the relative alignment of magnetizations in the two layers: It's low when the magnetizations in both layers are parallel and high when they are opposed. Commonly used magnetic tunnel junctions support only parallel and opposed states. However, when the components of the analog device have only two resistance states, its usefulness for artificial intelligence purposes is severely limited. Therefore, the analog device which the team will develop and fabricate will be based on novel magnetic tunnel junctions that support a larger number of magnetic states, greatly enhancing the precision, speed, and energy efficiency of the analog device.

Researchers from Bar-Ilan University, along with the Israeli company SpinEdge, the Portuguese INESC institute, UCLouvain and VUB universities in Belgium, and the Italian company I-FEVS will work together for approximately three years in order to develop and fabricate an initial prototype of a device and prove its usefulness and efficiency. Such hardware will be particularly important for Edge computing, where it's crucial for computations to be performed near the data source obtained, for example, through various sensors. Such applications include autonomous vehicles, real-time analysis of security camera footage, urgent medical procedures requiring immediate decisions, and more.