The amount of data that circulate on the web exponentially increases every day. To extract knowledge from this huge amount of data, one needs novel computing concepts which are able to overcome conventional computers based on Boolean logic and von Neumann architecture. The Nanodevice Lab at Politecnico di Milano develops novel electron devices and novel in-memory computing concepts, which are able to execute computation (summations and multiplications) directly via device physics, with high parallelism and in the analogue domain. In particular, the group has developed a novel crosspoint computing, which can solve a linear system of equations (Ax= b) in a single operation lasting few tens of nanoseconds. The novel circuit outperforms both classical digital computers and futurable quantum computers: soon, i twill be possible to develop a new generation of computing accelerators that will revolutionize the technology of artificial intelligence.
The new circuit, which was developed in the frame of the European project ERC RESCUE (Resistive Switch CompUting bEyond CMOS), solves linear systems of equations (Ax=b) by an innovative methodology of in-memory computing, where the coefficient of matrix A are stored in a novel device known as the memristor. The memristor is able to store analogue values, thus a memristor array can map a coefficient matrix A within the circuit, thus enabling the high-speed solution of the problem.
The memristor array (Fig. 1) has been developed in the Clean Room of the Centre for Micro and Nano fabrication PoliFab of the Politecnico di Milano. The memristor circuit has been experimentally validated on a broad range of algebra problems, such as the ranking of websites and the solution of advanced differential equations, such as the Schrödinger equation for the calculation of the quantum eigenfunctions of electrons. All these operations have been executed in just one operation step.
These results have been published on the renowned journal PNAS of the National Academy of Science of the USA.
(The study: https://www.pnas.org/content/116/10/4123)