A gaggle at Politecnico di Milano has developed an digital circuit in a position to resolve a system of linear equations in a single operation within the timescale of some tens of nanoseconds. The efficiency of this new circuit is superior not solely to classical digital computer systems, but in addition to quantum computer systems. It will likely be quickly potential to develop a brand new technology of computing accelerators that can revolutionize the expertise of synthetic intelligence.
Fixing a system of linear equations means discovering the unknown vector X which satisfies the equation Ax = b, the place A is a matrix of coefficients and b is a identified vector. To resolve this downside, a traditional digital laptop executes an algorithm that takes a number of operations, thus translating into appreciable time and power consumption.
The brand new circuit, which has been developed within the body of the ERC European venture Resistive Change Computing Past CMOS (RESCUE), solves techniques of linear equations (Ax=b) because of an revolutionary methodology of in-memory computing, the place the coefficients of matrix A are saved in a particular gadget referred to as a memristor. The memristor is ready to retailer analogue values, so a memristor matrix can bodily map a coefficient matrix A throughout the circuit, thus strongly accelerating the computation.
The memristor circuit has been examined and validated on a large set of algebraic issues, such because the rating of web web sites and the answer of difficult differential equations together with the Schrödinger equation for the computation of the quantum wavefunction for an electron. All these issues are solved in a single operation.
These outcomes have been revealed within the Proceedings of the Nationwide Academy of Science.
Reminiscence-processing unit might deliver memristors to the lots
Zhong Solar et al. Fixing matrix equations in a single step with cross-point resistive arrays, Proceedings of the Nationwide Academy of Sciences (2019). DOI: 10.1073/pnas.1815682116