Litcius/Paper detail

An SRAM-based reconfigurable analog in-memory computing circuit for solving linear algebra problems

Piergiulio Mannocci, Enrico Melacarne, Andrea Pezzoli, Giacomo Pedretti, Cristiano Villa, Flavio Sancandi, Umberto Spagnolini, Daniele Ielmini

202310 citationsDOI

Abstract

Analog in-memory computing (AIMC) is attracting strong interest for accelerating data-intensive tasks such as artificial intelligence by overcoming the memory wall. Recently, closed-loop AIMC circuits have been proposed to accelerate the solution of problems with high complexity, such as matrix inversion and pseudoinversion with typical O(N <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> ) complexity. This work presents an integrated circuit for AIMC capable of solving linear systems and regression with 0(1) time. Reconfiguration is possible by block mapping within the memory array. The circuit is demonstrated for two signal analytics applications, namely baseband processing in massive multiple-in multiple-out (MIMO) and Kalman filter.

Topics & Concepts

Computer scienceStatic random-access memoryApplication-specific integrated circuitKalman filterMixed-signal integrated circuitSignal processingControl reconfigurationParallel computingBlock (permutation group theory)Computer engineeringTheoretical computer scienceAlgorithmComputer hardwareEmbedded systemArtificial intelligenceIntegrated circuitDigital signal processingMathematicsOperating systemGeometryAdvanced Memory and Neural ComputingFerroelectric and Negative Capacitance DevicesNeural Networks and Reservoir Computing