Efficient FeFET Crossbar Accelerator for Binary Neural Networks

Taha Soliman¹, Ricardo Olivo², Tobias Kirchner¹, Cecilia De la Parra¹, Maximilian Lederer², Thomas Kämpfe², Andre Guntoro¹ and Norbert Wehn³

¹ Robert Bosch GmbH, Renningen, Germany ² Fraunhofer IPMS, Center Nanoelectronic Technologies (CNT), Dresden, Germany ³ TU Kaiserslauten, Germany

Abstract

This paper presents a novel ferroelectric field-effect transistor (FeFET) in-memory computing architecture dedicated to accelerate Binary Neural Networks (BNNs). We present inmemory convolution, batch normalization and dense layer processing through a grid of small crossbars with reduced unit size, which enables multiple bit operation and value accumulation. Additionally, we explore the possible operations parallelization for maximized computational performance. Simulation results show that our new architecture achieves a computing performance up to 2.46 TOPS while achieving a high power efficiency reaching 111.8 TOPS/Watt and an area of 0.026 mm2 in 22nm FDSOI technology.

Video

[The authors opted for not publicly sharing a presentation video.]